Jump to Main Content
An evaluative study of TRMM precipitation estimates over multi-day scales in a semi-arid region, Iran
- Mahbod, Mehdi, Veronesi, Fabio, Shirvani, Amin
- International journal of remote sensing 2019 v.40 no.11 pp. 4143-4174
- equations, prediction, rain, rain gauges, regression analysis, remote sensing, semiarid zones, Iran
- Data from the Tropical Rainfall Measuring Mission (TRMM) rainfall estimations have been evaluated at different time scales in the previous research, in particular, sub-daily, monthly, seasonally and annually. However, in arid and semi-arid regions water balance may be reached several days after a rainfall event. Hence, it becomes of crucial importance to investigate sub-monthly time periods (i.e. multi-day periods). For this reason, TRMM precipitation data version 3B42 (3B42) were evaluated and calibrated for 1, 2, 3, 5, 7, 10, 15, 20 days and monthly time scales using rain gauges data in Fars province, Islamic Republic of Iran, 1 January 2000 to 31 December 2014. Pearson’s correlation coefficient (r), Root Mean Square Error (RMSE), Normalized Root Mean Square Error (NRMSE), Mean Bias (MB), Prediction of Detection (POD), False Alarm Ratio (FAR) and Critical Success Index (CSI) were used for the purpose of evaluation. The results showed that with a logarithmic trend, r, NRMSE, and FAR values decreased while RMSE, CSI, and POD values increased with increasing time scales. Moreover, the spatial average MB was almost constant for various time scales, although the percentage of grid cells with over-estimated rainfall increased from 1 day to 1 month. By fitting logarithmic functions over the values of r, RMSE, NRMSE, POD, FAR, and CSI at 1, 10 days, and monthly time scales, the corresponding values of these measures were predicted for other time scales with the relative error (NRMSE) of less than 0.1, which indicates the accurate performance of these functions. Through linear regression analysis, the slope (M) and interception (B) of the equations for calibrating 3B42 precipitation estimates at various time scales were obtained. Furthermore, the results showed that the obtained values of M and B in 1, 10 days, and monthly time scales can be estimated with a high accuracy at 2, 3, 5, 7, 15, and 20 days.