Assessing daily and seasonal satellite rainfall estimates using local gauges for the anomalous 2012 monsoon season in the African East Sahel

Abstract

Satellite rainfall estimates (SRFEs) are essential for characterizing the seasonal delivery of monsoon rainfall to sub-Saharan Africa. Such data are particularly robust when composited over sufficiently large areas of the Earth's surface (≥2500 km²), and sufficiently long time periods (≥1 month). However, the application of SRFEs to early warning flash-flood and drought-monitoring systems at the hamlet level is less well documented, particularly for areas dominated by intense, small-scale convective storms, which require a higher spatial and temporal resolution than usually expected from standard SRFE products. This report assesses the potential of one of the primary contenders for such applications – the Climate Prediction Center (CPC) Rainfall Estimation Algorithm Version 2.0 (or RFE2) product – posing the specific challenge: How well does the RFE2 product perform on a site-by-site, storm-by-storm basis as a potential tool for early warning flood and drought applications in the East Sahel? Daily gridded (0.1° × 0.1°) RFE2 estimates are compared with daily time series from nine gauges provided by local agencies in the state of Gedaref, Sudan (area of gauge coverage ≈ 160 km × 160 km). Seasonal composites of multiple gauges and multiple RFE2 cells for 2012 (1 April to 31 October) agree quite well. However, daily time series of individual pairs of gauge data and corresponding RFE2 estimates show a remarkable lack of temporal coherence − typically only 10−20% of the daily variance of the RFE2 time series is associated with the gauge data. Moreover, low-pass filtered (30 day cut-off), smoothed versions of the daily RFE2 data do not conform well to the corresponding gauge distributions over the season, and fail to capture a seasonal bimodal rainfall pattern recorded by a cluster of rain gauges. Finally, there is a characteristic difference in the intensity-versus-frequency statistics of RFE2-produced events compared with corresponding gauge observations: RFE2 estimates imply that rain should have occurred on 70% of the days, whereby gauges recorded rainfall on only 35% of the days. On the other hand, extreme storms recorded by the gauge data are typically twice as strong as those estimated by the RFE2 algorithm. These results circumscribe specific limits on the RFE2 product for the purpose of site-by-site, storm-by-storm early warning monitoring of rainfall events, and for resolving details on the intra-seasonal modulation of rainfall patterns.

Acknowledgements

Brown University's International Advanced Research Institutes (BIARI) programme, under the aegis of Brown's Watson Institute for International Studies, facilitated the collaboration of the authors. We would like to thank the Mechanized Farming Corporation, the Sudan Meteorological Authority, and local station operators for providing the gauge data used in this report. Of course, the quasi-real-time satellite rainfall estimates (CPC-RFE2) archived on the web in GIS format by NOAA's CPC were essential for our project, along with the guidance of Pingping Xie, Wassila Thiaw, and Nick Novella. Of particular help regarding the historical data was NOAA's Howard Diamond, who helped us thread our way through the GHCN/WMO databases from NOAA's National Climate Data Center.