Universidade Federal Do Rio Grande Do Sul Instituto de Pesquisas Hidráulicas Programa de Pós-Graduação em Recursos Hídricos E Saneamento Ambiental
Universidade Federal Do Rio Grande Do Sul Instituto de Pesquisas Hidráulicas Programa de Pós-Graduação em Recursos Hídricos E Saneamento Ambiental
Universidade Federal Do Rio Grande Do Sul Instituto de Pesquisas Hidráulicas Programa de Pós-Graduação em Recursos Hídricos E Saneamento Ambiental
Porto Alegre
2019
ERIK SCHMITT QUEDI
Porto Alegre
2019
2
CIP - Catalogação na Publicação
3
ERIK SCHMITT QUEDI
_______________________________________________
Prof. Dr. Fernando Manardi Fan – IPH/UFRGS
Orientador
_______________________________________________
Prof. Dr. Walter Collischonn – IPH/UFRGS
Examinador
_______________________________________________
Prof. Dr. Juan Martin Bravo – IPH/UFRGS
Examinador
_______________________________________________
Prof. Dr. Leonardo Calvetti – UFPEL
Examinador
4
Agradecimentos
Os agradecimentos são direcionados a todos que tiveram alguma contribuição,
direta ou indireta, ao longo do desenvolvimento da pesquisa.
Ao Prof. Fernando Fan pela orientação e paciência ao longo do mestrado, em que
sempre de forma brilhante e objetiva, sanou meus questionamentos sobre a pesquisa e
forneceu todo apoio necessário para a continuidade da mesma.
Ao Prof. Walter Collischonn, pelas contribuições ao longo da pesquisa, pelas aulas
de simulação hidrológica e momentos no grupo de pesquisa, em que, de forma simples,
porem genial, foi capaz de iluminar ideias e aprendizados sobre a hidrologia.
Ao Prof. Juan Martín Bravo, por ser uma figura de muita clareza, paciência,
conhecimento e humildade, que desde as aulas de Algoritmos Evolucionários, foi uma
referência dentro do IPH.
Ao Prof. Leonardo Calvetti, por participar da banca e poder contribuir com seus
conhecimentos na área de hidro meteorologia.
Ainda, a todos os outros professores com quem tive contato, que expresso aqui
minha admiração e respeito, em especial: Prof. Guilherme Marques; Profa. Maria Lucia;
Prof. André Silveira; Prof. Olavo Pedrollo; Prof. Rodrigo Paiva; Prof. Anderson Ruhoff
e a Profa. Edith Schettini.
Aos colegas de grupo de pesquisa Hidrologia de Grande Escala, deixo aqui minha
admiração a vocês, jovens e excepcionais pesquisadores, sejam algumas das pessoas:
Ayan, Sly, Hugo, João Paulo, Otávio, Cleber, Pedro, Vinicius, Arthur, Aline, Renata,
Matheus, Rafael, entre outras das quais não tive maior contato. Não obstante, àquelas
pessoas de outros grupos de pesquisa, como o pessoal da Hidrologia, Tubarão, NET e
demais.
À CAPES pelo apoio financeiro, e à UFRGS e ao Instituto de Pesquisas
Hidráulicas, seus funcionários e à sua estrutura que fomentaram a realização deste
trabalho.
Agradeço imensuravelmente à minha família. A meus pais, Nagib e Cláudia,
sobretudo, pelo apoio e amor incondicional e pelo orgulho demonstrado ao longo de todo
este caminho; A meus “manos”, Kim e Mark, pela parceria, amizade e confiança; A
aqueles que foram meus segundo pai e mãe em Porto Alegre, Tia Chris e Tio Dunga, por
me acolheram de maneira tão carinhosa, muito obrigado pela confiança e lições de valores
cotidianos, além dos deliciosos almoços. Ainda, as minhas primas Maju e Clarinha, que
mesmo sendo ainda jovens, ensinam muito sobre maturidade.
Por fim, à minha querida companheira ao longo desta jornada, Juliana, a qual
compartilhamos e suportamos momentos difíceis e desfrutamos dos bons. Expresso aqui
minha gratidão, respeito, carinho e reconhecimento do seu companheirismo, obrigado!
5
“O que é, pois, a verdade? ...
... após uma longa utilização,
parecem a um povo consolidadas,
canônicas e obrigatórias:
as verdades são ilusões das quais
se esqueceu que elas assim o são...”
(Nietzsche)
Das previsões, o que é a verdade,
senão uma referência?
(o autor)
6
Resumo
As previsões hidrológicas feitas em horizonte sub sazonal, definido neste estudo
como um prazo de antedecência máximo de 46 dias, ainda estão em fase de pesquisa
científica no contexto brasileiro e sul-americano. Espera-se que essas previsões tragam
benefícios para setores como energia hidrelétrica e aplicações operacionais. Neste
sentido, no presente trabalho foram realizados experimentos com o objetivo de avaliar a
qualidade e utilidade das previsões de precipitação sub sazonal, originadas do Centro
Europeu de Previsão Meteorológica de Médio Prazo (ECMWF), na bacia do rio Paraná,
uma das mais importantes bacias hidrográficas da América do Sul, em termos de geração
de energia, contemplando a usina hidrelétrica binacional de Itaipu e mais de outras 150
grandes usinas hidrelétricas.
O Software Ensemble Verification System (EVS) foi aplicado para verificação
estatística das previsões de precipitação sub sazonal, através de métricas determinísticas
e probabilísticas, no período que vai dos anos de 2015 e 2016, utilizando os dados brutos
e em um formato com viés corrigido.
Também foram realizadas simulações de previsões de vazões por conjunto, em
horizonte sub sazonal, usando o modelo semi-distribuído MGB, testando os dados do
ECMWF como entrada - resultando no desenvolvimento de um Sistema de Previsão
Hidrológica por Conjunto (ou Hydrologic Ensemble Prediction System, H-EPS). As
previsões de vazão foram comparadas com as simulações usando como entrada a
informação climatológica, conhecida como previsão estendida de vazão (extended
streamflow prediction, ESP).
Este estudo é uma das primeiras avaliações dos dados de previsões sub sazonais,
para fins de previsão hidrológica realizados no contexto da América do Sul. A avaliação
da precipitação futura permitiu a inferência do desempenho dessas previsões de
precipitação para posterior utilização em modelagem hidrológica, além de quantificar
alguns dos erros associados aos dados.
Os resultados apontaram que, em geral, as previsões de vazões usando os dados
de previsão de precipitação sub sazonal apresentam maior previsibilidade (skill) do que
aquelas que usam informações históricas, até aproximadamente 30 dias de previsão. A
esse respeito, essas previsões podem ser exploradas particularmente em aplicações como
as afluências em reservatórios, o controle de enchentes e secas e a evolução de eventos
extremos várias semanas à frente.
7
Abstract
The hydrological prediction done within sub-seasonal timescale, defined in this
study as a lead-time of 46 days, is still incipient in Brazilian and South American context.
It is expected that these predictions will bring benefits to sectors such as hydro-energy
for scientific research and operational applications. In this sense, it is performed
assessments aiming on evaluating the quality and usefulness of sub-seasonal precipitation
forecasts, originated from the European Centre for Medium-Range Weather Forecast
(ECMWF), in Paraná River Basin, one of the most important basins in Brazil in terms of
energy generation, contemplating the bi-national Itaipú hydroelectric power plant and
more than others 150 large reservoirs plants.
The Ensemble Verification System (EVS) Software was applied to perform a
statistical verification of the sub-seasonal precipitation forecasts, through deterministic
and probabilistic metrics on the period ranging the years of 2015 and 2016, using the raw
data and in a bias corrected format.
Was also proceeded simulations of sub-seasonal streamflow forecasts using a
semi-distributed hydrological model, the MGB-IPH, using the data as input - resulting in
the development of an H-EPS. The streamflow forecasts were benchmarked against the
simulations using as input the climatological information, or the extended streamflow
prediction (ESP).
This study is one of the first evaluation of the sub-seasonal forecasts data, for
hydrologic forecasting purposes, done in Brazilian context. The assessment of the future
rainfall allows the inference of the performance of these precipitation forecasts for later
use in hydrological modeling, also quantifies some of the errors associated with the data.
It is indicated that the streamflow predictions using the sub-seasonal precipitation
forecast data presents greater predictability, or better skill, than those using historical
information until approximately lead-day 30. In this regard, these forecasts may be
explored particularly in applications such as reservoir inflow, flood and drought control
and the onset and evolution of events several weeks ahead.
8
Sumário
1. Apresentação ........................................................................................................... 11
1.1. Introdução e Justificativa ................................................................................. 11
1.2. Hipóteses .......................................................................................................... 13
1.3. Objetivos .......................................................................................................... 13
Objetivo geral ............................................................................................................. 13
Objetivos específicos .................................................................................................. 13
1.4. Organização do trabalho .................................................................................. 13
2. Revisão bibliográfica............................................................................................... 14
2.1. Previsão de variáveis hidrológicas ................................................................... 14
2.2. Previsão a longo prazo (clima ou sazonal) ...................................................... 17
2.3. Previsão a curto prazo (tempo) ........................................................................ 17
2.4. Horizonte sub sazonal como instrumento de previsão hidrológica ................. 19
2.5. Previsão por conjunto (ensemble) .................................................................... 22
2.6. H-EPS .............................................................................................................. 25
2.7. Verificação de previsões por conjunto............................................................. 28
3. Materiais e Métodos ................................................................................................ 30
3.1. S2S ................................................................................................................... 30
3.2. Dados de Precipitação Observada: MSWEP v2.1 ........................................... 32
3.3. Modelo Hidrológico MGB-IPH ....................................................................... 32
3.4. MGB-IPH aplicado a Bacia do Rio Paraná ..................................................... 34
4. Evaluation of Sub seasonal to Seasonal (S2S) Precipitation Forecasts for
Hydrologic Forecasting Purposes Within Paraná River Basin – Brazil. ........................ 37
4.1. Introduction ...................................................................................................... 38
4.2. Materials and Methods ..................................................................................... 41
4.2.1. Study Area .................................................................................................... 41
4.2.2. Data .............................................................................................................. 42
4.2.3. Data Processing ............................................................................................ 43
4.2.4. Bias Correction............................................................................................. 44
4.2.5. Verification Metrics ..................................................................................... 46
4.3. Results and Discussion .................................................................................... 48
4.3.1. Accumulated Precipitation Fields ................................................................ 48
4.3.2. Deterministic Metrics ................................................................................... 50
4.3.3. Probabilistic Metrics .................................................................................... 53
9
4.4. Conclusions ...................................................................................................... 54
5. Potential of Hydrological Sub-Seasonal Forecasts in Tropical Large-Scale Basins56
5.1. Introduction ...................................................................................................... 57
5.2. Study Catchment .............................................................................................. 61
5.3. Methodology .................................................................................................... 63
5.3.1. Precipitation Datasets ................................................................................... 63
5.3.2. Hydrological Model ..................................................................................... 63
5.3.3. Data interpolation ......................................................................................... 65
5.3.4. Ensemble Streamflow Forecast .................................................................... 65
5.3.5. Results Assessments .................................................................................... 66
5.4. Results .............................................................................................................. 68
5.4.1. Individual Assessments and Spread Analysis .............................................. 68
5.4.2. Skill Against Reference Forecast ................................................................. 73
5.5. Discussion ........................................................................................................ 74
5.6. Conclusion ....................................................................................................... 76
6. Conclusão ................................................................................................................ 77
7. Referências .............................................................................................................. 79
8. Anexo: Descrição do modelo meteorológico ipsis litteris portal S2S .................... 88
1. Ensemble version........................................................................................................ 89
2. Configuration of the EPS ........................................................................................ 89
3. Initial conditions and perturbations ........................................................................ 90
4. Model Uncertainties perturbations:......................................................................... 92
5. Surface Boundary perturbations: ............................................................................ 92
6. Other details of the models: .................................................................................... 92
7. Re-forecast Configuration....................................................................................... 93
8. References: .............................................................................................................. 93
10
1. Apresentação
1.1. Introdução e Justificativa
11
muito curto para a variabilidade de componentes do sistema terrestre que evoluem com
maior inércia, tenham forte influência (White et. al., 2017).
12
compreender a performance (qualidade) destas previsões. Por fim, o modelo utilizado foi
calibrado e validado para a Bacia do Rio Paraná, um dos mais importantes sistemas
hidráulicos da América do Sul, e que, no âmbito de geração de energia hidroelétrica e
navegação fluvial, é potencial beneficiária das previsões por conjunto em horizonte sub
sazonal.
1.2. Hipóteses
1.3. Objetivos
Objetivo geral
Objetivos específicos
13
Foram feitos dois estudos (artigos), que compõem a metodologia e os resultados
deste trabalho, ambos realizados na bacia do Rio Paraná. O primeiro destes é apresentado
no capítulo 4, no qual foi feito um estudo sobre os erros nos dados de precipitação,
oriundos do banco de dados S2S, em relação a um referencial observado. Neste estudo,
também apresenta uma investigação a respeito da correção de erros sistemáticos (viéses),
na qual explorou previsões sub sazonais feitas para tempos passados (hindcasts ou re-
forecasts).
2. Revisão bibliográfica
2.1. Previsão de variáveis hidrológicas
Figura 1. Enquadramento da previsão sub sazonal (tarja em azul) em relação ao seu tempo de antecedência máximo.
15
partir destes modelos numéricos obtém-se informações de diversas variáveis
meteorológicas, incluindo a precipitação, originando as previsões quantitativas de
precipitação (ou Quantitative Precipitation Forecast – QPF).
Em geral, estes modelos são resolvidos em uma grade espacial de três dimensões,
a latitude, longitude e altitude, e uma dimensão temporal. É comum nestes modelos a
representação de processos físicos que ocorrem em escalas pequenas, como a microfísica
das nuvens (fluxos convectivos) por meio da parametrização (Kalnay, 2003). Ainda,
apresentam como desvantagem a necessidade de grande esforço computacional,
modelagem complexa e usualmente estes modelos possuem resolução espacial grosseira,
que normaliza grandes áreas em valores médios.
Figura 2. Modelo de Circulação Global de previsão, esquematizado em grade vertical e horizontal. Tipicamente, são
definidos um número de camadas verticais, resolução horizontal, fluxos de energia e parametrizações, havendo fluxos
horizontais e verticais indicados pelas setas (Sampaio e Dias, 2014).
16
2.2. Previsão a longo prazo (clima ou sazonal)
17
Figura 3. Diagrama da acurácia de metodologias em previsão de curto prazo em relação ao seu tempo de antecedência
máximo (adaptada de Pierce et. al., 2012).
18
usos da água, e operação de barragens com segurança (Bravo, 2006; Bravo et. al., 2008;
Bravo et. al., 2009).
19
2008; O’Donnel e Colby, 2008; Shah et. al., 2017); no planejamento de operações na
geração de energia hidroelétrica (e.g. antecipando demandas, manutenção de
reservatórios, planejamento de manobras operacionais, Robertson et. al., 2014; Foster et.
al., 2017, Turner et. al., 2017); na preparação para eventos de grande magnitudes (Braman
et. al., 2013; Perez et. al., 2017; Vitart e Robertson, 2018); no planejamento de navegação
e transporte fluvial (Meibner et. al., 2017); na produção de safras e comercialização de
commodities (Hansen et. al., 2006; Breuer et. al., 2010; Mase e Prokopy 2014); na
conscientização antecipada de risco à sociedade (O'Connor et. al., 2005; Patt et. al., 2007;
Goddard et. al., 2010; Srinivasan et. al., 2011); à saúde pública (Zhou et. al., 2004; Kuhn
et. al., 2005; Thomson et. al., 2006; Kelly-Hope e Thomson, 2008; Jancloes et. al., 2014;
Thomson et. al., 2014); entre outros (NASEM, 2016). O setor hidroelétrico pode
beneficiar-se em relação às premissas de volume sazonal mantido em um reservatório,
que são frequentemente incorporadas ao processo de tomada de decisão, neste sentido, a
previsão sub sazonal pode ser usada para fazer os ajustes necessários entre longo prazo e
curto prazo, a medida que informações adicionais são derivadas das previsões.
