Overcoming "Physics Shock" in Earth Observation A Heteroscedastic Uncertainty Framework for PINN-based Flood Inference

Jagrati Talreja; Leila Hashemi-Beni; Matilda Anokye; Tewodros Syum Gebre

arxiv: 2605.24106 · v1 · pith:VKNLKBB3new · submitted 2026-05-22 · 💻 cs.LG · cs.AI

Overcoming "Physics Shock" in Earth Observation A Heteroscedastic Uncertainty Framework for PINN-based Flood Inference

Tewodros Syum Gebre , Jagrati Talreja , Matilda Anokye , Leila Hashemi-Beni This is my paper

Pith reviewed 2026-06-30 16:34 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords PINNflood mappingheteroscedastic uncertaintySARphysics shockearth observationuncertainty quantification

0 comments

The pith

Modeling sensor uncertainty lets PINNs enforce flood physics only where SAR data is reliable, raising IoU by 25 percent.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that standard PINNs fail on real SAR flood data because enforcing physical equations like the shallow water model on noisy inputs produces gradient divergence, termed physics shock. By adding a warm-start protocol and a heteroscedastic uncertainty term in the loss, the network learns to down-weight the physics penalty in high-noise regions and keep it strong in reliable regions. This stabilizes training and produces more accurate flood maps. A reader would care because flood mapping from satellite radar is used in disaster response, where physically impossible outputs can mislead decisions.

Core claim

By integrating a dynamic Warm-Start protocol and modeling heteroscedastic aleatoric uncertainty via a negative log-likelihood objective, the network learns to dynamically relax physical constraints in regions of high sensor noise while strictly enforcing them in high-confidence areas, stabilizing multi-objective optimization on noisy SAR inputs and yielding higher-fidelity flood extent predictions.

What carries the argument

Heteroscedastic aleatoric uncertainty modeled in the loss function that scales the weight of the physics residual term according to local data confidence.

If this is right

Training remains stable on real SAR flood data instead of diverging.
Flood extent maps achieve higher overlap with ground truth than deterministic PINNs.
Deep ensembles separate sensor noise from terrain ignorance to give calibrated uncertainty maps.
The outputs remain consistent with governing hydrological equations in high-confidence zones.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same uncertainty-weighting pattern could be tested on other physics-informed remote-sensing tasks such as soil moisture retrieval where observation noise also varies spatially.
An ablation that removes the warm-start protocol would show whether the uncertainty term alone is sufficient to prevent divergence.
Operational agencies could use the resulting per-pixel is high enough to trigger automated alerts without manual review.

Load-bearing premise

The assumption that catastrophic gradient divergence in PINNs comes mainly from applying rigid spatial derivatives to unconditioned noisy SAR data, and that uncertainty weighting can selectively relax those constraints without introducing new inconsistencies.

What would settle it

Training the deterministic baseline on the Sen1Floods11 dataset and checking whether gradient norms explode during optimization while the uncertainty version stays stable and the reported IoU gap disappears when input noise is synthetically lowered.

Figures

Figures reproduced from arXiv: 2605.24106 by Jagrati Talreja, Leila Hashemi-Beni, Matilda Anokye, Tewodros Syum Gebre.

**Figure 1.** Figure 1: Conceptual illustration of the Physics Shock phenomenon. When rigid physical constraints are enforced on unconditioned [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Figure 2: Schematic overview of the Uncertainty-Aware Physics [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Figure 3: Training dynamics illustrating the ‘Physics Shock’ and its mitigation. (A) In a baseline deterministic PINN, the immediate [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Training dynamics alternative perspectives with loss scaling and gradient behavior. (A) The physics loss trajectories [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 5.** Figure 5: Quantitative Performance Summary: Probabilistic vs. Deterministic Scaling. A) Mean Validation IoU across ablated architectures, highlighting the +25% relative improvement of the Uncertainty-Aware PINN over the baseline. Error bars denote ± 1 standard deviation. B) Performance vs. Robustness scatter plot demonstrating that the SOTA model is both the most accurate and the most stable (lowest standard deviati… view at source ↗

**Figure 6.** Figure 6: Qualitative flood mapping results (Part 1 of 2). D. Peak Operational Performance and Training Dynamics To fully evaluate the peak capabilities of the proposed uncertainty-aware framework, the optimal model configuration (Uncertainty-Aware PINN) was scaled to a broader data distribution, utilizing both the hand-labeled imagery and the extensive corpus of weakly-labeled samples from the Sen1Floods11 datase… view at source ↗

**Figure 7.** Figure 7: Qualitative flood mapping results (Part 2 of 2). the loss components (Figure 8A-B) demonstrate smooth, stable convergence, entirely avoiding the gradient divergence typical of standard PINNs. The final aggregated quantitative performance across the validation holdout is detailed in Table II. The validation metrics highlight a significant leap in mapping accuracy: the model achieves an exceptional peak Inte… view at source ↗

