Integration of multi-seasonal remotely-sensed images for improved landuse classification of a hilly watershed using geographical information systems

Information on existing landuse pattern and its spatial distribution is a pre-requisite for any watershed management programme. With the advent of remote sensing tools, with their inherent characteristics, it has been possible to prepare this dynamic resource map at various levels of confidence. But the application of multi-spectral classification techniques in vegetation discrimination has been the subject of discussion for many years. The paper describes a methodology developed to compile agricultural landuse map of a hilly watershed (where the rainfall erosivity is an ever-present threat) using geographical information systems (GIS). Multi-seasonal (monsoon-Kharif and post-monsoon-Rabi) and multi-sensor remotely-sensed data were used for mapping a landuse pattern of the watershed. The rasterised classified images, and the relevant watershed resources, were input and stored as separate layers in the GIS and then geometrically co-registered to a regular 30m grid. Knowledge-based rules were developed from the informed opinion of multi-disciplinary experts and field checkings, in addition to the knowledge of local landuse patterns. These expert rules were used to manipulate the information databases to discriminate various spectrally inseparable information classes, finding out the landuse/cover categories of the Kharif season under the cloud and its shadow areas. Finally, the improved and the agricultural landuse pattern map was classified into thick forest, sparse forest, degraded pastures, open (with/without scrub) areas, cropped (Kharif + Rabi) and fallow (Kharif + Rabi) lands. The areal extent of improved/final landuse map classes compared favourably with the natural conditions of the agro-climatic region of the watershed. The paper also envisages some future studies for watershed management policies.