Distributed validation of hydrological model using field and remotely sensed data

The assessment of the performance of (semi)distributed hydrological models has traditionally depended on parameters monitored at a gauging station usually located at the lowest end of a basin regardless of the size, complexity and spatial- temporal variations. The result of such an approach is that the processes in the basin are lumped by compositing the catchment processes over time and space. The Soil and Water Assessment Tool (SWAT) model gives various outputs which are distributed all over the basin by use of the hydrologic response units (HRU). However, due to lack of physical location for the HRU, and their possible large number in a single watershed or even subbasin, it is physically and almost practically impossible to monitor the flow, nutrients and sediments at all the outlets of these HRUs. A new concept of land use soil units (LUSU) created from overlaid soil and land use classes makes it possible to spatially compare outputs. The use of geographic information system (GIS) to overlay datasheets and the availability of gridded remotely sensed data for biomass, evapotranspiration, leaf area index (LAI) and yields in real time makes it possible to perform a dynamic distributed model validation. The SWAT model is used to test the applicability of remotely sensed variables on a 2905 Km2 basin. The watershed is data scarce, geologically difficult, topographically challenging with dynamic land management practices, and drastically changing climatic conditions from semi- Arid to humid tundra/montane conditions. When actual management practices like fertilizer application are considered in the SWAT model, the simulated yields compare favourably with observed field yields (p=0.05). Although remotely sensed Leaf Area Index values are higher than the simulated LAI, it mirrors to a great extent the timing and shape of the simulated LAI for the annual landuse classes. The seasonality characteristics for these land classes are also consistent for both RS and simulated LAI.