Seeing through water: diffuse image-based depth measurements in three-dimensional dam-break flows

This paper presents a validated experimental facility and image-based methodology using dyed water and bi-exponential modeling to accurately measure water depth in three-dimensional dam-break flows, with results confirmed by high statistical repeatability and independent volume estimates.

The Gist

Imagine trying to measure the depth of a rushing river, but you can't stick a ruler in it because the water is moving too fast and splashing everywhere. Now, imagine doing this for a massive, sudden flood caused by a dam breaking, spreading out in all directions like a giant, wet pizza dough being tossed in the air. That is exactly what this paper describes: a clever new way to "see" how deep the water is without ever touching it.

Here is the story of how the researchers did it, broken down into simple parts:

1. The Problem: The "Invisible" Flood

When a dam breaks, the water rushes out in a wild, three-dimensional wave. Engineers need to know exactly how deep the water is at every single spot to predict where it will go and how dangerous it will be.

• The old way: You could use point sensors (like tiny rulers), but you'd need hundreds of them to get a full picture, and they often break or get confused by splashing water.
• The new way: The researchers decided to use cameras and light to measure the depth, turning the water itself into a giant, living ruler.

2. The Setup: A Giant Light Box

To make this work, they built a massive, custom-made laboratory that looks like a giant photography studio.

• The Stage: They have a huge, flat floor (about 20 feet by 11 feet) that can be tilted.
• The Dam: At one end, there is a tank holding water. When they pull a pin, a gate drops, and the water rushes out.
• The Lighting Trick: This is the most important part. Instead of shining a spotlight at the water (which would create blinding reflections), they built a giant box around the whole floor. Inside the box, they hung 60 bright LED lights pointing up at the ceiling. The ceiling bounces the light back down, creating a soft, even, shadow-free glow that washes over the entire floor. It's like being inside a giant, glowing cloud.

3. The Secret Ingredient: Colored Water

To measure depth with light, they needed the water to act like a filter.

• The Dye: They added a special green food coloring to the water. Think of this like adding a tint to a window. The deeper the water, the "darker" the tint becomes.
• The Test: Before the big experiment, they tested different colors (red, yellow, blue, and green) to see which one blocked the most light. They found that a green mixture was the best "light blocker" for their specific cameras and lights.

4. The Magic Formula: From Gray to Depth

Here is how they turned a picture into a depth map:

1. The Camera: Two high-speed scientific cameras sit on the ceiling, looking straight down. They take pictures of the green water flowing over the floor.
2. The Logic:
   • Where the water is shallow, the light passes through easily, and the camera sees a bright image.
   • Where the water is deep, the green dye absorbs more light, and the camera sees a darker image.
3. The Math: The researchers realized that a simple math rule (used for single-color light) wasn't accurate enough because their lights were "broadband" (containing many colors). So, they invented a new, slightly more complex math formula (a "bi-exponential model") that perfectly translates the darkness of the pixel into the exact depth of the water.

5. The Proof: Did It Work?

They ran the experiment 15 times with different amounts of water in the tank.

• Repeatability: They checked if they got the same result every time. The answer was yes; the measurements were incredibly consistent.
• The "Volume" Check: To be absolutely sure, they did a second check. They used ultrasonic sensors (like bat sonar) inside the tank to measure how much water left the tank. They compared this to the total volume calculated from their camera images.
• The Result: The two numbers matched almost perfectly. This proved that their camera method was accurate.

The Bottom Line

The researchers built a giant, glowing room where they could watch a dam break in slow motion. By adding green dye and using a special math formula, they turned a video camera into a precise 3D depth scanner. They proved that you can measure the depth of a fast-moving, chaotic flood wave with high accuracy, just by looking at how dark the water looks in a photo.

This gives engineers a powerful new tool to understand floods better, without needing to stick hundreds of sensors into the dangerous, rushing water.

Technical Summary

Technical Summary: Seeing through Water – Diffuse Image-Based Depth Measurements in 3D Dam-Break Flows

Problem Statement
Dam-break flows represent a critical challenge in hydraulic engineering and risk assessment. While real-world dam failures are inherently three-dimensional, particularly for small-to-medium reservoirs releasing water onto unconfined, gently sloping terrains, experimental research has historically focused on two-dimensional channel configurations. A significant gap exists in the systematic experimental characterization of long-term, three-dimensional dam-break propagation over wide, effectively two-dimensional surfaces. Furthermore, measuring the free surface in such rapidly varying, unsteady flows remains a major challenge. Traditional intrusive methods (e.g., pressure transducers) fail under accelerating flow conditions where hydrostatic assumptions break down, while non-intrusive point sensors (e.g., ultrasonic gauges) struggle with steep or aerated surfaces and require dense arrays to reconstruct spatial fields. Existing optical methods like LiDAR or stereo photogrammetry often face trade-offs between spatial coverage and temporal simultaneity, limiting their utility for capturing the instantaneous evolution of dam-break waves.

