⚡ electrical engineering

Revisiting mesoscopic traffic flow simulation in SUMO: Limitations, analysis, and an alternative

This paper identifies limitations in SUMO's current mesoscopic traffic model regarding its adherence to LWR principles and queue dynamics, and proposes a new discrete-time Link Transmission Model that explicitly incorporates backward traveling spaces to more accurately simulate congestion onset, recovery, and spillback.

Original authors: Ying-Chuan Ni, Alina Akopian, Anastasios Kouvelas, Michail A. Makridis

Published 2026-06-09

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Ying-Chuan Ni, Alina Akopian, Anastasios Kouvelas, Michail A. Makridis

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine traffic simulation as a way to predict how cars move through a city without actually putting real cars on the road. There are three main ways to do this:

Microscopic: Like watching every single ant in a colony. It's incredibly detailed but takes a massive amount of computer power.
Macroscopic: Like looking at a river from a helicopter. You see the water flowing as a whole, but you can't see individual drops. It's fast but lacks detail.
Mesoscopic: The "Goldilocks" zone. It tracks individual cars (like the ants) but groups them together to save time (like the river).

This paper argues that the current "Goldilocks" tool used in the popular open-source software SUMO is broken. It's trying to be a hybrid, but it's mixing up the rules of the game, leading to unrealistic predictions.

Here is the breakdown of the problem and the proposed fix, using simple analogies.

The Problem: The Broken "Queue" Logic

The current SUMO mesoscopic model (called MESO) tries to manage traffic by dividing roads into segments and deciding when a car can move from one segment to the next. The authors found two major flaws in how it handles traffic jams:

1. The "Blind Spot" Analogy (Backward Traveling Spaces)
Imagine a line of people waiting to get into a concert. In reality, if the person at the very front of the line gets in, a "gap" opens up. That gap travels backward instantly, allowing the next person to step forward.

How MESO fails: MESO is like a bouncer who only notices the gap after the entire line has stopped moving completely. It waits until the line is totally jammed before it realizes a space has opened up.
The Result: Because it doesn't see the "gap" traveling backward fast enough, it thinks traffic clears up too quickly. It underestimates how long a jam lasts and how far back it stretches.

2. The Wrong "Cause of Jam" Analogy
In the current model, traffic jams on city streets are mostly blamed on cars slowing down because the road gets too crowded (a "capacity drop").

The Reality: In cities, jams are usually caused by turning. Imagine a straight road where cars are trying to turn left into a side street. If that side street is full, the turning cars get stuck, blocking the straight-moving cars behind them. This is called "queue spillback."
How MESO fails: The model doesn't account for this "blocking from the side" effectively. It assumes jams happen mostly because cars are too close together, missing the fact that a jammed side street can choke the main road.

The Consequence: When the authors tested this, the software predicted that traffic jams would appear later, disappear sooner, and be much smaller than they actually are. It was like a weather forecast that said, "It will rain for 10 minutes," when in reality, it was a three-hour downpour.

The Solution: A New "Link Transmission" Model

The authors propose a new model called LIFT (which they adapted into a discrete-time version for this study).

The "Elevator" Analogy
Instead of guessing when a car can move, LIFT explicitly tracks the empty spaces (vacancies) moving backward through the traffic.

Think of a link (a road segment) as an elevator with a fixed number of spots.
The model counts: How many people are inside? + How many empty spots are moving backward?
If the total (people + empty spots) is less than the elevator's capacity, a new person can get in.
This allows the model to see the "gap" traveling backward instantly, just like in real life.

Why it works better:
By explicitly tracking these backward-moving gaps, the new model captures "queue spillback" perfectly. If a side street is full, the model immediately knows the main road is blocked.

The Proof: The Test Drive

The authors ran two tests to prove their new model works:

The City Intersection Test: They simulated a busy street with traffic lights and a side street joining in.
- MESO (Old): Predicted the jam would clear up quickly and wouldn't stretch very far back.
- LIFT (New): Predicted the jam would build up slowly, stretch far back, and clear up gradually.
- The Verdict: The new model matched the "Microscopic" (super detailed) simulation almost perfectly, while the old one was way off.
The Highway Bottleneck Test: They simulated a highway where two lanes merge into one.
- MESO (Old): The traffic data looked chaotic and broke the laws of physics (showing traffic states that shouldn't exist).
- LIFT (New): The traffic data followed the smooth, predictable curves of traffic flow theory.

The Bottom Line

The paper concludes that the current mesoscopic tool in SUMO is fundamentally flawed because it mixes up different traffic theories and ignores how "empty space" travels backward through a jam.

Their new model, LIFT, fixes this by treating traffic like a fluid where empty spaces move backward just as real as cars move forward. This results in a simulation that is fast enough to run on big networks but accurate enough to trust when predicting real-world traffic jams. The authors hope to eventually put this new code directly into the SUMO software so everyone can use it.

