Machine Learning-Based Drought Classification Using Meteorological Data: Toward Smarter Environmental Models for Site Exploration

Abstract

Drought presents a significant challenge to sustainable water management, agriculture, and geotechnical site assessment. Accurate and timely classification of drought severity is essential for anticipating environmental changes that impact soil behavior, moisture conditions, and subsurface dynamics. This study introduces a machine learning framework utilizing the Random Forest algorithm to classify drought intensity based on multi-year meteorological datasets. The dataset, comprising over 3 million records and 18 meteorological features, includes variables such as temperature, humidity, wind speed, and precipitation, serving as environmental indicators for drought conditions. The model predicts drought severity across six discrete classes (0–5), evaluated using precision, recall, F1 score, and confusion matrices. While achieving an overall accuracy of 75%, the model reveals performance challenges in detecting minority drought classes, underscoring the importance of class balancing and feature selection. Beyond drought prediction, this work supports the integration of environmental intelligence into geotechnical and site exploration processes. By informing early-stage terrain assessment with climate-responsive data, the framework lays groundwork for adaptive modeling in subsurface analysis, simulation, and exploration planning. The study demonstrates the value of machine learning in developing scalable, data-driven environmental models that enhance decision-making in both civil and environmental engineering applications.