Synthetic hydrological data consistent with climate reanalysis to enable long-term hydrological modelling

High-density gauging station networks and complete hydrological time-series are needed to adequately model and manage water resources and assess the effects of climate change on hydrological processes. In data scarce regions however, remote sensing data has proven to be a viable alternative, but before the year 2000 satellite records often contain gaps or are not available at all.

We propose to create synthetic images of precipitation, temperature, evapotranspiration, and terrestrial water storage products to complete and extend past data availability to pre-satellite periods. Ideally, the synthetic images should be indistinguishable from real satellite acquisitions. The approach used is based on the relation between meteorological predictors and available satellite images, and the hypothesis that, under similar meteorological conditions, patterns of a particular process may be repeated over the years. Using ERA5 reanalysis data as meteorological predictor, a K-Nearest Neighbor algorithm associated with a process-specific similarity metric is applied to create synthetic images of the different satellite products.

The approach is tested on the Volta River Basin in West Africa, where water resources for millions of people are critically stressed by the effects of climate change. For calibration and validation, the synthetic images are fed to a spatially-distributed hydrological model (the mesoscale Hydrologic Model mHM). Their quality is assessed by their capacity to reproduce historical streamflow time series. This test phase allows improving the generation technique to obtain synthetic imagery that can be considered a reasonable approximate of unobserved processes consistent with the available climate data, and which will help improve modelling accuracy.

Keywords: Remote sensing, Climate reanalysis, Satellite time series, Hydrological modelling