Technical Summary: Revisiting Mesoscopic Traffic Flow Simulation in SUMO

Problem Identification
The paper addresses critical deficiencies in the mesoscopic traffic simulation model currently implemented in the open-source tool Simulation of Urban MObility (SUMO), referred to as MESO. This model, based on Eissfeldt (2004), aims to balance the computational efficiency of macroscopic models with the individual vehicle tracking of microscopic models. However, the authors identify that MESO fails to fully comply with the principles of the Lighthill–Whitham–Richards (LWR) kinematic wave theory.

Specific problems identified include:

Conceptual Mixture: The model inappropriately blends concepts from the Cell Transmission Model (CTM), which assumes homogeneous traffic states within cells, and the Link Transmission Model (LTM), which handles heterogeneous within-link dynamics.
Invalid Queue Dynamics: MESO relies on a binary distinction between "free" and "jammed" segments. This logic delays the activation of backward traveling spaces (vacancies) until a segment is fully jammed. Consequently, the model fails to accurately simulate queue spillback phenomena, particularly in interrupted traffic flow scenarios like urban networks.
Underestimation of Congestion: Due to the delayed propagation of congestion and the incorrect handling of backward traveling spaces, MESO systematically underestimates the magnitude of congestion and produces unrealistic onset and recovery patterns.
Misattribution of Congestion Causes: The model attributes congestion growth primarily to capacity drop (capacity reduction upon jamming), whereas urban congestion is often driven by imbalanced turning movements and queue spillback, which do not necessarily require a capacity drop to propagate.

Methodology
To address these issues, the authors propose an alternative mesoscopic implementation: a discrete-time version of the Link Transmission Model (LIFT). Unlike the original MESO, this approach:

Explicitly Tracks Backward Traveling Spaces: Instead of relying on binary state transitions, the model explicitly simulates the movement of vacancies (spaces) traveling upstream. A vehicle can enter a link only if the sum of current vehicles and propagating spaces does not exceed the link's storage capacity.
Adopts LWR Principles: The model follows the first-order traffic dynamics of the LWR model, discretized via Newell's formulation, ensuring consistency with kinematic wave theory.
Discrete-Time Implementation: To ensure a fair comparison with the discrete-time MESO, the authors developed a discrete-time version of LIFT (Algorithm 1) rather than using the continuous-time event-based version previously proposed. This version utilizes only two headway parameters: minimum headway ( $\tau_f$ ) and the time required for a backward traveling space to move a unit vehicle distance ( $\tau_j$ ).

Case Studies and Results
The authors evaluated the proposed LIFT model against the existing MESO and a microscopic SUMO simulation (using the Krauss car-following model) across two scenarios:

Signalized Urban Corridor: A six-link corridor with a bottleneck caused by a merging flow from a minor street.
- Result: MESO underestimated link densities during congestion, showed delayed congestion onset, and premature dissipation. In contrast, LIFT results were consistent with the microscopic simulation, accurately capturing the gradual onset and dissipation of congestion and the upstream propagation of queue spillback.
Motorway Lane-Drop Bottleneck: A two-lane motorway segment with a downstream reduction to one lane.
- Result: Traffic states generated by LIFT adhered closely to the theoretical Fundamental Diagram (FD) and LWR kinematic wave theory. MESO, however, produced states that exceeded the theoretical FD envelope and exhibited excessive state heterogeneity within edges.

Key Contributions

Critical Analysis: The paper provides a detailed theoretical and practical critique of the Eissfeldt (2004) model as implemented in SUMO, highlighting the incompatibility of its logic with LWR principles.
Alternative Implementation: It introduces a discrete-time mesoscopic LTM (LIFT) that explicitly incorporates backward traveling spaces to capture queue spillback and congestion propagation more accurately.
Validation: Through comparative case studies, the paper demonstrates that the proposed mesoscopic model yields results highly consistent with both kinematic wave theory and microscopic simulations, correcting the systematic underestimation of congestion found in the current SUMO-MESO.

Significance and Claims
The paper claims that the theoretical validity of many mesoscopic models has been overlooked due to a limited historical understanding of macroscopic traffic flow theory. The authors argue that simplified traffic dynamics significantly impact simulation accuracy. By demonstrating that a mesoscopic model following strict traffic flow theory principles can generate outcomes nearly identical to microscopic simulations, the work validates the potential of mesoscopic modeling for large-scale network analysis.

The authors modestly state that the proposed model is an alternative designed to demonstrate the drawbacks of the current implementation and verify theoretical accuracy. They express the hope that this model will eventually be integrated into the SUMO source code, but do not claim immediate integration or broader applications beyond the scope of the presented validation.

The Problem: The Broken "Queue" Logic

The Solution: A New "Link Transmission" Model

The Proof: The Test Drive

The Bottom Line

Technical Summary: Revisiting Mesoscopic Traffic Flow Simulation in SUMO

More like this