0% found this document useful (0 votes)
97 views11 pages

Seasonal Prediction of Crop Yields in Ethiopia Using An Analog Approach

1. The study develops an analog approach to predict crop yields in Ethiopia before planting season using historical soil moisture and yield data. 2. An hydrological model is used to simulate historical soil moisture levels at planting time, and an agricultural model is used to simulate corresponding historical crop yields. 3. The analog approach identifies historical years with similar soil moisture conditions to the target year and uses the simulated yields from those historical years to predict the target year's yield, providing forecasts one to two weeks before typical planting times to inform farmers' decisions.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
97 views11 pages

Seasonal Prediction of Crop Yields in Ethiopia Using An Analog Approach

1. The study develops an analog approach to predict crop yields in Ethiopia before planting season using historical soil moisture and yield data. 2. An hydrological model is used to simulate historical soil moisture levels at planting time, and an agricultural model is used to simulate corresponding historical crop yields. 3. The analog approach identifies historical years with similar soil moisture conditions to the target year and uses the simulated yields from those historical years to predict the target year's yield, providing forecasts one to two weeks before typical planting times to inform farmers' decisions.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 11

Agricultural and Forest Meteorology 331 (2023) 109347

Contents lists available at ScienceDirect

Agricultural and Forest Meteorology


journal homepage: www.elsevier.com/locate/agrformet

Seasonal prediction of crop yields in Ethiopia using an analog approach


Meijian Yang a, Guiling Wang a, *, Shu Wu b, Paul Block c, Rehenuma Lazin a, Sarah Alexander c,
Jonathan Lala c, Muhammad Rezaul Haider a, Zoi Dokou d, Ezana Amdework Atsbeha e,
Marika Koukoula a, Xinyi Shen a, Malaquias Peña a, Efthymios Nikolopoulos f,
Amvrossios Bagtzoglou a, Emmanouil Anagnostou a
a
Department of Civil and Environmental Engineering, University of Connecticut, Storrs, CT, United States
b
Nelson Institute Center for Climatic Research, University of Wisconsin-Madison, Madison, Wisconsin, United States
c
Department of Civil and Environmental Engineering, University of Wisconsin-Madison, Madison, Wisconsin, United States
d
Civil Engineering Department, California State University, Sacramento, California, United States
e
Department of Sociology, University of Connecticut, Storrs, Connecticut, United States
f
Department of Mechanical and Civil Engineering, Florida Institute of Technology, Melbourne, Florida, United States

A R T I C L E I N F O A B S T R A C T

Keywords: As Ethiopia’s population grows, crop yield predictions are becoming increasingly important for national food
Analog approach security. Accurate, easy-to-implement, and computationally efficient forecast approaches are desirable for broad
Crop yield forecast applications in emerging regimes like Ethiopia. In this study, we develop and test an analog approach for pre-
Ethiopia
season crop yield prediction conditioned on antecedent precipitation and planting time soil moisture content
Precipitation
Soil moisture
indices to guide cultivation decision making. Historical planting time soil moisture at four selected sites were
DSSAT simulated using the Coupled Routing and Excess STorage (CREST) hydrological model and classified into five
levels. Likewise, a historical crop yield database for each of the five classes of planting time soil moisture were
constructed using the Decision Support System for Agrotechnology Transfer (DSSAT) agricultural model.
Adopting maize as a representative crop, analog models based on different indexes or predictors with various
lead times were constructed and used to conduct hindcasts during 1979–2014 and real-time forecast in 2018 and
2019. Both the hindcast and real-time forecasts were then evaluated against yield observations. To verify the
applicability at locations with various environments, the analog models were then applied in different Agro­
ecological Zones. The analog models were shown to be accurate and easy to implement, which may incentivize
adoption by local extension agents and regional agricultural agencies to inform farmers’ crop choices.

1. Introduction learning and the hybrid of these technologies to generate crop yield
predictions at various spatiotemporal scales (Basso et al., 2013; Bolton
Crop yield predictions are increasingly important to every sector of and Friedl, 2013; Chlingaryan et al., 2018; Kang et al., 2009; Lobell
public and private agricultural systems and the supply chain, in terms of et al., 2009). However, great data demand, high model complexity, and
aiding crop selections, optimizing management practices, and antici­ high computational expense have limited their application especially in
pating market fluctuations (Lecerf et al., 2019; Macdonald and Hall, areas with scarce data, labor, education, and economic resources (Basso
1980). Historically, farmers make crop yield predictions following the et al., 2013; Filippi et al., 2019).
instructions of local experts who provide their forecast mainly by eval­ Season-ahead prediction of crop yield is particularly important in
uating the crop status throughout the growing season (Basso et al., countries prone to climate-driven food insecurity, such as Ethiopia,
2013). Yet, advanced data collection techniques and increased compu­ where the agricultural sector comprises over 50% of the country’s GDP
tational capacity have facilitated the development of a wide variety of (FSIN, 2019; Welteji, 2018). However, large rural populations and
crop yield forecast methods in recent decades. Recent studies utilize highly fragmented farmland result in a farming system dominated by
statistical models, process-based models, remote sensing, machine smallholders in Ethiopia with limited access to modern farm

* Corresponding author.
E-mail address: [email protected] (G. Wang).

https://doi.org/10.1016/j.agrformet.2023.109347
Received 26 June 2022; Received in revised form 28 December 2022; Accepted 24 January 2023
Available online 31 January 2023
0168-1923/© 2023 Elsevier B.V. All rights reserved.
M. Yang et al. Agricultural and Forest Meteorology 331 (2023) 109347

run for the historical years to simulate past crop yield (Yang et al.,
2020), and a hydrologic model, the Coupled Routing and Excess STorage
model (CREST, Shen and Anagnostou 2017), is used to simulate the
pre-season soil moisture for the same period. These model-generated
output form the database that supports the development of our analog
approach.
This study aims at providing an accessible tool for a real-time sea­
sonal forecast of meteorological effects on crop yield, which will support
and guide local communities in their effort to prepare for and mitigate
weather- and climate-related crop loss. This is especially useful for re­
gions where the capacity in process-based modeling is limited. Accord­
ing to our survey of the local farming communities, the regular planting
time is late May and early June. Hence, our forecast (around May 20th)
has a lead time of one to two weeks to inform farmers before their
regular planting time. Farmers and agencies could get prepared for the
upcoming season with a variety of measures including but not limited to
adjusting planting time and crop types.

