Spatio-Temporal Performance of 2D Local Inertial Hydrodynamic Models for Urban Drainage and Dam-Break Applications

This paper demonstrates that the HydroPol2D model, utilizing 2D local-inertial approximations, offers a computationally efficient alternative (23 times faster) to full-momentum solvers for urban and dam-break flood forecasting, achieving high accuracy in subcritical flows and peak depth predictions while highlighting the critical need to account for urban infrastructure to avoid significant discharge errors.

The Gist

Imagine you are trying to predict how a massive wave of water will rush through a city after a dam breaks or during a terrible storm. You need to know exactly where the water will go, how deep it will get, and how fast it will arrive to save lives and property.

This paper is about building a super-fast, smart calculator for these floods. The authors created a new computer model called HydroPol2D and tested it to see if it can do the job just as well as the "gold standard" models, but much, much faster.

Here is the breakdown of their work using simple analogies:

1. The Problem: The "Slow vs. Fast" Dilemma

Think of flood modeling like driving a car.

• The "Gold Standard" (Full Momentum Models like HEC-RAS): This is like driving a high-performance race car with a super-accurate GPS. It calculates every tiny bump, wind gust, and turn in the road perfectly. It gives you the most accurate route, but it takes a long time to process the map. If you need to know the route right now (like during a real-time emergency), this car might be too slow.
• The "Simple" Models: These are like riding a bicycle. They are super fast and easy to pedal, but they might miss some details, like a pothole or a steep hill, leading to a slightly wrong route.

The authors wanted to build a "Smart Scooter." It's faster than the race car but smarter than the bicycle. They wanted to see if they could skip the complicated physics (the "race car" details) without losing too much accuracy.

2. The Secret Sauce: Ignoring the "Push"

In physics, water moving fast has two types of energy:

1. Gravity/Friction: Water flowing downhill and rubbing against the ground.
2. Inertia (The "Push"): The force of the water pushing itself forward because it's already moving fast (like a heavy truck that can't stop quickly).

The "Gold Standard" models calculate both. The new model (HydroPol2D) mostly ignores the "Push" (inertia) because, in most city floods, the water isn't moving fast enough for that "Push" to matter. By ignoring the "Push," the math becomes much simpler, making the computer run 23 times faster.

3. The Three "Flavors" of the Model

The authors didn't just build one version; they built three slightly different versions of their "Smart Scooter" to see which one handles bumps best:

• The Original: The basic version.
• The "Centered" Version: Tries to guess the water's path by looking at what's happening on both sides.
• The "Upwind" Version: Looks at where the water is coming from to predict where it's going next.

They tested these against the "Gold Standard" in four different scenarios.

4. The Four Tests (The Real-World Trials)

Test 1: The Flat Pool (The Theory Check)

• Scenario: A wave moving across a perfectly flat, smooth surface.
• Result: The "Original" version got a bit wobbly (unstable), but the "Upwind" and "Centered" versions stayed steady. The "Upwind" version was the most accurate here.

Test 2: The Dam with a Pipe (The Detention Pond)

• Scenario: A pond that catches rain and releases it slowly through a pipe (culvert) and a spillway.
• The Twist: The model had to pretend the pipe existed without having a detailed map of the pipe inside the computer. It used a simple "rule of thumb" (a rating curve) to guess how much water flows through.
• Result: It worked amazingly well! It predicted the water flow with less than 5% error compared to the slow, complex model. It proved you don't need a billion details to get a good answer; you just need the right rules.

Test 3: The City Without Drains (The Urban Trap)

• Scenario: A busy city in Brazil during a heavy storm.
• The Problem: If you just look at a satellite map of a city, you see buildings and streets. You don't see the underground pipes, manholes, or tunnels. If the computer thinks water hits a building and stops, it creates a fake lake.
• The Fix: The authors told the model, "Hey, pretend there are invisible tunnels here."
• Result: This was huge. Without the "invisible tunnels," the model thought the city would flood 17.5% more and the water would arrive at the wrong time. With the tunnels included, the model was accurate. It also showed that ignoring city drains doubles the time the computer takes to run because the water gets "stuck" in fake puddles.