20
Os termos associados aos tempos de antecedência máximo (curto, médio e longo
prazo) evidenciam diferenças nas abordagens metodológicas (hipóteses) e aplicabilidades
da previsão, conforme ilustra a Figura 4. No eixo vertical o tempo de antecedência (lead
time) e no eixo horizontal os níveis de tomada de decisão (decision timescales). Em seu
escopo, na cor bege, a incerteza vai sendo amplificada na medida em que o tempo de
antecedência aumenta. Períodos mais distantes do dia inicial da previsão apresentam
menor confiabilidade.
Na figura 4 a cor vermelha indica aquelas previsões à curto prazo, que vai de
minutos (nowcasting) a uma semana, em que se destacam atividades como sistemas de
alertas, monitoramento hidrológico, previsões em tempo real, planos diários de
navegação, avaliação de disponibilidade energética (e.g. períodos com maior insolação
ou precipitação). Em verde é indicado a previsão a longo prazo ou sazonal (seasonal),
este horizonte de previsão fomenta a geração de informações sobre tendências ou
anomalias do clima, gerenciamento de produção agrícola futura em vista de períodos
futuros desfavoráveis, na irrigação, manejo de água em períodos de seca, planejamento
do uso de recurso hídrico, operação de manobras marítimas em regiões recobertas por
gelo. A cor preta indica atividades a serem desenvolvidas em períodos muito longos, ou
vários horizontes a longo prazo, isto é, pode ser entendido como projeções que baseiam
o planejamento das atividades nos outros horizontes. Por fim, em azul se insere a previsão
sub sazonal (sub seasonal to seasonal, ou S2S, abrangendo o período de duas semanas a
alguns meses).
21
Figura 4. Potenciais aplicações da previsão hidrológica em função do seu horizonte de antecedência e incerteza
(NASEM, 2016).
22
Estas previsões determinísticas podem resultar em prognósticos bastante distintos da
realidade, uma vez que um sistema de previsão contém diversas incertezas, como
aproximações (imperfeições) nos modelos hidrológicos e meteorológicos utilizados, ou
registros de dados observados em números insuficientes. Estas incertezas são oriundas da
natureza caótica da atmosfera que transfere ao modelo meteorológico grande
sensibilidade as condições de inicialização (ou estado inicial adotado), conforme
observado inicialmente por Lorenz (1963, 1965, 1969).
A Figura 5 ilustra a previsão por conjunto (ensemble), em que são realizadas múltiplas
simulações numéricas, inicializadas com condições iniciais ou mesmo utilizando modelos
(parametrizações) diferentes entre si. O resultado é um conjunto de trajetórias para a
varável prevista no horizonte de antecedência máximo. Pode-se pensar cada membro do
conjunto (linha preta) como uma previsão determinística iniciada com determinada
condição inicial e o resultado da previsão é função do conjunto de membros.
23
Figura 5. Previsão por conjunto. Cada linha preta corresponde a uma simulação inicializada com diferentes condições
iniciais, e a análise dos resultados é em função do conjunto, isto é, da distribuição de probabilidades.
24
relacionadas a meteorologia, como a hidrologia. A associação de chuvas previstas (ou
QPF) a um modelo hidrológico para obter-se séries futuras de vazões constitui um tipo
de sistema de previsão, denominado Sistema de Previsão Hidrológica por Cojunto (ou
Hydrologic Ensemble Prediction System, H-EPS).
2.6. H-EPS
25
o sistema em estudo (por exemplo, bacia hidrográfica) são definidas – sendo que estas
informações podem ser provenientes de diversas fontes: observações, produtos de
sensoriamento remoto e/ou estimativas numéricas. Após as rodadas da simulação
hidrológica, tantas quantas forem os membros do conjunto (ensemble), é realizado a etapa
de pós-processamento e interpretação dos resultados. No contexto de operação de
reservatórios (geração de energia hidroelétrica), pode-se incluir outras etapas na
modelagem que utilizariam das saídas do modelo hidrológico como informação de
entrada, para geração de informações no planejamento de ações operacionais.
26
Figura 7. Exemplo de resultado de uma previsão hidrológica por conjunto, com horizonte de 10 dias, gerada pelo
MARFC Hydro Ensemble Forecast Service – National Weather Service (NOAA). Os limiares representam a
concentração dos membros do conjunto em cada faixa de valor.
Ainda sobre os desafios acerca da previsão por conjunto, alguns autores destacaram
ao longo de suas pesquisas questões a serem resolvidas (Schaake et. al., 2006; Cloke e
Pappenberger, 2009; Ramos et. al., 2010; Boucher, et. al., 2011; Meller, 2012; Wetterhall
et. al., 2013; Pagano et. al., 2014; Emerton et. al., 2015; Fan, 2015):
27
• A inclusão de sistemas baseado em ensembles em operações de empreendimentos
(tomadas de decisões) pode se tornar tortuosa, uma vez que esta abordagem requer
maior conhecimento sobre o tema, sobretudo para interpretação dos resultados.
28
• Correlação linear (correlation): medida de correlação entre as previsões e
observações
• Destreza (skill): esta medida de desempenho é aplicada entre sistemas de previsão,
no qual mensura-se qual a destreza ou qualidade do sistema estudado em
comparação a referência.
• Resolução (resolution): refere-se à habilidade do sistema de previsão em atingir boa
probabilidade de previsão em eventos diferentes, por exemplo em períodos secos e
chuvosos. Um sistema capaz de obter bom desempenho períodos distintos é tido de
boa resolução.
• Discriminação (discrimination): utilizado para avaliar a capacidade de prever a
ocorrência de um evento discreto ou dicotomo, por meio da atribuição de um
liminar de exedência (treshold).
• Agudeza (sharpness): mensura a capacidade na tendência de prever eventos
extremos (probabilidades extremas, ~0% ou ~100%), é dito agudo o sistema que
sugere eventos extremos com alta probabilidade, isto é, se o sistema indica
probabilidades extremas com pouca frequência ou alta frequência.
29
Tabela 1. Compilação de métricas em relação ao atributo, tipo da previsão e evento discreto.
Atributo Tipo da
Nome da Métrica Evento discreto
Avaliado previsão
3. Materiais e Métodos
3.1. S2S
30
O banco de dados é apresentado e descrito em Vitart et. al., 2017. Esta base
disponibiliza previsões quasi tempo real desde o início de 2015 até dias atuais, com
tempos de antecedência máximo de até 60 dias, geradas por 11 centros de pesquisa:
Australian Bureau of Meteorology (BoM), China Meteorological Administration (CMA),
European Centre for Medium Range Weather Forecasts (ECMWF), Enviroment and
Climate Change Canada (ECCC), Institute of Atmospheric Sciences and Climate of the
National Research Council (CNR-ISAC), Hydrometeorological Centre of Russia
(HMCR), Japan Meteorological Agency (JMA), Korea Meteorological Administration
(KMA), Météo-France/Centre National de Recherche Meteorologiques (CNRM),
National Centers for Environmental Prediction (NCEP), e Met Office (UKMO). Cada
modelo desenvolvido nos respectivos centros de pesquisa possui configurações e
parametrizações distintas, bem como quantidade de membros do conjunto e horizontes
de previsão.
A utilização de dados do banco de dados S2S também é motivada por uma série de
estudos recentes que instigam o uso da previsão em horizontes sub sazonal para sazonal
e que promovema melhoria da qualidade destas previsões e viabilização do uso destas
informações em atividades econômicas. (e.g. Brunet et. al., 2010; Vitart et. al., 2012;
NASEM, 2016).
Para esta pesquisa foi escolhido o modelo oriundo do European Centre for Medium
Range Weather Forecasts (ECMWF), reconhecidamente produtora de previsões de alta
qualidade. As previsões meteorológicas deste centro são frequentemente exploradas em
estudos, por exemplo, no contexto de grandes bacias brasileiras, Fan et. al. (2015c)
utilizou previsões de curto-médio prazo (15 dias), e demonstraram bom desempenho
como forçante do modelo hidrológico MGB-IPH.
31
quais são utilizadas para calibrar as previsões em tempo atual, ou seja, correção de erros
sistemáticos (ECMWF, 2017b).
O modelo Modelo de Grandes Bacias (MGB-IPH; Collischonn et. al., 2005, 2007;
Paz et. al., 2007; Paiva et. al., 2013) é um modelo hidrológico conceitual, semi-
distribuído que é capaz de calcular a vazão a partir da precipitação. Na versão aplicada
neste trabalho, a bacia hidrográfica é dividida em unidades elementares sempre que haja
confluência entre trechos de rios ou em pontos específicos, estas unidades são
denominadas de minibacias. A variabilidade do tipo de solo e vegetação em cada
minibacia é representada por uma Unidade de Resposta Hidrológica – URHs, ilustrado
na figura 8.
32
de solo saturado (Collischonn e Tucci, 2005). A estimativa da evapotranspiração é
calculada pelo método de Penman-Monteith e a interceptação vegetal é representada por
um reservatório cujo volume depende da cobertura vegetal. Os escoamentos subterrâneos
e subsuperficial são obtidos por relações lineares e não lineares de acordo com a
quantidade de umidade do solo (Paiva, 2009).
A vazão resultante de cada uma das unidades elementares até o curso d’água
principal é propagada como por meio de reservatórios lineares enquanto que a propagação
do rio pode ser propagada utilizando o método de Muskingum-Cunge, ou ainda, em
versões mais recentes do modelo, pelo método inercial.
Figura 8. Representação esquemática do procedimento de combinação de mapas de uso e tipos de solo para elaboração
de Unidades de Resposta Hidrológica – URH (Collischon et. al., 2014).
33
parâmetros pode ser realizada manualmente ou, ainda, de forma automática, a partir de
um algoritmo de otimização multi-objetivo.
Nesta bacia, estão concentrados mais de 150 grandes reservatórios para geração de
energia ao longo do Rio Paraná (rio principal) e seus tributários – alguns dos mais
importantes são os Rios Grande, Tietê, Paranapanema e Iguaçu (ANEEL, 2008). Além
do potencial hidroelétrico, alguns destes cursos d’água constituem importantes hidrovias
brasileiras, por exemplo, o trecho Tietê-Paraná.
A BRP tem uma área total de aproximadamente 907,000.00 km², com limites
localizados nas coordenadas 15.450 S até 26.850 S, 43.580 O até 56.110 O, conforme
mostra a figura 9.
34
Figura 9. Bacia do Rio Paraná (BRP), em destaque alguns dos principais cursos d’água, relevantes a esta dissertação.
O modelo utilizado neste estudo é uma versão do modelo preparada para bacia do
Rio Paraná e que vem sendo aprimorada para aplicações diversas (Fan et. al., 2012;
Pontes et. al., 2012; Pontes et. al., 2013; Adam et. al., 2014; Adam et. al., 2015; Colossi
et. al., 2017a; Colossi et. al., 2017b) utilizou a mesma configuração dos parâmetros que
(Fan et. al., 2012; Pontes et. al., 2012; Collischonn et. al., 2014).
35
Na calibração do modelo foi utilizado o período de 1975 até 1995 dada a melhor
disponibilidade de dados e para verificação foi feita para 1995 até 2010.A Figura 10
ilustra a bacia do Paraná discretizada em 1424 minibacias e um mapa de Unidade de
Resposta Hidrológica, demais definições e configuração do modelo MGB-IPH aplicado
nos estudos que compõem esta dissertação são descritos em Collischonn et. al. (2014).
Figura 10. a) Discretização em 1424 minibacias e seu respectivo trecho de rio associado. b) As URHs que compõem a
bacia.
36
4. Evaluation of Sub seasonal to Seasonal (S2S) Precipitation
Forecasts for Hydrologic Forecasting Purposes Within Paraná River
Basin – Brazil.
37
Este capítulo apresenta-se na forma de um artigo científico, escrito em língua
inglesa e submetido na Revista Brasileira de Meteorologia (RBmet) e ainda se encontra
em processo de revisão.
Abstract
The sub seasonal timescale is nowadays considered a predictability desert in hydrological
forecasting systems, although it is still a potential tool to be employed in many
applications, for both scientific and operational purposes. More recently, the community
of forecasters is engaged towards devising a seamless prediction system, encompassing
projections within sub seasonal lead times as well. In this study, assessments aimed
towards gauging the quality and usefulness of sub seasonal precipitation forecast for
further hydrologic forecasting purposes are performed. The case study is conducted
within the Paraná River Basin (PRB) – one of the most important hydroelectric producing
regions in Brazil. The sub seasonal forecast data were evaluated in a raw format and in
an unbiased format, by using the quantile mapping technique and the mean error method.
Resulting in the evaluation of sub seasonal forecasting and demonstrating the added value
that the bias correction step can have within the studied basin.
4.1. Introduction
Precipitation plays a key role in the hydrologic cycle, and it is considered one of
the most difficult variables to forecast. Over the last few decades, the use of quantitative
precipitation forecasts (QPF), which originated from numerical weather prediction
(NWP), through both scientific and operational purposes, has increased (Buizza et. al.,
1999a; Krzysztofowicz, 2001; Cuo et. al., 2011; Fan et. al., 2014; Fan et. al., 2015c).
38
forecasting operations (Buizza, 2008; Cloke and Pappenberger, 2009; Fan et. al., 2015a;
Fan et. al., 2015b). Ensemble forecasting is commonly used at many operational weather
prediction facilities such as, for example, the European Centre for Medium-Range
Weather Forecasts (ECMWF – Europe) and the Centro de Previsão de Tempo e Estudos
Climáticos (CPTEC – Brazil).
An ensemble prediction system may represent the uncertainties that arise from
inaccuracies in the forecast model, by engaging each individual element, assessing the
effect of random errors (Palmer et. al. 2007; ECMWF, 2017a). However, the models can
still be contaminated by systematic errors, or biases, that affect the entire ensemble.
Therefore, post processing of one’s QPFs constitutes an important step in previously
operational or scientific use of the forecast (Buontempo et. al. 2014; Cui et. al. 2011,
ECMWF 2017a).
Over the past decades some progress has been made in short and medium-range
forecasts, up to two weeks, which is also called weather prediction. Likewise, seasonal-
range predictability (long-term, or climate prediction), ranging from one to several
months, primarily aims to predict the averages over the target period, which are more
related to slowly evolving components of the climate system, such as sea surface
temperature and soil moisture. Moreover, it includes, as predictability drivers, large-scale
climate patterns such as El Niño-Southern Oscillation (ENSO), the North Atlantic
Oscillation (NAO), the Pacific-North American pattern (PNA), the Southern Annular
Mode (SAM), and the Indian Ocean Dipole (IOD) (Yuan et. al., 2015; Tian et. al. 2016;
Arnal et. al. 2017; NASEM, 2010; NASEM, 2016).
In more recent years, the forecasting research community’s endeavors are seeking
to ‘bridge the gap’ between weather and climate, moving toward to a ‘seamless’
prediction and development of ‘Ready-Set-Go’ systems (Vitart et. al. 2008; Brunet et. al.
2010; Hoskins, 2013; Vitart and Roberston, 2018; Wetterhall and Giuseppe 2018). The
prediction time-frame within the ‘gap’ (over two weeks and less than a few months), is
also called as sub seasonal to seasonal forecast. Its main objectives are to include the well-
established expertise of weather and climate researchers along with fulfilling the demands
for applications that fall within that time range (Vitart et. al. 2012; Vitart et. al. 2015;
Vitart et. al. 2017).
The sub seasonal time range is often considered challenging: it is situated between
the weather (deterministic predictable or initial condition problem) and climate
39
(probabilistic predictability from slowly varying conditions); for many years, it has been
referred to as a ‘predictability desert’. Once the lead time is long enough that the memory
of initial hydro-meteorological conditions is virtually lost, where it is too short for the
variability of slowly evolving components of the earth system have a strong influence
(Vitart et. al. 2017). Although, the researches on sources of sub seasonal predictability
such as the dominants climate drivers (e.g. Madden-Julian Oscillation, Baldwin et. al.,
2003; Waliser, 2011), better acknowledgment of earth systems initial conditions (e.g.
