Modelling of changes in hydrological balance in Gambia river basin using two lumped models

This study investigated the calibration performance of hydrological models applying a series of split-sample to crash-test potential combinations of calibration-validation periods under drought type (dry/wet) using lumped models: BILAN and GR2M. A sub-period focused on the drought was systematically selected for model calibration based on a particular climate characteristic (precipitation, temperature, runoff) and a 7-year moving window. This approach gives perception into calibrated parameters transferability overtime under similar or different climate conditions (drought). 

Both lumped models yielded similar results over a set of 6 catchments in a main West African river basin located in Senegal: the Gambia river basin. The Kling-Glupta Efficiency (KGE) was the objective function to assess models’ efficiency. A dependency was found between the model performance and the extent of input data. 

Results have shown that the calibration performance decreases within an extending simulation period width. A focus on the impact of drought type on calibration performance revealed models simulating better dry than wet years. The analysis on how model performance would be affected when calibrated in a climate condition different to the validation (e.g. calibrated in dry(wet) and validated into wet (dry) revealed that calibration over a wetter or dryer condition than the validation and vice-versa may lead to an over(under)estimation of the simulated runoff. 

The results also indicate a general performance loss due to the transfer of calibrated parameters to independent validation periods of −5 to −25%, on average. The shift of model parameters in time (validation) may generate a significant level of errors. The outcome of this study may lead to a master of the uncertainty associated with one hydrological model and a better assessment of runoff in a real-world application.

Keywords: Gambia river basin; calibration; crash test; rainfall-runoff model; BILAN; GR2M; lumped hydrological models;