Spatio-Temporal Analysis of the Accuracy of TRMM Satellite Data to Estimate the Severity of a Drought Based on Precipitation in Central Iran

Introduction
Precipitation data have been widely used in many earth science applications ranging from crop yield estimates, tropical infectious diseases, drought and flood monitoring. However, in many tropical regions and parts of the mid-latitudes, rainfall estimates still remain a major challenge due to sparse rain gauges. To better develop applications for these regions, it is necessary to have rainfall data with adequate spatial and temporal resolutions. Precipitation data plays the key role in drought monitoring. Rain gauges are the main measuring methods for precipitation but they are concentrated in developed countries and are spare in developing countries and remote areas in the world. Researchers have shown that remote sensing techniques using space-borne sensors can provide an excellent complement to continuous monitoring of rain events both spatially and temporally. Microwave and Visible/Infrared are the main forms of remote sensing technologies; both have varied advantages in terms of imaging accuracy and spatial-temporal resolutions. Thus, the fine spatial-temporal precipitation products need the coalescence of both. Tropical Precipitation Measuring Mission (TRMM) carrying sensors on precipitation provides the opportunity for fine spatial-temporal precipitation products. In this research for Central Iran, the precipitation data of TRMM satellite was evaluated and used to estimate the severity of a drought based on precipitation.
 Materials and Methods
Central Iran is located between 27N-37N latitudes and 48E-61E longitudes and has an area of about 837,184 km². There are 50 synoptic stations within the area. The data are including monthly precipitation depth from both synoptic stations and TRMM data (3B43 V.7, in ASCII format). A five year (2001–2005) period were chosen for the analysis. The accuracy of precipitation data that are used from synoptic stations and TRMM satellite are provided by the source provider. Firstly, the evaluation of TRMM satellite data was measured using coefficient of determination (R²), mean error (ME), mean absolute error (MAE) and root mean square error (RMSE) in 95% confidence levels. Then, TRMM remote sensing data are used to provide the required data for precipitation drought index in central Iran to make a mapping of the spatial distribution of drought. Finally, accuracy of the PDI drought index based on satellite data carried out using the evaluation criteria was compared with drought spatial distribution map of the PDI based on ground-based precipitation measurements data and soil moisture values of 50 synoptic stations.
Results and Discussion
In this study, monthly rainfall values estimated by satellite products were copared with those of rain gauge observations in Central Iran. The validation of TRMM 3B43 data were performed at monthly, season and annual scales. The average of monthly, seasonal and annual rainfall for all selected synoptic stations and TRMM data were compared during the period 2001-2005. TRMM data at all time steps, except August, estimates the average of monthly rainfall more than observed data. The correlation coefficient between the average of monthly rainfall, seasonal and annual rain gauge and TRMM has shown that this ratio is variable from 0.45 to 0.94 for all time steps and the average of this ratio is equal to 0.76. The highest and lowest values of R² at monthly time step obtained 0.92 for April and 0.45 for June. In this time step, the lowest and highest values of statistical error criteria are obtained for June and January, respectively. In seasonal time step, the highest and lowest correlation is related to the spring and summer with determination coefficient (R²) of 0.94 and 0.64, respectively. In this time step, the lowest and highest values of statistical error criteria obtained for summer and winter, respectively. Generally, TRMM data performs best in summer, but worst in winter, which is likely to be associated with the effects of snow/ice-covered surfaces and shortcomings of precipitation retrieval algorithms. The correlation coefficient for the annual time step is equal to 0.83. The results of statistical criteria showed that TRMM rainfall data in all time steps overestimated for all months except for August. The lowest to the highest values of statistical error criteria were obtained for monthly, seasonal and annually rainfall data, respectively. In the next step, spatial distribution of drought based on measured data from ground stations and TRMM data in the period 2001-2005 is obtained from Precipitation Drought Index (PDI) method in study area. The results of the statistical criteria of conformity assessment in PDI spatial distribution map based on TRMM data with corresponding pixels of spatial distribution map based on the synoptic stations precipitation data showed that the drought severity map had a high precision and good conformity with ground data (R²=0.89, ME=0.08, MAE=0.14, RMSE=0.19). Also, the results of the evaluation criteria showed that PDI index in accordance with soil moisture values had the significant correlation (0.71) and the lowest estimation error (2.33).   
Conclusion
In this research, for estimation of drought severity index based on precipitation, the monthly precipitation data of TRMM satellite (3B43) was evaluated. The evaluation was measured using coefficient of determination (R²), mean error (ME), mean absolute error (MAE) and Root Mean Square Error (RMSE). This analysis has demonstrated that the TRMM rainfall products show very good agreement with gauge data over the selected area of Central Iran on monthly timescales and 0.25° space scales. said it can be concluded that the satellite-based rainfall, e.g. TRMM data, have good potential for useful application to hydrological simulation and water balance calculations at monthly or seasonal time steps. This can be useful  for the regions where rain gauge observations are sparse or of bad quality. However, the TRMM can overestimate the rainfall in some years and areas and underestimate in other years and areas, and failed to detect the extreme rainfall. This can reduce the accuracy of stream flow simulation at short time steps and other applications including drought monitoring and flood forecasting. The conclusions indicate that it is necessary to further develop algorithms of satellite-based rainfall estimation in terms of both the accuracy and spatiotemporal resolutions of rainfall estimates.