2. Methodology and data

2.1. Reconstruction of historical yield dataset

Ethiopia’s Agricultural Sample Survey (AgSS) dataset provides crop


yield data at the field scale for the whole country (CSA 2019), yet the
observational record is short (13 years from 2007 to 2019). Thus, we
first reconstruct a 36-year crop yield series from 1979 to 2014, by
Fig. 1. Topography of Lake Tana Basin and the location of study sites. implementing a process-based crop model – DSSAT (Yang et al., 2020).
DSSAT is a widely used crop model that simulates the daily growth and
technologies and irrigation infrastructure (Mellander et al., 2013; Pas­ development of a variety of crops under prescribed environment con­
sarelli et al., 2018; Welteji, 2018). Crop production in the country de­ ditions, management strategies, and cultivar physiology and phenotype
pends highly on precipitation and other climate factors, with less than (Hoogenboom et al., 2012; Jones et al., 2003, 2017). Based on the 8-year
5% of the cropland in Ethiopia under irrigation (Passarelli et al., 2018). available observations, Yang et al. (2020) calibrated the physiological
Pre-season yield prediction may facilitate farmers’ decision making to parameters of major cereal crops (maize, wheat, millet, barley) in
maximize output by adjusting crop choices. As climate variability in DSSAT at our focus communities, Dangishta, Gaita, Kudmi, and Reem
Ethiopia is shown to affect crop yields and the national economy, sea­ (Fig. 1). The four sites have identical primary soil type - Chromic
sonal predictions are critical for developing adaptation capacity and Luvisols, a soil with a sandy-loam texture and reddish color (Getachew
building resilience to climate shocks (Block et al., 2008; Lala et al., 2020; and Manjunatha, 2022). In addition to meteorological forcing, crop
Lewis, 2017; MacLeod, 2019). Studies also found that preseason soil yield is also heavily influenced by the interannual variability of factors
moisture has a significant impact on crop yields and can act as a key that cannot be captured by DSSAT, including for example labor input,
index in forecasting the total seasonal yield (Mjelde and Keplinger, seed quality, pests, and events that may prevent timely harvesting of
1998; Yang et al., 2021). crops. For this reason, the DSSAT calibration focused on minimizing the
Analog approaches have long been favored for its ease of application multi-year mean biases and not on capturing the observed inter-annual
and low data demand in making predictions (Bergen and Harnack, 1982; variability (Yang et al., 2020). Here we use the crop yields simulated by
Bessafi et al., 2002; Monache et al., 2013), and index selection is a core the calibrated DSSAT during 1979–2014 as the database for developing
element of the approach. Historically, large scale features such as the analog forecast approach.
regional sea surface temperature (SST) and the Southern Oscillation The historical climate reanalysis (1979–2014) including solar radi­
Index (SOI) were commonly selected as the index(es) for choosing ation (SR), maximum temperature (Tmax), minimum temperature
analog years in generating crop yield predictions (De Jager et al., 1998; (Tmin) and precipitation (P), etc. are used for the calibration and
Mjelde and Keplinger, 1998; Hansen and Indeje, 2004; Mcdermott and simulation of DSSAT and CREST, as well as the hindcast verification of
Wikle, 2016). Such studies took advantage of the well-established tele­ analog approach. The data for SR, Tmax, and Tmin are from European
connections between ENSO and local climate to develop skillful fore­ centre for Medium-Range Weather Forecasts (ECMWF) ERA-Interim
casts with a seasonal (i.e., 3+ months) lead time. However, these indices (Dee et al., 2011), and data for P was from Multi-Source Weight­
are not directly related to crop growth and cannot account for the spatial ed-Ensemble Precipitation (MSWEP) (Beck et al., 2017). Both data have
heterogeneity of cropland in a large domain like Ethiopia. During the 0.25◦ spatial and 3-hourly temporal resolutions. Other data such as the
Kiremt (rainy season) in Ethiopia, cereal crops are mostly rainfed, and soil properties dataset is derived from SoilGrids and ISRIC-AfSIS (Hengl
local climate variables (that can differ significantly over a short dis­ et al., 2014; International Research Institute for Climate and Society
tance) are the key factors that drive yield variabilities. Hence, in this (IRI) et al., 2015). Agricultural management information was obtained
study, we choose different climate variables (i.e. solar radiation, from in-situ fieldwork and survey.
maximum and minimum temperature, precipitation) to assess and As planting time soil moisture is found to have a notable influence on
compare analog skill, ultimately selecting indices that most skillfully crop yield in this region (Yang et al., 2021), information on soil moisture
predict yields of major cereal crops (i.e. maize) in four kebeles – the condition should benefit our seasonal crop yield forecast. Hence, we
smallest administrative unit of Ethiopia, located in the Lake Tana Basin, apply the CREST hydrologic model to simulate planting time (April 21 -
Ethiopia (Fig. 1). To supplement the very limited observational data in May 10) soil moisture at three soil layers, 0–15 cm, 15–40 cm, 40–190
this region, a process-based crop model, the Decision Support System for cm (Lazin et al., 2020). CREST is a fully-distributed hyper-resolution
Agrotechnology Transfer (DSSAT, Jones et al., 2003), is calibrated and model that can simulate long-term hydrological processes with high
computational efficiency (Shen and Anagnostou, 2017). In this study,

2
M. Yang et al. Agricultural and Forest Meteorology 331 (2023) 109347

Fig. 2. Classification of soil moisture in 2018. The five-colored panel indicates the five initial soil moisture categories; the blue line is the CDF of historical initial soil
moisture, and the red line is the CDF of the initial soil moisture in the target year. The category that the majority of the red line locate is determined as the initial soil
moisture level in the target year.

