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

Description

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).

Steps to reproduce

1- Run separate GEE code for each ML model into pixel-based and object based categories based on both temporal periods. 2- Bring it into ArcGIS Pro for Wetland Change Analysis. 2- For DNN model (L5) export training data as CSV tables for DL analysis in Google Colab. 3- For U-Net semantic segmentation model (S1/S2) export 64×64 image patches as TFRecord datasets DL for analysis in Google Colab.

Institutions

• University of Missouri System

  Missouri, Columbia

Categories

Licence