A Synthetic One-Year, 5-Minute Resolution Traffic–Air Quality–Meteorology Time Series Generated from SUMO Microsimulations on the Real OpenStreetMap Network

This dataset provides a synthetic full calendar year (1 January – 31 December 2023) of network-level traffic conditions for central Casablanca, Morocco, generated from SUMO (Eclipse SUMO 1.26.0) microsimulations on the real OpenStreetMap road network, and coupled with a parametric air-quality and meteorology model. The data comprise 105,120 observations at 5-minute resolution (365 days × 288 intervals), 27 variables, no missing values. The dataset is a hybrid simulated–synthetic product: traffic dynamics originate from SUMO microsimulation of five demand regimes on the real network; the annual series is then produced by a fully reproducible stochastic generative model (fixed seed) that schedules these regimes over the year and injects calendar effects specific to a North African metropolis — Ramadan iftar demand surges, Friday-prayer troughs, weekend patterns — together with physically informed air-pollutant and weather covariates. Widely used traffic benchmarks (METR-LA, PEMS-BAY) and public SUMO scenarios (TAPAS Cologne, Bologna, Monaco/MoST, Luxembourg LuST) cover exclusively North American or European networks and none encodes such cultural-calendar regimes. Production pipeline (fully scripted, seed = 42) Network: OpenStreetMap extract of central Casablanca (final extent −7.6567, 33.5312 → −7.5428, 33.6197, i.e. ≈10.6 × 9.9 km; UTM zone 29 / WGS84), filtered to the drivable hierarchy (trunk → residential) and converted with netconvert into a SUMO network of 34,841 edges, 40,921 junctions, 42,808 lanes totaling 2,562 lane-km, and 461 traffic-light systems (predominant speed limit: 50 km/h). Demand regimes: five route sets generated with SUMO randomTrips.py at regime-specific insertion rates — rush_hour (6,000 veh/h), ramadan_iftar (6,546), normal (3,600), friday_prayer (1,125), night (900). Demand is synthetic (random origin–destination trips on the real network); it is not calibrated against field traffic counts. Microsimulation: one 1-hour SUMO run per regime (1 s step), aggregated into 5-minute network-level indicators (running vehicles, mean speed, waiting time, halting vehicles). Annual synthesis: a stochastic generator schedules the five regimes over 2023 (peak hours 07:00–09:00 and 17:00–20:00; nights; Fridays 12:00–14:00; iftar windows in the Ramadan period), applies multiplicative calendar modifiers (Ramadan +15–45%, Friday prayer −60–72%, weekends −15–30%) and Gaussian noise, and derives a composite congestion index (weighted speed degradation, halting ratio, and normalized waiting time) discretized into 4 ordinal classes. Air quality module: parametric emission–dispersion model — urban background concentrations anchored on IQAir 2023 values for Casablanca; traffic-proportional emissions with stop-and-go amplification; dispersion modulated by wind speed, atmospheric boundary-layer height (diurnal coastal profile, 150–1,384 m), and precipitation washout; AQI computed from US-EPA sub-index breakpoints (PM2.5, NO₂, O₃). These are modeled covariates, not monitoring-station measurements. Meteorology module: synthetic climatology built from monthly normals for Casablanca (temperature, humidity, wind, rain probability; ONMT 2010–2023) with sinusoidal diurnal cycles, a sea-breeze wind profile, and stochastic precipitation. Data structure (27 columns) Temporal/calendar (9): timestamp, date, hour, minute, day_of_week, month, is_weekend, is_friday_prayer (Fri 12:00–14:55), is_ramadan. Regime (1): scenario_sumo — normal (33.6%), night (32.6%), rush_hour (27.8%), ramadan_iftar (4.2%), friday_prayer (1.8%). Traffic (4): vehicles_running (mean 387, peak 1,902), mean_speed_kmh (26.0–57.1), waiting_time_s, halting_vehicles. Air quality (5): pm25_ugm3, no2_ugm3, co_mgm3, o3_ugm3, aqi_composite (21–275). Correlations with congestion: NO₂ 0.80, PM2.5 0.75, AQI 0.82. Meteorology (5): temperature_c (3.1–33.2 °C), humidity_pct, wind_speed_ms, precipitation_mm, boundary_layer_m. Targets (2): congestion_level ∈ [0, 0.63] and congestion_class (quantile-calibrated thresholds 0.095 / 0.22 / 0.365; distribution 47.1% / 27.9% / 18.9% / 6.0%; the severe class occurs exclusively in rush-hour and Ramadan-iftar regimes at peak hours). Network id (1): osm_network. Design notes and limitations All traffic values are simulation outputs, not field measurements; absolute magnitudes reflect the synthetic demand, and the dataset should be used for methodological benchmarking (forecasting, imbalanced ordinal classification, joint traffic–air-quality modeling, RL environments, digital-twin prototyping) rather than as an empirical record of Casablanca traffic. The is_ramadan flag covers March–April 2023 in full (61 days) as a simplifying design choice; the actual Ramadan 1444 spanned 23 March – 21 April 2023. The annual series extrapolates one simulated hour per regime; intra-regime variability beyond the injected stochastic components is therefore limited. Air-quality and weather variables are model-generated covariates with clipped physical ranges. Reproducibility The complete generation pipeline (network extraction, route generation, SUMO runs, annual synthesis) is a single Jupyter notebook with fixed random seed, included in this record. https://github.com/attiouimehdi/Casablanca-SUMO-v2 import pandas as pddf = pd.read_csv("casablanca_sumo_OSM_365days_v2.csv", parse_dates=["timestamp"]) How to cite Attioui, M. (2026). Casablanca SUMO v2: A Synthetic One-Year Traffic–Air Quality–Meteorology Time Series Generated from SUMO Microsimulations on the Real OpenStreetMap Network [Data set]. Zenodo. https://doi.org/10.5281/zenodo.21539315