Use of Drone Technology for Early Detection of Crop Stress

Abstract

Abiotic stressors such as drought, nitrogen deficiency, and heat stress that are induced by climate are increasingly threatening crop productivity across the world, but traditional field monitoring systems do not detect these stressors until their effects are seen, when it is too late and the damage to yield has been done. This paper fills this gap of knowledge by designing and testing a drone-based, multisensor fusion system to detect early signs of abiotic stress in maize during two growing seasons. The high-resolution images were obtained with a hexacopter carrying synchronized multispectral, thermal infrared, and RGB cameras at 24 controlled plots that were drought-stricken, nitrogen-deficient, heat-stressed, and healthy. An original Crop Stress Fusion Index was developed that combines the differentials of canopy temperature with NDVI, and a random forest classifier was developed that incorporates spectral, thermal, and textural measures to determine the presence and type of stress. The fusion model had a classification accuracy of 94.6 percent and AUC-ROC of 0.972, which was far much better than single-sensor models, which had 79.5 percent and 74.2 percent in multispectral only and thermal only, respectively. The intervention lead times were predicted with a root mean square error of between 5.4 and 6.8 hours, with the intervention lead times predicted by a support vector regression model, which is capable of proactively managing the intervention 4.6 days before visual scouting on average. Temporal decay analysis demonstrated that stress-specific health half-lives of drought are 4.0 days, heat stress are 4.8 days, and nitrogen deficiency are 7.1 days. End-to-end latency between drone landing and farmer alert was less than 12 minutes with cloud-based processing. Early intervention averted yield losses of 3.7 to 4.6 metric tons per hectare, which is equal to savings of 740 to 920 US dollars per hectare. These findings show that machine learning-based multi-sensor drone fusion can be used to reliably identify early abiotic stress and a scalable route to improving food security in the world by precision agriculture.