Near-Real Time Crop Progress Estimation using Remote Sensing in Regions without Ground Survey Data

In this study a method for near-real time (nRT) crop progress estimation (CPE) for data-poor regions - those without large scale crop surveys - is proposed. The method utilizes Long Short-Term Memory and is pre-trained on USDA corn crop progress data for the US Midwest using weather and MODIS-derived vegetation index products. Performance of the method is evaluated in different growing zones of Argentina, a major corn exporter, using Bolsa de Cereales corn crop progress data. To establish how the proposed nRT CPE method would perform in regions any ground survey data, evaluation is conducted without prior access to or fine-tuning on Argentinian ground truth crop progress. Initial results from a single growing zone in Argentina indicate that pre-training an LSTM-based nRT CPE method using data from regions with high ground truth data availability may translate to effective nRT CPE in regions where ground survey data are unavailable.