Inertial memory in soil moisture; Snow cover; Ocean-Atmosphere flux exchanges,
Woolnough et. al., 2007; Koster et. al., 2010; Sobolowski et. al., 2010) and technologies,
are allowing this paradigm to change (Hoskins, 2013; Vitart et. al. 2015).
Although the potential possibilities are highlighted, prediction on this sub seasonal
time range faces inherent challenges: determining the economic value of weather and
climate forecasts (Wilks and Hamill, 1995; Richardson, 2000; Buizza, 2008); the lack of
understanding and systematic credibility in the evaluation of regional forecasts (i.e.
verification metrics are often not directly relevant to users, Morss et. al., 2008); shaping
the willingness to adopt research progress into operational activities (Pagano et. al., 2002;
40
Lemos 2008); currently available products may not fit immediately into an institutional
framework, including the need for customization of ‘raw’ forecasted information to fulfill
mismatches between available products and the actual need of end users (Pagano et. al.,
2002); it is also worth pointing out very few communication pathways between decision
makers (end-users) and the community of forecasters (Lemos and Morehouse, 2005;
Suarez and Tall, 2010; Kirchhoff et. al., 2013; Buontempo et. al., 2014; Perez and Mason,
2014). One can find more information on sub seasonal to seasonal prediction in NASEM
(2010) and NASEM (2016).
This study aims primarily to investigate the potential quality and usefulness of sub
seasonal time-range QPFs, defined in this work as a forecast horizon of 46 days, for future
hydrological forecasting in Brazilian territory. The tested data were originated from the
European Centre for Medium-Range Weather Forecasts (ECMWF), embedded in the S2S
(Sub seasonal to Seasonal) project context (Vitart et. al. 2017). Additionally, this study
was conducted within the Paraná River watershed, where the Itaipu hydropower plant is
located, one of the most important regions for hydroelectricity production in Brazil, which
could potentially benefit from an ensemble forecasting system to complement the
operational prediction chain.
The Paraná River basin is one of the most important hydraulic systems in all of
South America. In Brazilian territory, it covers regions from South-Central Brazil to the
Itaipu hydropower dam, denoted Paraná River Basin (PRB), shown in Figure 1. The PRB
is responsible for more than 50% of Brazilian hydroelectric production, concentrating in
the Paraná River and main tributaries (Grande, Paranaíba, Tietê, Paranapanema and
Iguaçu) more than 150 large reservoir plants for power generation (ANEEL 2008).
The PRB has a total area of approximately 907,000 km², situated at coordinates
15.450 S to 26.850 S, 43.580 W to 56.110 W. The PRB covers a large area which presents
distinct hydrological regimes. The variation of precipitation in the basin is related to the
action of air masses over Brazil and presents a seasonal character. In the far north, rainfall
is tropical and convective, while in the extreme south it is convective and frontal. The
region at the northern end of the basin is characterized by the tropical climate, with dry
41
winters and rainy summers; while in the southern end it presents regular rainfall, more
well-distributed over the seasons (Siqueira and Martins, 2018).
4.2.2. Data
The QPF was provided by the ECMWF CY43R3 model, including the real-time
forecasts for 2015 and 2016. In its latest version, the model is based on a global coupled
ocean-atmosphere integration of 51 ensemble members, one starting from initial
undisturbed conditions (control member) and 50 are disturbed with small perturbations
of small amplitude, it runs twice a week (Monday and Thursday, UTC 00) up to day 46
(ECMWF, 2017b). The forecasts used in this study were acquired with 0.125º grid
resolution and accumulated daily.
42
One may note that the model version changed over time and for the real time
forecasts before 14/05/2015 model runs for a 32-day lead-time and 51-member ensemble,
as for the hindcasts, the ensemble consists of 5-members, and 32-day lead-time. More
detailed information about this model can be found in ECMWF (2017b).
The QPF were evaluated with the Multi Source Weighted Ensemble Precipitation
(MSWEP v2.1) as the observed benchmark, this global dataset covers the years from 1979
to 2016 with a 3-hourly temporal and 0.1o spatial resolution (Beck et. al., 2017a; Beck
et. al., 2017b). The use of a gridded precipitation dataset has advantages over local in-
situ gauges, by using complementary information acquired through gauges, satellite and
reanalysis data, providing estimates for large areas (even the entire globe) and for long
temporal availability.
Given the fact that the objective of this study is to verify the QPF for hydrologic
forecasting, it was carried out by following a distributed hydrologic model framework,
driven by the precipitation forecast on daily timesteps (Yuan et. al., 2015; Fan et. al.,
2014; Tucci et. al., 2003; Collischonn et. al., 2005a, 2005b, 2007).
The main river basin is discretized into 1424 unit catchments (i.e. elemental areas
based on drainage area and stream confluences, Siqueira et. al., 2016), displayed in Figure
2, areas ranging in 108.07 km² to 1100.43 km², in which the centroid coordinate of each
of these was used to interpolate representative precipitation series from the gridded S2S
data, by taking the nearest QPF grid pixel from the centroid coordinate as its value. The
same interpolation had to be done for the MSWEP dataset (observed benchmark), since
this dataset was accumulated in daily timestep increments at each centroid.
43
Figure 2. Discretization of the PRB into 1424 unit catchments. The grid corresponds the S2S data over the unit
catchments.
One method used to adjust the bias of the QPF model to the observations, was the
application of the quantile mapping technique at point-wise locations (centroids). This
technique is based on the empirical transformation of Panofsky and Brier (1968), often
considered to be the best execution of bias correction methods and has been applied to
hydro-meteorological studies (Xu, 1999; Wood et. al., 2002; Hay and Clark, 2003; Boe
et. al., 2007; Das et. al., 2008; Maraun et. al., 2010; Piani et. al., 2010; Bardossy e
Pegram, 2011; Kallache et. al., 2011; Themebl et. al., 2011; Berg et. al., 2012;
Teutschbein and Seibert, 2012; Gudmundsson et. al., 2012; Maraun, 2013; Fan et. al.,
2014; Cannon et. al., 2015; Reiter et. al., 2015, 2017).
44
distributive function of the observations. A correction function can be used, so that the
probability associated with each future rainfall forecast (i.e. raw or forecast) is identified
from the cumulative distribution function within the data of past simulations, then, the
bias corrected value is obtained by corresponding to the same probability in the
cumulative distribution function of climatological observations, demonstrated in equation
1.
−1
𝑋𝑋 𝑄𝑄𝑄𝑄 = 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑜𝑜𝑜𝑜𝑜𝑜 ( 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑠𝑠𝑠𝑠𝑠𝑠 (𝑋𝑋 𝑟𝑟𝑟𝑟𝑟𝑟 ) ) (1)
Where: XQM is the quantile-mapped rainfall value; Xraw is the biased forecast;
eCDF-1 is the inverse eCDF, a data quantile.
Also, it is worth pointing out that this technique may require some caution when
applied in a nonstationary context, as it assumes a stationary bias form (e.g. not verified
in analysis of future climates, Kallache et. al., 2011). As a historical re-forecast (hindcast)
was used from same model version of real-time forecast to construct the cumulative
curves of simulations, we consider the bias as having the same structure as its
climatology.
Figure 3. Sampled eCDFs of climatological simulations (gray line) and observations (black line).
45
Mean Error (ME)
A second bias correction method was applied using the estimated average bias
over the historical past forecasts, hindcasts, similar to Cui et. al. 2011, but assigning the
same weight to all data, from now on identified as ME. The bias was estimated for each
date associated with lead time t, each unit catchment centroid i and defined as the
difference between the observation and re-forecast, averaged over the past 20 years of
real time simulation, following equation 2.
The averaged bias over the trained period, that corresponds to the 20 years of past
forecasts is then applied as a correction group to the current forecast at each unit
catchment centroid and lead time, the forecast value was estimated following equation 3:
1
𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏(𝑡𝑡,𝑖𝑖) = n ∑ni=1 obs(t,i) − refcst (t,i) (2)
Where: bias(t,i) is the estimated bias on centroid i and lead time t, averaged over n
= 20 years; obs(t,i) is the observation value; refcst(t,i) is the past forecast or hindcast;
fcst(t,i) is the raw forecast and fcst(t,i)corr is the mean error bias corrected forecast.
Where: PPRB is the area-weighted precipitation; Ai is the area of the unit catchment
i; Pi is the precipitation series of the unit catchment i; n is the number of unit catchments.
46
Following on, using all available real-time forecasts over verification period (N =
209), at each timestep (t), considering the mean of all 51-ensemble members, the forecasts
were evaluated by the Mean Absolute Error (MAE, equation 3) and the Root Mean
Squared Error (RMSE, equation 4), against the observed benchmark. The MAE gives the
same weight to all errors, measuring the average magnitude of errors at each time step.
Meanwhile, the RMSE as a quadratic scoring rule penalizes larger errors, measuring the
variability of the error magnitudes. Both metrics are commonly employed in model
evaluation studies and provides different insights on model errors, therefore, a
combination of metrics is recommended to assess model performance (Chai and Draxler,
2014).
∑𝑁𝑁
1 �PPRB − PPRBo �
f(t) (t)
MAE(t) = (5)
N
2
∑N
1 �PPRB − PPRBo �
� f(t) (t)
RMSE(t) = (6)
N
47
Finally, the evaluation of ensemble performance was executed through the Mean
Continuous Ranked Probability (Mean CRPS) Score and the Mean Brier Score (BS).
The Mean CRPS summarizes the quality of one ensemble prediction into a single
value. It is obtained by measuring the integrated square difference between the cumulative
distribution function (CDF) of the forecast value and the corresponding CDF of the
observed (equation 5), where lower values correspond to optimal results. The average
across multiple pairs of forecast and observation leads to the mean CRPS.
The numerical value of this score is not directly interpretable as a measure of error,
but it can be used as comparative across ensemble and/or deterministic forecasts.
∞ 2
∑n
1 �∫−∞ CDF(ft ) − 1{f≥ot }� df
����������
CRPS(t) = (7)
N
The brier score is a measure of the average square error of a probability forecast
that accounts for the distribution of the ensemble, analogous to the mean squared error of
a deterministic forecast but in probabilistic terms. This score measures the error with
which a discrete event, such as a given threshold, is predicted.
∑N
1 (Ft (TS)−1{TS≥ot })
2
BS(t) = (8)
N
Figure 4 displays the accumulated precipitation fields on wet (a) and dry (b)
seasons. A comparison between the raw S2S data and the observed benchmark suggest
that the ECMWF model may not appropriately represent the rainfall variability within the
basin. Especially in regard to the wet season, it tends to overestimate the precipitation. In
addition, the figure indicates that the forecast model slightly overestimated downstream
regions on dry season.
48
This assessment enabled an overall verification, on a spatial basis, of the applied
bias correction methods. Since the bias corrected accumulated fields displayed more
cohesion to the observed fields, it might have been able to correct some variability and
magnitudes of precipitation within the study basin at unit catchment centroid locations.
49
Figure 4. Seasonal averaged total precipitation fields, ranging the years of 2015 and 2016. a) Wet Season (October to
to March). b) Dry Season (April to September). The first illustration of the basin is the observed dataset, following the
raw (without bias correction), the quantile map (QM) and mean error (ME) datasets.
The descriptive statistics of the entire basin average (PPRB) climatology are shown
in tables 1 to 4 for each precipitation dataset. The mean standard deviation (STD) of both
raw and bias corrected S2S data suggests little variation in daily rainfall values, which
may not represent the observed precipitation that presented larger variability in almost
every other month.
For some applications it is valuable to know the dry days’ (i.e. precipitation less
than 1 mm) frequency over a month. The tables show that the S2S raw model heavily
underestimates the number of dry days (NDD). This can be attributed to the averaging
procedures (e.g. area-weighting, and ensemble mean), to the overly brief wet-threshold
applied, or even the S2S model itself. Though a more robust correction method might be
able to improve this matter.
50
As for monthly mean total precipitation volumes, the bias correction seems to
adjust the forecast values to observations more accordingly, reflecting on more accurate
mean annual precipitation.
Table 3. Monthly Climatology for the bias corrected Quantile Map Dataset.
JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
MIN 0 0 0 0 0 0 0 0 0 0 0 0
MAX 21 24 23 16 17 14 12 17 25 18 24 17
MEDIAN 7 6 5 2 2 1 1 1 3 4 5 6
MED 7 6 5 2 3 2 1 1 3 4 5 6
STD 2 2 2 2 2 2 2 2 2 2 2 2
NDD* 1% 1% 1% 14% 11% 32% 41% 47% 17% 6% 1% 0%
VOL 211 182 156 74 78 48 43 35 76 129 165 195
Calculated 2-year Mean Annual Precipitation = 1399.3 mm
* Dry Days (< 1.0 mm)*100 / Total number of days on specified month
Table 4. Monthly Climatology for the bias corrected Mean Error Dataset.
JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
MIN 0 0 0 0 0 0 0 0 0 0 0 0
MAX 20 21 18 15 17 13 13 14 22 17 22 18
MEDIAN 6 6 5 2 2 1 1 1 1 4 5 6
MED 7 6 5 3 3 2 2 2 2 4 5 6
STD 2 2 2 2 2 2 2 1 2 2 3 2
NDD* 2% 1% 1% 10% 12% 33% 33% 50% 20% 5% 2% 1%
VOL 205 187 163 80 85 51 43 34 66 122 164 202
Calculated 2-year Mean Annual Precipitation = 1404 mm
* Dry Days (< 1.0 mm)*100 / Total number of days on specified month
Although this statistical description of the datasets may give insight for the
inference of the onset of seasons or anticipating mean sub seasonal (monthly) totals,
further analyses on point-wise precipitation series, or even splitting the whole basin into
smaller regions that share some hydrologic similarity (e.g. Themebl et. al., 2011), may be
necessary to obtain less misleading results.
Considering that the rainfall-runoff process over a large basin is a complex non-
linear process, a small reduction in errors may improve streamflow predictions.
The S2S model performed better on initial timesteps, when the initial conditions
have a stronger influence over the prediction.
This matter can constrain the hydrologic application of the method, demanding
more accurate information on the time-range over the period of a few weeks. However,
through data assimilation techniques (i.e. updating the initial conditions), one can yield
material of higher quality on shorter forecast lead times (e.g. Fan et. al., 2015d).
Despite, the analysis may also be improved by correcting the single area-average
precipitation series, using the same bias correcting method.
52
4.3.3. Probabilistic Metrics
Figure 10 shows results for the mean CRPS. The behavior of this metric is like
those that consider the mean ensemble, which generally tend to increase over the lead
times. The mean CPRS for ensemble means showed little variability, between 1.8 mm
and 2.5 mm. Directly comparing the mean CRPS of the ensemble’s averages with its
equivalent deterministic measure (control forecast MAE), indicates that the ensemble
forecast has better performance over the forecast lead times. In addition, the bias
correction mostly has not improved this metric, only in the initial timesteps is some
improvement visible.
The brier score (BS), showed in figure 11, which verifies the detection of discrete events,
computed the rainfall score to be 5 mm. For the 5 mm threshold the BS assumed values
ranging from 0.16 to 0.22. Also, the bias correction showed little improvement on the
brier score.
The results above demonstrate some of the potential uses of future forecast that
fall within the sub seasonal time range. In this context, some ways of presenting the results
53
are displayed, as in projecting the model uncertainty into a single value at each time step
(e.g. MAE, RMSE and mean CRPS) or averaging accumulated values within distinct time
ranges (e.g. seasonal and intra-seasonal total precipitation).
Furthermore, the sort of study presented here may help in answering questions
concerning the general usefulness of S2S predictions on a regional basis. Such as: in
which location the rainfall event is likely to occur; the wet and dry season onset; the
accuracy of longer lead times. Also, an aspect that is not explored within the frame of this
study, is association of predictions to vulnerability or expected loss in decision making.
4.4. Conclusions
This Sub Seasonal to Seasonal (S2S) evaluation has been one of the first for
hydrological forecasting purposes of future rainfall provided by the database, within one
of the most hydropower important regions in Brazil, the Paraná River Basin. In addition
to the assessment of raw forecasts, the bias correction method applied to the S2S
precipitation dataset remedied the intra-basin variability by following a distributed
hydrological model framework.
