Evaluation Of The Accuracy Of GPM/IMERG Over The Mainland Of China

In this paper, the accuracies of Integrated Multi-satellite Retrievals for Global Precipitation Measurement (GPM) (IMERG) products over the mainland of China,are evaluated with the quantitative method and the method of classification score using the China Gauge-Based Daily Precipitation Analysis (CGDPA) products,developed by the National Meteorological Information Center. The accuracy differences between IMERG and CMORPH (Climate Prediction Center morphing technique) and PERSIANN (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks) satellite precipitation products are compared. In addition, based on the high temporal-spatial resolution precipitation estimates of Nanjing University of Information Science & Technology C-Band Dual-Polarization Radar (NUIST-CDPR) that calibrated by the ground rain gauge,,the accuracies of IMERG products in the process of heavy rainfall are assessed. The main results are as follows:(1) Analysis of the accuracy of GPM/IMERG from September 2014 to August 2015 using CGDPA products show that, the continuity of the spatial distribution of precipitation is more reasonable in the IMERG than in the CGDPA. The IMERG can produce the characteristics of the spatial distribution of precipitation over the mainland of China, although precipitation over some regions is underestimated. The seasonally averaged precipitation in IMERG reveals good consistency of spatial distribution with CGDPA, especially over the middle, south of China with slight difference. However, biases over the regions are seasonally different. The correlation coefficients between IMERG and CGDPA over most areas range between 0.2 and 0.5. The correlation coefficients over some areas are still found to be negative,ranging between -0.3 and -0.5. Although much more accuracies are shown in IMERG over the east of China than those over the west of China, the daily precipitation estimation of IMERG is 10% —30% (even 35%) lower than the daily precipitation of CGDPA over the east of China. Over the west of China, the daily precipitation estimation of IMERG over some regions is lower than the daily precipitation of CGDPA, and the opposite signs are found over some regions.(2) The difference of the probability density of the small-amount precipitation between IMERG and CGDPA is the largest (especially over the west of China),followed by the moderate rain, while there is small difference between them in the other precipitation intensities. The probabilities of detections (PODs) of precipitation events for IMERG are low seasonally (less than 0.45), while the false alarm ratios (FARs) are relatively high (greater than 0.5). Meanwhile, the critical success indexes (CSIs) are relatively low(less than 0.35). The same trend is found in the POD and the CSI, with their highest values of both in summer, followed by spring and autumn,and the lowest in winter. However, the trend of FAR is opposite,with its highest value in winter, followed by autumn and spring, and the lowest in summer.(3) Comparative analysis results of the IMERG, CMORPH and PERSIANN precipitation products shows that, the IMERG products are closer to the observed precipitation, the correlation coefficient is 0.93, the root mean square error is 0.62mm/d, and the relative error is -2.78%, while the correlation coefficients of CMORPH and PERSIANN products are 0.84 and 0.67, respectively, the mean square errors are 1.1 mm/d and 1.36 mm/d, respectively, and the relative errors are-23.8% and -18.24%, respectively. Comparison of the accuracy differences between IMERG and CMORPH and PERSIANN satellite precipitation products in each sub region of China show that, poor performance is found in western China in three kinds of satellite precipitation products, especially in northwestern region and the Qinghai-Tibetan Plateau region. But generally speaking, the performance of IMERG products in these areas are better than CMORPH and PERSIANN products.(4) Compared IMERG precipitation products with the high temporal-spatial resolution precipitation estimates of NUIST-CDPR that calibrated by the ground rain gauge, the results show that, the correlation coefficients between IMERG and radar estimated precipitation in three heavy rainfall processes are 0.6, 0.83 and 0.78 respectively, the mean square errors are 1.36 mm/hhr, 0.52 mm/hhr and 0.99 mm/hhr respectively, and the relative errors are -50.63%, 1.08% and -34.72% respectively.With the increase of precipitation intensity threshold, the detection abilities of IMERG precipitation products weakened. In general, performance of weak precipitation is better than that of heavy rainfall.