Abstract
Arboviruses spread in urban tropics under climate change. At Aedes aegypti breeding sites in Recife, Brazil, we linked surveillance and climate data from the Pernambuco Water and Climate Agency (APAC), the Brazilian National Institute of Meteorology (INMET), Rapid Survey of Indices for Aedes aegypti (LIRAa), and Recife's Open Data Portal. We modeled 2013-2021 cases and 2009-2017 breeding sites. We generated spatial fields with inverse distance weighting. We built bimonthly training grids with 5000 points and validation grids with 50,000 points. We tested linear regression, random forests, multilayer perceptrons, support vector regressors, and extreme learning machines in the Weka platform and Python Reservoir Computing Networks (PyRCNs). We ran 30 repetitions with cross-validation. The random forests performed well. Multilayer perceptrons reached very high correlations but needed longer training. Polynomial Support Vector Machines (SVMs) reached near-perfect accuracy but required very high computation. Single-layer extreme learning machines delivered the best trade-off, with low errors, correlations near 1.0, and short training times. The models produced fine-scale risk predictions and highlighted priority areas. The findings support earlier, targeted control and guide public health plans in Recife.