Scribd
Upload
45 views21 pages

Integrating Top-Down and Bottom-Up Modelling Approaches: David Post CSIRO Land and Water, Townsville

The document discusses integrating top-down and bottom-up modelling approaches to reduce predictive uncertainty in hydrological models. It describes applying a physically-based bottom-up model (LISEM) to a monitored hillslope to derive parameters for a simpler top-down conceptual model. This allowed applying the conceptual model across an entire catchment to predict travel times and sediment delivery. The study demonstrates how bottom-up and top-down models can complement each other to provide robust predictions at the catchment scale.

Uploaded by

SudharsananPRS
Copyright
© Attribution Non-Commercial (BY-NC)
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
45 views21 pages

Integrating Top-Down and Bottom-Up Modelling Approaches: David Post CSIRO Land and Water, Townsville

The document discusses integrating top-down and bottom-up modelling approaches to reduce predictive uncertainty in hydrological models. It describes applying a physically-based bottom-up model (LISEM) to a monitored hillslope to derive parameters for a simpler top-down conceptual model. This allowed applying the conceptual model across an entire catchment to predict travel times and sediment delivery. The study demonstrates how bottom-up and top-down models can complement each other to provide robust predictions at the catchment scale.

Uploaded by

SudharsananPRS
Copyright
© Attribution Non-Commercial (BY-NC)
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/ 21

Integrating top-down and bottom-up modelling approaches

David Post CSIRO Land and Water, Townsville USA PUB Workshop 16-19 October 2006
www.csiro.au

The top-down working group (Croke, Littlewood, Post)


Sits in Theme 1 catchment classification although feeds heavily into Theme 6 new model approaches.
The overall aim of the TDWG mirrors that of PUB : A focus on the reduction of predictive uncertainty. To achieve this aim, we intend to use the top-down approach. A variety of techniques may be useful, as long as the modeller is prepared to consider and tackle the sources of uncertainty in their model predictions.

Website and newsletter: http://www.stars.net.au/tdwg/


Special sessions are held at the MODSIM (2005, 2007) and iEMSs (2004, 2006, 2008) conferences.

What is the top-down approach?

It is an empirical or data based approach, involving learning about a catchments hydrologic functioning from patterns in the observed data.
Sivapalan et al. Hydrol. Proc. 17: 2099 (2003).

Importantly, it is not a rigid methodology, but rather represents a generic framework.


Sivapalan et al. Hydrol. Proc. 17 : 2101-2111 (2003).

Top-down models
Have these characteristics:
Model structure is derived based only on available data. New processes are only included where the available data supports the identification and parameterisation of these processes. Tend to be lumped conceptual models. Are parametrically parsimonious models (models that have as few parameters as possible in order to reproduce only the dominant response characteristics).

Pros
Fewer parameters leads to those parameters being more readily identifiable.

Cons
May not capture all of the important processes occurring in a catchment.

Bottom-up models
Have these characteristics:
Model structure is defined a-priori by the modeller. New processes can be included if the modeller thinks that they are important. Tend to be physics-based models. Are usually parametrically generous (models that have many parameters in order to explicitly reproduce the processes occurring in a catchment).

Pros
More process-based approach.

Cons
Problems with parameter identifiability and equifinality (many parameter sets lead to similar model predictions).

Top-down vs bottom-up models


Most models are a mixture of approaches:
Lumped conceptual top-down models may make a-priori assumptions about streamflow behaviour (such as a division into quick and slowflow processes. Physics-based bottom-up models are based on some data, such as that macropore flow is an important process in a particular catchment.

Both approaches have problems with regionalisation:


Top-down models may provide an adequate fit to the observed hydrograph but have no physical meaning. Bottom-up models may describe hydrologic response at the bottom of a hillslope but cannot be scaled up to provide a catchment response.

Both problems may produce model parameters which have little or no relationship to catchment attributes.

Theme 6 new model approaches (Post, Croke, Pomeroy)


Create a new suite of hydrologic models which are physicallybased, parametrically efficient, and applicable to catchments where we have little or no hydrologic data.
Requires a merging of top-down and bottom-up approaches.

How to do this?
Use the outputs from a bottom-up model to constrain the range of parameter values in a top-down model (or vice-versa). Run numerous models in order to produce an optimal model fit (Reichl, 2006). Use process-based understanding gained from bottom-up models to modify the representation of processes in top-down models (Post, 2006).

A new type of model?

