Pest outbreaks are a growing threat to farmers in Asia, especially smallholders who rely on rice and corn for their livelihoods. These pests are becoming harder to manage as weather patterns become more unpredictable due to climate change. To help address this, we developed a forecasting system that uses data from multiple sources to predict when and where pest outbreaks are likely to happen.

The system focuses on two major pests: the Brown Planthopper (BPH), which affects rice in China, and the Fall Armyworm (FAW), which affects corn in India. It combines field data collected by agronomists, weather data such as temperature and humidity from The Weather Company (IBM), and satellite information like vegetation health and crop cycles from Sentinel-1 and Sentinel-2.

For BPH, we used a type of deep learning called Graph Convolutional Networks (GCNs) that can learn patterns across different locations, even when some data is missing. For FAW, we used a classification model to predict pest severity right after planting.

These models help give farmers early warnings, allowing them to take action before damage occurs. This approach can be expanded to other pests and crops, helping make farming more resilient to climate change across Asia.