Monitoring and modelling of irrigation practices and its effect on catchment hydrology for fragmented landscapes

Abstract

It is known that irrigation can significantly alter the distribution of water in a terrestrial system. So it is of importance to include the presence of irrigation in hydrological modelling studies. However, in many countries, consistent and reliable documentation on artificial water applications is not available. Previous studies successfully mapped irrigation with remote sensing by using a pixel-based approach on NDVI information. Object-based image analysis has often proved to produce more accurate results, especially when working with high resolution imagery in spatially and spectrally complex areas. Therefore, an object-based remote sensing method to identify irrigated objects in a fragmented landscape, located in the Ethiopian Rift Valley, was developed and tested. SPOT imagery with a resolution of 1.5 meter was used and resulted in a highly satisfying irrigation map. The same irrigation mapping method was performed on coarser resolution imagery (Landsat). Evidently, SPOT was able to capture the small fragmented agricultural areas more accurately than Landsat. However, despite the coarser resolution of Landsat imagery, a visual inspection of the SPOT and Landsat irrigation maps revealed that Landsat also had its advantages. It performed better in the classification of heterogeneous agricultural fields and bare agricultural plots. The object-based irrigation mapping method developed in this study has proven to be an adequate approach for identifying irrigated agriculture, with the use of both 1.5 and 30-meter resolution imagery. It enables the production of irrigation maps with a relatively high temporal frequency, causing it to be a promising input for hydrological modelling studies. Remotely sensed irrigation maps were implemented in the Spatial Processes in HYdrology (SPHY) model to examine the impact of irrigation on the outcomes of the model. Two different scenarios were set up for a case study in the Tikur Woha catchment in Ethiopia. The reference scenario did not consider the application of irrigation while the irrigation scenario incorporated spatio-temporal irrigation information. Increases in soil moisture content, evapotranspiration, root percolation, and to a lesser extent runoff, were observed in the catchment when the presence of irrigation was taken into account. The remotely sensed irrigation maps produced in this study indicated that the extent of irrigated agriculture is highly variable over time. So the modelling results emphasized that when irrigation is used as input for hydrological models it must be properly characterized, both spatially and temporally, as it can contribute considerably to the uncertainty of model outcomes.