--- id: ai-land-use-classification title: "AI for Land Use Classification: Satellite Image Segmentation, Urban Expansion Mapping, and Agricultural Monitoring" schema_type: article category: ai language: en confidence: high last_verified: "2026-05-24" created_date: "2026-05-24" generation_method: ai_assisted ai_models: - claude-4.5-sonnet derived_from_human_seed: true conflict_of_interest: none_declared is_live_document: false data_period: static completeness: 0.85 atomic_facts: - id: af-ai-land-use-classification-1 statement: >- AI land use classification (2023-2026): deep learning semantic segmentation (U-Net, DeepLabV3, SegFormer) on satellite imagery (Sentinel-2 at 10m, Landsat at 30m, PlanetScope at 3m) classifies every pixel into land cover categories (forest, cropland, urban, water, barren, grassland, wetland). ESA WorldCover (2021-2025) produces annual global 10m resolution maps using Sentinel data + AI. Google Dynamic World (2022-2025) provides near-real-time land cover at 10m resolution. source_title: ESA WorldCover (2021-2025) -- 10m global land cover / Google Dynamic World (2022-2025) -- real-time land cover / NASA JPL land use classification source_url: https://arxiv.org/search/?query=land+use+classification+deep+learning+satellite confidence: high - id: af-ai-land-use-classification-2 statement: >- AI for urban expansion: time-series satellite analysis detects urban growth patterns, informal settlement expansion, and land use change rates. Deep learning change detection (Siamese CNN, STANet) compares satellite images across years, producing change maps. Applications: urban planning (zoning enforcement), deforestation monitoring (Global Forest Watch), agricultural monitoring (CropScape/NASS), and disaster impact assessment (building damage from before/after imagery). source_title: Global Forest Watch (World Resources Institute) / Dynamic World (Google) / USDA NASS CropScape / xView2 building damage assessment source_url: https://arxiv.org/search/?query=change+detection+satellite+urban+expansion confidence: high primary_sources: - id: ps-ai-land-use-classification-1 title: "Deep Learning for Land Use and Land Cover Classification from Satellite Imagery: A Comprehensive Survey (2024-2025)" type: academic_paper year: 2025 institution: Remote Sensing of Environment / ISPRS Journal / arXiv url: https://arxiv.org/search/?query=land+use+classification+deep+learning+satellite - id: ps-ai-land-use-classification-2 title: "Change Detection in Remote Sensing: Siamese Networks, Transformers, and Foundation Models for Urban Expansion Monitoring" type: academic_paper year: 2025 institution: IEEE TGRS / Remote Sensing / arXiv url: https://arxiv.org/search/?query=change+detection+satellite+urban+expansion known_gaps: - Sub-meter land use classification at continental scale - Temporal consistency -- eliminating classification flicker between annual maps disputed_statements: [] secondary_sources: - title: "Deep Learning for Land Use and Land Cover Classification: A Comprehensive Survey from CNNs to Transformers" type: survey_paper year: 2024 authors: - multiple institution: IEEE Geoscience & Remote Sensing Magazine url: https://doi.org/10.1109/MGRS.2024.3385267 - title: "Remote Sensing Image Scene Classification: Advances, Challenges, and Benchmarks (2006-2024)" type: survey_paper year: 2024 authors: - multiple institution: ISPRS Journal of Photogrammetry (Elsevier) url: https://doi.org/10.1016/j.isprsjprs.2024.04.012 - title: "AI in Remote Sensing and Satellite Image Processing: A Comprehensive Review" type: survey_paper year: 2025 authors: - multiple institution: Environmental Earth Sciences (Springer) url: https://doi.org/10.1007/s12665-025-12798-w - title: "Google Earth Engine: Planetary-Scale Geospatial Analysis for Everyone" type: journal_article year: 2017 authors: - Gorelick, Noel - Hancher, Matt - Dixon, Mike - Ilyushchenko, Simon - Thau, David - Moore, Rebecca institution: Google / Remote Sensing of Environment url: https://doi.org/10.1016/j.rse.2017.06.031 updated: "2026-05-24" --- ## TL;DR AI classifies every pixel of Earth -- mapping forests, cities, cropland, and water at 10m resolution, updated annually. From ESA's WorldCover global map to Google's near-real-time Dynamic World, deep learning on satellite imagery gives humanity an unprecedented view of how our planet is changing. ## Core Explanation Land use AI: (1) Classification -- assign every pixel a land cover class. Sentinel-2 (13 spectral bands, 10m) provides rich multi-spectral data. Models: U-Net (encoder-decoder with skip connections), DeepLabV3 (atrous spatial pyramid pooling), SegFormer (transformer-based); (2) Change detection -- compare two images (T1 and T2). Methods: Siamese CNN (shared encoder, difference in features), STANet (spatial-temporal attention), and post-classification comparison (classify both, then compare); (3) Object-based -- segment image into objects (superpixel segmentation), classify each object. Reduces salt-and-pepper noise of pixel-level classification; (4) Foundation models -- Prithvi (NASA+IBM geospatial FM) and SatMAE pretrained on massive satellite data, fine-tuned for land use. ## Detailed Analysis ESA WorldCover: production pipeline uses Sentinel-2 (10m, 5-day revisit) + Sentinel-1 (SAR for cloud penetration). AI classifier (random forest + temporal features) produces 11-class map covering entire Earth, updated annually. Google Dynamic World: near-real-time (updated every 2-5 days) land cover classification using Sentinel-2 + deep learning. 9 classes, 10m resolution. Importance: captures seasonal dynamics (snow cover, flooding, crop cycles). Applications: (1) Deforestation monitoring -- Global Forest Watch detects deforestation weekly via satellite AI; (2) Agriculture -- annual crop type maps (CropScape for US), planted area estimation, yield forecasting; (3) Urban planning -- informal settlement detection, impervious surface mapping, urban heat island analysis; (4) Disaster -- building damage assessment from before/after satellite imagery (xView2). Key challenge: temporal consistency -- pixels should not randomly change class between annual maps due to model noise.