Regional-scale evaluation of 14 satellite-based precipitation products in characterising extreme events and delineating rainfall thresholds for flood hazards

Abstract

Gridded satellite-based rainfall products have not been so far evaluated for flood hazards monitoring through empirical methods, especially over large areas. Therefore, the main aims of this study are (i) to assess the quality of satellite-based precipitation products for identifying extreme rainfall events able to produce flood events, and (ii) to evaluate the use of satellite-based precipitation products for creating rainfall thresholds to support decision making for issuing flood warnings. Eight products fully based on satellite data (i.e., uncorrected) and six gauge-corrected products were evaluated based on ground-based data obtained from 583 sub-daily rainfall gauges and considering a catalogue of 551 flood occurrences in the state of São Paulo, Brazil, for a period of five years (2015–2019). The gauged values were compared with the precipitation products for six different time steps (3 h, 6 h, 12 h, 1 d, 3 d, and 10 d) and considering rainfall-duration thresholds for six non-exceedance probabilities (5%, 10%, 20%, 30%, 40%, and 50%). Results show that all analysed products tend to large underestimate the extreme rainfall events (i.e., when and where flood events were registered), mainly for sub-daily scales, with the best results found for two uncorrected products (i.e., PDIR-Now and GPM + SM2RAIN) considering 10-days accumulated precipitation. Considerable underestimations were also identified for the rainfall thresholds delineated by the satellite-based products, with the best performances obtained by CHIRP V2.0 (uncorrected) and IMERG-F (corrected) considering tolerance levels of 20%. Based on our findings, the rainfall satellite-based products dataset, even less accurate than the ground-based observations, can be applied, when multi-daily accumulated data are considered, as an alternative source of data for determining precipitation thresholds in some regions that present low-density of RAIN gauges but not replace the gauged data in regions with a high-density of rain gauges with sub-daily data available