A Quantitative Assessment of the NESDIS Auto-Estimator

Abstract

A systematic evaluation of the National Environmental Satellite, Data and Information Service Auto-Estimator (A-E), a satellite-based algorithm for the estimation of rainfall, was performed for a 13-month period ending December 1998. This effort was intended to serve as an independent benchmark for the assessment of A-E products, as well as to provide a measure of satellite-based precipitation algorithms in general. It is important that the accuracy of the A-E be established, since this algorithm is currently being proposed as the successor to the operational Interactive Flash Flood Analyzer technique used by Satellite Analysis Branch forecasters at the National Centers for Environmental Prediction (NCEP). Furthermore, the results of this assessment will be used to determine whether these satellite-based estimates should be integrated into the current state of NCEP multisensor precipitation analyses. The NCEP stage III precipitation analysis was chosen as the primary ?ground truth? for this evaluation following a comparison of available radar-, gauge-, and multisensor-based estimates. This intracomparison demonstrated a good correspondence in the mean precipitation between the stage III and gauge-only datasets. This agreement was consistent for all skill measures evaluated. Radar estimates generally underforecasted the amount of precipitation relative to either the stage III or gauge-only analyses. This underestimation was most dramatic during the cool season when radar estimates were about 50% of the multisensor and gauge amounts. The results from this study show that the A-E exhibits high-amplitude month-to-month fluctuations in its representation of mean areal precipitation compared to those from bias-corrected, multisensor (stage III) and 24-h gauge-only analyses. Overall, the verification results indicate that the skill of the A-E was greatest over 24-h periods and lowest for 1-h periods of accumulations. The algorithm demonstrated little skill overall with the most skill occurring during the warm season (May?August). Moreover, the A-E skill scores displayed a high degree of variability from one accumulation period to the next across the entire suite of temporal and spatial scales evaluated in this study. This variability makes it difficult to place any confidence in the accuracy of precipitation estimates for operational use.