GROUND VALIDATION OF MULTI-SATELLITE PRECIPITATION DATA PRODUCTS

Abstract

Precipitation is the key element of the hydrological cycle and the accuracy of measurement of precipitation plays vital role in hydrological modelling, climate change studies, extreme event predication, agriculture and drought monitoring. Tropical Rainfall Measure Mission (TRMM) and Global Precipitation Measurement (GPM) Precipitation data are of high spatio-temporal resolution; Ground Validation or Ground truth – Validation parameters like Percentage Bias (PBIAS), Correlation coefficient (CC), Mean Error (ME), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), False Alarms ratio (FAR), Frequency Bias Index (FBI), Probability of Detection (POD), Accuracy, Critical Success Index (CSI) etc have been used to access the accuracy of multi satellite precipitation estimates. In addition, a new methodology has been used for cross-validation of satellite precipitation estimates which is recommended by NASA i.e. using rainfall-runoff model depicting the true scenario in the hydrological cycle. In the present study, TRMM_3B42_V7 and GPM_3IMERGHHv03 satellite precipitation data products have been validated over two study areas i.e. Himachal Pradesh (from March, 2014 to January, 2016) and Dal Lake, J&K (from 2001 to January, 2016) in India using statistical techniques and HEC – HMS rainfall – runoff model as cross-validation methodology. The study over Himachal Pradesh shows slight overestimation by GPM against raingauge, very good consistency upto High Hill Zone with improvement from hourly to daily time steps, however the variations were more erratic beyond altitude of 2100 m. Average success index for the rainy day (precipitation ≥2.5 mm/day) is 0.79 which shows very good capability of GPM in capturing rainy days. It is inferred that GPM 3IMERGHHv03 are not reliable on lower rainfall intensities i.e., ≤2 mm/hr, whereas, extreme precipitation events at 98th, 99th and 99.99th percentile thresholds show accurate capturing and satisfactory performance upto 99th percentile. The study over Dal lake reveals that the TRMM precipitation estimates are underestimating the actual precipitation by 48.90%, while the GPM estimates are nearly equal to actual precipitation i.e 9.08% underestimation. HEC-HMS model has simulated the rainfall-runoff process nearly equal to the observed rainfall-runoff relation with GPM precipitation estimates. In addition, the model has correctly simulated the extreme flow condition i.e. 93.6 m3/s with GPM data. iv GPM 3IMERGHHv03 precipitation estimates are excellent in terms of precipitation volume measurement with slight overestimation across different altitude ranges and may be utilised as an alternative of observed precipitation estimates. The evaluation concluded that GPM precipitation estimates may be used for hydrological analysis on daily time step, while for half hourly time step GPM retrieval algorithm needs further refinement over Himalayan region.