Forecasting the frequency and magnitude of hurricanes in the Yucatan Peninsula, Mexico, in the period from 2025 to 2034 using convolutional neural networks (CNNs), Long Short-Term Memory networks (LSTMs) and statistical models

Climate change has significantly increased the frequency and severity of extreme weather events, a trend recognized under the United Nations Sustainable Development Goal 13: Climate Action. This study forecasts hurricane activity in the Yucatan Peninsula, Mexico, for the period 2025–2034 using advanced computational models, including Convolutional Neural Networks (CNNs), Long Short-Term Memory networks (LSTMs), Autoregressive Integrated Moving Average models (ARIMA), and Linear Regression (LR). Historical hurricane data were extracted from the HURDAT2 database kept by the National Hurricane Center (NHC) and spatially analyzed in QGIS to assess storm trajectories and wind intensities.

The data were processed using Python, and each model was trained to predict hurricane frequency within three wind speed categories: <50 knots, 50–100 knots, and >100 knots. Results reveal divergent performance among the models. CNN exhibited high variability for low-speed events, peaking at 4.21 events in 2027 and dropping to 1.27 by 2034. In contrast, LSTM and ARIMA maintained stable forecasts: LSTM fluctuated between 2.7 and 3.0, and ARIMA ranged from 1.5 to 1.8. For the 50–100 knot range, CNN reached an anomalous high of 8.14 events in 2032, while LSTM and ARIMA remained within narrower bands (1.85–2.01 and 1.32–1.99, respectively). At the >100 knot level, ARIMA showed a rising trend from 0.21 in 2025 to 0.57 in 2034, suggesting a potential increase in high-intensity cyclones.

These findings emphasize the need for adaptive forecasting systems that account for nonlinear behavior under climate change conditions.

The model outputs offer valuable insights for risk management, contingency planning, and infrastructure resilience in the hurricane-prone Yucatan Peninsula.