In general, the quality of the assessed hydrological forecasting showed that the
ensemble average MAE ranged 2.6mm to 3.7mm, whilst the control member ranged
2.7mm to 4.8mm. The RMSE varied from 3.6mm to 5.26mm.
On the other hand, the bias correction did not show notable improvement on the
metrics, therefore it may be an unnecessary step to further hydrologic modeling.
54
future studies within the same database will include evaluation of the results of
hydrological forecasts for hydropower plants located in the Paraná watershed.
55
5. Potential of Hydrological Sub-Seasonal Forecasts in Tropical
Large-Scale Basins
O foco deste capítulo é a respeito dos sistemas de previsão hidrológica por conjunto
(H-EPS, sigla em inglês), os quais utilizam as previsões quantitativas de precipitação
como forçantes em um modelo hidrológico. Tradicionalmente, as previsões hidrológicas
em prazo sub sazonal (usualmente denominadas de prazo estendido) são obtidas
utilizando informações históricas de variáveis meteorológicas, constituindo o método
Extended Streamflow Prediction (ESP; Day, 1985). Todavia, devido ao crescente avanço
e melhorias na previsão quantitativa destas variáveis, sobretudo a precipitação, oferecem
a oportunidade de realizar estas previsões aplicando dados de modelos atmosféricos
globais. Baseado nesta motivação, esta pesquisa explora os dados de precipitação
quantitativa oriundos do European Centre for Medium-Range Weather Forecasts
(ECMWF), disponíveis no banco de dados mantido pelo projeto subseasonal-to-seasonal
(S2S), como entradas ao modelo MGB-IPH.
56
Abstract
The sub-seasonal forecast horizon, defined in this study as up to 46 days, gives rise to
many opportunities for research and operational applications. This work is one of the first
evaluations of ensemble sub-seasonal meteorological inputs in large-scale basin
hydrological modelling within a tropical climate in South America – where one could
potentially explore the forecasts for hydropower generation purposes. The precipitation
forecast data were provided by the European Centre for Medium-Range Weather
Forecasts (ECMWF), within the sub-seasonal-to-seasonal (S2S) project’s context. To
assess the quality of the forecasts, a statistical evaluation was performed, including a
comparison with the traditional extended streamflow prediction (ESP) technique. Results
allowed for an estimation of the error magnitudes and the potential to the benefit of the
ensemble over the deterministic reference forecast and the ESP approach. We showed
that for all locations, the S2S forecasts have advantages over the ESP, although generally
their skill deteriorates in lead times after day 30. The evaluation across multiple locations
considering its respective drainage area and hydrological conditions, suggests that they
are related to the statistical performance outcome.
5.1. Introduction
Over the last decades, the improvement of numerical weather predictions (NWP)
enabled the use of quantitative precipitation forecasts (QPF) for many applications, both
operational and scientific (Cuo et. al., 2011, Golding 2009, Fan et. al., 2014). Despite the
errors and uncertainties associated with the QPF data, it is increasingly being adopted as
the preferred input for hydrologic models, aiming towards streamflow prediction.
57
shorter lead times, valuing the prediction on a point of interest, especially near the outlet.
These aspects make the H-EPS the best alternative for systems in which it could
potentially extend lead times and more efficiently quantify the system’s predictability (or
unpredictability) (Cloke and Pappenberger, 2009; Cuo et. al., 2011; Cloke et. al.,2013).
The H-EPS have been extensively used for short-range forecasting in flood warning
systems and natural disaster prediction, and more recently, the development of
technologies has led to large-scale, global flood forecasting (Pappenberger et al., 2013).
In addition, medium and seasonal range forecasting constitutes a potential tool for issuing
river flow conditions and optimizing water resource management and decision-making
(Cloke and Pappenberger, 2009; Emerton et. al., 2016).
58
horizon (Palmer and Webster, 1993; Palmer et. al., 2008; Vitart et. al., 2008; Brunet et.
al., 2010; Hoskins, 2013; Vitart and Roberston, 2018; Wetterhall and Giuseppe 2018),
mainly based on the idea of a concatenation of “the best” forecast at each lead time.
Although the sub-seasonal prediction method may not be the best option to detect
short-term extreme events (e.g. floods), it could benefit sectors such as hydropower by
predicting inflow volumes, and anomalous conditions within longer lead times, such as
low flows and droughts (Dutra et. al., 2014; Meibner et. al., 2017). It could also be applied
to draw the necessary adjustments between long-term and short-term planning, as new
information is made available (for example, assumptions regarding the volume held in a
reservoir), also delay or anticipated start of rainy/dry season.
Specifically, regarding the sub-seasonal time range, it is noticeable that there are
few research projects concerned with the applications of numerical weather predictions
within the sub-seasonal time range (up to 3 months, once called extended-range).
Subsequently we discuss a few of the currently available works. White et. al., (2017)
reviews potential applications of sub-seasonal-to-seasonal (S2S) predictions, discussing
new research and operational opportunities. It also outlines the key challenges for
advancing this new frontier – identifying sources of predictability; improving forecasting
skill; operationalization of predictions and communication with decision makers. Despite
59
being a review paper, it gives insights into the possibilities of applications and
categorizing the primary challenges in this current, early stage of practices in S2S
forecasting.
The study of Shah et. al., (2017) is an application of sub-seasonal (of lead time up
to day 45) precipitation forecasts, tailored to benefit the agricultural sector. The authors
evaluated the meteorological prediction outputs from CFSv2, GEFSv2 and a local Indian
Agency (IITM) against observations from the India Meteorological Department. The total
runoff and the root-zone soil moisture was estimated using the Variable Infiltration
Capacity (VIC) model, also applying bias correction methods on forecast variables. Their
results show that the IITM products outperform the CFSv2 and GEFSv2 models; using
the forcing variables from the IITM ensemble and the VIC model, they have predicted
45-day accumulated hydrologic variables with satisfying performance, in terms of
correlation coefficient, Mean Absolute Error (MAE) and critical success index; providing
useful information to water (such as total runoff) and agriculture (root-zone soil moisture
and precipitation anomalies) managers.
The authors Monhart et. al., (2019) compared the traditional ESP approach to a sub-
seasonal (with a lead time up to day 32) hydrometeorological ensemble prediction system,
within three catchments with distinct hydroclimatic conditions. They used ECMWF sub-
seasonal temperature and precipitation inputs, with raw and bias corrected (pre-
processed) datasets, aiming towards investigating the predictability of streamflow and the
importance of the complex interactions between precipitation and temperature in a sub-
seasonal timescale. The results, expressed in standard performance scores, including the
Continuous Ranked Probability Skill Score (CPRSS), indicated that the sub-seasonal
inputs showed improved predictability over ESP, specially at early lead times. Regarding
skill in comparison with the ESP, they revealed that the raw forecasts only display
improvement for the first 5 lead days, and pre-processing the precipitation improves it on
early lead times. As for correcting the temperature, it enhances the skill on later lead
times. One interesting aspect tested by the authors is the seasonal variation in skill, in this
matter forecasts generally initiated in winter and spring show the highest benefits over
the reference climatology. Furthermore, the research discusses some limitations of the
quantile-mapping technique for distributed hydrological modelling (i.e. the variation
inflation issue may be influenced by convective precipitation events, influencing all the
results), it also provides insights in enhancing the generation of ESPs by suggesting that
60
only individual years be considered, possessing similar initial conditions. Another
important conclusion is that solely pre-processing precipitation data is not enough to
enhance the forecast performance, and it is by highlighting their combination with other
variables (temperature) which improves the reliability, however, this verification went
against a reference simulation as a replacement for real observations.
Regarding the testing of sub-seasonal forecasts, we could not find any that used
quantitative precipitation forecasts, for application in tropical climates, for the evaluation
of hydropower benefits, nor for South American locations. On the other hand, there is an
increasing urge to systematically explore and evaluate the sub-seasonal forecasts across
a wide range of activities and locations (Brunet et. al., 2010; Vitart et. al., 2012; Vitart et.
al., 2017; White et. al., 2017).
The objective in this study is to perform one of the first evaluations of sub-seasonal
hydrological forecasts applied to hydropower dams’ locations in large tropical and sub-
tropical basins, based on ECMWF QPFs which originated from the S2S project. The
research is conducted within the Paraná River Basin, one of the most important
hydropower producing regions in the South American continent. This study aims
primarily to evaluate the ensemble streamflow through statistical scores, as well as
comparing it to the well-established extended range prediction ESP technique, potentially
providing information concerning the use of S2S data for hydrological modelling.
The researched basin is the Paraná River Basin (PRB), located within Brazilian
territory, covering regions from South-Central Brazil to the Itaipú hydropower dam. The
PRB is one of the most important hydraulic systems in South America, being responsible
for more than 50% of the hydroelectric production in the entire country (Adam et. al.,
2015). Contemplating around 150 large power generating reservoirs along its main river
and tributaries – Paraná River, Grande, Tiête, Paranapanema and Iguaçu; ANEEL, 2008).
These rivers also double as waterways for navigation, such as the Tietê-Paraná. The PRB
has total area of approximately 907,000 km², localized within coordinates 15.450 S to
26.850 S, 43.580 W to 56.110 W, as shown in figure 1.
The PRB concentrates approximately one third of the population of Brazil and
covers several of the country federative units. Within the basin it is included the region
with greater population and industrialization of Brazil. The main uses of water and its
61
conflicts are around public, industrial, agriculture and irrigation water supply, therefore,
also suffers from the degradation of this resource in its main tributaries – originating from
organic and inorganic pollution (mainly effluents).
The studied basin presents different hydrological regimes across its large territory,
exhibiting a seasonality caused primarily by the action of air masses in Brazil. The region
at the north end of the basin presents a tropical climate, with dry winters and more rain in
the summer; the rainfall is tropical and convective. While the southern region presents
steadier rainfall, evenly distributed throughout the seasons (Siqueira and Martins, 2018).
62
5.3. Methodology
The observed precipitation dataset used to run the hydrological simulations was the
Multi Source Weighted Ensemble Precipitation (MSWEP v2.1). This is a global dataset,
covering the years from 1979 to 2016, with grid resolution of 0.1o and provided as 3-
hourly accumulated value, which had to be accumulated from daily values (Beck et. al.,
2017a; Beck et. al., 2017b). Employing a gridded dataset was the selected method in the
current study due to its advantages over local in-situ gauges, as it uses complementary
information from diverse sources such as satellite, reanalysis and data gauges; also
providing information for long temporal availability without missing data.
63
The utilized version of this model discretizes the basin into unit catchments
whenever there are confluences of streams or at specific points, these units are
denominated mini-basins (Paiva et. al., 2013). The soil and vegetation variability are
classified by Hydrologic Response Units (HRU), whose parameters are associated with
each HRU type (Kowen et. al., 1993); for the evapotranspiration calculation, the model
is based on Penman-Monteith (Shuttleworth, 1993); the surface runoff and soil water
balance follow the Arno model approach (Todini, 1996); also, the streamflow routing is
computed using the Muskigum-Cunge method (in the case of this study) or the full Saint-
Venant when necessary (Paiva et. al., 2013).
In this study, the basin was discretized into 1424 catchments, while the rainfall-
runoff process was simulated using a daily time step. Model calibration was carried out
using data from the National Electric System Operator (ONS), spanning the years
between 1975 to 1995 for the calibration period and 1995 to 2010 for validation due to
data availability and occurrences of extreme events Collischonn et. al., (2014).
64
Table 2. Performance of MGB model on the study case gauges in the calibration and validation periods.
Calibration (1975-1995) Verification (1995-2010)
Location
NS log-NS dV(%) NS log-NS dV(%)
Água Vermelha 0,85 0,83 -10,10 0,85 0,88 -6,10
Barra Bonita 0,76 0,67 5,30 0,76 0,75 -6,50
Furnas 0,81 0,82 -9,90 0,84 0,87 -2,50
Itaipú 0,86 0,85 -2,80 0,85 0,85 -3,90
Jupiá 0,87 0,85 -3,50 0,87 0,88 -2,70
Porto Primavera 0,86 0,82 -3,60 0,88 0,87 -1,50
Based on Table 1, we believe that the hydrologic model calibration and verification
was satisfactory for further hydrological applications since the NS and log-NS were above
0.75 and the volume error (dV) between ±10%.
5.3.3. Data interpolation
Since the main river basin is discretized into 1424 unit catchments (elemental
areas), the centroid coordinate of each of these was used to interpolate representative
precipitation series from the gridded S2S data, by taking the nearest QPF grid pixel from
the centroid coordinate as its value. The same interpolation had to be done for the
MSWEP dataset.
Figure 2. Visual analysis on sampled sub-seasonal streamflow forecasts at Itaipú HPP. It is noted that the S2S based
forecasts followed in more accordance to the benchmark than the ESP based, providing more accurate information on
inflows to the Itaipú dam location. On the left the S2S-based; on the right the ESP-based. Blue line refers to the
streamflow benchmark, red line is the control member of the ensemble, gray line is an ensemble member and dashed
black line is the ensemble mean.
The hydrological simulations produced results in the streamflow series for all 1424
catchments, however, only the 6 case studies locations were assessed. One of the analyzed
primary results is the comparison of the “perfect” rainfall simulation against the observed
natural streamflow, provided by ONS, the Nash-Sutcliffe score was also computed at each
location.
The ensemble forecasting was evaluated using the Ensemble Verification System
(EVS) software, described by Brown et. al., 2010. We applied some scores that are usual
in ensemble forecasting, such as the Mean Absolute Error (MAE), Root Mean Squared
Error (RMSE), Continuous Ranked Probability Skill Score (CRPSS) and Rank Histogram
(Stanski et al., 1989; Hersbach, 2000; Wilks, 2006; Brown et al., 2010; Bradley and
Schwartz, 2011; Jolliffe and Stephenson, 2012). These scores are usually employed in
66
ensemble assessments and are highly documented within specific statistical literature
(Wilks, 2006; Brown et al., 2010; Jolliffe and Stephenson, 2012).
The Mean Absolute Error (MAE) is a score that reduces the ensemble to its mean
value for each time step, measuring the difference from the observed benchmark (i.e.
“perfect” rainfall simulations). This score assigns the same weight to all errors and
primarily measures the magnitude of the errors, the optimal value is equal to zero.
Meanwhile, the Root Mean Squared Error (RMSE), as a quadratic score rule, it penalizes
larger errors, providing a measurement on the variability of the errors’ magnitudes,
serving as a measure of accuracy.
The Mean Continuous Ranked Probability Score (CRPS) summarizes the quality of
the ensemble forecast reducing it to a single value, which is obtained by measuring the
integrated square difference between the cumulative distribution function of the forecast
value, and the corresponding function of the observed. The CRPS, when averaged for all
pair of forecasts and observations, leads to the Mean CRPS, where lower values
correspond to the optimal. It is important to highlight that the Mean CRPS may be used
to compare the ensemble performance to a reference forecast, since it reduces to the MAE
for deterministic forecasts. In addition, the Mean Continuous Ranked Probability Skill
Score (CRPSS) measures the performance of the forecast system relative to another in
terms of Mean CRPS.
In order to allow a comparison between different locations, errors scores (MAE and
CPRS) were gathered after a normalization procedure. The calculated value of the
respective scores were divided by the long-term average of the “perfect” rainfall
simulated flow series of 2015 and 2016. Thusly, the result may disregard the magnitudes
and express the errors relative to the mean flow.
67
The Rank Histogram (RH) is a measure of reliability (statistical consistency
between, measurements and simulations, Hamill, 2000; Candille and Talagrand, 2005)
and whether uncertainty is correctly represented in the forecast. It scopes out the fraction
of observation that may fall between any two ranked ensemble members in the forecast
distribution. This score is optimal (i.e. the forecast system is reliable in terms of Rank
Histogram) when the probability of the observation falls between any two ranked
members and is approximately uniform. A lack of spreading of the ensemble is indicated
by a “U” shaped rank histogram (High probabilities in one or both tails), on the other
hand an inverted “U” shape indicates an excessive spreading.
5.4. Results
Figure 3 displays the results of the MAE and CRPS metrics at each of the six dam
locations for the ensemble mean and control forecast, after normalizing of the scores with
the long-term average of each location.
