Accuracy and Prediction of Hopperburn by Brown Planthopper (Nilaparvata Lugens Stal) with Sentinel-2 Images

Abstract

Forecasting of brown planthopper attack or BPH (Nilaparvata lugens Stal) using artificial intelligence and vegetation index of Sentinel-2 Satellite Imagery improves forecasting the incidence of hopperburn. This study aimed to determine the accuracy and correlation of the random forest classification of Sentinel-2 imagery to the incidence of hopperburn reported by Plant Pest Organisms Observer (PPOO) and determine the best method for predicting it. The study was done through observation and secondary data processing about the age of the plant, the incidence of hopperburn by BPH, interviews with farmers, and PPOO. The results showed that the hopperburn NDVI index ranged from 0.23 - 3.8. The random forest classification accuracy was high (Kappa Index = 0.82). The relationship between the hopperburn area from the PPOO report and the predicted area from Sentinel-2 images classified as (R² = 0.53, R = 0.728) with the equation Y = -1.5 + 0.82 X. The correlation can be improved using spatial regression Geographically Weighted Regression (GWR4) with the best gaussian distance of 1.76 km (R₂ = 0.6, R = 0.77). The best prediction for the NDVI stage of hopperburn attack time series with random forest (RMSE = 0.12819) was better than the prediction of the hopperburn attack time series with the exponential smoothing method from the PPOO report (RMSE 3.302184).