CREST is run for the whole Upper Blue Nile Basin at 500 m spatial cubic spline interpolation to make it consistent with the baseline data for
resolution and 3-hourly time step. The climate forcing data used to better comparison (Haider et al., 2020). Also, DSSAT requires daily
initiate CREST for 2018 and 2019 include SR, Tmax, Tmin obtained from maximum and minimum temperature data as input, which can be more
Global Data Assimilation System (GDAS) (GDAS, n.d.), and precipitation accurately retrieved from 3-hourly than 6-hourly data. Finally, given the
from Integrated Multi-satelliE Retrievals for GPM (IMERG) (Huffman consistency between IMERG (available 2013 – present) and MSWEP
et al., 2014). Note that we used GDAS and IMERG for the forecast years (1979 - 2014), IMERG is applied for the analog year selection.
(2018 and 2019) instead of ERA5, because our first real-time seasonal Cumulative distribution mapping, or quantile mapping, is used to
forecast was conducted (and provided to the farming communities in correct the biases of the seasonal climate forecast by matching the
Ethiopia) in 2018, before the release of ERA5 in 2019. The climate probability distribution of the forecast values with the historical refer­
prediction data were obtained from National Oceanic and Atmospheric ence distribution. Below are the three steps of the quantile mapping
Administration (NOAA) Climate Forecasts System (CFSv2) and input to approach (see the example illustrated in Supplementary Figure S1):
CREST as a forcing for the soil moisture prediction (Saha et al., 2014).
CFS assimilates observational data from multiple sources (including Step 1. We first estimate the CDF of both biased/forecast data and
surface instruments, upper air balloons, aircrafts, and satellites) into a reference data. For a given value in the biased data, we find the
model of the Earth’s ocean, land, and atmosphere system (Saha et al., corresponding CDF value of this variable. In this example, with a
2010). CFS forecasts key climate variables up to seven months including biased data value of 25.6, the corresponding CDF value is 0.47.
precipitation, temperature, shortwave radiation, etc. at 1◦ spatial and Step 2. We identify the data point with the same CDF value in the
6-hourly temporal resolution. We simultaneously bias correct and reference data, which is 19.
spatially downscale (to 0.1◦ ) the solar radiation and temperature in Step 3. We then take the reference data value (19) at the corre­
CFSv2 data based on GDAS using a cumulative distribution function sponding CDF value (0.47) as the corrected data. In this example, the
matching approach, and then temporally downscaled to 3-hourly by forecast value of 25.6 is corrected to the value of 19.

3
M. Yang et al. Agricultural and Forest Meteorology 331 (2023) 109347

Table 1 five representative soil moisture scenarios defined by the five percentiles
Criteria for soil moisture and crop yield classification. as an initial condition. This was done to account for the potential impact
Percentile Soil Moisture Crop Yield of different soil moisture scenarios in each year and to increase the in­
ventory of combinations between initial soil moisture and growing
< 10th Extreme dry Very low
10th - 25th Dry Low season meteorological forcing. Hence, for each site, the crop yield
25th - 75th Normal Normal database has an array size of 5 (scenarios) × 36 (years).
75th - 90th Wet High
> 90th Extreme Wet Very high 2.2. Analog selection and crop yield forecast

This approach removes the systematic bias of the forecast dataset For each target year, the crop yield forecast involves the determi­
relative to the observed dataset. The quantile mapping approach has nation of initial soil moisture category and the identification of analog
been widely used in the bias correction of hydrological and climate years. We first plot the CDF of planting time soil moisture of the target
variables (Teutschbein and Seibert 2012; Ahmed et al., 2013; Tahir year, and choose the category that most part of the CDF falls in as the
et al., 2021, Panjwani et al., 2021). Here we applied this approach in planting time soil moisture level for that year. Taking 2018 for example
correcting the biases in solar radiation and temperature (see Supple­ (Fig. 2), the planting time soil moisture at Dangishta in 2018 was clas­
mentary Figure S2). sified as Extreme Wet. The analog years for a target year are selected
The simulated sub-daily soil moisture at the four sites are then based on climate variables at a certain period of time. Taking 2018 again
aggregated to daily resolution. At each site, we use historical soil as an example, if we choose the total precipitation from January to May
moisture data for 20 days of the planting season for each of the 36 years (J2M) as the index, we would then find the top three years with J2M
to derive soil moisture statistics. A cumulative distribution function precipitation closest to 2018 (Fig. 3), and 2005, 2013, and 2001 would
(CDF) is derived based on the historic three-layer averaged soil moisture be selected as the analog years for 2018. The forecasted crop yield is
data (Fig. 2), and historical soil moisture series (20 days × 36 years) is determined as the mean of crop yield in the three analog years corre­
classified into five levels according to the percentile criteria shown in sponding to the planting time soil moisture level of the target year. The
Table 1. These categories were defined after consulting local farmers and analog skill is defined as the correlation coefficient between two crop
agencies to best reflect the stakeholders’ needs and preference. we yield series – the DSSAT-simulated yields of the 36 target years and the
define 50% of the data as “normal” category to allow enough room for average yield of the three corresponding analog years for each of the 36
climate and yield to fluctuate. If the range of “normal” category is too target years. We compare the analog skills of using different climate
small, most of the years would be categorized as “abnormal”, which variables at different time period through hindcast to decide which
could be misleading to those who use our report without understanding combination should be chosen, as detailed in the Results section.
the statistical details. Given that we only have 13 years of data, making Here we employ two commonly used metrics R2 and RMSE to
the range of “extreme dry” or “extreme wet” too wide would categorize quantify the performance of our forecast system.
many years as “extreme” events although they may not be truly Coefficient of determination (R2) is defined as the square of Pearson
∑n
(xi − x)(yi − y)
“extreme”. Soil moisture values at 10th, 25th, 50th, 75th, and 90th Correlation Coefficient rxy = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ̅ . Root Mean
∑n i=1 2 √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
∑n 2
(xi − x) (yi − y)
percentiles are taken to be representative of the five categorical classi­ i=1 i=1

fications. Next, we linearly interpolate the three-layer soil moisture Square Error (RMSE) is the standard deviation of the residuals (predic­
√∑̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
simulated by CREST from three layers to nine layers required by DSSAT, n
(y − xi )2
tion errors), RMSE = i=1 i
. Here, xi is the estimated time series,
and run DSSAT for the historic 36 years five times, each using one of the n

Fig. 3. Procedure of the selection of top three analog years based on J2M precipitation. a) J2M total precipitation in Dangishta, b) Top three historical years with the
least J2M precipitation difference from 2018, and c) Comparison between target precipitation and analog precipitation.

