HYDROLOGICAL MODELLING OF A RIVER BASIN USING MULTI-SATELLITE PRECIPITATION ESTIMATES

Abstract

Distributed hydrological models have become important tools for understanding the hydrological processes and achieving solutions for practical hydrological and water resources management problems. High spatial resolution precipitation data availability is of immense use in many important applications in meteorology, hydrology and ecology. The Kopili River basin (=7198 km2) which is a part of Brahmaputra basin has been selected as study area which lies between 25° 05’ 52.44” N and 25° 51’ 45.72” N latitude and 92° 07’ 49.44” E and 93° 27’ 47.16” longitude. In this study, the observed daily runoff data was available at Kherunighat site (2003-2010). The J2000 model was calibrated for its sensitive parameters i.e. linear reduction co-efficient for AET, outflow coefficient for LPS, maximum percolation rate to groundwater, recession coefficient for overland flow, recession coefficient for Interflow, adaptation for RG1 flow, adaptation for RG2 flow, flood routing coefficient, etc. High values of coefficient of correlation (CC), Nash-Sutcliffe Coefficient (NSE), and low values of percent bias (PBIAS) and RMSE-observations standard deviation ratio (RSR) indicated overall good calibration and validation of the J2000 model. The results of the sensitivity analysis indicated soilConcRD1 as the most sensitive parameter to the model output, followed by soilConcRD2 and soilLinRed, whereas the least sensitive parameters were soilConcRD1Flood and gwRG2Fact. Various components of water balance (sub-basin and HRU-wise) were estimated using the J2000 model. The satisfactory results show the reliability of the J2000 model for understanding the hydrological system dynamics of the Kopili River basin. Hydrological response of river basin with hypothetical variations in precipitation and temperature were studied in context of the global climate change. Results reveal that the ET process is largely affected due to global climate change and resulted in low surface runoff and ground water contribution in the study area. The rigorous assessment of TRMM precipitation for hydrological application in the region was performed by creating four simulation scenarios i.e. gauged precipitation (Scenario-1), raw TRMM precipitation (Scenario-2), TRMMBC precipitation (Scenario-3), and combined gauge and TRMMBC precipitation (Scenario-4). It was concluded that the Satellite derived vi rainfall estimate (TRMM) with suitable bias correction techniques, can serve as an alternative to the measured rainfall dataset where data are insufficient for runoff simulation. Six models i.e. ANN, RBFNN, LS-SVR, MLR and decision tree models such as CART and M5 model tree were employed for suspended sediment modelling. Comparative results showed that the LSSVR and ANN models were able to produce good simulation than the other investigated models. Within Decision tree models, M5 model tree is better in simulating suspended sediment than its near counterpart, the CART model, and marginally inferior when compared to LS-SVR and ANN models. The use of TRMM dataset for modelling suspended sediment using Neural Network with different training function i.e. Levenberg-Marquardt, Scaled Conjugated Gradient and Bayesian Regulation were explored. Analysis suggests use of LM algorithm for rainfallrunoff- sediment modelling and forecasting sediment concentration due to its prominent characteristics of quicker process learning and convergence. Moreover, the TRMM precipitation is promising for suspended sediment simulation in the absence of ground-based measurements. The TRMM data can be employed for rainfall-runoff-sediment modelling where the gauges are sparsely distributed and the radar measurements are rarely available.