An improved error decomposition scheme for satellite-based precipitation products

Abstract

The conventional error decomposition schemes of satellite-based precipitation products (SPPs) suffer from significant drawbacks such as misrepresentation of hit bias (H) component and magnification of random error, which limits their accuracy in deducing reliable conclusions. We propose a novel error decomposition technique by disintegrating the total bias (TB) in SPPs into four different independent components - over-hit (OH), under-hit (–UH), missed (–M) and false (F) precipitation. Further, the systematic and random parts are identified by adopting a multiplicative error model. The efficacy of the proposed scheme is demonstrated by comparing six different SPPs over India with a high-resolution gridded observed reference rainfall dataset by the India Meteorological Department, at daily-time scale for 16 years (2001–2016). The results of the study highlight that the traditional hit component is mostly underestimated due to the cancelling of opposite natured OH and –UH bias components and hence is unreliable. Often, the magnitudes of OH and –UH are found to be higher than TB and H. The proposed OH and –UH bias accurately estimates the “intensity” error in satellite precipitation retrieval process and can be used by both the data producers and users for deriving reliable conclusions. The findings of the study will aid in a more reliable and accurate characterization of error in SPPs, which is a fundamental step in the development of uncertainty modelling and bias-reduction algorithms.