Assessment of Soil Moisture Budget Using a Water Balance Model and Use of Model Results for Drought Early Warning. Case of a Moderate Semi Arid Watershed Northern Tunisia

A water balance model presented by Kobayashi et al. (2001) modified in parameterisation by Bargaoui and Houcine (2010) is adopted to assess soil moisture content evolution through the hydrological year. This is the main model aim. The model is composed by a single soil layer reservoir, daily time scale and lumped. Three model parameters out of seven (Active soil layer, moisture retaining capacity and stomatal resistance of vegetation to evapotranspiration) are estimated using observed daily discharge data, the other being estimated out of the model through adopting pedotransfer functions and using basin soil texture information. The study analyzes the effect of adopting regional evapotranspiration a priori knowledge to constraint the set of parameters solutions obtained using runoff residual criteria. Effectively in a first step fitting criteria linked to model efficiency are adopted to select valuable parameters sets. These performance measures are: absolute annual relative error less than 20% associated to acceptable runoff Nash coefficients at monthly and decadal time scales. The case study is Wadi Oussafa (Tunisia), area 397 km˛, elevation between 508 m and 1294 m, intermittent river coming from Makthar elevations feeding the deep aquifer of Siliana underflow during winter season. Data division method for model testing is k-fold partitioning, with k=4. Thus, daily discharge series from 1928 to 1932 are used for training. Periods 1933 to 1938, 1960 to 1963 and 1966 to 1972 are used for testing. Before using data for model calibration and testing, data control process was undertaken identifying flood periods (minimum daily runoff volume = 0.5 Mm3 and separation time interval=2 days). Observed daily hydrographs were found coherent with observed daily hyetographs and resulting runoff coefficients. Parameters sets selected this way, result in good visual agreement between observed and predicted discharges in calibration and testing periods. They are then adopted to run the model. Subsequently, monthly predicted percolation fluxes, evapotranspiration fluxes and soil moisture contents are compared between solutions. Surprisingly, they do report a wide variability from one set of solutions to another. Moreover, they do display bimodal distributions. Using the Budyko model (1956) and its Hsuen Chen (1988) approximation to estimate regional evapotranspiration for this basin, as well as using ecological constraints dictated by the basin climatic conditions (semi arid area) (Eagleson 1994), the number of acceptable solutions is therefore reduced as well as the variance of model responses. Using the retaining solutions after this second step of model building, it is found that the average soil water content predicted by the model is a good proxy of the hydrological year dryness. Thus, a drought index is adopted (representing a defined soil moisture content quantile) to evaluate drought occurrence conditions and to help drought early warning.