Adapting parameterization

21
Dynamic averaging of rainfall-runoff model simulations within non stationary climate conditions Nicolas Le Moine & Ludovic Oudin Univ. Paris 6 1 IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

description

Dynamic averaging of rainfall-runoff model simulations within non stationary climate conditions Nicolas Le Moine & Ludovic Oudin Univ. Paris 6. Coping with non stationary behaviors: models with more constraints (and robustness) or more freedom (and flexibility)?. Adapting parameterization - PowerPoint PPT Presentation

Transcript of Adapting parameterization

Page 1: Adapting parameterization

Dynamic averaging of rainfall-runoff model simulations within non stationary climate

conditions

Nicolas Le Moine & Ludovic Oudin

Univ. Paris 6

1IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Page 2: Adapting parameterization

Coping with non stationary behaviors: models withmore constraints (and robustness) or more freedom (and flexibility)?

Adapting parameterization Flexibility: Dynamic recalibration with climate analogs (de

Vos et al., 2010). Robustness: Constraining model parameter with multi-

objective approach (with e.g. more weights on bias criterion)

Adapting model structure Flexibility: Multi-model approach Robustness: Choice of a fixed model structure that is

relevant for more arid catchments and/or that is efficient when performing DSST

2IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Page 3: Adapting parameterization

Reconciling robustness and flexibility

Multi-model / Dynamic averaging / fuzzy comittee : A good idea involving arbitrary choices

Complementary objective functions for calibrating individually the models

A weighting function to average the simulated flows from the models

Is there a way to reduce the number of arbitrary choices?

3IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Page 4: Adapting parameterization

Data and models

3 catchments with non-stationnary climate:

Axe Creek Gilbert Bani

One daily conceptual model: GR4J

4IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Page 5: Adapting parameterization

Rainfall-Runoff Model

P PE

Methodology: Identifying long-term shifts of the hydric state of a catchment through modelling

5IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Page 6: Adapting parameterization

Methodology: Identifying long-term shifts of the hydric state of a catchment through modelling

6IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Mean of the period

Low frequency signal

Page 7: Adapting parameterization

Methodology: Designing a weighting function

7IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Page 8: Adapting parameterization

8IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Methodology: Designing a weighting function

Page 9: Adapting parameterization

Methodology: Designing a weighting function

9IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Page 10: Adapting parameterization

Methodology: Designing a weighting function

10IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Prob. of non exceedance of Low Freq. anomaly

Page 11: Adapting parameterization

Methodology: Designing a weighting function

11IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Prob. of non exceedance of Low Freq. anomaly

Page 12: Adapting parameterization

Methodology: Designing a weighting function

12IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Prob. of non exceedance of Low Freq. anomaly

Page 13: Adapting parameterization

Methodology: Calibrating bi-polar models

13IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Page 14: Adapting parameterization

Methodology: Using Bi-polar models in validation

14

Page 15: Adapting parameterization

Detailed Results on Axe Creek: calibration period 1

15

Page 16: Adapting parameterization

16

Detailed Results on Axe Creek: validation period 4

Page 17: Adapting parameterization

Comparative results for Bias

17IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Gilbert River Axe Creek

Page 18: Adapting parameterization

Comparative results for Bias

18IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Bani River

Page 19: Adapting parameterization

Comparative results for KGE

19IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Gilbert River Axe Creek

Page 20: Adapting parameterization

Comparative results for KGE

20IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

Bani River

Page 21: Adapting parameterization

Conclusion

21IAHS Joint Assembly Gothenburg. Hw15 Testing simulation and forecasting models in non-stationary conditions

A methodology focused on long-term variability Robustness: each pole has a behavioural parameter set that

works by itself Flexibility: The weights may vary largely on a subperiod but

smoothly in time

Need to test other settings Assessing the methodology on stationary catchments Effect of time series length Objective functions