The figure shows that the CRPS for the ensemble is always lower than MAE of the
Ensemble Mean and the Control Member in all locations. The CRPS compared to the
control forecast (deterministic reference) indicates better predictability from the ensemble
mean. The MAE for the ensemble is always lower than the control member, and the errors
readily increase from initial time steps onwards. It illustrates that for the Barra Bonita and
Furnas locations, the errors are more prominent in relation to the long-term average.
In general, for the Itaipú (827,000 km²), Porto Primavera (574,000 km²), Jupiá
(479,000 km²) and Água Vermelha (139,000 km²) locations, the three first being located
along the Paraná River (the main river of the basin), and the forth on the tributary Grande
river, the MAE and CRPS magnitudes for the ensemble are placed around 0.20 and 0.15
times the long term flow average. As for the control member, until day 8 the MAE value
was akin to the ensemble, and after this lead time, the values were around 0.3 times the
long-term average.
The Furnas (52,000 km²) location, situated on the Rio Grande river, and the smallest
drainage area of all locations in the ensemble, presented a MAE with values around 0.3
times the long-term average, and 0.2 times for the CPRS. Regarding the control member,
68
the MAE resulted in values ranging 0.5 times, and, aside from that, around day 6 it
showed disparity with the ensemble.
The Barra Bonita (33,000 km²) location, situated on the Tietê river, presented the
greatest variability of MAE and CPRS, ranging values around 0.6 times the long-term
average for the first score and 0.4 times for the second, within the ensemble. The control
member presented MAE values over 0.8 times, and until day 8 the score remained close
to ensemble.
Figure 3. Evolution on the lead time up to 46 days of the MAE, RMSE and CRPS for the forecasts issued.
Figure 4 shows the normalized Root Mean Squared Error (RMSE) of the ensemble
mean on all locations. In the figure it is indicated that the errors tend to increase as the
drainage area decreases. The three locations situated along the main river (Itaipú, Porto
Primavera and Jupiá) presented the smallest RMSE relative to the long-term average. As
for the ones located on the tributary Grande River (Água Vermelha and Furnas), this
69
behavior appears once more, in addition, for Furnas the errors increase promptly from
initial time steps. Finally, in the Barra Bonita location, which represents the smallest
drainage area, located on tributary Tietê River, it presented the largest normalized RMSE
as well.
In general, the locations with larger drainage areas performed better, and it is duly
noted that the hydrological forecasts in locations with a drainage area above 479,000 km²
the RMSE score does not present significant improvements, suggesting that this value
may be an area threshold in which the performance of hydrological forecast results cannot
be enhanced.
Figure 4. Normalized RMSE on all locations, figure shows that the error tends to increase as the drainage area of the
location decreases.
Figure 5, 6 and 7 show the 51-member rank histogram for Itaipú, Furnas and Barra
Bonita dam locations, at sampled lead times 1, 15, 20 and 46, showing the rank
histogram’s behavior throughout the forecast horizon. From Itaipú location, the general
behavior of the ensemble spread is demonstrated, which is similar to that of the other dam
localities over the locations along the Paraná River (Porto Primavera and Jupiá), as well
as the Água Vermelha location.
The Itaipú rank histograms indicate that in early lead-times, approximately from
day 1 up until day 15, there is a lack of spread, represented by the “U” shape of the
histograms, and as the forecast lead times increase, more spread is presented (in the flat
form of the histograms). These results are compatible with ensemble forecasts, in which
70
the lack of spread in early lead-times is noted as the only uncertainty considered is the
precipitation (Fan et. al., 2014).
For the Furnas rank histograms, the lack of spread on early lead times and a small
tendency towards an overestimation of the forecast streamflow over the entire horizon is
also perceived. The Barra Bonita rank histograms show that in early timesteps there is a
lack of spread (“U” shape), but along with lead time evolution a tendency to overestimate
the streamflow is noted (positive bias).
Figure 5. Rank Histograms indicating a lack of spread in early lead times. This behavior it is akin to the other locations
along the main river of the basin (Paraná River) – Porto Primavera, Jupiá and Água Vermelha.
71
Figure 6 Rank Histograms indicating right from initial time step, a tendency towards overestimation on the forecast
streamflow.
Figure 7. Rank Histograms indicating lack of spread in early lead times, but a tendency to overestimation (positive
bias) along the development of the forecast.
72
.
Figure 8 shows the CPRSS results for the six dam locations, delineating the
performance of the sub-seasonal streamflow forecast in comparison to the ESP, based on
historical rainfall data (reference forecast). As the CPRSS perfect score is 1, indicating a
higher performance skill of S2S based forecasts, and negative values indicating that the
reference forecast performs betters in terms of CPRS.
This result shows that for the Itaipú location, up until day 3 the reference ESP
forecast outperforms the sub seasonal ensemble. At the Porto Primavera, Jupiá and Água
Vermelha locations, the score presents similar behavior, in which only after day 3 did the
sub seasonal ensemble outperform the reference forecast; however, it is verified that
approximately after a month (day 30) the ESP performs superior skill. Broadly, these
results concour with studies investigating the quality of ESP, which claims that many
ensemble predictions systems have difficulties in outperforming ESP after a month of
forecasting (Arnal et. al., 2018; Lucatero et. al., 2018; Monhart et. al., 2019). As for the
deficient sub-seasonal ensemble skill until day 3, this is possibly caused by the good ESP
skill on forecasting the initial lead-days on large basin, in which the inertial memory of
hydrological conditions has more influence on the prediction.
The Barra Bonita location presented a better skill of the sub-seasonal ensemble
across all time steps. Finally, in Furnas, the sub-seasonal forecasts were slightly better
than the ESP until the third week (day 21).
73
Figure 8. Mean Continuous ranked probability skill score (CPRSS) for the sub seasonal based forecasts against ESP
as reference.
5.5. Discussion
The results from the statistical verification indicates that ensemble hydrological
forecasting presents many advantages when compared to comparison to the deterministic
(control member), in terms of MAE and CPRS. After a week of lead time, the direct errors
of the ensemble are always lower than the deterministic reference, for all six locations.
Also, it is suggested that the error scores may be dependent on the basin scale, since
smaller basins presented greater amounts of errors (Barra Bonita and Furnas), as it may
be attenuated by the streamflow generation processes on larger basins.
74
hydrological forecasting were very similar, suggesting a threshold area for bearing the
best results.
Regarding Rank Histograms, the statistical metrics showed that for the early lead
times, results are more dependent on the initial observed conditions in the basin rather
than on meteorological inputs, as indicated by the lack of spread in the ensemble across
all locations. The locations along the Paraná River and the Água Vermelha location, show
an increased reliability along with further lead times. On Barra Bonita, despite a lack of
spread on initial lead times, a positive bias is noted along the forecast. Furthermore,
Furnas presented a lack of spread on early lead times and a small tendency to positive
bias – indicating a systematic overestimation of the streamflow on these two last
locations, suggesting that some bias removal technique may be applied.
The CPRSS results suggest that the S2S based forecast have more skill than the
ESP based, therefore, it may provide more valuable information on decision making
processes than the forecasts that relies on historical series. The inflow estimation and
reservoir management may benefit from the modelled framework used on this study, once
we highlighted the associated errors within the forecast horizon and provided estimates
of streamflow several weeks ahead.
Another important aspect investigated is the rainfall seasonality effect on the ESP
performance, one may note that on Barra Bonita location is within a climate region that
presents little seasonality, therefore, resampling historic rainfall from distinct years
toward the past may not be representative of future conditions.
The results suggest that the ECMWF ensemble may perform differently at distinct
sub-basins among the study area. It can be noted specially for the locations that aren’t
along the main river of the studied basin, that the performance results on one location
may not be transposed to another. This evidence is also perceived within the studies of
Fan et. al., (2015) that tested ensemble forecasts on distinct watersheds within same
hydroclimatic region.
75
The lack of spread on early lead times could be improved by applying a data
assimilation method and by considering the uncertainties in initial hydrological
conditions. Also, we did not carry out any post-processing on the forecast streamflow,
which could be assessed in future studies, considering each location’s particularities. For
an operational application of the S2S forecasts, it may be necessary to account not only
for the meteorological uncertainties. This matter may be resolved by adopting the
aforementioned strategies. Moreover, as the interpretation of the ensemble is not entirely
straight-forward, a better understanding would be attained through the use of optimization
models.
Our work brings light to the sub seasonal hydrological forecast, in one of the most
important hydrographic regions of South America, the Paraná River Basin. These
forecasts have been shown to provide medium-to-long term valuable information,
especially for inflow assumptions in hydroelectric power plants, and have outperformed
the traditional ESP approach. Also, the results are in agreement with other studies that
have evaluated the performance of ensemble forecasts in relation to deterministic
reference, which generally confer the best quality of forecast to the ensemble.
5.6. Conclusion
This research presented one of the first statistical assessments of the sub-seasonal
to seasonal (S2S) meteorological data as input for hydrological modelling in a tropical
and sub-tropical context, considering the location of hydropower plants which could
potentially benefit over the streamflow predictions.
The results showed that the performance of the issued ensemble forecasts possess
some advantages in applications where total inflows is necessary when compared to the
deterministic (control member) and to a climatological reference (ESP).
The locations with greater drainage area situated along the main river of the basin
presented the best performance, when taking into account MAE, CPRS, and CPRSS,
especially for the Itaipú dam, which presented a higher skill than climatology after day 3
until the end of the forecast horizon. Following the Paraná River upstream, the direct
errors (MAE and CPRS) presented slight deteriorations, concerning the skill score the
results indicated similar behavior to Itaipú, except that after a month of lead time, it has
underperformed the historical reference.
76
The two other locations, situated in a tributary of the main river, with smaller
drainage areas, presented a higher degree of divergence in the results. On Barra Bonita
and Furnas, the MAE and CPRS resulted on the largest values, possibly caused by the
bias. On the other hand, the first location outperformed the ESP on all timesteps; and the
second presented skill above the ESP until week 3 (day 21).Ultimately, we expect that
our results will contribute to the usage of sub-seasonal horizon forecasts, and its benefits
to hydrological forecasting, consolidating ensemble forecasting experiments in large
South American basins as well. For hydropower generation purposes in large-scale
basins, the sub-seasonal streamflow predictions proved to be promising, with better skill
than the forecasts generated using the historical rainfall. In addition, the efforts addressed
some issues related to the main incitements of the sub-seasonal to the seasonal (S2S)
project (Vitart et. al.,2017).
Regarding future works, we believe that more test cases may be useful, considering
the meteorological inputs originated from different research centers, as they are freely
available from the S2S database – potentially providing a ‘grand’ ensemble with more
statistically reliable results. Likewise, succeeding recent efforts in the integrated
modeling of complex hydraulic systems in South America (e.g. Siqueira et. al., 2018), it
is expected that value can be extracted from these models towards the use of sub-seasonal
forecasting (precipitation and other variables) in hydrological forecasts, especially when
regarding techniques that only employ historical information.
6. Conclusão
A previsão de variáveis hidrológicas, como a precipitação e a vazão em recursos
hídricos, feita em prazo de antecedência sub sazonal, possui valor e utilidade em diversas
aplicações, apesar das incertezas associadas a este horizonte. Ainda, nota-se que o
benefício potencial destas previsões está vinculado ao uso final, isto é, deve-se adequar o
produto (previsão) à aplicação desejada.
77
Além disso, testou-se a eficácia de dois métodos de correção de erros sistemáticos,
os quais não obtiveram melhorias em nenhuma das métricas estatísticas (MAE, RMSE e
BS). Todavia, pode-se constatar que ambos os métodos testados introduziram
variabilidade na climatologia intra-anual da precipitação. Estes resultados podem ajudar
nas estimativas de períodos mais secos e outros com maiores volumes de chuvas.
78
sub sazonais tendem a crescer, em vista de novas tecnologias de monitoramento e o
engajamento da comunidade cientifica (e operacional) a respeito desta temática.
7. Referências
(ANEEL). Agência Nacional de Energia Elétrica. (2008). Atlas de energia elétrica do Brasil. Brasília: ANEEL.
ADAM, K. N.; FAN, F. M. ; PONTES, P. R. M. ; BRAVO, J. M. ; COLLISCHONN, W. (2014) . Climate Change and
Floods in Paraná River Basin. In: 6th International Conference on Flood Management - ICFM6, 2014, São
Paulo - SP. PROCEEDING OF THE 6th International Conference on Flood Management - ICFM6. Porto
Alegre - RS: ABRH, 2014. p. PAP014788.
ALLASIA, D. G. (2007). “Avaliação da previsão hidro-climática no Alto Paraguai”. Tese de doutorado, Universidade
Federal do Rio Grande do Sul, Instituto de Pesquisas Hidráulicas, Porto Alegre, Brasil, 208 p.
ARNAL, L., CLOKE, H. L., STEPHENS, E., WETTERHALL, F., PRUDHOMME, C., NEUMANN, J.,
KRZEMINSKI, B., PAPPENBERGER, F. (2018). Skilful seasonal forecasts of streamflow over Europe?
Hydrology and Earth System Sciences, 2057-2072.
BALDWIN, M. P., STEPHENSON, D. B., & THOMPSON, D. W. (2003). Stratospheric Memory and Skill of
Extended-Range Weather Forecasts. SCIENCE, 636-640.
Bárdossy, A., & Pegram, G. (2011). Downscaling precipitation using regional climate models and circulation patterns
toward hydrology. WATER RESOURCES RESEARCH, 1-18.
BARTHOLMES, J., & TODINI, E. (2005). Coupling meteorological and hydrological models for flood forecasting.
Hydrol. Earth Syst. Sci., 9, 333-346. https://doi.org/10.5194/hess-9-333-2005
BECK, H. E., MING PAN, TIRTHANKAR ROY, GRAHAM P. WEEDON, FLORIAN PAPPENBERGER, ALBERT
I. J. M. VAN DIJK, GEORGE J. HUFFMAN, ROBERT F. ADLER, AND ERIC F. WOOD. (2019). Daily
evaluation of 26 precipitation datasets using Stage-IV gauge-radar data for the CONUS. Hydrol. Earth Syst.
Sci., 23, 207–224, https://doi.org/10.5194/hess-23-207-2019
BECK, H. E., VERGOPOLAN, N., PAN, M., LEVIZZANI, V., DIJK, A. I., & WEEDON, G. P. (2017). Global-scale
evaluation of 22 precipitation datasets using gauge observations and hydrological modeling. Hydrology and
Earth System Science, 6201-6217.
BECK, H., E. WOOD, M. PAN, C. FISHER, D. MIRALLES, A. VAN DIJK, T. MCVICAR, AND R. ADLER, (2018):
MSWEP V2 global 3-hourly 0.1° precipitation: methodology and quantitative assessment. Bull. Amer. Meteor.
Soc. doi:10.1175/BAMS-D-17- 0138.1, in press.
BELL, V. A., DAVIES, H. N., KAY, A. L., BROOKSHAW, A., & SCAIFE, A. A. (2017). A national-scale seasonal
hydrological forecast system: development and evaluation over Britain. Hydrol. Earth Syst. Sci., 21, 4681-
4691. https://doi.org/10.5194/hess-21-4681-2017
BERG, P., FELDMANN, H., & PANITZ, H.-J. (2012). Bias correction of high resolution regional climate model data.
Journal of Hydrology, 80-92.
BOÉ, J., TERRAY, L., HABETS, F., & MARTIN, E. (2007). Statistical and dynamical downscaling of the Seine basin
climate for hydro-meteorological studies. INTERNATIONAL JOURNAL OF CLIMATOLOGY, 1643–1655.
BOUCHER, M.-A.; TREMBLAY, D.; DELORME, L.; PERREAULT, L.; ANCTIL, F. (2012). “Hydroeconomic
assessment of hydrological forecasting systems”. J. Hydrol., v. 416 417, np. 133–144, 2012
79
BRAMAN, L. M., AALST, M. K., MASON, S. J., SUAREZ, P., AIT‐CHELLOUCHE, Y., & TALL, A. (2012).
Climate forecasts in disaster management: Red Cross flood operations in West Africa, 2008. Disasters, 144-
164.
