Forecasting Inundation Extents using REOF analysis (FIER) over Lower Mekong Basin

Abstract:

Forecasted inundation extent is vital information for communities and responders to reduce property damage and save lives. However, existing flood forecasting systems give little indication to spatial inundation extents. Although hydrodynamic models can transform discharges from in-situ gauges or rainfall-runoff models to distributed inundation extents, its heavy computational burden, especially for a large-scale forecasting framework, could affect forecast lead-time. A non-modeling approach such as Height Above Nearest Drainage (HAND) that employs a planar approximation has also been used but may be less skillful over relatively flat terrain, such as of the Lower Mekong (LM).

A recent innovative approach of forecasting inundation extents was developed and demonstrated over the Tonle Sap Lake (TSL) floodplains in Cambodia (Chang et al., RSE, 2020), utilizing regression analysis between temporal patterns extracted from a timeseries stack of historical Sentinel-1 Synthetic Aperture Radar (SAR) images using Rotated Empirical Orthogonal Function (REOF) analysis and historical discharge or water level data (Forecasting Inundation Extents using ROEF – FIER). Forecasted discharges obtained from a rainfall-runoff model, along with the corresponding temporal patterns obtained from the regression model is integrated with REOF-extracted spatial patterns of SAR images to generate synthesized SAR intensity images from which forecasted inundation extents can be produced with water classification method.

Here, we expand the study region to the floodplains in LM and apply FIER to generate forecasted inundation extents using (1) forecasted streamflows along the Mekong mainstem from HYPE model developed over the Greater Mekong region (Du et al., J Hydrol, 2020), (2) forecasted TSL levels using ENSO Index (Chang et al., RSE, 2020), (3) forecasted streamflows over the Mekong Delta using the Ensemble Learning Regression (Kim et al., RS, 2019), which are correlated with specific extracted temporal patterns. The pseudo-forecasted inundation extents are validated with those from historical Sentinel-1 SAR and MODIS imagery. It is expected that FIER can complement the existing flood forecasting system in the region for more effective flood risk mitigation and preparation by predicting spatial flood hazard.