MACHINE LEARNING BASED IMPROVED PRECIPITATION DATESETS OVER UPPER GANGA BASIN, INDIA

Abstract

The upper Ganga basin (UGB) constituting Bhagirathi and Alakananda basins, had been considered as a single hydrological unit so far with comparable climate forcings with common schemes of spatial distribution. As a downside of this assumption, wide scale adoption of uniform temperature lapse rate is frequently exercised in numerous hydrological studies over this region. The manifestations of spatial inhomogeneity of temperature and moisture distribution on present and future aspects of local and regional climate, basin run-off, snow cover and glacier response have been frequently overlooked here. The quantification of these phenomena form the standard input of numerous hydrological modelling schemes which when run with erroneous basis, can lead to misleading climate and weather projections. Certainly, the magnitude of these errors propagates with the duration of the projection period. Principally, a clear representation of the interplay between topography, precipitation, temperature and orographic discontinuity over UGB is missing. Such an explicit depiction is necessary to feed the glacio–hydrological models with correct information to achieve reliable information. Further, it is well perceived that the monsoons play the dominant role in the cryospheric setup here and this region is often included into the definitions of ‘monsoon accumulation type’ glaciers. But, the sustenance of major glacier resources in this region under a declining monsoonal influence remains uninvestigated. Another major aspect of the certainty of the glacio–hydrological research exercises over UGB, is the accuracy of the available precipitation data. It has been observed that topography dependent distribution of precipitation is missing or has a vague representation in various climate or precipitation datasets available over UGB. Though, some prominent precipitation datasets are available here, they bear a compromise between (a) spatial resolution, (b) accuracy, (c) spatial distribution, (d) temporal resolution, and (e) temporal coverage of precipitation. Hence, the requirement of precipitation datasets capable of closely capturing the spatio-temporal phenomenon, is crucial for determining a reliable environmental response over the study area. Further, station data is deficiently available and the performance of numerous datasets is crucially dependent on the sparse gauge network. A well performing precipitation dataset independent of gauge network is a scarce resource here.