4
M. Yang et al. Agricultural and Forest Meteorology 331 (2023) 109347

Table 2 This could weaken DSSAT’s performance in capturing the actual yield at
Correlation coefficients between maize yield and annual mean climate variables; these two sites.
pre-season precipitation is also listed as a comparison.
Dangishta Kudmi Reem Gaita 3.2. Determining time period for analog index based on crop yield
SR 0.56* 0.53* 0.29 0.31 hindcast
Tmax 0.62* 0.56* 0.37* 0.23
Tmin 0.47* 0.36* 0.22 0.22
Precipitation during different sub-periods may influence crop yields
P − 0.42* − 0.58* − 0.43* − 0.43*
Pre-season P − 0.63* − 0.68* − 0.65* − 0.41* differently (e.g., pre-season precipitation that impacts soil moisture or
growing season precipitation). Here we first identify the months when
Note: * significant at 0.05 level.
precipitation influences the analog skill the most. Fig. 4 shows a contour
plot of correlation coefficient between maize yield and total precipita­
yi is the actual observation time series, x and y are the mean value of tion during different time periods. At all sites, except Gaita, the highest
each time series, n is the number of data points. correlation (< − 0.6) is found for total precipitation before May
With the developed analog approach, we perform seasonal forecast (Table 2). We also conducted hindcast for all past years using precipi­
for crop yields in Kiremt (wet season) over the period 2007–2019 and tation during different periods as index and the analog skills are shown
verify the forecast against observational data both numerically and in Fig. 5. At Dangishta, Kudmi, and Reem, the highest analog skill (>
categorically. 0.6) is achieved when the pre-season (January-May) precipitation is
used as the index. This is particularly remarkable for Kudmi and Reem,
3. Seasonal forecast model development and verification which are located in a mountainous area where excessive pre-season
rainfall often makes the field too wet to be workable and increases the
3.1. Identifying analog indexes among climate variables risk of flooding damages.
As indicated in our earlier studies (Yang et al., 2020), planting time is
To apply the analog approach, we must first determine what hydro- the only period when the correlation between antecedent precipitation
climate variables should be included in the analog index. Maize, the and maize yield is significant. Due to the requirement from our project
most widely planted crop in this area, is selected as the representative and local communities, the objective of this study is to release the sea­
crop for testing our methodology. The correlation coefficients between sonal forecast before planting with a reasonable lead time. Although
maize yield and annual mean climate variables are shown in Table 2. precipitation in anthesis and seed set stages is positively correlated with
Among the four climate variables, precipitation is the only variable with yield, the insignificant correlation suggests a relatively small influence
a significant correlation (P < 0.05) to maize yield at all four sites. The of precipitation in these stages. Other climate variables have significant
negative correlation between P and maize yield is primarily due to and positive correlation with yield only in the flowering stage. To use
excessive rainfall resulting in a shortage of sunshine and low tempera­ flowering stage climate variables as indexes, we need to rely on climate
ture during the Kiremt season (rainy season, June to September) (Yang forecasts, which would add more uncertainties to our forecast system.
et al., 2020). Albeit negatively correlated, precipitation stands out as the Therefore, cumulative precipitation up to planting-time is selected as
best index for selecting analog years. One of the main reasons that leads our analog index.
to the less significant correlation between climate variables and yield in
Reem and Gaita than Dangishta and Kudmi is topographic complexity.
Reem and Gaita are located in areas with more complex topography 3.3. Seasonal forecast and sensitivity test
where crop yield is often more prone to natural hazard such as flooding.
Here we describe how to perform the analog-based seasonal forecast

Fig. 4. Correlation coefficient between maize yield and accumulated precipitation during different time periods (i.e. from the beginning to ending months).

5
M. Yang et al. Agricultural and Forest Meteorology 331 (2023) 109347

Fig. 5. Analog skills based on precipitation during different periods (from the beginning to ending months), showing the highest skill with pre-season (January-May)
precipitation.

Fig. 6. Selection of the top three analog years for 2018 and 2019 based on the January-May precipitation from IMERG.

using the years 2018 and 2019 as examples. Based on the above ana­
Table 3
lyses, the top 3 analog years for 2018 and 2019 are selected according to
Planting season soil moisture level in 2018 and 2019.
the bias corrected IMERG precipitation from January to May in each site
2018 2019 (Fig. 6). Planting season soil moisture in 2018 and 2019 is each classified
Dangishta Extreme Wet Normal using the CDF matching method (Table 3). At most of our sites, the
Kudmi Extreme Wet Normal majority of the CDF curve for the planting season soil moisture falls in
Reem Normal Dry
the Extreme Wet range in 2018 and Normal range in 2019.
Gaita Extreme Wet Normal
Based on the constructed historical maize yield series for each

