Assessing the accuracy of SM2RAIN (Soil Moisture to Rainfall) products in poorly gauged countries: the case of Burkina Faso in the West African Sahel.

Abstract

<p>Effects of climate change and variability in West African countries are heightening the vulnerability of local populations, which heavily rely on agriculture and natural resources through ecosystem services. Developing effective water management strategies for mitigation of these impacts requires knowledge of weather, especially rainfall, built upon continuous long-term records (in both time and space). Yet, most West African countries are poorly gauged, with a low density of reliable gauging stations, hampering applications such as water planning and weather forecasting. Recently, some daily global precipitation products have been developed, providing rainfall estimates derived from soil moisture observations using the innovative SM2RAIN (Soil Moisture to Rain) bottom-up inversion algorithm, hence treating the soil as a natural rain gauge. Since these products are gridded, they also provide continuous spatial information regarding rainfall. In this study, the accuracy of such three typical SM2RAIN products (SM2RAIN-CCI, GPM+SM2RAIN, SM2RAIN-ASCAT) at depicting rainfall estimates in Burkina Faso (West Africa, area of 272,200 km²) at 10 synoptic stations over the period 2007-2017 (9 years), at the daily, dekadal (10-days accumulation), monthly and annual timescales. The results reveal that at the daily timescale, all products performance is poor to moderate (KGE: 0.18 to 0.36). At higher time scales, however, both products performed satisfactorily to very good (<em>dekadal</em>: KGE: 0.61 to 0.79; <em>monthly</em>: KGE: 0.63 to 0.91; <em>annual</em>: KGE: 0.44 to 0.81), with SM2RAIN-CCI being consistently superior. Overall, SM2RAIN-CCI presented the lowest volumetric hit and miss bias at all stations, whereas SM2RAIN-GPM presented the lowest false bias. Also, SM2RAIN-CCI featured the highest ability at picturing the timing of occurrence of daily rainfall events (probability of detection, false alarm ratio, threat score) for various thresholds in the range of 0 to 25 mm. Finally, the ability of these products at picturing observed rainfall extremes have been evaluated through various ETCCDI climate indices, at which SM2RAIN-CCI and SM2RAIN-GPM presented equal performance, SM2RAIN-ASCAT being less good. These results provide a quantitative assessment of the SM2RAIN approach in the context of Burkina Faso and might help in the selection of an optimal product for further applications.</p><p><strong>Keywords: </strong>ASCAT, Burkina Faso, CCI, GPM, rainfall, Sahel, SM2RAIN.</p>

Future climate or land use? Attribution of changes in surface runoff in a typical Sahelian landscape

“…The SWAT model has been previously calibrated for the Tougou watershed (over the period 2004-2018) with satisfactory performance (Kling-Gupta efficiency: KGE = 0.95/0.94; percent bias: PBIAS = −2.30%/2.90% for calibration/validation periods). The Table 2 is adapted from Yonaba et al [2021a] and shows the calibrated values (and uncertainty range) for the model parameters, which were used in this study for hydrological simulation. Among these parameters, CN2 (SCS Curve Number), SOL_AWC (soil water content), ESCO (soil evaporation compensation factor), OV_N (Manning's n roughness for overland flow), SOL_K (saturated hydraulic conductivity) are the ones which are affected straight by land use [Amogu et al, 2015, Descroix et al, 2018, Yonaba et al, 2021a.…”

“…The units provided in the description stand for the absolute parameter value [Neitsch et al, 2011]. The calibration method refers to the strategy used for updating the model parameters during the calibration phase: r (for relative) is the relative deviation applied to the original parameter value, which therefore preserves its spatial variability; v (for replace) is the absolute and unique value given to the parameter in the model, regardless of its location [Abbaspour et al, 2007, Yonaba et al, 2021a.…”

“…As this total relative change is both affected by climate and land use change, the isolated contributions of each factor are evaluated using a "fixingchanging" method, where a single factor (either land use or climate) is changing [Yonaba et al, 2021a]. This needs intermediate simulations summarized in Table 3.…”

“…The separation method however assumes that land use changes are independent of climate change, and the land-use change affects only actual evapotranspiration (ET). The latter assumption holds since actual ET is mostly dependent on vegetation (cover fraction, growing stage and development) and land use management [Dey and Mishra, 2017, Gbohoui et al, 2021, Yonaba et al, 2021a.…”

“…Another difficulty in achieving accuracy in simulating the water balance in hydrological models resides in a full acquaintance of land use/land cover changes (LULCC) during long-term runs in modelling experiments. Usually, a single and static LULC map of the landscape at a given time point is used, which might lead to failure in picturing the spatial and temporal patterns of evolution of hydrological 3 processes [Wagner et al, 2016, Yonaba et al, 2021a. Some hydrological models, such as the Soil and Water Assessment Tool [Arnold et al, 1998] provide mechanisms to integrate a dynamical update of LULC conditions during simulations, through a dedicated land use update (LUP) module, which reportedly helped in achieving significant higher model performance [Aghsaei et al, 2020, Yonaba et al, 2021a.…”