Assessing the applicability of TMPA-3B42V7 precipitation dataset in wavelet-support vector machine approach for suspended sediment load prediction

In the present study, the latest Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA) research product 3B42V7 has been evaluated over gauge-based India Meteorological Department (IMD) gridded dataset employing statistical and contingency table methods for two South Indian watersheds. A comparative analysis of TMPA-3B42V7 with IMD gauge-based gridded dataset was carried out on daily, monthly, seasonal and yearly basis for 16 years (1998-2013). The study revealed that TMPA estimates performed reasonably well with the gauge-based gridded dataset, however, some significant biases were also observed. It has been observed that TMPA overestimates at very light rain, but underestimates at light, moderate, heavy and very heavy rainfall intensities. Further, the TMPA estimates was evaluated for prediction of daily suspended sediment load (SL) employing Support Vector Machine (SVM) with wavelet analysis (WASVM). Initially, 1-day ahead SL prediction was performed using best WASVM model. The results showed that 1-day predictions were very precise and shows a better agreement with the observed SL data. Later, the developed WASVM model was used for the prediction of SL for the higher leads period. The statistical analysis shows that the developed WASVM model could predict the target value successfully up to 6-days lead and found to be not suitable for higher lead specifically in the selected watersheds with similar hydro-climatic conditions like the ones selected in this study. Predictions results of WASVM model is superior to conventional SVM model and could be used as an effective forecasting tool for hydrological applications. The study suggest that the use of TMPA precipitation estimates can be a compensating approach after suitable bias correction and have potential for SL prediction in data-sparse regions. (C) 2017 Elsevier B.V. All rights reserved.