Integrating Random Forest and Gradient Boosting for Predictive Wildfire Analytics using Environmental and Satellite data

Wildfires are an existential threat in the modern era, so data science and machine learning tools are needed to combat this issue. This paper showcases a wildfire prediction system integrating satellite-based image analysis, machine learning, and environmental input parameters to produce accurate risk assessments. Specifically, this paper uses NASA Earth imagery and FIRMS data to collect and analyze real time atmospheric and environmental parameters. Using these parameters, a Random Forest classifier with 100 estimators coupled with an XD Gradient Boost Classifier, optimized using Research with 3-fold cross-validation across hyperparameters (estimators, learning rate, adept, subsample), is used to predict wildfire risk using a feature set from CSV-based environmental datasets. These features include temperature, wind speed, humidity, FFMC, DMC, DC, ISI, rainfall, and more. Risk labeling is based on a threshold of area burned greater than 10 hectares. Furthermore, the tool can generate scatterplots across all numerical variable pairs, correlation heatmaps, and a comprehensive PDF report compiling the entire project. The system also fetches and displays real-time NASA satellite imagery at 0.1° spatial resolution in 512×512 tile format to visually assess wildfire zones. Moreover, the GUI is built with Tainter allowing users to input values, toggle between high- and low-risk presets, and interactively explore model insights. Experimental testing with this tool shows 85.3% accuracy using the Random Forest classifier, making the system a practical solution for wildfire management in cloud-based deployments.