Rapid and accurate mosquito abundance forecasting with Aedes-AI neural networks

We present a method to convert weather data into probabilistic forecasts of Aedes aegypti abundance. The approach, which relies on the Aedes-AI suite of neural networks, produces weekly point predictions with corresponding uncertainty estimates. Once calibrated on past trap and weather data, the model is designed to use weather forecasts to estimate future trap catches. We demonstrate that when reliable input data are used, the resulting predictions have high skill. This technique may therefore be used to supplement vector surveillance efforts or identify periods of elevated risk for vector-borne disease outbreaks.