Evaluating downscaling methods of GRACE (Gravity Recovery and Climate Experiment) data: a case study over a fractured crystalline aquifer in southern India

Claire Pascal (1) , Sylvain Ferrant (1) , Adrien Selles (2, 3) , Jean-Christophe Maréchal (3, 2, 4) , Abhilash Paswan (5) , Olivier Merlin (1)

1 CESBIO - Centre d'études spatiales de la biosphère
2 UMR G-EAU - Gestion de l'Eau, Acteurs, Usages
3 BRGM - Bureau de Recherches Géologiques et Minières (BRGM)
4 UM - Université de Montpellier
5 IICT - CSIR Indian Institute of Chemical Technology [Hyderabad]

GRACE (Gravity Recovery and Climate Experiment) and its follow-on mission have provided since 2002 monthly anomalies of total water storage (TWS), which are very relevant to assess the evolution of groundwater storage (GWS) at global and regional scales. However, the use of GRACE data for groundwater irrigation management is limited by their coarse (≃300 km) resolution. The last decade has thus seen numerous attempts to downscale GRACE data at higher – typically several tens of kilometres – resolution and to compare the downscaled GWS data with in situ measurements. Such comparison has been classically made in time, offering an estimate of the static performance of downscaling (classic validation). The point is that the performance of GWS downscaling methods may vary in time due to changes in the dominant hydrological processes through the seasons. To fill the gap, this study investigates the dynamic performance of GWS downscaling by developing a new metric for estimating the downscaling gain (new validation) against non-downscaled GWS. The new validation approach is tested over a 113 000 km2 fractured granitic aquifer in southern India. GRACE TWS data are downscaled at 0.5∘ (≃50 km) resolution with a data-driven method based on random forest. The downscaling performance is evaluated by comparing the downscaled versus in situ GWS data over a total of 38 pixels at 0.5∘ resolution. The spatial mean of the temporal Pearson correlation coefficient (R) and the root mean square error (RMSE) are 0.79 and 7.9 cm respectively (classic validation). Confronting the downscaled results with the non-downscaling case indicates that the downscaling method allows a general improvement in terms of temporal agreement with in situ measurements (R=0.76 and RMSE = 8.2 cm for the non-downscaling case). However, the downscaling gain (new validation) is not static. The mean downscaling gain in R is about +30 % or larger from August to March, including both the wet and dry (irrigated) agricultural seasons, and falls to about +10 % from April to July during a transition period including the driest months (April–May) and the beginning of monsoon (June–July). The new validation approach hence offers for the first time a standardized and comprehensive framework to interpret spatially and temporally the quality and uncertainty of the downscaled GRACE-derived GWS products, supporting future efforts in GRACE downscaling methods in various hydrological contexts.

Domaines: Sciences de la Terre

Claire Pascal, Sylvain Ferrant, Adrien Selles, Jean-Christophe Maréchal, Abhilash Paswan, et al.. Evaluating downscaling methods of GRACE (Gravity Recovery and Climate Experiment) data: a case study over a fractured crystalline aquifer in southern India. Hydrology and Earth System Sciences, 2022, 26 (15), pp.4169-4186. ⟨10.5194/hess-26-4169-2022⟩. ⟨hal-03760943⟩