Methodology
The authors present a dedicated large-scale experimental facility and an image-based measurement technique designed to reconstruct water depth fields, $h(x, y, t)$, in three-dimensional dam-break flows.

• Experimental Facility: The setup features a prismatic stainless-steel reservoir ($3.38 \text{ m} \times 0.96 \text{ m} \times 0.6 \text{ m}$) with a removable vertical breach ($0.05 \text{ m}$ wide). Water is released onto a $6.4 \text{ m} \times 3.4 \text{ m}$ composite plane floor, which can be inclined from $0^\circ$ to $30^\circ$. The entire domain is enclosed within a light box ($4.3 \text{ m} \times 6.8 \text{ m} \times 6 \text{ m}$) lined with highly reflective white panels. Illumination is provided by an array of 60 LED floodlights mounted 1 meter above the floor and tilted $30^\circ$ upward to ensure highly uniform diffuse illumination while avoiding direct specular reflections off the free surface.
• Imaging System: Two scientific CMOS cameras (Andor Zyla, 5.5 MP) are mounted on the ceiling of the light box, capturing 16-bit grayscale images at $2560 \times 2160$ resolution.
• Measurement Principle: The technique relies on light attenuation through dyed water. The water is mixed with a soluble colorant, and the transmitted light intensity is recorded.
  • Spectral Characterization: Preliminary dry calibration tests evaluated four food-grade dyes (red, yellow, blue, and a green mixture) against the broadband LED emission and CMOS sensor sensitivity. A green mixture (5% Tartrazine, 5% Patent Blue V) was selected for its superior attenuation within the sensor's effective spectral band.
  • Calibration Model: Due to the broadband nature of the illumination and the spectral sensitivity of the sensor, the standard monochromatic Beer-Lambert law was found insufficient. Instead, a bi-exponential model was introduced to relate the normalized gray level ($g_n$) to the optical path length ($l_p$):
    $$g_n = C_1 \exp(-c_1 l_p) + C_2 \exp(-c_2 l_p)$$
    This model was calibrated in situ using standing water levels ranging from 1.3 mm to 44.6 mm, with corrections applied for optical path length based on Snell's law to account for non-normal incidence.
• Validation: The reconstructed depth fields were validated against independent volume estimates. These included:
  1. An array of 12 ultrasonic level sensors in the reservoir to track volume loss.
  2. A simplified analytical emptying model based on Bernoulli's equation and continuity, calibrated with a discharge coefficient ($C_c$).

Key Contributions

1. Novel Facility Design: The construction of a large-scale, enclosed light box capable of providing homogeneous diffuse illumination over a $6.4 \text{ m} \times 3.4 \text{ m}$ area, overcoming the limitations of conventional backlighting for large domains.
2. Advanced Imaging Technique: The development of a robust image-based depth measurement method that explicitly accounts for the spectral mismatch between broadband LEDs and camera sensors. The introduction of the bi-exponential attenuation model significantly improves accuracy over traditional monochromatic approximations.
3. Systematic 3D Characterization: The execution of a preliminary campaign of 15 experiments (three initial reservoir levels, five repetitions each) to assess statistical repeatability and characterize the spatial and temporal evolution of fully three-dimensional dam-break waves.

Results

• Repeatability: The ensemble average depth maps ($\langle h \rangle$) and standard deviation maps ($\sigma_h$) demonstrated high consistency across repeated tests. The largest variability was confined to the advancing wave front and lateral regions where surface instabilities occur, while the central propagation sector exhibited very low variability ($\sigma_h \sim 0.3 \text{ mm}$), confirming the robustness of the measurement procedure.
• Flow Dynamics: The results showed that front propagation velocity increases almost linearly with the initial reservoir level ($H_0$). The free surface was observed to be very steep near the breach but became significantly flatter downstream, justifying the assumption of a locally horizontal free surface for depth reconstruction in the main propagation zone.
• Volume Conservation: The total volume of the wave calculated from the imaged depth fields ($V^c$) showed excellent agreement with the theoretical emptying model and the volume loss measured by ultrasonic sensors. The discrepancy was minimal, with the in-tank volume varying by less than 5% over time, supporting the validity of the simplified boundary conditions used in the analytical model.

Significance
The paper claims that this work provides a reliable, high-resolution, and spatially distributed method for measuring water depth in rapidly varying, three-dimensional free-surface flows. By validating the imaging-derived depths against independent continuity-based estimates, the authors establish the methodology as a robust benchmark. This capability is presented as essential for the physical interpretation of complex dam-break phenomena and, crucially, for the validation of advanced numerical models (such as 3D lattice Boltzmann, SPH, and VoF solvers) which currently lack detailed experimental data for unconfined, three-dimensional configurations. The facility and technique offer a practical route for future investigations into the effects of breach geometry, bed inclination, and surface roughness on dam-break flows.