Multi-temporal downscaling of daily to sub-daily streamflow for flash flood watersheds at ungauged stations using a hybrid framework

Abstract

Unprecedented flash floods (FFs) in urban regions have been rising due to the increasing intensity and magnitude of heavy rainfall resulting from human-induced climate and land-use changes. Modelling of FF along different spatiotemporal scales is extremely complex since FF models require multi-resolution forcing and observed information. In the absence of sub-daily, and multi-site streamflow data, multi-temporal downscaling (MTD) plays a crucial role for FF modeling. While a wide range of methods are available for the spatio-temporal downscaling of climate data, the applicability of the MTD strategy for streamflow has not been investigated yet. The current study proposed a MTD methodology for yielding the daily to sub-daily streamflow of gauged and ungauged stations using adaptive emulator modelling concepts. The proposed MTD framework for ungauged stations is a hybrid model that draws on conceptual and machine learning-based approaches to analyze catchment behavior. The study selected the Peachtree Creek watershed (United States) since it frequently experiences FFs. Results suggest that model-derived sub-daily streamflow had minimal uncertainty in capturing hydrological signatures and nearly 95% accuracy in predicting flow attributes over ungauged stations. The proposed framework can be useful for planning, mitigation, and management, where the fine resolution data is required.