Spatial correlation in precipitation trends in the Brazilian Amazon

Abstract

A geostatistical analysis of variables derived from Amazon daily precipitation records (trends in annual precipitation totals, trends in annual maximum precipitation accumulated over 1–5 days, trend in length of dry spell, trend in number of wet days per year) gave results that are consistent with those previously reported. Averaged over the Brazilian Amazon region as a whole, trends in annual maximum precipitations were slightly negative, the trend in the length of dry spell was slightly positive, and the trend in the number of wet days in the year was slightly negative. For trends in annual maximum precipitation accumulated over 1–5 days, spatial correlation between trends was found to extend up to a distance equivalent to at least half a degree of latitude or longitude, with some evidence of anisotropic correlation. Time trends in annual precipitation were found to be spatially correlated up to at least ten degrees of separation, in both W–E and S–N directions. Anisotropic spatial correlation was strongly evident in time trends in length of dry spell with much stronger evidence of spatial correlation in the W–E direction, extending up to at least five degrees of separation, than in the S–N. Because the time trends analyzed are shown to be spatially correlated, it is argued that methods at present widely used to test the statistical significance of climate trends over time lead to erroneous conclusions if spatial correlation is ignored, because records from different sites are assumed to be statistically independent.