GeoAI-Driven Wetland Change Analysis in the Sangamon River Watershed (2000–2025): A Comparative Assessment of Machine Learning and Deep Learning Approaches.

This study performs a spatiotemporal wetland change analysis for the Sangamon River Watershed, Illinois, using Landsat 5 TM between 2000–2008, Sentinel-2 Surface Reflectance, Sentinel-1 Synthetic Aperture Radar (SAR) between 2017–2025, and terrain data through an integrated machine learning (ML) and deep learning (DL) frameworks in Google Earth Engine (GEE), Google Colab (Python3) and ArcGIS Pro 3.6. We conducted a comparative assessment of DL Dense Neural Networks (DNN) and U-Net semantic segmentation, as well as three ML models including Random Forest (RF), Gradient Tree Boosting (GTB), and Support Vector Machine (SVM) through pixel-based and object-based image analysis (OBIA) methods. Reclassified National Land Cover Database (NLCD) datasets were used for training and validation using stratified random sampling for five categories namely Wetlands (1), Forest (2), Agriculture/ Grassland/ Barren land (3), Urban/Developed (4) and Water (5).本研究针对美国伊利诺伊州萨加莫尔河汇水区开展时空湿地变化分析，所用数据源包括2000–2008年的Landsat 5 TM数据、2017–2025年的Sentinel-2地表反射率数据、Sentinel-1合成孔径雷达（Synthetic Aperture Radar，SAR）数据及地形数据；研究依托谷歌地球引擎（Google Earth Engine，GEE）、谷歌Colab（Python3环境）与ArcGIS Pro 3.6平台，结合集成机器学习（Machine Learning，ML）与深度学习（Deep Learning，DL）框架完成分析流程。本研究对比评估了两类深度学习模型：密集神经网络（Dense Neural Networks，DNN）与U-Net语义分割模型，同时纳入随机森林（Random Forest，RF）、梯度树提升（Gradient Tree Boosting，GTB）与支持向量机（Support Vector Machine，SVM）三类机器学习模型，分析方法涵盖基于像素与面向对象影像分析（Object-Based Image Analysis，OBIA）两种路径。本研究采用重分类后的美国国家土地覆盖数据库（National Land Cover Database，NLCD）数据集作为训练与验证数据源，通过分层随机抽样对5类地物进行标注，具体类别为：湿地（1类）、林地（2类）、农业/草地/裸地（3类）、城市/建成区（4类）与水体（5类）。