This comment regards the error analysis of satellite precipitation products by Mei et al. (2014). The authors of the original paper quantified the error characteristics of four widely utilized satellite rainfall estimates [TMPA 3B42, version 7 (3B42-V7); TMPA 3B42 in real time (3B42-RT); Climate Prediction Center (CPC) morphing technique (CMORPH); and PERSIANN] in capturing heavy precipitation events over different basin scales. The primary objective of their study was to investigate the hydrological potential of satellite precipitation products in natural hazard warning, such as flash floods, shallow landslides, and debris flows, triggered by heavy precipitation events over mountainous areas. They analyzed 3249 precipitation events during the period of 2003–10 over the mountainous upper Adige River basin in Italy. Their analysis is of great interest to hydrologic users of the satellite precipitation datasets.

However, the real-time availability of satellite precipitation estimates is quite important for satellite-derived flood prediction and landslide detection. As for the three post-real-time satellite products used in their study, CMORPH and PERSIANN are made available 18 h and 2 days following real time, respectively, while 3B42-V7 has about 2 months latency. In practice, in terms of the basin-scale heavy precipitation events, hydrologic users would prefer to see the comparative results between those near-real-time satellite rainfall products, such as the latest QMORPH (a variation on CMORPH), Global Satellite Mapping of Precipitation in near–real time (GSMaP_NRT), 3B42-RT, and the uncalibrated 3B42-RT (another purely satellite-derived real-time TMPA dataset as an additional field in 3B42-RT). The basin-scale assessment of Yong et al. (2013) has revealed that the employment of the historical gauge data and the smooth-fill scheme in the 3B42-RT’s monthly climatological calibration algorithm (CCA) could homogenize the highly variable local rainstorm characteristics despite that the CCA can effectively reduce the substantial negative biases of the TMPA system. Our evaluations revealed that this characteristic directly results in an increase of the dynamic range with the daily 3B42-RT, particularly at higher rain rates. That means the daily or 3-hourly RMSEs of 3B42-RT became worse after the CCA correction. This finding was further confirmed by the CPC gauge-based validation over the globe (Yong et al. 2015). From Table 4 in the original paper, one can see that the RMSE values of 3B42-V7 are lower than those of 3B42-RT for heavy precipitation events. In fact, the monthly CCA tends to decrease the TMPA’s systematic bias at long temporal scales but increase the random errors at short duration. Because of the dynamic balance between systematic and random errors caused by the CCA, we speculate that the RMSE values of uncalibrated 3B42-RT might also be lower than 3B42-RT in this Italian basin. Inclusion of the real-time uncalibrated 3B42-RT dataset in the Mei et al. (2014) assessments would have reached out to a much broader audience of users, thus providing essential and comprehensive information in understanding the error characteristics of current satellite-based precipitation in monitoring the storm events over different basin scales.

Results in Table 4 of Mei et al. (2014) show that the 3B42-RT has a relatively higher correlation coefficient (CC) value (0.34) than 3B42-V7 (0.04) in the September–December period. The authors explained that during cold periods the monthly climatological gauge-adjusted feature from the real-time 3B42-RT has a higher degree of influence on event-based precipitation accumulation compared to the actual monthly gauge adjustments from the post-real-time 3B42-V7. But, such an explanation is a speculation that is not very convincing to the readers. Our experience is that currently both passive microwave and geostationary infrared retrievals are not necessarily reliable over snow-covered or icy land surfaces, especially for complex terrain regions (Yong et al. 2015). The statistics in the original paper’s Table 4 are just regarded as a reflection of the large uncertainties for TRMM-based multisatellite precipitation estimates during the cold season. Yong et al. (2012) showed that several random snow events might significantly change those statistical values. In practice, we can pretty much guarantee that the CCA made the real-time estimates statistically closer to the post-real-time product, but not to the ground gauge observations. By comparing the 3B42-V7 and the two 3B42-RT datasets before and after the climatological calibration is applied, some potential causes for the results can be clearly revealed.

The authors claimed that the CMORPH and PERSIANN algorithms seem to lack accuracy for heavy precipitation events over complex terrain, while two 3B42 products can provide a more accurate estimation of the three storm parameters (rainfall accumulation, maximum rainfall rate, and duration), which was presented in Figs. 7 and 5 of Mei et al. (2014). We note that these findings counter those found over some Chinese regions. The assessment results of Shen et al. (2010) indicated that CMORPH exhibits the best performance in depicting the spatial pattern and temporal variations of precipitation (with the highest correlation and smallest RMSE differences). Furthermore, they found that the difference between the correlation/RMSE for the CMORPH and those for other products decreases with an increase of the time scales. Thus, they concluded that a Lagrangian approach like CMORPH can work effectively at a higher temporal resolution (e.g., 3 h or even 1 h). Additionally, our preliminarily evaluation over the upper–middle Huai River basin with complex terrain in mideastern China also showed that CMORPH generally has a better skill during the summer relative to 3B42-RT and 3B42-V7. We also found that the newly released PERSIANN Climate Data Record (PERSIANN-CDR; Ashouri et al. 2015) exhibits a significant improvement on previous versions in our study area. Therefore, we believe that the TMPA algorithm is not always superior to the CMORPH and PERSIANN algorithms in estimating heavy precipitation events over different regions. This result asserted in Mei et al. (2014) might be only limited to the eastern Italian Alps or similar mountainous basins, as discussed by the authors in their conclusions.

In summary, we expect that the above discussion can give hydrologic users a better understanding of the global error characteristics and calibration algorithms associated with currently available satellite precipitation estimates.