Behind the early warning: Improving impact-based forecasting of riverine floods in Malawi using passive microwave remote sensing

Abstract: This thesis investigates whether freely available, coarse-resolution, Passive Microwave Remote Sensing (PMRS) data (37 GHz) can be effectively used for early warning systems for floods in Malawi. The Shire River Basin in Chikwawa and the smaller-scale North Rukuru River Basin in Karonga were studied using two alternative, ratio-based satellite indices that make use of the signal difference between wet and dry pixel cells: The m index and the rcmc index. The m index is directly related to the rcm index, introduced by Brakenridge et al. (2007), and divides the brightness temperature in a relatively stable, dry calibration cell by the brightness temperature measured in the cell with the river. The rcmc index is an adaptation of this, and uses an additional wet calibration cell. It was investigated whether these indices could aid in the detection and forecasting of flood events and their magnitude. The findings pertaining to detection skill showed that at both study sites, rcmc and m detected a similar seasonality to the observed discharge hydrographs, as long as the downstream virtual gauging station was located at a sufficient distance from a large water body. A regression analysis showed that the indices’ relationship with observed discharge had a moderately strong, positive correlation in Chikwawa, but not in Karonga. Flood occurrence detection skill was assessed using an impact database. A flood threshold corresponding to a return period of 5 years was determined to see if the indices could simulate historical flood events. Both indices did not detect the majority of registered floods, which is likely a consequence of the method used to determine the trigger threshold. There were no upstream virtual gauging stations present that had a sufficient lag time with the downstream satellite signal. A possible forecasting system using merely the downstream satellite signals was shown to have a sufficient accuracy at a lead time of up to nearly 3 days, although in an operational setting, the forecasts would not reach the calculated trigger threshold at this lead time. Overall, the PMRS-model showed a better performance in Chikwawa when compared to the global runoff model GloFAS. As it also does not require extensive input data when used as an Early Warning System (EWS), as many smaller-scale EWS do, we suggest that when perfected, the PMRS-method is implemented in a coupled EWS solution, including a PMRS-model, a global forecasting model and a more detailed national model. This would offer early warnings in data-scarce regions and at a variety of lead times. In order for this to be effective, we suggest that more research be done on correctly setting the trigger threshold, and into the potential spatial interpretation of rcmc.