Présentations Colloques

Oral Presentation
Session 6.03: Innovative tools to improve understanding of groundwater recharge processes
Moeck Christian
Crash test for groundwater recharge models- The effects of model complexity and calibration period on groundwater recharge predictions
Reliable groundwater recharge predictions are indispensable. However, the choice of what recharge model to use often seems to be subjective. Furthermore, an implicit assumption is made that model parameters calibrated over historical periods are also valid for the predictions. To the best of our knowledge, there has not yet been a systematic investigation of the effects of the calibration strategy on the performance of recharge models. In addition, the missing link between the mechanics of the model structure and changing climatic forcing functions is still unclear. In our study we used a unique data set from a large-scale lysimeter in order to perform a differential split sample test with four groundwater recharge models with varying complexity and six climatically contrasting calibration periods.**We demonstrate that an acceptable model performance during the calibration period does not ensure reliable predictions under dissimilar climatic conditions. The deviation of simulated from observed recharge, however, is a function of the chosen model complexity. We also show that the more complex, physically-based models best reproduced observed recharge, even when calibration and prediction periods had contrasting climatic conditions. In contrast, the soil-water balance model and the lumped model perform relatively poor and a strong dependency on the chosen calibration period was evident. It can also be shown that the uncertainty in model parameters is generally less important than the model structure itself, so that the robustness of each individual model follows the degree of model complexity. It can be argued that physically-based models have a greater potential to obtain predictions beyond the range of conditions during calibration. It is still difficult to provide general guidelines on how to choose an optimal calibration period, since model performance seems to depend more on the model complexity and structure rather than on the calibration period. **