Multimodal Large Language Models for Global Desert Locust Risk Prediction
LocustLens is a multimodal large language model designed for global desert locust (Schistocerca gregaria) presence risk prediction. Built upon Qwen3-VL-4B, it is fine-tuned through a two-stage training pipeline combining lightweight supervised fine-tuning and reinforcement learning on a curated 43-year locust event dataset.
Intended use
- Early warning systems for locust outbreak risk assessment
- Decision support for desert locust surveillance and monitoring
- Multimodal reasoning research on spatiotemporal environmental data
