A comparison of Amazon rainfall characteristics derived from TRMM, CMORPH and the Brazilian national rain gauge network

This paper compares four selected characteristics (mean annual rainfall; mean number of wet days in the year; mean number of days with 2mm or more of rain; and mean 95% quantile of daily rainfall) computed from the Brazilian rain gauge network in the Amazon region, and from the satellite-derived data sets TRMM 3B42 and CMORPH. Comparisons were made at 488 sites over the years 2003-2005, the period for which all three sources provide data, and rainfall characteristics were calculated only from those days in the year with data from all three sources. TRMM and rain gauges mean annual rainfalls were fairly similar, but CMORPH estimates were greater than either, although the statistical significance of the differences were greatly reduced when spatial correlation was allowed for. However, differences between the mean numbers of days with rain (and those with 2mm or more) calculated from the three sources, were very marked, and the differences usually persisted when the errors in the differences included the effects of spatial correlation. Mean 95% quantile of daily rainfall calculated from CMORPH and from rain gauges were shown to be very similar. Over all 488 sites, however, mean annual rainfall calculated from CMORPH was considerably greater than that calculated from rain gauge records, and this result was explained by an analysis of daily rainfalls that exceeded their 95% quantile. The paper emphasizes that, unless spatial correlation is allowed for, uncertainty in rainfall characteristics will be under-estimated, and the apparent statistical significance of differences between values obtained from alternative data sets will be over-estimated.