BRAVO, J. M. (2006). “Otimização da operação de um reservatório para controle de cheias com base na previsão de
vazão”. Porto Alegre: UFRGS. Dissertação de mestrado em Recursos Hídricos e Saneamento Ambiental,
Instituto de Pesquisas Hidráulicas, Universidade Federal do Rio Grande do Sul. 155p,
BRAVO, J. M.; COLLISCHONN, W.; PILAR, J. V.; TUCCI, C. E. M. (2008). “Otimização de regras de operação de
reservatórios com incorporação da previsão de vazão”. Revista Brasileira de Recursos Hídricos, v. 13(1), p.
181-196.
BRAVO, J.; PAZ, A.; COLLISCHONN, W.; UVO, C.; PEDROLLO, O.; CHOU, S. (2009). “Incorporating Forecast
of Rainfall in Two Hydrologic Models Used for Medium-Range Streamflow Forecasting”. Journal of
Hydrologic Engineering, v. 1, n. 5, p. 435-445.
BREUER, N. E., FRAISSE, C. W., & CABRERA, V. E. (2010). The Cooperative Extension Service as a Boundary
Organization for Diffusion of Climate Forecasts: A 5-Year Study. Journal of Extension, 1-5.
BRUNET, G., SHAPIRO, M., HOSKINS, B., MONCRIEFF, M., DOLE, R., KILADIS, G. N., KIRTMAN, B.,
LORENC, A., MILLS, B., MORSS, R., POLAVARAPU, S., ROGERS, D., SCHAAKE, J., SHUKLA, J. S.
(2010). Collaboration of the weather and climate communities to advance subseasonal-to-seasonal prediction.
American Meteorological Society, 1397-1406.
BUIZZA, R. (1997). “Potential forecast skill of ensemble prediction and spread and skill distributions of the ECMWF
Ensemble Prediction System”. Monthly Wheater Review, v. 125, p. 99-119.
BUIZZA, R., HOLLINGSWORTH, A., LALAURETTE, F., & GHELLI, A. (1999a). Probabilistic Predictions of
Precipitation Using the ECMWF Ensemble Prediction System. Weather and Forecasting, 14, 168–189.
https://doi.org/10.1175/1520-0434(1999)014<0168:PPOPUT>2.0.CO;2.
BUIZZA, R., MILLER, M., & PALMER, N. (1999b). Stochastic representation of model uncertainties in the ECMWF
Ensemble Prediction System. Quarterly Journal of the Royal Meteorological Society, 2887-2908.
BUIZZA, R. (2008). The value of probabilistic prediction. Atmospheric Science Letters, 36-42.
BUIZZA, R.; LEUTBECHER, M.; ISAKSEN, L. (2008). “Potential use of an ensemble of analyses in the ECMWF
Ensemble Prediction System”. Quart. J. Roy. Meteor. Soc, 134, 2051-2066, doi: 10.1002/qj.346.
BUIZZA, R., & LEUTBECHER, M. (2015). The forecast skill horizon. Quarterly Journal of the Royal Meteorological
Society, 3366-3382.
BUONTEMPO, C., HEWITT, C. D., DOBLAS-REYES, F. J., & DESSAI, S. (2014). Climate service development,
delivery and use in Europe at monthly to inter-annual timescales. Climate Risk Management, 1-5.
CALHEIROS, A. J. P.; ENORÉ. D. P.; MATTOS, E. V.; COSTA, I. C. C.; MACHADO, L. A. T. (2016). “Sistema de
Previsão Imediata: Descrição dos Produtos”. Instituto Nacional de Pesquisas Espaciais – INPE. São José dos
Campos – SP.
CANNON, A. J., SOBIE, S. R., & MURDOCK, T. Q. (2015). Bias Correction of GCM Precipitation by Quantile
Mapping: How Well Do Methods Preserve Changes in Quantiles and Extremes? JOURNAL OF CLIMATE,
6938 6959.
CARVALHO F.; SILVA, D. A.; SANTOS, J. P. L.; LIMA, V. C.G.R. (2005). “Aplicação de redes neurais artificiais
na previsão de precipitação na bacia do rio Mundaú – Alagoas”. In: Simpósio Brasileiro de Recursos Hídricos,
16, João Pessoa – PA. Anais 1 CD.
CHAI, T., & DRAXLER, R. R. (2014). Root mean square error (RMSE) or mean absolute error (MAE)? – Arguments
against avoiding RMSE in the literature. Geoscientific Model Development, 1247-1250.
CHANGNON, S. A.; VONNAHME, D. R. (2003). “Impact of spring 2000 drought forecast on Midwestern water
management”. Journal of Water Resources Planning and Management, v129(1), p. 18-25.
CHIEW, F. H. S.; ZHOU, S. L.; MCMAHON, T. A. (2003). “Use of seasonal streamflow forecasts in water resources
management”. Journal of Hydrology, 270, p. 135-144.
CHOU S. C.; NUNES, A. M. B.; CAVALCANTI, I. F. A. (2000). “Extended range forecasts over South America using
the regional ETA model. Journal of Geophysics Research, Vol. 105, n. 8, p. 10147-10160
80
CLOKE, H. L., & PAPPENBERGER, F. (2009). Ensemble flood forecasting: A review. Journal of Hydrology, Volume
375, Issues 3–4, 15 September, Pages 613-626. 2009. https://doi.org/10.1016/j.jhydrol.2009.06.005
CLOKE, H. A , PAPPENBERGER, F., VAN ANDEL, S. J., SCHAAKE, J., THIELEN, J., RAMOS, MH..
(2013).Preface: Hydrological ensemble prediction systems. Hydrological Processes, 27, 1-4. DOI: 10.1002/hyp.9679
COLLISCHONN, W., & TUCCI, C. E. (2005a). Previsão Sazonal de Vazão na Bacia do Rio Uruguai 1: Ajuste e
Verificação do Modelo Hidrológico Distribuído. RBRH – Revista Brasileira de Recursos Hídricos, 43-59.
COLLISCHONN, W., TUCCI, C. E., & CLARKE, R. T. (2005b). Previsão Sazonal de Vazão na Bacia do Rio Uruguai
2: Previsão Climática-Hidrológica. RBRH – Revista Brasileira de Recursos Hídricos, 61-72.
COLLISCHONN, W., TUCCI, C. E., CLARKE, R. T., CORBO, M. D., SILVA, B. C., COLLISCHONN, B., . . . PAZ,
A. R. (2007). Modelo Hidrológico Distribuído para Previsão de Vazões Incrementais na Bacia do Rio Paranaíba
entre Itumbiara e São Simão. RBRH — Revista Brasileira de Recursos Hídricos, 43-55.
COLLISCHONN, W.; TUCCI, C. E. M. Simulação hidrológica de grandes bacias. Revista Brasileira de Recursos
Hídricos, v. 6, n. 1, p. 95-118, 2001. http://dx.doi.org/10.21168/rbrh.v6n1.p95-118.
Collischonn, W.; Bravo, J. M.; da Silva, B. C.; Rodriguez, D. A. (2014). Capítulo 3: “Modelagem Hidrológica”. Em:
LIMA, J. W. M., COLLISCHON, W., MARENGO, J. A, 2014. Efeitos das mudanças climáticas na geração de
energia elétrica. Editora BH. São Paulo, Brasil.
CUI, B., TOTH, Z., ZHU, Y., & HOU, D. (2011). Bias correction for global ensemble forecast. Weather and
Forecasting, 396-410.
CUO, L., PAGANO, T. C., & WANG, Q. J. (2011). A review of quantitative precipitation forecasts and their use in
short-to-medium streamflow forecasting. Journal of Hydrometeorology. 12, 713–728,
https://doi.org/10.1175/2011JHM1347.1.
CZAJA, A. P.; VAART, V.D.; MARSHALL, J. (2002). “A diagnostic Study of the Role of Remote Forcing in Tropical
Atlantic Variability”. Journal of Climate, v. 15, p. 3280-3290.
DALE, M.; WICKS, J.; MYLNE, K.; PAPPENBERGER, F.; LAEGER, S.; TAYLOR, S. Probabilistic flood
forecasting and decision making: an innovative risk-based approach. Nat. Hazard. ISSN:1573-0840, Springer
Netherlands, 2012.
DANCE, S.L. & ZOU, Q.P. (2010). “Ensembles, uncertainty and flood prediction”. Hydrol. Earth Syst. Sci. Discuss.,
7, p.3591-3611, 2010.
Das, T., Bárdossy, A., Zehe, E., & He, Y. (2008). Comparison of conceptual model performance using different
representations of spatial variability. Journal of Hydrology, 106– 118.
DIETRICH, J.; SCHUMANN, A.H.; REDETZKY, M.; WALTHER, J.; DENHARD, M.; WANG, Y.; PFUTZNER,
B.; BUTTNER, U. Assessing uncertainties in flood forecasts for decision making: prototype of an operational
flood management system integrating ensemble predictions. Nat. Hazards Earth Syst. Sci., v. 9, p.1529–1540,
2009.
ECMWF, E. C.-R. (2017a). SEAS5 User Guide.
ECMWF, E. C.-R. (2017b). IFS DOCUMENTATION - Cy43r3. PART V: ENSEMBLE PREDICTION SYSTEM.
FAN, F. M., COLLISCHONN, W., MELLER, A., & BOTELHO, L. C. (2014). Ensemble streamflow forecasting
experiments in a tropical basin: The São Francisco river case study. Journal of Hydrology, Volume 519, Part
D, 27 November 2014, Pages 2906-2919. https://doi.org/10.1016/j.jhydrol.2014.04.038
FAN, F. M., COLLISCHONN, W., QUIROZ, K. J., SORRIBAS, M. V., BUARQUE, D. C., & SIQUEIRA, V. A.
(2015d). Flood forecasting on the Tocantins River using ensenble rainfall forecasts and real-time satellite
rainfall estimates. Journal of Flood Risk Management, 1-11.
81
FAN, F. M., PONTES, P. R., COLLISCHONN, W., & BUARQUE, D. C. (2015a). Sobre o uso da persistência de
previsões determinísticas de vazão para a tomada de decisão. Revista Brasileira de Meteorologia, 218-228.
FAN, F. M., RAMOS, M.-H., & COLLISCHONN, W. (2015b). Sobre o uso de previsões hidrológicas probabilísticas
para a tomada de decisão. Revista Brasileira de Recursos Hídricos, 914-926.
FAN, F. M., SCHWANENBERG, D., COLLISCHONN, W., & WEERTS, A. (2015c) Verification of inflow into
hydropower reservoirs using ensemble forecasts of the TIGGE database for large scale basins in Brazil. Journal
of Hydrology Volume 4, Part B, September 2015, Pages 196-227. https://doi.org/10.1016/j.ejrh.2015.05.012
FOSTER, K., UVO, C. B., & OLSSON, J. (2017). The development and evaluation of a hydrological seasonal forecast
system prototype for predicting spring flood volumes in Swedish rivers. Hydrology and Earth System Sciences,
1-29.
GEORGAKAKOS, K.P., KRZYSZTOFOWICZ, R. (2001). Probabilistic and ensemble forecasting. Volume 249,
Issues 1–4, 1 August 2001, Page 1. https://doi.org/10.1016/S0022-1694(01)00455-3
GODDARD, L., AITCHELLOUCHE, Y., BAETHGENA, W., DETTINGER, M., GRAHAM, R., HAYMANE, P., &
CONRAD, E. (2010). Providing Seasonal-to-Interannual Climate Information for Risk Management and
Decision making. Procedia Environmental Sciences, 81–101.
GODDARD, L., MASON, S. J., ZEBIAK, S. E., ROPELEWSKI, C.F., CANE, M. A. (2001). Current approaches to
seasonal-to-interannual climate predictions. International Journal of Climatology, v21, p.1111-1152.
GOLDING, B. W. (2014). “Regional Prediction Models”. Em: North, G. R.; Pyle, J. A.; Zhang, F. (2014).
“Encyclopedia of Atmospheric Sciences”. Academic Press, 2nd Edition, 2008.
GOLDING, B. W. Long lead time flood warnings: reality or fantasy? Met. Apps, v. 16, p. 3– 12, 2009doi: =
10.1002/met.123.
GREUELL, W., FRANSSEN, W. H., BIEMANS, H., & HUTJES, R. W. (2018). Seasonal streamflow forecasts for
Europe – Part I: Hindcast verification with pseudo- and real observations. Hydrology and Earth System
Sciences 22 (2018)6. - ISSN 1027-5606 - p. 3453 - 3472. https://doi.org/10.5194/hess-22-3453-2018
GUDMUNDSSON, L., BREMNES, J. B., HAUGEN, J. E., & ENGEN-SKAUGEN, T. (2012). Technical Note:
Downscaling RCM precipitation to the station scale using statistical transformations – a comparison of
methods. Hydrology and Earth System Sciences, 3383–3390.
HAMILL, T., HAGEDORN, R., WHITAKER, J., (2008). Probabilistic forecast calibration using ECMWF and GFS
ensemble reforecasts. Part II: Precipitation. Mon. Weather Rev. 136 (7), 2620–2632.
HAMLET, A. F.; HUPPERT, D.; LETTENMAIER. (2002). “Economic values of long-lead streamflow forecasts for
Columbia River Hydropower”. Journal of the Water Resources Planning and Management”, v. 128(2), p. 91-
101.
HANSEN, J. W., CHALLINOR, A., INES, A., WHEELER, T., & MORON, V. (2006). Translating climate forecasts
into agricultural terms: advances and challenges. CLIMATE RESEARCH, 27–41.
HARTMANN, H. (2005). Use of Climate Information in Water Resources Management. Encyclopedia of Hydrological
Sciences, 1-15.
HAY, L., & CLARK, M. (2003). Use of statistically and dynamically downscaled atmospheric model output for
hydrologic simulations in three mountainous basins in the western United States. Journal of Hydrology, 56–75.
HOSKINS, B. (2013). The potential for skill across the range of the seamless weather-climate prediction problem: a
stimulus for our science. Quarterly Journal of the Royal Meteorological Society, 573–584.
HSIEH, W. W.; YUVAL, L. J.; SHABBAR, A. SMITH, S. (2003). “Seasonal prediction with error estimation of
Columbia River Streamflow in British Columbia”. Journal of water Resources Planning and Management, v.
129(2), p. 146-149.
HSU, M. H.; FU, J. C.; LIU, W. C. (2003). “Flood routing with real-time stage correction method for flash flood
forecasting in the Tanshui River, Taiwan”. Journal of Hydrology, v. 283, p. 287-280.
82
JANCLOES, M., THOMSON, M., COSTA, M. M., & HEWITT, C. (2014). Climate Services to Improve Public Health.
International Journal of Environmental Research and Public Health, 4555-4559.
KALLACHE, M., VRAC, M., NAVEAU, P., & MICHELANGELI, P.‐A. (2011). Nonstationary probabilistic
downscaling of extreme precipitation. JOURNAL OF GEOPHYSICAL RESEARCH, 1-15.
KALNAY, E. (2003). “Atmospheric Modeling, Data Assimilation and Predictability”. Cambridge University Press.
UK. 276p.
KAYANO, M. T.; ANDREOLI, R. V. (2004) “Decadal Variability of Northern Brazil rainfall and its relation to tropical
sea surface temperature and global sea level pressure anomalies”. Journal Geophys. Res., v. 109, p. 21-32.
KELLY-HOPE, L., & THOMSON, M. C. (2008). Climate and Infectious Diseases. Em M. C. Thomson, R. Garcia
Herrera, & M. Beniston, Seasonal Forecasts, Climatic Change and Human Health (p. 232). Springer.
KIRCHHOFF, C. J., LEMOS, M. C., & DESSAI, S. (2013). Actionable Knowledge for Environmental Decision
Making: Broadening the Usability of Climate Science. The Annual Review of Environment and Resource, 393–
414.
KOSTER, R. D., ET. AL. (2010). Contribution of land surface initialization to subseasonal forecast skill: First results
from a multi-model experiment. GEOPHYSICAL RESEARCH LETTERS, 1-6.
