Do the Satellite-based Rainfall Estimates “truly-hit” the Gauge-based Observations?

Abstract:

Verification of satellite-based precipitation datasets with respect to observed datasets often involves identifying hit rainy events (observed rainy events well detected by satellite), false alarm events (observed non-rainy events falsely detected as rainy events in satellite dataset), missed rainy events (an observed rainy event not detected by satellite) and true-null events (non-rainy events in both observed and satellite datasets). The frequency of these identified events also well correlates with the error in the satellite rainfall retrieval process, wherein the first stage is the detection of rainy/non-rainy events and the second step is determining the magnitude of detected (hit) events. Although the frequency of hit events quantifies the discrepancies in the first stage of detection, however, it misses to provide the information about the rainy events correctly detected in both the stages, i.e., events detected in the first stage and their magnitude is accurately estimated in the second stage. We term these events correctly detected at both stages as “true-hit”. Ideally, the satellite should “truly-hit” the observed dataset. Additionally, it would be also beneficial to quantify that if not true hit then whether the satellite overestimates or underestimates the magnitude of rainfall events. Therefore, we propose to expand the hit events into over-hit (OH), true-hit (TH) and under-hit (UH) events based on the magnitude of detected events. TH represents the frequency of those hit events which have the same magnitude. OH (UH) represents the frequency of those hit events wherein the satellite dataset overestimates (underestimates) the observed rainfall. The significance of expanding the hit events into OH, TH and UH is depicted by using the Tropical Rainfall Measurement Mission (TRMM) satellite and India Meteorological Department (IMD) observed dataset over India. The results of the study show that out of 63% rainfall events detected by TRMM, 30 % of them over-estimate (OH), 10 % of them accurately estimate (TH) and 23 % of them under-estimate (UH) the IMD rainfall. Only 10% of the total events are “truly detected”. Such information will not only aid the retrieval algorithm developers but also provide significant information to the data-users about the accuracy of the dataset before their application.