Study site

Burdekin catchment

Weany Creek (14 km2)

Application of a parametrically-parsimonious (top-down) model (Ferro) Travel time, ti for each pixel is given by:

di ti = vi
where di is hydraulic distance to stream and vi is the velocity of water:

vi = a i S i
where Si is slope of the flowpath to stream and ai is a coefficient related to landuse

Application of a parametrically-generous (bottom-up) model (LISEM)


LISEM (the Limburg Soil Erosion Model), is a physically based hydrological and soil erosion model which routes water and sediment over terrain surfaces using a gridbased routing scheme.
Patch Type Ksat (mm/h) Psi (cm) Init. moisture content
(cm3/cm3)

B 20 15 0.04 0.33 0.17 0.8 34%

C 2 20 0.1 0.32 0.1 0.47 57%

D 1 30 0.06 0.16 0.03 0.02 9%

Random Roughness
(StdDev of relief cm)

Mannings (n) %Cover


(vegetation and litter)

% of flume

LISEM calibration and simulation


100 90 80 70 60 50 40 30 20 10 0 21:00 22/01/05 09:00 23/01/05 21:00 23/01/05 09:00 24/01/05 0 20 40 60 80 100 120 140 21:00 24/01/05 Rainfall (mm/hr)

Discharge (L/s)

Observed discharge = 96 m3 Modelled discharge = 102 m3 r2 = 0.91 Simulation hydrograph


0 50 100 150 200 250 300 09:00 13/01/04 21:00 13/01/04 09:00 14/01/04 Rainfall (mm/hr)

Modelled

Observed

Rainfall

Observed discharge = 376 m3 Modelled discharge = 392 m3 r2 = 0.85

Discharge (L/s)

Calibration hydrograph

100 90 80 70 60 50 40 30 20 10 0 21:00 12/01/04

Modelled

Observed

Rainfall

Comparison of LISEM and Ferro model results


4000

3000

y = 4.0526x
LISEM Travel Times (Seconds) 2000

1000

0 0 200 400 Ferro Travel Time (seconds) 600 800

Our landuse co-efficient, a is too small

Observed travel times

Flume 1

Flume 2

Flume 3

Comparison with observed travel times


0.8 Cross correlation (observed data) 0.7 0.6 35 0.5 0.4 0.3 0.2 10 0.1 0 0 500 1000 1500 2000 Travel time (s) Observed LISEM model Ferro model (corrected) 2500 3000 3500 5 0 4000 30 25 20 15 50 45 Frequency (modelled data) 40

Application at whole-of-catchment scale Travel times

Relating sediment delivery to travel time

Sediment delivery from a pixel on the hillslope is given by:

HSDRi = e

( t i )

where and are parameters and ti is travel time.

Deriving the values of and


1.0 0.9 0.8 0.7 0.6 QDR 0.5 0.4 0.3 0.2 0.1 0.0 0 500 1000 1500 2000 2500 3000 3500 Travel time, t (seconds) QDR = e-0.002t

= 0.002

is derived by assuming that sediment delivery decreases with travel time in the same way that discharge delivery does;

is derived such that total modelled sediment delivery equals total observed sediment delivery for the catchment.

Total delivery of sediment = 275 t (from hillslope flumes) Total modelled erosion = 7577 t (from RUSLE) = 0.1

Sediment delivery at the whole-of-catchment scale

HSDR

Conclusions
This study represents one way in which a bottom-up model can be used in conjunction with a top-down model to the benefit of both:
A relatively complex, physics-based model, LISEM was applied to a monitored hillslope to help parameterise a relatively simple algorithm which could then be applied across the whole of the catchment.

On its own, the bottom-up model is too computationally expensive to be applied across the whole of the catchment; Conversely, on its own, the top-down model does not have sufficient process representation for us to be able to derive the value of . Together, the approaches have led to a simple, yet robust approach which can be applied across an entire catchment.

Where do we put our efforts?


Improving model performance
This study. Australian Hydrological Modelling Initiative (Chiew et al, CSIRO). E2 (Catchment Modelling Toolkit).

Improved catchment characterisation


***Rainfall*** Defining the minimum dataset required to characterise catchment hydrologic response. Improved data collection techniques (remote sensing, sensor networks, the Water Resources Observation Network, CSIRO).

Improved regionalisation procedures


PCD/HRC relationships. PCD/model parameter relationships.

A proposal
The Regionalisation Inter-Comparison Experiment (RICE)
Provide participants with everything they need to predict the hydrologic response of a catchment (except streamflow data to calibrate against!). Needs to be a comparison of techniques, not just my R2 is bigger than your R2! How to do it? Data for a number of catchments or just one? Predict daily streamflow / FDC / MAQ? Details, details

Are we ready for this???

You might also like