KOUSSIS, A. D; LAGOUVARDOS, K.; MAZI, K.; KOTRONI, V.; SITZMANN, D.; LANG, J.; ZAISS, H; BUZZI,
A.; MALGUZZI, P. (2003). “Flood Forecasts for Urban Basin with integrated Hydro-Meteorological Model”.
Journal of Hydrologic Engineering, v. 8(1), p. 1-11.
KRZYSZTOFOWICZ, R. (2001). The case for probabilistic forecasting in hydrology. Volume 249, Issues 1–4, 1
August 2001, Pages 2-9. https://doi.org/10.1016/S0022-1694(01)00420-6
KUHN, K., CAMPBELL-LENDRUM, D., HAINES, A., & COX, J. (2005). Using climate to predict infectious disease
epidemics. Geneva: World Health Organization Press.
LEMOS, M. C. (2008). What influences innovation adoption by water managers? Climate information use in brazil
and the united states. JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION, 1388-1396.
LEMOS, M. C., & MOREHOUSE, B. J. (2005). The co-production of science and policy in integrated climate
assessments. Global Environmental Change, 57–68.
LETTENMAIER, D. P.; WOOD, E. F. (1993). Hydrologic Forecasting. In: Maidment (Org.), Handbook of Hydrology.
Ed McGraw-Hill
LIN, Y., RAY, P.; JOHNSON, K. (1993). “Initialization of a modelled convective storm using Doppler radar derived
fields”. Monthly Weather Review, v.121, p. 2757-2775, 1993.
LORENZ, E. N. (1965). “A study of the predictability of a 28-variable atmospheric model”. Tellus, v. 17, p. 321-333.
LORENZ, E. N. (1969). “The predictability of a flow which contains many scales of motion. Tellus, v. 21A, p. 289
307.
LUCATERO, D., MADSEN, H., REFSGAARD, J. C., KIDMOSE, J., AND JENSEN, K. H (2018).: Seasonal
streamflow forecasts in the Ahlergaarde catchment, Denmark: the effect of preprocessing and post-processing
on skill and statistical consistency, Hydrol. Earth Syst. Sci., 22, 3601–3617, https://doi.org/10.5194/hess-22-
3601-2018, 2018b.
MARAUN, D. (2013). Bias Correction, Quantile Mapping, and Downscaling: Revisiting the Inflation Issue. American
Meteorological Society, 2137-2143.
MARAUN, D., WETTERHALL, F., IRESON, A. M., CHANDLER, R. E., KENDON, E. J., WIDMANN, M., . . .
THIELE‐EICH, I. (2010). PRECIPITATION DOWNSCALING UNDER CLIMATE CHANGE: RECENT
83
DEVELOPMENTS TO BRIDGE THE GAP BETWEEN DYNAMICAL MODELS AND THE END USER.
Reviews of Geophysics, 1-34.
MASE, A., & PROKOPY, L. (2014). Unrealized Potential: A Review of Perceptions and Use of Weather and Climate
Information in Agricultural Decision Making. American Meteorological Society, 47-61.
MCCOLLOR, D.; STULL, R. Hydrometeorological Short-Range Ensemble Forecasts in Complex Terrain. Part II: =
Economic Evaluation. Weather and Forecasting, v. 23, p. 557-574, 2008, doi: 10.1175/2007WAF2007064.1
MCGUFFIE, K.; HENDERSON-SELLERS, A. (1997). “A climate Modelling Primer”. England: John Wiley & Sons.
253p.
MEIßNER, D., KLEIN, B., & IONITA, M. (2017). Development of a monthly to seasonal forecast framework tailored
to inland waterway transport in central Europe. Hydrology and Earth System Science, 6401-6423.
MELLER, A.; COLLISCHONN, W.; FAN, F.M.; BUARQUE, D.C.; PAIVA, R.C.D.; DIAS, P.; MOREIRA, D.
Previsão de Cheias por Conjunto em Curto Prazo. Revista Brasileira de Recursos Hídricos, v. 19, p. 33-49,
2014.
MONHART, S., ZAPPA, M., SPIRIG, C., SCHAR, C., BOGNER, K. (2019). Subseasonal hydrometeorological
ensemble predictions in small- and medium-sized mountainous catchments: benefits of the NWP approach.
Hydrol. Earth Syst. Sci., 23, 493–513, https://doi.org/10.5194/hess-23-493-2019
MOORE, R. J.; BELL, V. A.; JONES, D. A. (2005). “Forecasting for flood warning”. Computes Rendus Geoscience,
v. 337, p. 203-217.
MORSS, R. E., LAZO, J. K., BROWN, B. G., BROOKS, H. E., GANDERTON, P. T., & MILLS, B. N. (2008). Societal
and economic research and applications for weather forecasts. American meteorological society, 335-346.
MURPHY, A. H. (1993). “What is a good forecast? An essay on the nature of goodness in weather forecasting”. Wea.
Forecasting, v. 8, p. 281-293, 1993
NAMIAS, J. (1972). “Influence of Northern Hemisphere Ciculation on Drought in Northes Brazil”. Tellus, v.24, p.
336-342.
NASEM. The National Academies of Sciences, E. a. (2010). Assessment of Intraseasonal to Interannual Climate
Prediction and Predictability. Washington, D.C.: The National Academy Press.
NASEM. THE NATIONAL ACADEMIES OF SCIENCES, E. A. (2016). Next Generation Earth System Prediction:
Strategies for Subseasonal to Seasonal forecasts. Washington, D.C.: The National Academy Press.
NEAL, E. G.; WALTER, M. T.; COFFEN, C. (2002). “Linking the Pacific oscillation to seasonal stream discharge
patterns in Southest Alaska”. Journal of Hydrology, v. 263, p. 188-197.
NILSSON, P.; UVO, C. B.; BERNDTSSON, R. (2006). “Monthly runoff simulation: comparison of a conceptual
model, neural networks anda combination of them”. Journal of Hydrology, v. 321, p. 344- 363.
NOBERT, S., DEMERITT, D. & CLOKE, H. (2010). “Informing operational flood management with ensemble
predictions: lessons from Sweden. Journal of Flood Risk Management”. 3, p.72 79, 2010.
NOBRE, P., SHUKLA, J. (1996). “Variations of sea surface temperature, wind stress and rainfall over the tropical
Atlantic and South America”. Journal of climate, v. 10(4), p. 2464-2479.
O’CONNOR, R. E., YARNAL, B., DOW, K., & JOCOY, C. L. (2005). Feeling at Risk Matters:Water Managers and
the Decision to Use Forecasts. Risk Analysis, 1265-1275.
O’DONNELL, M., & BONNIE COLBY. (2009). Dry-Year Water Supply Reliability Contracts: A Tool for Water
Managers. Arizona: The University of Arizona.
OGALLO, L. A.; BOULAHYA, M. S.; KEANE, T. (2000). “Applications of seasonal to interannual climate prediction
in agricultural planning and operations”. Agricultural and Forest Mteorology, v. 103, p. 159-166.
OLSSON, J.; LINDSTROM, G. Evaluation and calibration of operational hydrological ensemble forecasts in Sweden.
J. Hydrol., v. 350, p.14-24, 2008.
OLSSON, J.; UVO, C. B.; JINNO, K.; KAWAMURA, A.; NISHIYAMA, K; KOREEDA, N.; NAKASHIMA, T;
MORITA, O. (2004). “Neural Networks for rainfall forecasting by atmospheric downscaling”. Journal of
Hydrologic Engineering, v. 9(1), p. 1-12.
84
ONS – Operador Nacional do Sistema, 2011. Procedimentos de Rede Submódulo 9.5: Previsão de Vazões e Gerac¸ ão
de Cenários de Afluências, vol. 2., pp. 9.
ONS – Operador Nacional do Sistema, 2012a. Diretrizes para as Regras de Operac¸ ão de Controle de Cheias – Bacia
do Rio São Francisco (Ciclo 2012–2013). ONS RE 3/166/2012., pp. 158.
ONS – Operador Nacional do Sistema, 2012b. Inventário das Restric¸ ões Operativas Hidráulicas dos Aproveitamentos
Hidrelétricos (Revisão 1 de 2012). ONS RE 3/0105/2012., pp. 154.
ONS – Operador Nacional do Sistema, 2014. Diretrizes para as Regras de Operac¸ ão de Controle de Cheias – Bacia
do Rio Iguaçu¸(Ciclo 2013–2014). ONS RE 3/0064/2014., pp. 48.
PAGANO, T. C., HARTMANN, H. C., & SOROOSHIAN, S. (2002). Factors affecting seasonal forecast use in Arizona
water management: a case study of the 1997-98 El Niño. CLIMATE RESEARCH, 259–269.
PAGANO, T.C.; WOOD, A. W.; RAMOS, M.-H.; CLOKE, H. L.; PAPPENBERGER, F.; CLARK, M. P.;
CRANSTON, M.; KAVETSKI, D.; MATHEVET, T.; SOROOSHIAN, S.; VERKADE, J. S. (2014).
“Challenges of Operational River Forecasting”. Journal of Hydrometeorology, v. 15, n. 4, p. 1692-1707, doi:
10.1175/JHM-D-13 0118.1.
PALMER, R. B.-Y. (2007). The Ensemble Prediction System - Recent and Ongoing Developments. Fonte: Series:
ECMWF Technical Memoranda: http://www.ecmwf.int/publications/
PANOFSKY, H. A., & BRIER, G. W. (1968). Some Applications of Statistics to Meteorology. Pennsylvania, US.
PAPPENBERGER, F.; BEVEN, K.J.; HUNTER, N.M.; BATES, P.D.; GOUWELEEUW, B.T.; THIELEN, J.; DE
ROO; A.P.J. (2005). “Cascading model uncertainty from medium range weather forecasts (10 days) through a
rainfall runoff model to flood inundation predictions within the European Flood Forecasting System (EFFS)”.
Hydrol.Earth Syst. Sci., v.35, p.381-393, 2005.
PAPPENBERGER, F., STEPHENS, E., THIELEN, J., SALAMON, P., DEMERITT, D., VAN ANDEL, S. J.,
WETTERHAL, F., ALFIERI, L. (2013). Visualizing probabilistic flood forecast information: expert
preferences and perceptions of best practice in uncertainty communication. Hydrol. Process.27, 132–146
(2013). DOI: 10.1002/hyp.9253
PAPPENBERGER, F.; BOGNER, K.; WETTERHALL, F.; HE, Y.; CLOKE, H. L.; THIELEN, J. (2011a). “Forecast
convergence score: a forecaster’s approach to analyzing hydro-meteorological forecast systems”. Adv. Geosci.,
v. 29, p. 27-32, doi: 10.5194/adgeo-29-27-2011.predictions. Hydrol. Earth Syst. Sci., v. 16, p. 815–831, 2012.
PAPPENBERGER, F., STEPHENS, E., THIELEN., DEL MEDICO, M. (2011b). The impact of weather forecast
improvements on large scalehydrology: analysing a decade of forecasts of the EuropeanFlood Alert System.
Hydrol. Process.25, 1091 – 1113 (2011). DOI: 10.1002/hyp.7772
PATT, A. G., OGALLO, L., & HELLMUTH, M. (2007). Learning from 10 Years of Climate Outlook Forums in Africa.
SCIENCE, 49-50.
PEREZ, E. C., & MASON, S. J. (2014). Climate information for humanitarian agencies: some basic principles. Earth
Perspectives, 1-6.
PEREZ, E. C., STEPHENS, E., BISCHINIOTIS, K., & AALST, M. V. (2017). Should seasonal rainfall forecasts be
used for flood preparedness? Hydrology and Earth System Sciences, 4517-4524.
PIANI, C., HAERTER, J. O., & COPPOLA, E. (2010). Statistical bias correction for daily precipitation in regional
climate models over Europe. Theoretical and Applied Climatology, 187–192.
PIERCE, C.; SEED, A.; BALLARD, S.; SIMONIN, D.; LI, Z. (2012). “Nowcasting, Doppler Radar Observations
Weather Radar, Wind Profiler, Ionospheric Radar, and Other Advanced Applications, Dr. Joan Bech (Ed.),
ISBN: 978 953-51-0496-4.
85
PONTES, P. R. M. ; FAN, F. M. ; COLLISCHONN, W. ; PAIVA, R. C. D. (2013) . Análise da Sensibilidade da Vazão
do Rio Paraná a Pontencial Alteração da Precipitação. In: XX Simpósio Brasileiro de Recursos Hídricos, 2013,
Bento Gonçalves - RS. Anais XX Simpósio Brasileiro de Recursos Hídricos, 2013.
RAMOS, M.H.; MATHEVET, T.; THIELEN, J.; PAPPENBERGER, F. (2010). “Communicating uncertainty in hydro
meteorological forecasts: mission impossible?”. Meteorological Applications, v. 17, p. 223-235, 2010.
RAMOS, M.H.; VAN ANDEL, S.J.; PAPPENBERGER, F. Do probabilistic forecasts lead to better decisions? Hydrol.
Earth Syst. Sci., v. 9, n. 17, p. 2219-2232, 2013, doi:10.5194/hess-17-2219-2013.
REITER, P., GUTJAHR, O., SCHEFCZYK, L., HEINEMANN, G., & CASPER, M. (2015). Bias correction of
ENSEMBLES precipitation data with focus on the effect of the length of the calibration period.
Meteorologische Zeitschrift, 85–96.
REITER, P., GUTJAHR, O., SCHEFCZYK, L., HEINEMANN, G., & CASPER, M. (2017). Does applying quantile
mapping to subsamples improve the bias correction of daily precipitation? INTERNATIONAL JOURNAL OF
CLIMATOLOGY, 1-11.
RICHARDSON, D. S. (2000). Skill and relative economic value of the ECMWF ensemble prediction system. Quarterly
Journal of the Royal Meteorological Society, 649-667.
ROBERTSON, A. W., BAETHGEN, W., BLOCK, P., LALL, U., SANKARASUBRAMANIAN, A., FILHO, F. D.,
& VERBIST, K. M. (2014). Climate risk management for water in semi–arid regions. Earth Perspectives, 1-
12.
ROULIN, E. Skill and relative economic value of medium-range hydrological ensemble predictions. Hydrol. Earth
Syst. Sci., v. 11, p.725–737, 2007, doi:10.5194/hess-11 725-2007.
SAMPAIO, G.; DIAS, P. L. S. (2014). Evolução dos modelos climáticos e de previsão de tempo e clima. Revista USP,
São Paulo, n. 103, p. 41-54.
SCHAAKE, J.; FRANZ, K.; BRADLEY, A.; BUIZZA, R. (2006). “The Hydrologic Ensemble Prediction Experiment
(HEPEX)”. Hydrol. Earth Syst. Sci. Discuss., v. 3, p.3321–3332, 2006.
SCHELLEKENS, J.; WEERTS, A.H.; MOORE, R.J.; PIERCE, C.E; HILDON, S. (2011). “The use of MOGREPS
ensemble rainfall forecasts in operational flood forecasting systems across England and Wales”. Adv. Geosci.,
29, p.77–84, 2011.
SCHERRER, S. C.; APPENZELLER, C.; ECKERT, P.; CATTANI, D. Analysis of the spread-skill relations using the
ECMWF ensemble prediction system over Europe. Wea. Forecasting, v. 19, p. 552 – 565, 2004.
SCHMIDLI, J., FREI, C., & VIDALE, P. L. (2006). Downscaling from gcm precipitation: a benchmark for dynamical
and statistical downscaling methods. INTERNATIONAL JOURNAL OF CLIMATOLOGY, 679–689.
Schmidli, J., Goodess, C. M., Frei, C., Haylock, M. R., Hundecha, Y., Ribalaygua, J., & Schmith, T. (2007). Statistical
and dynamical downscaling of precipitation: An evaluation and comparison of scenarios for the European Alps.
JOURNAL OF GEOPHYSICAL RESEARCH, 1-20.
SHAH, R., SAHAI, A. K., & MISHRA, V. (2017). Short to sub seasonal hydrologic forecast to manage water and
agricultural resources in India. Hydrology and Earth System Science, 707–720.
SHAPIRO, M. A., AND COAUTHORS, 2010: An Earth-system prediction initiative for the twenty-first century. Bull.
Amer. Meteor. Soc., 91, 1377–1388.
