An Analysis of Rainfall Measurements over Different Spatio-Temporal Scales and Potential Implications for Uncertainty in Satellite Data Validation

Abstract

The work presented herein deals with the problems of inter-comparison of precipitation values at 8 km × 8 km and 4 km × 4 km scales (herein often referred to as 8 km and 4 km pixel or scale) that represent the approximate instantaneous field of view of thermal infrared measurements from geostationary satellites like Kalpana, Geostationary Meteorological Satellite (GMS), etc. A number of temporal, spatial and spatiotemporal precipitation series at 4 and 8 km scales from TOGA-COARE radar observations are subjected to autocorrelation analysis to study precipitation variability. It was found that autocorrelation of precipitation measurements drop to 0.45 and 0.58 in 10 minutes time for 4 and 8 km pixel, respectively. Similarly spatial autocorrelation drops to 0.64 and 0.61 for 4 km pixel with 4 km displacement and 8 km pixel with 8 km displacement, respectively. The drop in autocorrelation in the spatiotemporal series is shown to be even more rapid. The root mean square fractional error for precipitation shows a value of 0.63 to 0.90 and 0.71 to 0.94 for 10 to 60 minute time lag for 4 and 8 km pixel, respectively. The autocorrelation analysis of precipitation underlines the need for precise geolocation and time measurement at each pixel for meaningful intercomparison. We have also presented results on the precipitation comparison when two observations have different spatial scales. We analyzed the probability distribution P (r_i|R), of precipitation measurements r_i at 2 km scale within a 8 km pixel for a given pixel averaged rain rate R. It was found that there is only 40% probability that a 2 and 8 km pixel averaged rain will match with a percentage error less than 50% of R. We also presented a Monte Carlo simulated precipitation comparison measured at two different scales 8 and 2 km. This analysis highlights the importance of the comparable scales in precipitation validation.