Abstract

This presentation revisits the problem of forming a "better" hydroclimatological dataset through merging of existing reconstructions. This problem has existed since long and alternatives such as Triple Collocation (TC), amongst others, formulated. However, a simple extension of these alternatives to the case of precipitation, is not possible, as precipitation exhibits a mixed probability distribution and different treatments are necessary for the occurrence (or non-occurrence) and the actual precipitation amount. Here we present a markedly new take at this problem, revisiting the challenges involved, and proposing two new approaches, the first for reconstructing rainfall occurrence over time, and the second for reconstructing the rainfall amount on each day identified as "wet". The parent precipitation datasets that are merged include a reanalysis product (ERA5), a satellite product (IMERG) and another satellite product but one that is derived from retrieved soil moisture (SM2RAIN). The focus area is High Mountain Asia, for which measurement density is low but gauge measurements of precipitation are available for assessing performance. The results show that (i) the data triplet containing IMERG was the best performing among all merging products (ii) the merging method significantly enhanced dataset accuracy compared to the parent datasets; and, (iii) the SNR_opt merging approach (Kim et al, 2022) had the flexibility of estimating both low and heavy rainfall, with a reduced bias for extreme precipitation, using fewer inputs (two) compared to TC. These advantages were found to be especially notable for the high-mountain region in focus, given the significant variability and limited observational data available. While the focus here is precipitation, the conclusions drawn are extendable to other hydroclimatological datasets, especially soil moisture and evapotranspiration, for which ground measurements are scarce.

Kim, S., A. Sharma, Y.Y. Liu, and S.I. Young, Rethinking Satellite Data Merging: From Averaging to SNR Optimization. IEEE Transactions on Geoscience and Remote Sensing, 2022. 60.