Integrating satellite data and machine learning algorithms into a flood prediction system

Abstract

The objective of this study is to develop an intelligent flood forecasting system based on the integration of satellite and ground-based hydrometeorological data using machine learning algorithms in the Orange visual analytical environment. The aim of the article is to improve the accuracy and efficiency of forecasting extreme hydrological events, as well as to simplify the process of building forecasting models through the use of an interface that does not require programming. The methodological basis of the study is the formation of a synthetic multivariate dataset combining satellite vegetation indices NDVI and LST temperature with meteorological parameters such as precipitation, temperature, humidity, and water level. Random forest, gradient boosting (XGBoost), and multilayer perceptron (MLP) algorithms were used to build and validate the models. All stages from data loading and pre-processing to visualization and interpretation of results are implemented in the Orange environment using cross-validation and feature significance assessment. The results obtained demonstrated high forecast accuracy (up to 94%), especially when using ensemble and deep models. The significance of satellite data is confirmed by analyzing the contribution of features to the final classification. The developed forecasting model can be adapted to various geographical conditions and integrated into existing monitoring and early warning platforms. The proposed approach has high applied value and demonstrates the potential for using modern big data analysis technologies and artificial intelligence methods in the tasks of reducing the risk of natural floods.