Comparing the performances of Pedotransfer Functions with Hydrus 1D Inverse Parameters Estimation in a deep cultivated sahelian soil

Abstract

Knowledge about soil water balance and ecosystem water partitioning is crucial for managing soils in semi-arid areas like the Sahel, but hydraulic parameters are hardly available to run either parsimonious or detailed process models. This study aims at bridging this parameterization gap in a typical deep (> 2m) loamy sand soil from the groundnut basin in Senegal[1]. Five approaches of soil hydraulic parameterization with a range of different complexity were compared: (1) the lookup table of Carsel and Parrish (1988) that use only the soil texture class known as “Class PTFs”, (2) Rosetta PTFs from only topsoil characterization, (3) Rosetta PTFs with a detailed multilayer soil characterization, and inverse estimation from soil moisture using Hydrus-1D, considering the soil column either as (4) a single soil material and (5) with three-layered soil material. We compared the predicted (i) soil water content (SWC) with high-frequency measurements from 15 cm down to 200 cm deep and (ii) actual evapotranspiration (ET) with Eddy Covariance (EC) data during four consecutive growing seasons under a rotation of pearl millet and peanut crops. The simplest methods (1 & 2) resulted in a significant bias of the predicted SWC, with, however, some predictive ability of Method 2 to simulate the general trends of Swc, especially under peanut crops. Method 3 behaved reasonably with average RMSE for SWC, varying between 0.029 and 0.023 cm^-3 cm^-3. Method 4 further improved the predictions with RMSE ranging from 0.013 to 0.020 cm^-3 cm^-3. The best agreement was found under peanut using Method 5 (RMSE ≤ 0.013 cm³ cm^-3). Methods 3, 4 or 5 behaved satisfactorily for predicting ET whatever the crop, e.g. Method 4 (RMSE= 0.05 cm day-1, NSE= 0.9 and R²= 0.93) for pearl millet.