Evaluation of TRMM Satellite Data for Mapping Monthly Precipitation in Pakistan by Comparison with Locally Available Data

Increase in global average temperatures, widespread snowmelt and rising sea levels are important evidences of climate change, causing severe changes in the spatial and temporal rainfall patterns and increasing the frequency of related natural disasters like floods and droughts. This scenario increases the importance of satellite data for effective monitoring and forecasting of such disasters. Such data become more valuable in case of less developed countries where there is limited availability of local data. The open source and most commonly used data for detecting critical rainfall events are taken from Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA). This paper presents the evaluation of TRMM monthly rainfall data (3B43) for Pakistan, using 15 years data from 1998 to 2012 in comparison to the ground rainfall data for the same period. For calibration, regression analysis was performed for each month on the data of rain gauge stations and their corresponding pixel values from the satellite datasets. Regression equations were used to develop the calibrated seasonal and annual rainfall maps, as well as the monthly rainfall maps on district level. The calibrated monthly datasets were validated by calculating Nash-Sutcliffe Efficiency of TRMM estimates before and after calibration. The original TRMM 3B43 data was found quite reliable for its direct use with NSE values ranging from 0.73 to 0.92 for different months, while the calibration further improved the NSE by about 5% on average. However, the NSE values were decreased after calibration for the months of July, August and September, indicating that high accuracy and care is required alongwith much dense rain gauge network to perform calibration in hilly areas having local gradients and heavy orographic rainfalls during these rainy months. For dry months of winter, i.e. October to December, TRMM rainfall estimates were found relatively less accurate, but NSE values were improved by 10-15% after calibration in these months