Test 4: The Dam Break (The Big Disaster)

• Scenario: A massive dam breaks, sending a tsunami-like wave toward a city of 200,000 people.
• The Challenge: This is the "hard mode." The water is moving super fast (supercritical flow), which is exactly where simple models usually fail.
• Result:
  • Speed: The new model was 23 times faster. While the old model took 43 hours to run, the new one took less than 2 hours.
  • Accuracy: It was very good at predicting where the water would go (95% accuracy for the "Original" version).
  • The Flaw: Because it ignored the "Push" (inertia), the water arrived in the city a little bit too fast in the simulation. It's like a runner who starts too quickly but doesn't account for their own weight slowing them down later. However, for emergency planning (knowing where to evacuate), this speed is worth the tiny timing error.

5. The Big Takeaway

This paper proves that we can build emergency flood tools that are incredibly fast without sacrificing too much accuracy.

• For City Planners: You don't need a perfect map of every single pipe to get a good flood forecast. You just need to tell the computer where the drains are using simple rules.
• For Emergency Managers: You can run hundreds of "what-if" scenarios (like "What if the dam breaks at 2 PM?" vs "What if it breaks at 4 PM?") in the time it used to take to run just one. This allows for better, faster decisions to save lives.

In short: The authors built a "Smart Scooter" that drives 23 times faster than the "Race Car" and gets you to the right destination almost as accurately, making it perfect for saving lives during real-time floods.

Technical Summary

1. Problem Statement

Accurate and rapid flood modeling is critical for forecasting and risk management. However, full momentum hydrodynamic models (solving the complete Saint-Venant equations) often require prohibitive computational times, especially for high-resolution domains or real-time applications. Conversely, low-complexity models, such as local-inertial approximations, offer significant computational speed but are traditionally hypothesized to perform poorly under supercritical flows (Froude number > 1) or in complex urban environments where infrastructure (culverts, pumps, spillways) dictates flow dynamics.

The paper addresses two primary gaps:

1. Urban Infrastructure Representation: How to accurately simulate urban drainage (culverts, spillways, pumps) in 2D models when detailed sewer network data is unavailable, and how omitting these features affects flood predictions.
2. Numerical Scheme Performance: The spatio-temporal accuracy of different local-inertial numerical schemes (original, s-centered, s-upwind) in complex scenarios, specifically dam-break events with supercritical flow regimes, compared to a full-momentum benchmark.

2. Methodology

Model Development: HydroPol2D

The authors utilized HydroPol2D, a 2D hydrodynamic solver implemented in MATLAB with GPU acceleration. It solves the shallow water equations using a Godunov-like approach on a staggered grid.

• Governing Equations: The model solves the momentum equation by neglecting the convective (advective) inertia term, retaining only local inertia, pressure, and friction.
• Numerical Schemes: Three specific formulations were tested:
  1. Original Local-Inertial (lim): The standard formulation by Bates et al. (2010).
  2. s-centered: Introduces numerical diffusion to stabilize solutions.
  3. s-upwind: Uses an upwind scheme to handle fluxes, designed to improve stability in variable flows.
• Internal Boundary Conditions (IBC): To address the lack of detailed sewer data, the model implements spatial simulations of urban infrastructure (orifices, weirs, pumps, culverts) as internal boundary conditions. These are modeled using rating curves ($Q = k_1 \Delta h^{k_2}$) derived from geometric relationships (e.g., Google Earth data) rather than complex 1D-2D coupling.

