Advances in satellite technology have led to the development of many remote-sensing algorithms to estimate precipitation at quasi-global scales. A number of satellite precipitation products are provided at high spatial and temporal resolutions that are suitable for short-term hydrologic applications. Several coordinated validation activities have been established to evaluate the accuracy of satellite precipitation. Traditional verification measures summarize pixel-to-pixel differences between observation and estimates. Object-based verification methods, however, extend pixel based validation to address errors related to spatial patterns and storm structure, such as the shape, volume, and distribution of precipitation rain-objects.
In this investigation, a 2D watershed segmentation technique is used to identify rain storm objects and is further adopted in a hybrid verification framework to diagnose the storm-scale rainfall objects from both satellite-based precipitation estimates and ground observations (radar estimates). Five key scores are identified in the objective-based verification framework, including false alarm ratio, missing ratio, maximum of total interest, equal weight and weighted summation of total interest. These scores indicate the performance of satellite estimates with features extracted from the segmented storm objects.
The proposed object-based verification framework was used to evaluate PERSIANN, PERSIANN-CCS, CMORPH, 3B42RT against NOAA stage IV MPE multi-sensor composite rain analysis. All estimates are evaluated at 0.25°x0.25° daily-scale in summer 2008 over the continental United States (CONUS). The five final scores for each precipitation product are compared with the median of maximum interest (MMI) of the Method for Object-Based Diagnostic Evaluation (MODE). The results show PERSIANN and CMORPH outperform 3B42RT and PERSIANN-CCS. Different satellite products presented distinct features of precipitation. For example, the sizes of rainfall objects acquired from PERSIANN-CCS were smaller, while rainfall objects obtained from PERSIANN typically cover larger area. It is concluded that the discrepancies between various satellite precipitation estimates can be identified through the proposed verification framework.