6
M. Yang et al. Agricultural and Forest Meteorology 331 (2023) 109347

Table 4 low at Dangishta (Table 4). It can also be noticed that the overall analog
Analog forecast maize yield and its comparison with observations in 2018 and forecast in 2019 is less accurate than 2018, particularly in Gaita where
2019 (unit: kg/ha). Yields in the three analog years are listed in parenthesis. the yield was significantly overestimated. Such an overestimation is
Dangishta Kudmi Reem Gaita primarily due to the flooding in the harvest season that caused extensive
2018 Fcts 3388 - Normal 2547 - 2673 - 4172 - High
crop damage in this region reported by the local community.
(3285, 3303, Normal Normal (3772, 4199, As the planting time often varies from year to year, we employ an
3576) (2432, 2593, (2548, 2555, 4546) automatic planting scheme in determining the planting time in this
2617) 2915) study. DSSAT determines the planting time in a prescribed planting
Obs 3697 - Normal 2851 - 2851- 3697-
month according to soil moisture conditions. Usually, it leads to a
Normal Normal Normal
2019 Fcts 2806 - Very 2488 - 2603 - 3759 - planting time in the early or middle of May for the sites of interest in this
low Normal Normal Normal study. To examine the impact of planting time on our forecast frame­
(2616, 2666, (2381, 2416, (2302, 2734, (3589, 3697, work, we specify three planting windows – early planting, usual
3135) 2666) 2773) 3990) planting, and late planting for maize according to our field survey and
Obs 2396 - Very 2668 - 2668 - 2396 - Very
low Normal Normal low
performed a sensitivity test (Table 5 using the forecast for 2019 as an
example). Results from these tests indicate that the sensitivity of forecast
maize yield to planting time is rather low. At most sites, the forecast falls
in the same or adjacent category under different planting times, which
Table 5
indicates the forecast with automatic planting still holds for planting
Forecast maize yield categories based on different planting time (using year
time that differs from farmers’ choices.
2019 as an example).
Early Usual Late
May 9 – May 18 May 19 – May 30 June 1 – June 12 3.4. Forecast verification
Dangishta Very low Very low Low
Gaita Low Normal Normal To verify the forecast, we compared the analog forecast results with
Kudmi Normal Normal Normal ground observation from Ethiopia’s Agricultural Sample Survey (AgSS)
Reem Normal Normal High from 2007 to 2019 (CSA, 2019). Fig. 7 depicts the comparison between
analog forecast and observation at the four selected sites.
Table 6 shows that Dangishta has the highest R2 value (0.60) while
scenario of planting season soil moisture, and the corresponding analog
Gaita has the lowest (0.36) (correlation is significant at 0.05 level for all
years determined according to preseason precipitation, we forecast the
maize yields for each site in 2018 and 2019. In addition to providing a
deterministic or numeric prediction, we classify yields into five different Table 6
Metrics for analog forecast verification at the four sites.
categories according to Table 2 to account for prediction uncertainty.
Based on the analog scheme, our forecast shows in 2018, maize yield Dangishta Kudmi Reem Gaita
will be normal at Dangishta, Kudmi, and Reem, while high at Gaita; in R2 0.60 0.40 0.51 0.36
2019, maize yield will be normal at Kudmi, Reem, and Gaita, while very RMSE 372.0 325.7 310.8 495.4

Fig. 7. Maize yield time series comparison between analog forecast and observation at the four selected sites.

7
M. Yang et al. Agricultural and Forest Meteorology 331 (2023) 109347

Table 7 forecast yield categorically following the same classification criteria in


Number of years (out of 13) with different levels of maize yield classification Table 1 and compare the maize yield classification in each year between
agreement between analog forecast and observation at the four sites: D0 for analog forecast and observation in the past 13 years for which obser­
exact match, D1 when forecast and observation differ by one category (e.g., very vational data is available. Table 7 summaries the performance of the
low vs. low, or normal vs. high), and D2 when forecast and observation differ by model in capturing the right category of crop yield, using the number of
two or more categories (e.g., normal vs. very high).
years when the analog-based forecast and observation agree (D0), when
Dangishta Kudmi Reem Gaita they differ by one category (D1), or by two or more categories (D2).
D0 9 9 6 10 Among the four sites, Gaita has the greatest number of years (10 out of
D1 4 3 6 2 13 years) with exact agreement between forecast and observation while
D2 0 1 1 1 Reem has the smallest number of years (6 out of 13 years). Dangishta is
the only site with zero D2 value while other sites each has one year with
sites). Reem has the smallest RMSE value (310.8 kg/ha) while Gaita has D2 performance. Overall, the analog-based model achieves considerable
the biggest RMSE (495.4 kg/ha). The overall analog skills for Dangishta, forecast skills at the four selected sites.
Kudmi, Reem and Gaita respectively are 0.775, 0.633, 0.713 and 0.603.
In addition to quantitative or numeric comparison, we also express the

Fig. 8. Classification of Agroecological Zones (AEZs) and the location of selected woredas.

Fig. 9. Maize yield comparison between analog forecast and observation at different woredas.

8
M. Yang et al. Agricultural and Forest Meteorology 331 (2023) 109347

Table 8 0.670. Similar to Table 7, Table 9 depicts that AEZ4 woreda has the
Metrics for analog forecast verification at the four woredas in different AEZs. greatest number of years (8 out of 13 years) with exact agreement be­
AEZ3 AEZ4 AEZ5 AEZ7 tween forecast and observation while AEZ7 woreda has the least years
2 with exact match (4 out of 13 years). AEZ5 woreda is the only site with
R 0.50 0.51 0.48 0.45
RMSE 369.4 393.9 274.1 328.5 no D2 prediction while other sites each has one year when the forecast
and observation differ by more than one category. It can also be noticed
that this approach performs better in relatively dry places than wet
places, since precipitation and soil moisture used as indices in this study
Table 9
Number of years (out of 13) with different levels of maize yield classification
have stronger influences on crop growth in dry regions. Results indicate
agreement between analog forecast and observation at the four woredas in that although the performance of our analog approach decreases when
different AEZs. applied to locations out of the four sites, it is still acceptable as most of
the disagreements are within one category.
AEZ3 AEZ4 AEZ5 AEZ7
To further justify the choice of the analog model, we developed a
D0 6 8 5 4 multi-variable linear regression model following the conventional
D1 6 4 8 8
D2 1 1 0 1
approach by taking the four climate variables during the growing season
(May to September) as predictors, and compare it with the analog
approach (Fig. 10). Given our data availability, the climate data used for
3.5. Application in different agroecological zones yield forecasting here for 2007–2017 is historical observations, and for
2018 and 2019 is bias corrected CFS forecast. The analog model out­
The Upper Blue Nile Basin (UBNB) spans five different Agroecolog­ performs the linear regression model in seasonal yield forecast because
ical Zones (AEZ3–7) categorized by elevation, climate divisions and of three main reasons. First, the analog model uses observed preseason
water availability (HarvestChoice, 2011), while the four selected sites precipitation as index without relying on seasonal climate forecast, and
are located in AEZ6 (Fig. 8). To further verify the applicability of our seasonal forecast is subject to especially large uncertainties at the site
analog approach under different agroecological conditions, we selected level. Second, as shown in Fig. 4, compared with growing season pre­
four woredas (the third level administrative division in Ethiopia) with cipitation, the preseason precipitation has stronger correlation with
available maize yield observational data located in different AEZs (AEZ yield, which can better serve as forecast index. Third, yield forecast with
3, 4, 5 and 7) in the UBNB. Using the same procedures as above, we analog model is based on the ensemble of three analog years, which can
forecasted maize yield at the four selected woredas and compared it with reduce the uncertainties from climate in a single year.
observations. Fig. 9 illustrates the comparison between analog forecast
and observation at the four selected woredas. Table 8 shows that AEZ4 4. Conclusion
woreda has the highest R2 value (0.51) while AEZ7 woreda has the
lowest (0.45) (correlation is significant at 0.05 level for all woredas). In this study, an analog approach was developed for site-scale sea­
AEZ5 woreda has the smallest RMSE value (274.1 kg/ha) while AEZ4 sonal crop yield prediction and applied to four farming communities in
woreda has the biggest RMSE (393.9 kg/ha). The overall analog skills for the Lake Tana basin of Ethiopia. To address data limitations, we
the woredas in AEZ3, 4, 5 and 7 respectively are 0.708, 0.711, 0.695 and reconstructed the crop yield series from 1979 to 2014 using a process-

Fig. 10. Comparison between the performance of analog approach and linear regression model in yield forecast.