SILVESTRI, G. E.; VERA, C. S. (2003). “Antartic Oscillation signal on precipitation anomalies over southeastern
South America”. Geophysical Research Letters, v. 30(21), 4p.
SIQUEIRA, B., MARTINS, G., NERY, J.T. (2018). Variabilidade Sazonal da precipitação na bacia do Paraná no
território brasileiro. Revista Brasileira de Climatologia. 16p
SIQUEIRA, V. A., COLLISCHONN, W., FAN, F.M., CHOU, S. C. (2016b). Ensemble flood forecasting based on
operational forecasts of regional Eta EPS in the Taquari-Antas basin. RBRH. vol.21, n.3, pp.587-602. ISSN
2318-0331. http://dx.doi.org/10.1590/2318-0331.011616004.
86
SIQUEIRA, V. A., FLEISCHMANN, A., JARDIM, P. F., FAN, F. M., & COLLISCHONN, W. (2016a). IPH-Hydro
Tools: a GIS coupled tool for watershed topology acquisition in an opensource environment. RBRH - Revista
Brasileira de Recursos Hídricos, 274- 287.
SIQUEIRA, V. A, PAIVA, R. C. D., FLEISCHMANN, A. S., FAN, F. M., RUHOFF, A. L., PONTES, P., R., M.,
PARIS, A., CALMANT, S., COLLISCHONN, W. (2018). Toward continental hydrologic–hydrodynamic
modeling in South America. Hydrol. Earth Syst. Sci., 22, 4815-4842, 2018. https://doi.org/10.5194/hess-22-
4815-2018
SOBOLOWSKI, S., GONG, G., & TING, M. (2010). Modeled Climate State and Dynamic Responses to Anomalous
North American Snow Cover. American Meteorological Society, 785-799.
SRINIVASAN, G., RAFISURA, K. M., & SUBBIAH, A. R. (2011). Climate information requirements for community
level risk management and adaptation. CLIMATE RESEARCH, 5–12.
SUAREZ, P., & TALL, A. (2010). Towards forecast‐based humanitarian decisions: Climate science to get from early
warning to early action. Humanitarian Futures Programme, 1-10.
TEUTSCHBEIN, C., & SEIBERT, J. (2012). Bias correction of regional climate model simulations for hydrological
climate-change impact studies: Review and evaluation of different methods. Journal of Hydrology, 12–29.
THEMEßL, M. J., & LEUPRECHT, A. G. (2011). Empirical-statistical downscaling and error correction of daily
precipitation from regional climate models. INTERNATIONAL JOURNAL OF CLIMATOLOGY, 1530–
1544.
THOMSON, M. C., DOBLAS-REYES, F. J., MASON, S. J., HAGEDORN, R., CONNOR, S. J., PHINDELA, T., &
MORSE, A. P. (2006). Malaria early warnings based on seasonal climate forecasts from multi-model
ensembles. Nature, 576–579.
THOMSON, M. C., MASON, S., PLATZER, B., MIHRETIE, A., OMUMBO, J., & MANTILLA, G. (2014). Climate
and health in Africa. Earth Perspectives, 1-12.
TIAN, D., WOOD, E. F., & YUAN, X. (2016). CFSv2-based sub seasonal precipitation and temperature forecast skill
over the contiguous United States. Hydrology and Earth System Sciences, 1477-1490.
Tucci, C. E., Clarke, R. T., & Collischonn, W. (2003). Long-term flow forecasts based on climate and hydrologic.
WATER RESOURCES RESEARCH, 1-11.
TURNER, S. W., BENNETT, J. C., ROBERTSON, D. E., & GALELLI, S. (2017). Complex relationship between
seasonal streamflow forecast skill and value in reservoir operations. Hydrology and Earth System Sciences,
4841 4859.
VERKADE, J. S.; WERNER, M. G. F. Estimating the benefits of single value and probability forecasting for flood
warning. Hydrol. Earth Syst. Sci., v. 15, p. 3751-3765, 2011. doi:10.5194/hess-15-3751-2011.
VITART, F., & ROBERTSON, A. W. (2018). The sub seasonal to seasonal prediction project (S2S) and the prediction
of extreme events. Climate and Atmospheric Science, 1-7.
VITART, F., BUIZZA, R., BALMASEDA, M. A., BALSAMO, G., BIDLOT, J.-R., BONET, A., . . . PALMER, T. N.
(2008). The new Var-EPS-monthly forecasting system: A first step towards seamless prediction. Quarterly
Journal of the Royal Meteorological Society, 1789-1799.
VITART, F., BUIZZA, R., BALMASEDA, M. A., BALSAMO, G., BIDLOT, JR., BONET, A., FUENTES, M.,
HOFSTADLER, A., MOLTENI, F., PALMER, T. (2008). The new VarEPS–monthly forecasting system: A
first step towards seamless prediction. QUARTERLY JOURNAL OF THE ROYAL METEOROLOGICAL
SOCIETY. Q. J. R. Meteorol. Soc. 134: 1789–1799. DOI: 10.1002/qj.322
VITART, F., ROBERTSON, A. W., & ANDERSON, D. L. (2012). Subseasonal to Seasonal Prediction Project:
bridging the gap between weather and climate. WMO Bulletin, 23-28.
87
VITART, F., ROBERTSON, A. W., & GROUP, S. S. (2015). CHAPTER 20. SUB SEASONAL TO SEASONAL
PREDICTION: LINKING WEATHER AND CLIMATE. Em W. M. (WMO), SEAMLESS PREDICTION OF
THE EARTH SYSTEM: FROM MINUTES TO MONTHS (p. 483). Geneva, Switzerland: WMO.
WALISER, D. (2005). Predictability and Forecasting. Em; W. Lau, & D. Waliser, Intraseasonal Variability in the
Atmosphere-Ocean climate system (p. 457). Chichester, UK: Praxis Publishing.
WATTERSON, I. G. (2009) “Components of precipitation and temperature anomalies and change associated with
models of the Southern Hemisphere”. Hydrological Processeces, v. 29, p. 809-826.
WETTERHALL, F., & GIUSEPPE, F. D. (2018). The benefit of seamless forecasts for hydrological predictions over
Europe. Hydrology and Earth System Sciences, 3409–3420.
WHITE, C. J., CARLSEN, H., ROBERTSON, A. W., KLEIN, R. J. T., LAZO, J. K., KUMAR, A., VITART, F., =
COUGHLAN DE PEREZ, E., RAY, A. J., MURRAY, V., BHARWANI, S., MACLEOD, D., JAMES, R.,
FLEMING, L., MORSE, A. P., EGGEN, B., GRAHAM, R., KJELLSTRÖM, E., BECKER, E., PEGION, K.
V., HOLBROOK, N. J., MCEVOY, D., DEPLEDGE, M., PERKINS-KIRKPATRICK, S., BROWN, T. J.,
STREET, R., JONES, L., REMENYI, T. A., HODGSON-JOHNSTON, I., BUONTEMPO, C., LAMB, R.,
MEINKE, H., ARHEIMER, B., AND ZEBIAK, S. E. (2017). Potential applications of subseasonal-to-seasonal
(S2S) predictions, Meteorol. Appl., 24, 315 325. https://doi.org/10.1002/met.1654.
WILKS, D. (2006). STATISTICAL METHODS IN THE ATMOSPHERIC SCIENCES. London, UK: Elsevier.
WILKS, D. S., & HAMILL, T. M. (1995). Potential Economic Value of Ensemble-Based Surface Weather Forecast.
American Meteorological Society, 3565-3575.
WOOD, A. W., MAURER, E. P., KUMAR, A., & LETTENMAIER, D. P. (2002). Long-range experimental hydrologic
forecasting for the eastern United States. JOURNAL OF GEOPHYSICAL RESEARCH, 1-15.
WOOLNOUGH, S. J., VITART, F., & BALMASEDA, M. A. (2007). Implications for MJO prediction: The role of the
ocean in the Madden–Julian Oscillation: QUARTERLY JOURNAL OF THE ROYAL METEOROLOGICAL
SOCIETY, 117–128.
XU, C.-Y. (1999). From GCMs to river flow: a review of downscaling methods and hydrologic. Progress in Physical
Geography, 229–249.
YUAN, X., WOOD, E. F., & MA, Z. (2015). A review on climate-model-based seasonal hydrologic forecasting:
physical understanding and system development. WIREs Water, 523-536. https://doi.org/10.1002/wat2.1088
ZAPPA, M; JAUN, S.; GERMAN, U.; WALSER, A. & FUNDEL, F. (2011). “Superposition of three sources of
uncertainties in operational flood forecasting chains”. Atmospheric Research, 100, np.246–262, 2011.
ZHOU, G., MINAKAWA, N., GITHEKO, A. K., & YAN, G. (2004). Association between climate variability and
malaria epidemics in the East African highlands. Proceedings of the National Academy of Sciences of the
United States of America, 2375-2380.
The S2S database contains real-time forecasts from ECMWF from 1st January 2015, and
the associated re-forecasts.
88
1. Ensemble version
Ensemble identifier code: CY43R3
Short Description: Global ensemble system that simulates initial uncertainties using
singular vectors and ensemble of data assimilation and model uncertainties due to physical
parameterizations using a stochastic scheme. based on 51 members, runs twice a week
(Monday and Thursday at 00Z) up to day 46.
If yes, please describe ocean model briefly including frequency of coupling and any
ensemble perturbation applied: Ocean model is NEMO3.4.1 with a 0,25 degree horizontal
resolution, 75 vertical levels, initialized from ECMWF Ocean Analysis + 4 perturbed analyses
produced by perturbing the wind field during the ocean analysis. Frequency of coupling is
hourly.
If yes, please describe sea-ice model briefly including any ensemble perturbation
applied: Interactive sea-ice model (the Louvain-la-Neuve Sea Ice Model - LIM2). Initial
perturbations of sea-ice from the 5 ensemble ocean/sea-ice analysis/re-analysis. No
stochastic perturbations.
If yes, please describe wave model briefly including any ensemble perturbation
applied: ECMWF wave model. No perturbation. Resolution is 0.25 degrees up to day 15
and 0.5 degrees after day 15.
Horizontal resolution of the atmospheric model: Tco639 (about 16 km) up to day 15 and
Tco319 (about 32 km) after day 15
Integration time step: 12 minutes for day 0-15 and 20 minutes for day 15-46
89
3. Initial conditions and perturbations
Data assimilation method for control analysis: 4D Var (atmosphere) and 3DVAR
(ocean/sea-ice)
What is the land surface model (LSM) and version used in the forecast model, and
what are the current/relevant references for the model? Are there any significant
changes/deviations in the operational version of the LSM from the documentation of
the LSM? IFS Documentation, Physical Processes, Chapter 8 Surface parameterisation,
2016 http://www.ecmwf.int/en/elibrary/16648-part-iv-physical-processes
If “realistic”, does the soil moisture come from an analysis using the same LSM as is
coupled to the GCM for forecasts, or another source? Please describe the process of
soil moisture initialization.
LDAS-based (simplified EKF) as used in all IFS forecast and described
here: https://software.ecmwf.int/wiki/display/LDAS/LDAS+Home
Is there horizontal and/or vertical interpolation of initialization data onto the forecast
model grid? If so, please give original data resolution(s). Yes horizontal interpolations.
For soil moisture the interpolation is standardized on soil moisture index (to account for
different soil texture in input and target resolution grid).
Does the LSM differentiate between liquid and ice content of the soil? If so, how are
each initialized? Yes in a diagnostic wave using temperature and a latent heat barrier
(described in Viterbo et al. 1999, see IFS documentation)
If all model soil layers are not initialized in the same way or from the same source,
please describe. The LDAS is active on the top 1m of soil moisture (the first 3 layers) and
the forth layer (1 to 2,89 m deep) is not initialised.
If “realistic”, does the snow come from an analysis using the same LSM as is coupled
90
to the GCM for forecasts, or another source? Please describe the process of soil
moisture initialization. LDAS-based (Optimal Interpolation) as used in all IFS forecast and
described here: https://software.ecmwf.int/wiki/display/LDAS/LDAS+Home
Is there horizontal and/or vertical interpolation of data onto the forecast model
grid? If so, please give original data resolution(s) horizontal interpolation
Are snow mass, snow depth or both initialized? What about snow age, albedo, or
other snow properties? Snow mass and snow temperature are initialized by the LDAS,
snow albedo and snow density are cycled from the model forecast (open loop).
If “realistic”, does the soil moisture come from an analysis using the same LSM as is
coupled to the GCM for forecasts, or another source? Please describe the process of
soil moisture initialization. LDAS-based (Optimal Interpolation) as used in all IFS forecast
and described here: https://software.ecmwf.int/wiki/display/LDAS/LDAS+Home
Is the soil temperature initialized consistently with soil moisture (frozen soil water
where soil temperature ≤0°C) and snow cover (top layer soil temperature ≤0°C under
snow)? Both the top soil temperature and the snow temperature (if present) are initialized.
Is there horizontal and/or vertical interpolation of data onto the forecast model
grid? If so, please give original data resolution(s) horizontal interpolation
If all model soil layers are not initialized in the same way or from the same source,
please describe. Only the first soil layer temperature is initialized, the other layers are
cycled from the model forecast (open loop).
If not, what is the source of vegetation parameters used by the LSM? Which time-
varying vegetation parameters are specified (e.g., LAI, greenness, vegetation cover
fraction) and how (e.g., near-real-time satellite observations? Mean annual cycle
climatology? Monthly, weekly or other interval?) Leaf Area Index and Albedo monthly
climatology both based on MODIS collection 5
What is the source of soil properties (texture, porosity, conductivity, etc.) used by the
LSM? FAO dominant soil texture class (as in Van Genuchten, 1980)
If the initialization of the LSM for re-forecasts deviates from the procedure for
forecasts, please describe the differences. The re-forecasts initialization is based on ERA-
91
Interim and ERA-Interim/Land datasets, while the real-time forecasts are based on the IFS
operational initial conditions of the ENS/EDA systems.
Do all ensemble members use exactly the same model version? Same
Additional comments
What kind of convective parameterization is used? Tiedtke 89, Bechtold et al 2004 (QJ)
92
7. Re-forecast Configuration
Number of years covered: 20 past years
Frequency: Produced on the fly twice a week to calibrate the Monday and Thursday 00Z
real-time forecasts. The re-forecasts consists of a 11-member ensemble starting the same
day and month as the Thursday real-time forecasts for the past 20 years.
Initial conditions: ERA interim (T255L60) + Soil reanalysis (Tco639) + ORAS5 ocean initial
conditions (0.25 degree)
Is the model physics and resolution the same as for the real-time forecasts: Yes
Is the ensemble generation the same as for real-time forecasts? Yes. Except for EDA
perturbations which are taken from the most recent year.
ECMWF re-forecasts are produced on the fly. This means that every week a 2 new set of re-
forecasts are produce to calibrate the Monday and Thursday real-time ensemble forecasts
of the following week using the latest version of IFS. The ensemble re-forecasts consist of a
11-member ensemble starting the same day and month as a real-time forecast (Monday
and Thursday), but covering the past 20 years. For instance the first re-forecast set archived
in the S2S database with this new version of the ECMWF model was the re-forecast used to
calibrate the real-time forecast of 14 May 2015 (a Thursday). This set consisted of a 11-
member ensemble starting on 1st January 1995, 1st January 1996, ... 1st January 2014 (20
years, 11 member ensemble = 220-member climate ensemble). The re-forecast dataset is
therefore updated every week in the S2S archive.
The ECMWF re-forecasts are archived in the S2S database using two dates: "date" and
"hdate" (see examples below): hdate is the actual date of the re-forecast (e.g. 19950101)
while date is the date of the real-time forecast (=ModelversionDate in grib2) associated to
the re-forecast (20150101). The reason we need 2 dates is because the ECMWF re-forecasts
are produced on the fly and we need to avoid the re-forecasts produced in the future years
to overwrite the re-forecasts currently produced. Therefore ModelversionDate allows us to
distinguish the re-forecasts produced in 2015 from those produced in 2016, 2017...
8. References:
Comprehensive description of the model physics: CY43R1 Official IFS Documentation
93