**Figure 8.** Figure 8: Training dynamics and validation performance during model training. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗

**Figure 9.** Figure 9: Flood prediction results visualization using a probabilistic framework. Each row corresponds to a different sample, with [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗

**Figure 10.** Figure 10: Uncertainty Calibration and Error Correlation. Empirical Mean Squared Error (MSE) of the predicted flood depth plotted against bins of the model’s predicted aleatoric variance (σ 2 ). Error bars denote the standard error of the mean (SEM) within each bin. The strong positive correlation (Linear Fit R2 = 0.9794) demonstrates that the probabilistic head is well-calibrated; regions where the network predicts… view at source ↗

**Figure 11.** Figure 11: Disentangling Aleatoric and Epistemic Uncertainty via Deep Ensembles. The top row establishes the predictive context: (A) the noisy Input SAR (VH) signal, (B) the hydraulically simulated ground truth depth, and (C) the ensemble’s mean predicted depth (µ∗). The bottom row isolates the sources of doubt: (D) Aleatoric uncertainty successfully maps intrinsic sensor noise (e.g., speckle and radar shadow bounda… view at source ↗

read the original abstract

Rapid and accurate flood extent mapping from Remote Sensing data, such as Synthetic Aperture Radar (SAR), is critical for operational disaster response, but standard Deep Learning models often produce physically impossible predictions due to a lack of hydrological constraints. While PhysicsInformed Neural Networks (PINNs) attempt to address this by embedding governing laws directly into the loss function, their application to real-world remote sensing data frequently fails. Enforcing rigid spatial derivatives (e.g., the 2D Shallow Water Equations) onto unconditioned latent spaces attempting to fit noisy SAR speckle causes catastrophic gradient divergence, a phenomenon we term Physics Shock. In this paper, we propose a novel Uncertainty-Aware PINN framework tailored specifically for applied Earth Observation that addresses this instability. By integrating a dynamic Warm-Start protocol and modeling heteroscedastic aleatoric uncertainty via a negative log-likelihood objective, the network learns to dynamically relax physical constraints in regions of high sensor noise while strictly enforcing them in high-confidence areas. Evaluated on the Sen1Floods11 dataset, our probabilistic Attention-Gated FNO-UNet successfully stabilizes multi-objective optimization, achieving a +25% relative improvement in Intersection over Union (IoU) compared to deterministic baselines. Furthermore, through Deep Ensembles, we successfully disentangle intrinsic sensor noise from out-of-distribution terrain ignorance, providing operational agencies with highly calibrated, physically consistent confidence bounds for robust disaster mitigation and real-time decision-making.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper offers a practical uncertainty-aware PINN setup for SAR flood mapping that claims to dodge gradient blow-ups, but the evidence does not confirm the physics constraints are being relaxed selectively.

read the letter

The core idea is to stabilize PINNs on real SAR data by letting the model down-weight the shallow-water physics terms where sensor noise is high. They do this with a warm-start schedule plus a heteroscedastic negative-log-likelihood loss, wrapped in an attention-gated FNO-UNet, and they report a 25% IoU lift on Sen1Floods11 over plain deterministic runs. They also run deep ensembles to separate aleatoric sensor noise from epistemic ignorance.

That combination is the actual new piece: applying heteroscedastic uncertainty specifically to gate physics enforcement in an Earth-observation setting, and giving the failure mode a name. The operational angle is reasonable; flood-mapping pipelines do need outputs that stay physically plausible under speckle.

The soft spot is exactly what the stress-test flagged. The claim that constraints are enforced only in low-uncertainty regions rests on the IoU number alone. No PDE residual maps or quantitative checks are described that would show lower residuals where uncertainty is low. Without those, the gain could come from the architecture, the NLL objective, or the ensembles rather than from any selective physics mechanism. The abstract also skips baseline details, ablations, and error bars, so the 25% figure is difficult to interpret.

This is for groups already working on physics-informed remote sensing who need a concrete recipe for noisy inputs. It is coherent on its own terms and engages the literature honestly, so it deserves a serious referee to check whether the full paper supplies the missing residual diagnostics and controls.

Referee Report

1 major / 2 minor

Summary. The manuscript proposes a heteroscedastic uncertainty-aware PINN framework for flood extent mapping from SAR imagery that combines a dynamic Warm-Start protocol with negative log-likelihood training to model aleatoric uncertainty. This is claimed to mitigate 'Physics Shock' (catastrophic gradient divergence when enforcing 2D Shallow Water Equations on noisy data) by selectively relaxing constraints in high-uncertainty regions while enforcing them in low-uncertainty regions. The Attention-Gated FNO-UNet is evaluated on Sen1Floods11 and reports a +25% relative IoU gain over deterministic baselines; Deep Ensembles are additionally used to separate sensor noise from epistemic uncertainty.

