Bayesian reversal of the liquid level trajectory in a draining tank for pollution forensics
Kyla D. Jones, Gbenga Fabusola, Alexander W. Dowling, Cory M. Simon ∗ · 2026
Read Bayesian reversal of the liquid level trajectory in a draining tank for pollution forensics by Kyla D. Jones, Gbenga Fabusola, Alexander W. Dowling, Cory M. Simon ∗ (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-29193.txt.
Introduction
Kyla D. Jones et al. present a Bayesian model for finding the initial liquid level in a partially drained tank after a spill. It infers the level from the final level and a drainage time guess. With drainage dynamics, parameters, and readings unsure, the Bayesian method joins physics with test data, sizing the doubt. A Torricelli's-law model covers the draining, with a hands-on gap for flawed physics. In water-tank experiments the found level proves accurate; uncertainty grows with drainage duration.
Catalog record
| CREATOR | Kyla D. Jones, Gbenga Fabusola, Alexander W. Dowling, Cory M. Simon ∗ |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45759 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.29193
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Bayesian reversal of the liquid level trajectory in a draining tank for pollution forensics Kyla D. Jones, Gbenga Fabusola, Alexander W. Dowling, Cory M. Simon ∗ Abstract Storage tanks for hazardous liquids are common in industry and agriculture. During a pollution incident, liquid may drain from a storage tank through a small hole, crack, or pipe. After containing the leak, estimating the discharged volume of liquid is essential for public safety, regulatory assessment, and remediation. When the original inventory of liquid is unknown, this constitutes an inverse problem. In this work, we present a framework for inferring the initial liquid level in a partially drained tank from the observed final liquid level after a pollution incident and an estimate of the drainage duration. Because the drainage dynamics, model parameters, and observations are uncertain, we employ Bayesian statistical inversion to combine prior physical knowledge with experimental liquid level time series data to predict the initial liquid level with quantified uncertainty. We use a physics-based model based on Torricelli’s law to describe the tank-draining dynamics and augment it with an empirical discrepancy function to account for missing or imperfectly modeled physics. In our experiments with a tank draining of water, we found that our inferred initial liquid level was accurate, although uncertainty increased with drainage duration. Beyond its application to pollution forensics, this work may also serve as a hands-on classroom project illustrating dynamic modeling, model discrepancy, and Bayesian inference. 1 Introduction Large above- or under-ground tanks are widely used in the chemical and petroleum industries to store hazardous liquids such as solvents, additives, precursors, intermediate or finished products, and waste [ 27 , 45 ] . In agriculture, above-ground tanks on farms store liquid fertilizers, fuel, pesticides, and livestock slurry [ 55 ] . The wine industry ferments and stores wine in large tanks [ 41 ] . Additionally, the U.S. Department of Energy’s Hanford Site has more than 100 underground tanks storing radioactive and chemical waste generated during the separation of plutonium from irradiated fuel rods for nuclear weapons production [ 43 ] . At Fukushima [ 54 ] , about 1000 tanks store cooling water, rainwater runoff, and groundwater contaminated with radioactive Tritium (as of 2022) [ 2 ] . Equipment failure, operator error, negligence, and sabotage have caused unintended liquid discharges from storage tanks [ 10 , 25 , 21 , 55 ] . These incidents can cause illness, injury, disability, or death among exposed personnel [ 6 , 12 ] . Further, soil, groundwater, and surface water can be contaminated, negatively affecting ecosystems, food, and drinking water [ 38 , 60 ] . Thus, regulatory, liability, and legal consequences can follow. Engineers can reduce the risk of accidental releases by designing tanks appropriately (e.g., materials selection) and performing routine maintenance and inspections [ 10 ] . Additionally, installing chemical sensors near storage tanks enables early leak detection and prompt responses such as discharge stoppage, spread containment, evacuation, and cleanup [ 48 , 23 , 35 ] . This paper concerns incident forensics for a storage tank that partially drained under gravity through a small hole, crack, or discharge pipe. Estimating the volume of liquid discharged is important for assessing the degree of environmental contamination. When the initial liquid inventory is unknown, reconstructing the initial liquid level from the observed final liquid level and the estimated drain time becomes an inverse problem [ 20 ] . Conceptually, we want to reverse the observed final liquid level backwards in time, to the estimated beginning of the drainage event. Quantifying uncertainty in the inferred initial liquid inventory is also important because decisions regarding remediation, fines, and safety may depend on it. To address this problem, we employ a probabilistic framework that combines a forward model of the drainage dynamics with a stochastic observation model. Bayesian calibration methods [ 40 , 26 , 39 , 28 ] provide a natural approach for incorporating prior physical knowledge, experimental liquid level time series data, model discrepancy, and observation uncertainty within a unified inference framework. Bayesian statistical inversion and Markov chain Monte Carlo methods [ 30 , 9 , 14 , 58 ] then enable joint inference of uncertain model parameters, observation noise, and the unknown initial liquid level associated with the pollution event. This work provides a practical framework for pollution forensics involving drainage incidents in storage tanks. More fundamentally, it contributes to the study of inverse problems involving draining tanks [ 20 , 1 , 16 ] . Given the accessibility of the experiment and problem setting, the problem may also serve as a hands-on classroom project illustrating concepts in dynamic modeling, inverse problems, model discrepancy, and Bayesian inference [ 53 , 36 , 44 ] . 2 Problem Statement Figure 1 : Reconstructing a tank drainage event. Liquid drains from a storage tank, under the force of gravity, through a small orifice (in the side of the tank and at the bottom) over the time interval [t_{0},t^{\prime}] . The objective is to infer the unknown initial liquid level h_{0} from the observed final liquid level h^{\prime}_{\rm obs} and the drainage duration. Fig. 1 summarizes the problem setup. An above-ground storage tank holds a hazardous liquid. The initial height of the liquid in the tank, h_{0} [cm], is unknown. At time t=t_{0} [s], the tank suddenly begins draining through a small orifice near its bottom due to the hydrostatic pressure generated by the liquid column caused by gravity, contaminating the surrounding area. Owing to a vent, the headspace in the tank remains at atmospheric pressure during draining. An in situ gas sensor outside the tank alerts onsite personnel to the discharge. At time t^{\prime}>t_{0} , the personnel seal the orifice to stop the drainage. The remaining height of liquid, h^{\prime}_{\rm obs} [cm], is measured. Inspection of sensor logs and video surveillance gives an estimate of the drainage time span [t_{0},t^{\prime}] . The objective is to estimate, with quantified uncertainty, the initial height of liquid h_{0} in the tank to calculate the volume of hazardous liquid released into the environment. 3 Methods 3.1 Experimental setup and water level time series data collection The tank used in the experiments was an open-top plastic kitchen container approximately shaped as a truncated inverted square pyramid. To allow drainage, a small hole was drilled into the side of the tank near its base. To monitor the liquid level over time, centimeter-scale markings were placed along the tank’s side. Each experiment began by filling the tank with tap water to a prescribed initial liquid level, with the drainage hole plugged. At time t=0 , the plug was removed to initiate gravity-driven drainage through the orifice. The liquid level was then recorded over time as the tank drained. Liquid level time series data. We conduct and generate liquid level time series data from three replicate tank-draining experiments. The first experiment aims to mimic the pollution scenario in Fig. 1 by generating a single data point concerning the final liquid level: \mathcal{D}^{\prime}=\{(t^{\prime},h_{\rm obs}(t^{\prime}))\}. We note the initial condition (0,h_{0,\rm obs}) , but hold it out from our inference procedure to test our ability to accurately solve the inverse problem. The purpose of the second and third experiments is to calibrate