用卫星数据和机器学习提升孟加拉国干旱分级精度
Enhanced Drought Analysis in Bangladesh: A Machine Learning Approach for Severity Classification Using Satellite Data
- 融合气象与土壤湿度数据，用聚类+分类模型识别干旱
- 覆盖38个地区，能区分不同严重程度的干旱情况
- 适合关注农业风险与气候适应的政策制定者
干旱对孟加拉国构成普遍环境挑战，受地理与人为因素影响，威胁农业、社会经济稳定与粮食安全。传统干旱指数如标准降水指数（SPI）和帕尔默干旱指数（PDSI）常忽略土壤湿度与温度等关键因子，分辨率有限。现有机器学习模型在孟加拉国干旱预测中应用不足，缺乏跨38个地区的多源卫星数据整合。为此，本文提出一种基于卫星数据的机器学习框架，用于分类该国38个地区的干旱程度。采用K-means与贝叶斯高斯混合等无监督算法进行聚类，再结合KNN、随机森林、决策树与朴素贝叶斯等分类模型，融合2012–2024年气象数据（湿度、土壤湿度、温度）。该方法有效实现干旱严重程度分级，且揭示区域间干旱脆弱性差异显著，体现机器学习在识别与预测干旱中的潜力。
原文摘要 · Abstract (English)
Drought poses a pervasive environmental challenge in Bangladesh, impacting agriculture, socio-economic stability, and food security due to its unique geographic and anthropogenic vulnerabilities. Traditional drought indices, such as the Standardized Precipitation Index (SPI) and Palmer Drought Severity Index (PDSI), often overlook crucial factors like soil moisture and temperature, limiting their resolution. Moreover, current machine learning models applied to drought prediction have been underexplored in the context of Bangladesh, lacking a comprehensive integration of satellite data across multiple districts. To address these gaps, we propose a satellite data-driven machine learning framework to classify drought across 38 districts of Bangladesh. Using unsupervised algorithms like K-means and Bayesian Gaussian Mixture for clustering, followed by classification models such as KNN, Random Forest, Decision Tree, and Naive Bayes, the framework integrates weather data (humidity, soil moisture, temperature) from 2012-2024. This approach successfully classifies drought severity into different levels. However, it shows significant variabilities in drought vulnerabilities across regions which highlights the aptitude of machine learning models in terms of identifying and predicting drought conditions.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。