A fuzzy-based flood warning system using 19-year remote sensing time series data in the Google Earth Engine cloud platform

Abstract

Many Flood Warning Systems (FWS) have been developed to date to reduce flood risk and properly manage this natural disaster. This study presents a novel method to create an FWS based on anomaly detection in remote sensing climate data from western Lorestan, Iran, from 2001-to 2019. To this end, the monthly time series of climate products related to floods (e.g., precipitation, soil moisture, soil and air temperatures, vegetation, snow, and evapotranspiration) were first processed in Google Earth Engine (GEE). Then, three algorithms – Median-Interquartile range (M−IQR), Multi-Layer Perceptron (MLP), and Recurrent Neural Network (RNN) – were applied to detect anomalies in the time series of each parameter. Finally, a rule-based Fuzzy Inference System (FIS) was designed to estimate the potential of floods per month by establishing the relationship between the observed anomalies and the occurrence of floods. The results of the proposed Fuzzy-based Flood Warning System (FFWS) using all three anomaly detection methods accurately showed the very high potential for floods in March and April 2019 (i.e., actual flood events). Two other floods occurred in October 2015 and April 2016 were also considered for further evaluation of the proposed method. The results indicated that the RNN method achieved the highest performance in flood forecasting with the overall accuracy and Kappa coefficient of 93.85% and 0.93, respectively. Moreover, the potential of floods at the beginning of 2019 (i.e., January and February) was also high, although not to the extent as in March and April, indicating that the proposed method correctly identified the potential of flooding in later months and can thus provide a warning to help mitigate the impact of flood damage.