Developing High-Resolution Crop Maps for Major Crops in the European Union Based on Transductive Transfer Learning and Limited Ground Data

Precise and timely information on crop spatial distribution over large areas is paramount to agricultural monitoring, food security, policy development. Currently, automatically classifying types at a scale challenging due the scarcity of ground data. Although previous studies have indicated that transductive transfer learning (TTL) promising method address this problem, it performs poorly within regions where compositions phenology differ largely. Here we transferred random forest classifiers trained in limited with diversified growing conditions land covers rest study area data are scarce, more than 130,000 Sentinel-2 images processed using Google Earth Engine (GEE) platform. We established 10 m maps for four major crops (i.e., maize, rapeseed, winter, spring Triticeae crops) across European Union (EU) countries from 2018 2019. The final had high accuracy overall generally greater 0.89, user’s producer’s ranging 0.72 0.98. Moreover, resulting were consistent NUTS-2 level official statistics, R2 consistently 0.9. further analyzed rotation patterns found intervals these EU least one year. Maize was dominantly rotated winter or converted other following Rapeseed grown crops, whereas diversified. Red Edge Position (REP) Normalized Difference Yellow Index (NDYI) played significant roles classification EU. This highlights potential developed TTL extents labeled differences relatively large.