Spatial Dependency Patterns in Weather-Enhanced Bike-Share Demand Forecasting in Washington DC

Accurate bike-share demand forecasts enable operators to rebalance networks before shortages occur. Building on Miao et al. (2025), we quantified how weather variables improve spatial model performance. We analyzed 251,633 Capital Bikeshare trips in Washington, D.C. during January 2026 using separate ConvLSTM models for members and casual users across a 64-zone grid. Weather covariates improved predictions for both groups, with member demand proving substantially more predictable (R² = 0.816) than casual demand (R² = 0.427). Members exhibited higher mean perturbation-based influence scores than casual users, and spatial influence followed a directional, non-proximal pattern consistent with New York City evidence. The magnitude of weather-related improvement may vary across urban contexts, requiring validation before generalization.