Can changes in precipitation regime explain the first Sahelian hydrological paradox ? An inquiry with a statistical emulator of sub-daily hydrological processes

The Sahel, the semi-arid fringe south of the Sahara, experienced severe meteorological droughts in the 70s-80s. During these droughts, several watersheds may have experienced a regime shift that led to an increase in the annual runoff coefficient (annual runoff divided by annual precipitation). This phenomenon, known as the first Sahelian hydrological paradox, has been attributed to soil crusting, a very typical feature of the sahelian region, which led to an increase in Hortonian runoff. The physical driver of this soil crusting is still debated in the literature. Standard explanations generally involve land use and cover changes (LUCCs). However, alternative explanations exist: soil crusting may have also been impacted by changes in precipitation regime (total precipitation, precipitation intensity, …). Here, we focus on the impact of precipitation regimes on annual runoff coefficient.

 

In this region, most hydrological processes occur at a sub-daily scale. However, existing observations of precipitation are only available at the daily scale. A classic way to resolve this issue is to downscale observations of precipitation at a sub-daily scale, and model processes at this scale. In practice, such downscaling always implies strong hypotheses concerning spatio-temporal dependencies of rainfall process at fine scale. Instead, we propose an alternative solution: to “upscale” sub-daily hydrological processes at a daily scale. Specifically, we use fine-scale rainfall series to force a simplified Green-Ampt (GA) infiltration model which predicts runoff at fine scale. Then we compute both annual statistics of rainfall regime from sub-daily rainfall series and annual runoff coefficients from the GA simulations ; and we infer a statistical link between annual rainfall statistics and annual runoff coefficients. This statistical emulator of the GA model predicts annual runoff coefficient based on several features: the saturated hydraulic conductivity of the soil (ksat), and several indicators of precipitation regime, such as the average and the maximum of daily precipitation.

 

In our results, this emulator is first trained and assessed using observations of precipitation from Sahelian stations. In this case, we show that this emulator can reproduce annual runoff coefficients produced by the GA model. We also note that ksat has more impact than annual indicators of precipitation, which may be due to the crystalline sedimentary context of our study. Then, we test this emulator on Sahelian watersheds where we have both rainfall series and observed runoff series (and thus runoff coefficients). Our preliminary results show that this emulator can reproduce observed trends in annual runoff coefficient.