Predicting within?field cotton yields using publicly available datasets and machine learning

Early detection of within-field yield variability for high-value commodity crops, such as cotton (Gossypium spp.), offers growers potential to improve decision-making, optimize yields, and increase profits. Over recent years, publicly available datasets have become increasingly at a resolution where prediction is possible. However, the viability using these with machine learning predict lint key growth stages are largely unknown. This study was conducted on two fields, located near Mungindi, New South Wales, Australia. Three years data, soil, elevation, rainfall, Landsat imagery were collected from each field. A total 12 models created using: (a) algorithms: random forest (RF) gradient boosting machines (GBM); (b) three stages: squaring, flowering, boll-fill; (c) different amounts variables: all variables optimal determined by recursive feature elimination (RFE). Results showed strong agreement between predicted observed yields flowering boll-fill when more information available. At boll-fill, root mean square error (RMSE) ranged 0.15 0.20 t ha?1 Lin's concordance correlation coefficient (LCCC) 0.50 0.66, RF providing superior results in most cases. Models RFE provided similar compared variables, allowing greater model accuracy targeted sampling. Overall, findings indicate significant guide decision-making in-season.