5. Drought Estimation from Vegetation Phenology Analysis of Maize in Indonesia Using Deep Learning Algorithm

The goal of this research was to collect visual information at the crop production that can be used for drought estimation. The study was completed to create an automated detection system of drought with high accuracy, low computing cost, and a lightweight deep learning model. Considering the advantages of YOLOv3, it was proposed to detect and localize vegetation phenology analysis under conditions of season in Indonesia. The study was planned to analyze the vegetation phenology to forecast drought during maize production at the central East Java areas of Indonesia. In the study, the vegetation index was utilized to produce the normalized difference vegetation index (NDVI) and normalized difference water index (NDWI) derived from Sentinel-2 to estimate water stress due to drought. According to the NDVI trajectory, the maize planting season was in April 2018, and the harvest was concluded in late August 2018. This study presents a convolutional neural network (CNN)-based you only look once (YOLO) model for detecting drought at the maize growth phases. The drought estimation was validated from the vegetation phenology analysis based on the growing season. The accuracy assessment of the deep learning model reported Intersection of Union (IoU) 83.4%, precision 98%, recall 99%, F1-Score 98%, and mean average precision 96% for the drought-prone areas. The deep learning analysis suggested that the proposed YOLOv3 model can perform robust and accurate detection of drought estimation from vegetation phenology.