Integrating satellite imagery and Machine Learning for infrastructure risk forecasting in Honduras

Abstract

Honduras and neighboring countries are known vulnerable territories to extreme weathers in terms of rainfalls and irregular areas with dense vegetaion, these factors along with the scarce access to studies, lack of economic resources and tabulated data makes it of particular interest to develop a method that helps assess risk infrastructure in areas of difficult access. This study integrates satellite imagery, AI and machine learning into training a model that helps establish an index of failure for given regions in Honduras, based on weather features and terrain characteristics that include vegetation indexes compared to real data about outages in a given period of time. With AUC = 0.857 this study shows a promising correlation between irregularity of terrain, temperature among other factors and power outages, however its Cross-validation AUC for this model 0.638 ± 0.229 highlights the need for further research with different training methods, more samples and potentially more features from the data sources.