Parameterization of a Semi Distributed Hydrological Model by using a combination of ground and satellite-derived data during the calibration process. A case study in the Wami River Basin.

In Tanzania, the uneven distribution in time and space of water resources throughout the country, the high population growth (2.9% per year) and the increase of extreme climate events (drought and flood) continuously increase the risk of food insecurity. To enhance food security and national economic growth in a country where agriculture is the primary economic activity for 80% of the population, and represents around 50% of the Gross Domestic Product, the Tanzanian government is developing research activities, infrastructures and regulations to develop irrigation in a sustainable manner and to improve water allocation frameworks at the watershed scale. The growing number of smallholder farmers requires efficient agricultural water management to ensure food security, rural incomes, health and nutrition. In this context, hydrological modeling provides a reliable tool for assessing the impacts and viability for developing and implementing water management practices.

The combined use of ground and remote sensing data in hydrological modeling applications significantly improve models’ performance and provide decision makers with timely, objective, accurate, and actionable information to improve water management and strengthen food security. This is especially the case for data scarce environments in which the lack of ground-based observations hinders hydrological models’ predictability capacity. Indeed, remote sensing products facilitate routine monitoring of hydrological processes, water resources, deficits and extremes. The incorporation of satellite-derived hydrological variables during a calibration processes add constraints to the model parameters, which results in reduced model uncertainties and a reduced number of equifinal solutions.

This study attempts to test the improvement of the semi distributed Soil & Water Assessment Tool (SWAT) model’s performance, expressed as the goodness-of-fit between observed and simulated streamflow, after assimilating the Integrated Multi-satellitE Retrievals for GPM (IMERG) precipitation products and evapotranspiration time series derived from the Moderate Resolution Imaging Spectroradiometer (MODIS). At the same time this research focuses on analyzing the differences in model parameters sensitivity between a streamflow-based calibration approach and a combined streamflow and satellite-derived evapotranspiration-based calibration approach.

This research comprises the following steps: First, the model is run using ground data as model inputs, followed by a streamflow-based calibration process. Second, the optimized model is recalibrated and rerun after replacing the ground precipitation inputs by satellite-derived precipitation data that was resampled at subbasin level. Finally, a multi-objective calibration approach will include different sources of observed data, discharge (ground) and evapotranspiration (satellite), as an attempt to add constraints to model parameters and to reduce model uncertainties and the number of equifinal solutions.

Preliminary results showed that SWAT model performance, based on the efficiency coefficient NSE, significantly improved after incorporating remote sensed precipitation data. The initial NSE value after the streamflow-based calibration approach was 0.13, but reached to 0.51 when the model was run and fully calibrated using satellite-derived data. It is expected that the multi-objective calibration approach highlights differences on the optimized model parameter values between the two calibration approaches.