9
M. Yang et al. Agricultural and Forest Meteorology 331 (2023) 109347

based crop model, DSSAT. Among different climate variables, precipi­ References
tation during the pre-season (January-May) was identified as the most
highly correlated climate variable influencing crop yield. By imple­ Ahmed, K.F., Wang, G., Silander, J., Wilson, A.M., Allen, J.M., Horton, R., Anyah, R.,
2013. Statistical downscaling and bias correction of climate model outputs for
menting preseason precipitation as the analog index, a fairly promising climate change impact assessment in the U.S. northeast. Glob. Planet. Change 100,
skill was obtained for the seasonal forecast of maize yield. Accounting 320–332. https://doi.org/10.1016/j.gloplacha.2012.11.003.
for the local hydrogeologic properties, the planting season soil moisture Basso, B., Cammarano, D., Carfagna, E., 2013. Review of Crop Yield Forecasting Methods
and Early Warning Systems. First Meet. Sci. Advis. Comm. Glob. Strateg. Improv.
was simulated, and the historical crop yield database was developed Agric. Rural Stat. 1–56. https://doi.org/10.1017/CBO9781107415324.004.
based on different prescribed soil moisture levels. With the proposed Beck, H.E., Van Dijk, A.I.J.M., Levizzani, V., Schellekens, J., Miralles, D.G., Martens, B.,
analog approach and database and based on pre-season precipitation, De Roo, A., 2017. MSWEP: 3-hourly 0.25◦ global gridded precipitation (1979-2015)
by merging gauge, satellite, and reanalysis data. Hydrol. Earth Syst. Sci. 21,
we forecast maize yield at four sites in 2018 and 2019 and verified our 589–615. https://doi.org/10.5194/hess-21-589-2017.
forecast with ground observations. The modeling approach was then Bergen, R.E., Harnack, R.P., 1982. Long-Range Temperature Prediction Using a Simple
applied to different Agroecological Zones to test its applicability under Analog Approach. Mon. Weather Rev. 110, 1083–1099. https://doi.org/10.1175/
1520-0493(1982)110<1083:LRTPUA>2.0.CO;2.
diverse environments. This study offers a simple but reasonably accurate
Bessafi, M., Lasserre-Bigorry, A., Neumann, C.J., Pignolet-Tardan, F., Payet, D., Lee-
crop yield forecast approach that could help guide farmers’ practices in Ching-Ken, M., 2002. Statistical Prediction of Tropical Cyclone Motion: an
mitigating meteorologically-induced crop losses, if appropriately Analog–CLIPER Approach. Weather Forecast 17, 821–831. https://doi.org/10.1175/
communicated for integration into decision-making. 1520-0434(2002)017<0821:SPOTCM>2.0.CO;2.
Block, P.J., Strzepek, K., Rosegrant, M.W., Diao, X., 2008. Impacts of considering climate
The analog approach developed in this study has multiple advan­ variability on investment decisions in Ethiopia. Agric. Econ. 39, 171–181. https://
tages. First, the approach exhibits high accuracy for both the four doi.org/10.1111/j.1574-0862.2008.00322.x.
selected sites and the woredas in different AEZs. Second, the approach Bolton, D.K., Friedl, M.A., 2013. Forecasting crop yield using remotely sensed vegetation
indices and crop phenology metrics. Agric. For. Meteorol. 173, 74–84. https://doi.
demonstrates simplicity by using historical data rather than complex org/10.1016/j.agrformet.2013.01.007.
process-based models; the constructed historical crop yield database Chlingaryan, A., Sukkarieh, S., Whelan, B., 2018. Machine learning approaches for crop
allows future applications to pull out the analog years based on pre- yield prediction and nitrogen status estimation in precision agriculture: a review.
Comput. Electron. Agric. https://doi.org/10.1016/j.compag.2018.05.012.
season precipitation and generate a crop yield forecast with a long CSA, 2019. Agricultural Sample Survey. Stat. Bull.
lead time. Third, the approach requires minimal data input, using only De Jager, J.M., Potgieter, A.B., Van Den Berg, W.J., 1998. Framework for Forecasting the
pre-season precipitation and soil moisture to forecast crop yields before Extent and Severity of Drought in Maize in the Free State Province of South Africa.
Agric. Syst.
the planting season. Dee, D.P., Uppala, S.M., Simmons, A.J., Berrisford, P., Poli, P., Kobayashi, S., Andrae, U.,
The approach developed here is not without limitations. For Balmaseda, M.A., Balsamo, G., Bauer, P., Bechtold, P., Beljaars, A.C.M., van de
example, finding precipitation analog years from historical data assumes Berg, L., Bidlot, J., Bormann, N., Delsol, C., Dragani, R., Fuentes, M., Geer, A.J.,
Haimberger, L., Healy, S.B., Hersbach, H., Hólm, E.V., Isaksen, L., Kållberg, P.,
climate stationarity, which limits the application of the analog approach
Köhler, M., Matricardi, M., Mcnally, A.P., Monge-Sanz, B.M., Morcrette, J.J., Park, B.
in future climate conditions. Nevertheless, this limitation can be miti­ K., Peubey, C., de Rosnay, P., Tavolato, C., Thépaut, J.N., Vitart, F., 2011. The ERA-
