MODIS-based Crop Classification and Yield Maps for the Paddy Rice Regions of Interest in South and North Korea

This study develops and openly provides a moderate-resolution geospatial dataset of paddy rice areas on the Korean Peninsula, demonstrating its application for spatially explicit classification and yield simulation. Using Moderate Resolution Imaging Spectroradiometer (MODIS) imagery from the Terra satellite, we generated raster and vector maps of paddy fields for 15 regions across South Korea (Cheorwon, Gimje, Icheon, Naju, Paju, Jeonbuk, and Jeonnam) and North Korea (South/North Hwanghae, South Pyeongan, and Pyeongyang) over the 2010-2022 period. Threshold (T)-based rules (elevation ≤800 m; slope ≤7°; normalized difference vegetation index [NDVI] max/ min, 0.55/0.40; NDVI increase rate, 4.0; May-June land surface water index [LSWI]+T > NDVI, LSWI T, 0.17) achieved an overall classification accuracy of 78.8% (κ=0.512), with paddy-user and paddy-producer accuracies of 69.7% and 64.1%, and non‑paddy accuracies of 82.7% and 86.1%. We then integrated a machine learning regression model into the remote sensing-integrated crop model to simulate leaf area index and yield. Simulated 2022 yields averaged 6.8±0.5 t ha-1 in major South Korean areas and 5.9±0.4 t ha<sup>-1</sup> in North Korean plains, with an inter-regional difference of 1.0 to 1.2 t ha<sup>-1</sup>. Our results confirm that intermediate-resolution remote sensing data combined with simple threshold rules and machine learning-enhanced process modeling can effectively map paddy fields and predict yields. The resulting geospatial datasets, provided with complete metadata in accordance with <i>GEO DATA</i> standards, are openly available for reuse and extension to diverse crops and agroecological contexts.