Introduction
This project investigates the potential of permanent inductive loop bicycle counters by analyzing traffic data collected in Barcelona from 2017 to 2023. It explores seasonal patterns, daily trends, and anomalies while emphasizing data cleaning and imputation techniques to improve data reliability.
Visualization 1: Evolution of the number of counters
The available data set is comprised of 54.037.979 15-minute observations collected using permanent counters in various locations in Barcelona between 2017 and 2024. The number of stations varies during the years and does not match on both datasets, as seen in in Figure 1. The counters are in both unidirectional and bidirectional bicycle lanes and treat each travel direction separately. Each observation contains the id of the counter, the date and time of the observation, the bicycle count, and the associated error. Each station contains the id, a string describing the counter, the type of vehicle that it counts, the number of lanes, the district and neighborhood number, the type of equipment and the coordinates.
Visualization 2: Error Analysis and Data Quality Assessment
In the Figure 2 we can observe the evolution in the number of observations as well as the presence of errors of varying seriousness. The number of observations increases overtime, with some notable drops that can last various days or may not recover at all.
Focusing on the categories of observations, we can see that most of the data falls under the "Valid" category, followed by "Invalid" and "Unknown" as shown in Table 1Table 2. The "Valid" category accounts for most observations (89.9%) and is present on almost every day (99.9%). In contrast, "Unknown" errors are relatively rare, making up only 2.5% of the total, and occur on fewer days (15.1%). The "Partial" category, though less frequent overall (1.4%), appears on most days (99.1%) but never exceeds 25.5% of the daily observations. "Invalid" observations are more common than "Unknown" (6.1%) and occur on nearly as many days as "Valid," with a significant maximum daily ratio of 61.6%.
Visualization 3: Estimated AADBT
This visualization presents anomalies detected in the data, such as unusually high or low counts. It also demonstrates how data cleaning enhanced the accuracy of the analysis.
Visualization 4: Evolution AADBT
Map of AADBT evolution by time shown using hexbins.
Evolution Bikelanes
See how bikelanes have evolved during the last 40 years .
Explore AMBici trips
See how bikelanes have evolved during the last 40 years .
Comparison of Models
Compare predicted AADBT using four different models across the city map.