gated by updating the historical database with data from more recent Interim reanalysis: configuration and performance of the data assimilation system.
years. Continuous development of agricultural technologies improves Q. J. R. Meteorol. Soc. 137, 553–597. https://doi.org/10.1002/qj.828.
Filippi, P., Jones, E.J., Wimalathunge, N.S., Somarathna, P.D.S.N., Pozza, L.E., Ugbaje, S.
crop yields, thus future crop yields may not be consistent with past U., Jephcott, T.G., Paterson, S.E., Whelan, B.M., Bishop, T.F.A., 2019. An approach
yields even if the local climate were to remain stationary. It is possible to forecast grain crop yield using multi-layered, multi-farm data sets and machine
that the performance skill of the analog forecast approach would be learning. Precis. Agric. 20, 1015–1029. https://doi.org/10.1007/s11119-018-
09628-4.
reduced with significant technology improvement. This approach also FSIN, 2019. Global Report on Food Crises.
has difficulty in anticipating the impacts caused by extreme events such GDAS, n.d. “Global Data Assimilation System”, Retrieved from https://www.ncdc.noaa.
as the flash flood in 2019 that took place near the end of the growing gov/data-access/model-data/model-datasets/global-data-assimilation-system-gdas
[WWW Document].
season and destroyed yield. This limitation could be mitigated by the Getachew, B., Manjunatha, B.R., 2022. Potential climate change impact assessment on
enhancement of seasonal climate forecasts. Alternative methods in crop the hydrology of the Lake Tana Basin, Upper Blue Nile River Basin, Ethiopia. Phys.
yield forecast include using within-season remote sensing products that Chem. Earth 127, 103162. https://doi.org/10.1016/j.pce.2022.103162.
Haider, M.R., Peña, M., Lazin, R., Khadim, F.K., Yang, M., Dokou, Z., Nikolopoulos, E.,
reflect crops’ real-time condition to assist yield forecast in addition to Wang, G., Anagnostou, E., 2020. Enabling Water and Agriculture Management in the
adjusting parameters in statistical and process-based crop models. Upper Blue Nile Basin through Numerical Seasonal Forecasts and High-Resolution
Sectoral Models. Clim. Predict. S&T Dig. 1, 154–159. https://doi.org/10.25923/
t4qa-ae63.
Declaration of Competing Interest
Hansen, J.W., Indeje, M., 2004. Linking dynamic seasonal climate forecasts with crop
simulation for maize yield prediction in semi-arid Kenya. Agric. For. Meteorol. 125,
The authors declare that they have no known competing financial 143–157. https://doi.org/10.1016/j.agrformet.2004.02.006.
HarvestChoice, 2011. AEZ Tropical (8-class). International Food Policy Research
interests or personal relationships that could have appeared to influence
Institute, Washington, DC., and University of Minnesota, St. Paul, MN. Available
the work reported in this paper. online at http://harvestchoice.org/node/4997.
Hengl, T., de Jesus, J.M., MacMillan, R.A., Batjes, N.H., Heuvelink, G.B.M., Ribeiro, E.,
Data availability Samuel-Rosa, A., Kempen, B., Leenaars, J.G.B., Walsh, M.G., Gonzalez, M.R., 2014.
SoilGrids1km — Global Soil Information Based on Automated Mapping. PLoS ONE 9,
e105992. https://doi.org/10.1371/journal.pone.0105992.
Data will be made available on request. Hoogenboom, G., Jones, J.W., Wilkens, P.W., Porter, C.H., Boote, K.J., Hunt, L.A.,
Singh, U., Lizaso, J.L., White, J.W., Uryasev, O., Royce, F.S., Ogoshi, R., Gijsman, A.
J., Tsuji, G.Y., Koo, J., 2012. Decision Support System for Agrotechnology Transfer
(DSSAT) Version 4.5 [CD-ROM]. University of Hawaii, Honolulu, Hawaii.
Acknowledgements Huffman, G., Bolvin, D., Braithwaite, D., Hsu, K., Joyce, R., Xie, P., 2014. Integrated
Multi-satellitE Retrievals for GPM (IMERG), version 4.4 [WWW Document]. NASA’s
Precip. Process. Center, accessed 31 March 2015, ftp//arthurhou.pps.eosdis.nasa.
This paper is based upon work supported by the National Science gov/gpmdata/. URL https://gpm.nasa.gov/data-access/citations (accessed 4.27.20).
Foundation [grant number 1545874]. The authors thank three anony­ International Research Institute for Climate and Society (IRI), Michigan State University
mous reviewers for their constructive comments towards an earlier (MSU), HarvestChoice (IFPRI), 2015. Global High-Resolution Soil Profile Database
for Crop Modeling Applications [WWW Document]. Harvard Dataverse, V2, Harvard
version of this paper. Dataverse, V2. https://doi.org/https://doi.org/10.7910/DVN/1PEEY0.
Jones, J.W., Antle, J.M., Basso, B., Boote, K.J., Conant, R.T., Foster, I., Godfray, H.C.J.,
Supplementary materials Herrero, M., Howitt, R.E., Janssen, S., Keating, B.A., Munoz-Carpena, R., Porter, C.
H., Rosenzweig, C., Wheeler, T.R., 2017. Toward a new generation of agricultural
system data, models, and knowledge products: state of agricultural systems science.
Supplementary material associated with this article can be found, in Agric. Syst. 155, 269–288. https://doi.org/10.1016/J.AGSY.2016.09.021.
the online version, at doi:10.1016/j.agrformet.2023.109347.