Case Studies

The study employed four distinct scenarios to validate the model against HEC-RAS 2D (a full-momentum solver):

1. Theoretical Validation: Propagation of a non-breaking wave on a flat surface with low roughness ($n=0.005$) to test numerical stability against an analytical solution.
2. Detention Pond (Case 1): Simulation of a reservoir in São Paulo with a culvert and spillway under a 1-in-100-year inflow hydrograph.
3. Urban Catchment (Case 2): A highly urbanized catchment in São Paulo subjected to a 1-in-50-year storm. This compared simulations with and without internal boundary conditions representing drainage infrastructure.
4. Dam-Break (Case 3): A high-risk scenario involving the Pirapama Dam (Brazil) failing and impacting the coastal city of Cabo Santo Agostinho (pop. ~200,000). This involved a peak discharge of ~49,000 $m^3/s$ and generated supercritical flows (Froude > 1).

3. Key Contributions

• Novel Urban Drainage Implementation: The paper demonstrates a method to integrate urban drainage infrastructure (culverts, pumps, spillways) into 2D models using simple rating curves as internal boundary conditions. This bypasses the need for detailed, often unavailable, 1D sewer network topology data.
• Comparative Numerical Analysis: It provides a rigorous spatio-temporal comparison of three local-inertial schemes (lim, s-centered, s-upwind) against a full-momentum solver, specifically focusing on the trade-off between computational speed and accuracy in supercritical flow regimes.
• Quantification of Infrastructure Impact: The study quantifies the error introduced by ignoring urban infrastructure, showing that omitting these features leads to significant overestimation of flood depths and incorrect hydrograph timing.

4. Key Results

Numerical Stability and Accuracy

• Non-breaking Wave: The s-upwind scheme showed the closest agreement with the analytical solution for flat, smooth surfaces. The original formulation (lim) exhibited instability near wetting/drying fronts.
• Dam-Break Performance:
  • Flood Extent: All local-inertial schemes performed well in predicting the maximum flooded area. The Critical Success Index (CSI) for maximum flood depth was high: 0.95 (lim), 0.92 (s-centered), and 0.89 (s-upwind).
  • Flood Wave Advance: Due to the neglect of convective inertia, all local-inertial models predicted a faster flood wave advance than the full-momentum HEC-RAS model, particularly in flat coastal areas.
  • Depth Prediction: The s-upwind scheme demonstrated superior performance in predicting temporal hydrographs and depth metrics (NSE, KGE, RMSE) at specific waterlogging points, especially near the dam breach where flows were supercritical.

Impact of Urban Infrastructure

• Peak Discharge: Omitting urban drainage structures resulted in a 17.5% difference in peak discharge at the outlet compared to simulations including them.
• Flood Depth: Without internal boundary conditions, the model incorrectly stored water in terrain depressions (due to LiDAR noise), leading to flood depths 1.8 meters higher than reality in certain areas.
• Computational Efficiency: Including the simplified internal boundary conditions actually reduced computational time by nearly 50% compared to the simulation without them, as it prevented the solver from resolving complex, artificial storage pockets.

Computational Speed

• The local-inertial models were significantly faster than HEC-RAS 2D. In the dam-break scenario, HydroPol2D was 23 times faster (114 minutes vs. 2,610 minutes for the slowest scheme) while maintaining acceptable accuracy for flood extent mapping.

5. Significance and Conclusions

This research validates that local-inertial models are viable for complex, real-world applications, including dam-breaks and urban flooding, provided the appropriate numerical scheme is selected.

• For Flood Extent Mapping: The models are highly accurate (CSI > 0.89) and offer massive computational gains, making them ideal for probabilistic forecasting, ensemble modeling, and real-time warning systems.
• For Urban Modeling: The proposed method of using rating curves for internal boundary conditions is a robust alternative when detailed sewer data is missing. It prevents the overestimation of flood depths and significantly improves hydrograph accuracy.
• Limitations: The lack of convective inertia causes an overestimation of flood wave speed (arrival time) in supercritical regimes. Therefore, while excellent for determining where flooding will occur, these models may be less accurate for predicting exactly when the water will arrive in high-velocity zones.

The study concludes that the s-upwind scheme offers the best balance for spatio-temporal analysis, and the integration of simplified urban infrastructure allows for wider application of these efficient models in urban planning and disaster management.