Using planetscope imagery to evaluate herbicide efficacy in maize (Zea mays) through post-application weed detection

Abstract Background Optimizing herbicide efficacy is increasingly critical due to the absence of new herbicide modes of action (MOA) and their widespread overuse. This study developed a satellite-based approach to map herbicide control failures for evaluating efficacy, based on the hypothesis that effective control reduces crop-weeds co-existence and thus spectral-spatial heterogeneity over time, whereas low efficacy increases it. Methods In controlled experimental maize plots (2022–2023), analysis of Unmanned Aerial Vehicle (UAV) multispectral imagery characterized weed-suppression dynamics following application of two herbicides with different MOA. Satellite imagery was processed to select gray-level co-occurrence matrix (GLCM) texture features sensitive to herbicide-induced pixel differences. Features selected were used to compare two satellite-based approaches for mapping herbicide control failures cover (%) across seven commercial maize fields: (i) a rule-based framework and (ii) UAV to satellite upscaling using a Random Forest model. Results Results showed that by 10 days after spraying (DAS), herbicide effects were uniformly expressed in both MOA, enabling separation between damaged and healthy weeds in UAV images. At the satellite scale, Near-Infrared (NIR)-based GLCM variance and mean were the most sensitive indicators of weed suppression. Using these features and Normalized Difference Vegetation Index (NDVI) from 10 and 14 DAS, a rule-based framework showed better performance compared to UAV to satellite upscaling, achieving within field precision and recall of 0.71 both, and between fields agreement with a coefficient of determination (R 2 ) of 0.97 and root mean square error (RMSE) of 1.86 (%). Conclusions These results demonstrate that rule-based GLCM framework provide an effective tool for evaluating herbicide control efficacy using PlanetScope satellite image.