10
M. Yang et al. Agricultural and Forest Meteorology 331 (2023) 109347

Jones, J.W., Hoogenboom, G., Porter, C.H., Boote, K.J., Batchelor, W.D., Hunt, L.A., Passarelli, S., Mekonnen, D., Bryan, E., Ringler, C., 2018. Evaluating the pathways from
Wilkens, P.W., Singh, U., Gijsman, A.J., Ritchie, J.T., 2003. The DSSAT cropping small-scale irrigation to dietary diversity: evidence from Ethiopia and Tanzania.
system model. European Journal of Agronomy. Elsevier, pp. 235–265. https://doi. Food Secur. 10, 981–997. https://doi.org/10.1007/s12571-018-0812-5.
org/10.1016/S1161-0301(02)00107-7. Saha, S., Moorthi, S., Pan, H.L., Wu, X., Wang, Jiande, Nadiga, S., Tripp, P., Kistler, R.,
Kang, Y., Khan, S., Ma, X., 2009. Climate change impacts on crop yield, crop water Woollen, J., Behringer, D., Liu, H., Stokes, D., Grumbine, R., Gayno, G., Wang, Jun,
productivity and food security - A review. Prog. Nat. Sci. https://doi.org/10.1016/j. Hou, Y.T., Chuang, H.Y., Juang, H.M.H., Sela, J., Iredell, M., Treadon, R., Kleist, D.,
pnsc.2009.08.001. Van Delst, P., Keyser, D., Derber, J., Ek, M., Meng, J., Wei, H., Yang, R., Lord, S., Van
Lala, J., Tilahun, S., Block, P., 2020. Predicting rainy season onset in the Ethiopian Den Dool, H., Kumar, A., Wang, W., Long, C., Chelliah, M., Xue, Y., Huang, B.,
Highlands for agricultural planning. J. Hydrometeorol. https://doi.org/10.1175/ Schemm, J.K., Ebisuzaki, W., Lin, R., Xie, P., Chen, M., Zhou, S., Higgins, W., Zou, C.
JHM-D-20-0058.1. Z., Liu, Q., Chen, Y., Han, Y., Cucurull, L., Reynolds, R.W., Rutledge, G.,
Lazin, R., Shen, X., Koukoula, M., Anagnostou, E., 2020. Evaluation of the Hyper- Goldberg, M., 2010. The NCEP climate forecast system reanalysis. Bull. Am.
Resolution Model-Derived Water Cycle Components Over the Upper Blue Nile Basin. Meteorol. Soc. 91, 1015–1057. https://doi.org/10.1175/2010BAMS3001.1.
J. Hydrol. 590, 125231 https://doi.org/10.1016/j.jhydrol.2020.125231. Saha, S., Moorthi, S., Wu, X., Wang, J., Nadiga, S., Tripp, P., Behringer, D., Hou, Y.T.,
Lecerf, R., Ceglar, A., López-Lozano, R., Van Der Velde, M., Baruth, B., 2019. Assessing Chuang, H.Y., Iredell, M., Ek, M., Meng, J., Yang, R., Mendez, M.P., Van Den
the information in crop model and meteorological indicators to forecast crop yield Dool, H., Zhang, Q., Wang, W., Chen, M., Becker, E., 2014. The NCEP climate
over Europe. Agric. Syst. 168, 191–202. https://doi.org/10.1016/j. forecast system version 2. J. Clim. 27, 2185–2208. https://doi.org/10.1175/JCLI-D-
agsy.2018.03.002. 12-00823.1.
Lewis, K., 2017. Understanding climate as a driver of food insecurity in Ethiopia. Clim. Shen, X., Anagnostou, E.N., 2017. A framework to improve hyper-resolution hydrological
Change 144, 317–328. https://doi.org/10.1007/s10584-017-2036-7. simulation in snow-affected regions. J. Hydrol. 552, 1–12. https://doi.org/10.1016/
Lobell, D.B., Cassman, K.G., Field, C.B., 2009. Crop Yield Gaps: their Importance, J.JHYDROL.2017.05.048.
Magnitudes, and Causes. https://doi.org/10.1146/annurev.environ.041008.093740. Tahir, Z.ul R., Asim, M., Azhar, M., Moeenuddin, G., Farooq, M., 2021. Correcting solar
Macdonald, R.B., Hall, F.G., 1980. Global Crop Forecasting. Science (80-.) 208, 670–679. radiation from reanalysis and analysis datasets with systematic and seasonal
MacLeod, D., 2019. Seasonal forecasts of the East African long rains: insight from variations. Case Stud. Therm. Eng. 25, 100933 https://doi.org/10.1016/J.
atmospheric relaxation experiments. Clim. Dyn. 53, 4505–4520. https://doi.org/ CSITE.2021.100933.
10.1007/s00382-019-04800-6. Teutschbein, C., Seibert, J., 2012. Bias correction of regional climate model simulations
Mcdermott, P.L., Wikle, C.K., 2016. A model-based approach for analog spatio-temporal for hydrological climate-change impact studies: review and evaluation of different
dynamic forecasting. Environmetrics 27, 70–82. https://doi.org/10.1002/env.2374. methods. J. Hydrol. 456–457, 12–29. https://doi.org/10.1016/J.
Mellander, P.-.E., Gebrehiwot, S.G., Gärdenäs, A.I., Bewket, W., Bishop, K., 2013. JHYDROL.2012.05.052.
Summer Rains and Dry Seasons in the Upper Blue Nile Basin: the Predictability of Welteji, D., 2018. A critical review of rural development policy of Ethiopia: access,
Half a Century of Past and Future Spatiotemporal Patterns. PLoS ONE 8, e68461. utilization and coverage. Agric. Food Secur. https://doi.org/10.1186/s40066-018-
https://doi.org/10.1371/journal.pone.0068461. 0208-y.
Mjelde, J.W., Keplinger, K., 1998. Using the Southern Oscillation to forecast Texas winter Yang, M., Wang, G., Ahmed, K.F., Adugna, B., Eggen, M., Atsbeha, E., You, L., Koo, J.,
wheat and sorghum crop yields. J. Clim. 11, 54–60. https://doi.org/10.1175/1520- Anagnostou, E., 2020. The role of climate in the trend and variability of Ethiopia’s
0442(1998)011<0054:UTSOTF>2.0.CO;2. cereal crop yields. Sci. Total Environ., 137893 https://doi.org/10.1016/j.
Monache, L.D., Anthony Eckel, F., Rife, D.L., Nagarajan, B., Searight, K., 2013. scitotenv.2020.137893.
Probabilistic weather prediction with an analog ensemble. Mon. Weather Rev. 141, Yang, M., Wang, G., Lazin, R., Shen, X., Anagnostou, E., 2021. Impact of planting time
3498–3516. https://doi.org/10.1175/MWR-D-12-00281.1. soil moisture on cereal crop yield in the Upper Blue Nile Basin: a novel insight
Panjwani, S., Naresh Kumar, S., Ahuja, L., 2021. Bias Correction of GCM Data Using towards agricultural water management. Agric. Water Manag. 243, 106430 https://
Quantile Mapping Technique. In: Proceedings of International Conference on doi.org/10.1016/j.agwat.2020.106430.
Communication and Computational Technologies. Singapore. Springer,
pp. 617–621. https://doi.org/10.1007/978-981-15-5077-5_55.

11

You might also like