forecast micromobility station rebalancing needs from gbfs station status history

Steps

1. Archive station_status.json snapshots (num_vehicles_available, num_docks_available, is_renting, is_returning) at a fixed polling interval to build a time series per station_id.
2. Join against the corresponding station_information.json capacity fields (capacity, and in GBFS 3.0 the per-vehicle-type vehicle_types_capacity/vehicle_docks_capacity) so raw counts are normalized against each station's actual capacity.
3. Compute simple empty/full-dock risk indicators (e.g. rolling minimum/maximum availability by hour-of-day and day-of-week) as a baseline before reaching for more advanced forecasting models.
4. Cross-check station polling frequency against each feed's own ttl/last_updated fields — forecasting off a feed you're over- or under-polling relative to its actual update cadence will bias the time series.
5. Validate any capacity assumptions against real vehicle_types.json data if the system serves multiple vehicle types, since a station's total capacity can be split unevenly across types.