Prediction of flood flows based on the combined solution of Google Earth Engine data and artificial intelligence models

Principal component analysis, Sarzab watershed, Bayesian linear regression, artificial neural network  Flooding is one of the unfortunate events in nature, which, if not predicted in time, can cause severe financial and life damages. Therefore, estimating flood peak discharge is a crucial issue in hydrological studies today. However, research on the use of remote sensing tools for predicting,  modeling, and managing floods in most of the country's watersheds has received less attention. This research aims to determine the factors affecting the flood flow discharge of the Sarbaz watershed and also evaluate the role of artificial intelligence methods, including the artificial neural network (ANN) model, to predict the flood flow of this watershed. In this research, rainfall data, soil moisture and temperature, evapotranspiration, base water flow, and Enhanced Vegetation Index (EVI) from the Google Earth Engine system, as well as observational data of flood event discharges of the studied area from 1380-1401, were used. Principal component analysis was then used to determine the factors affecting flood discharge. These factors were modeled using Bayesian linear regression to implement the artificial neural network models. Finally, artificial neural network modeling was performed for flood flow analysis. The results showed that the total rainfall of the current day and the previous day, soil moisture at a depth of 0 to 10 cm of the previous day, and soil temperature of the previous day were selected as the most appropriate input patterns for modeling. The artificial neural network designed had an efficiency factor of 0.90, a determination coefficient (R2) of 0.89, and a root mean square error (RMSE) of 50.37 for the training stage. For the validation stage, it had an efficiency factor of 0.76, an R2 of 0.83, and an RMSE of 46.86, demonstrating a good ability to estimate peak flood discharge. The results indicated that the calibrated model for predicting flood flow using remote sensing data is practical and has acceptable accuracy. Therefore, it can be an efficient tool to help managers predict floods on time and reduce the resulting damages.