Significance. If the selective constraint enforcement mechanism is verified, the framework could meaningfully extend PINN applicability to noisy real-world Earth-observation tasks by providing both improved segmentation accuracy and calibrated, physically consistent uncertainty estimates for operational use. The explicit treatment of heteroscedastic aleatoric uncertainty and the warm-start stabilization strategy address a documented practical failure mode of physics-informed models on SAR data.

major comments (1)

[Abstract] Abstract: the central claim that heteroscedastic uncertainty modeling plus warm-start enables selective relaxation of 2D Shallow Water Equation residuals (lower residuals in low-uncertainty pixels, higher in high-uncertainty pixels) is not supported by any reported PDE residual analysis, ablation on residual maps, or quantitative comparison against a standard PINN baseline. The sole quantitative result is the +25% relative IoU improvement, which could arise from the Attention-Gated FNO-UNet architecture, the NLL objective, or ensembling rather than from stabilized physics enforcement.

minor comments (2)

[Abstract] Abstract: baseline details, statistical tests, ablation results, and error-bar information for the reported IoU metric are absent, preventing assessment of the empirical claim.
The manuscript does not describe how the warm-start protocol is scheduled or how the uncertainty threshold for constraint relaxation is chosen; these implementation choices are load-bearing for reproducibility.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive feedback. We address the major comment regarding support for the selective relaxation claim below.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim that heteroscedastic uncertainty modeling plus warm-start enables selective relaxation of 2D Shallow Water Equation residuals (lower residuals in low-uncertainty pixels, higher in high-uncertainty pixels) is not supported by any reported PDE residual analysis, ablation on residual maps, or quantitative comparison against a standard PINN baseline. The sole quantitative result is the +25% relative IoU improvement, which could arise from the Attention-Gated FNO-UNet architecture, the NLL objective, or ensembling rather than from stabilized physics enforcement.

Authors: We agree that the current manuscript lacks explicit PDE residual analysis, residual map ablations, or direct quantitative comparison to a standard PINN baseline, making it difficult to isolate the contribution of selective constraint relaxation from other components. The +25% IoU gain is the primary operational metric, but additional physics-consistency metrics are needed to support the central claim. In the revised version we will add: (1) average PDE residual values for our method versus a deterministic PINN baseline, (2) residual maps, and (3) residual statistics conditioned on uncertainty level to verify lower residuals in low-uncertainty regions. These will help demonstrate that the heteroscedastic modeling and warm-start stabilize physics enforcement. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical performance claim with no self-referential derivation

full rationale

The paper presents an empirical ML method (heteroscedastic PINN with warm-start and NLL objective) evaluated on Sen1Floods11 for IoU gains. No equations, fitted parameters, or derivations are described that reduce to their own inputs by construction. No self-citations are invoked as load-bearing uniqueness theorems, and the central result is a reported performance delta rather than a mathematical identity. This matches the default case of a self-contained empirical contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies insufficient detail to enumerate concrete free parameters, background axioms, or new postulated entities beyond standard neural-network components and the descriptive label Physics Shock.

pith-pipeline@v0.9.1-grok · 5806 in / 1316 out tokens · 46328 ms · 2026-06-30T16:34:42.140598+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

27 extracted references · 12 canonical work pages

[1]

Residual wave vision u-net for flood mapping using dual polarization sentinel-1 sar imagery,

A. Jamali, S. K. Roy, L. H. Beni, B. Pradhan, J. Li, and P. Ghamisi, “Residual wave vision u-net for flood mapping using dual polarization sentinel-1 sar imagery,”International Journal of Applied Earth Obser- vation and Geoinformation, vol. 127, p. 103662, 2024

2024
[2]

Deepflood for inundated vegetation high-resolution dataset for accurate flood mapping and segmentation,

M. Fawakherji, J. Blay, M. Anokye, L. Hashemi-Beni, and J. Dorton, “Deepflood for inundated vegetation high-resolution dataset for accurate flood mapping and segmentation,”Scientific Data, vol. 12, no. 1, p. 271, 2025

2025
[3]

Flood detection with sar: A review of techniques and datasets,

D. Amitrano, G. Di Martino, A. Di Simone, and P. Imperatore, “Flood detection with sar: A review of techniques and datasets,”Remote Sensing, vol. 16, no. 4, p. 656, 2024

2024
[4]

Inundated vegetation mapping using sar data: A comparison of polarization configurations of uavsar l-band and sentinel c-band,

A. Salem and L. Hashemi-Beni, “Inundated vegetation mapping using sar data: A comparison of polarization configurations of uavsar l-band and sentinel c-band,”Remote Sensing, vol. 14, no. 24, p. 6374, 2022

2022
[5]

Flood extent mapping: An integrated method using deep learning and region growing using uav optical data,

L. Hashemi-Beni and A. A. Gebrehiwot, “Flood extent mapping: An integrated method using deep learning and region growing using uav optical data,”IEEE Journal of Selected Topics in Applied Earth Obser- vations and Remote Sensing, vol. 14, pp. 2127–2135, 2021

2021
[6]

Multi-head encoder-decoder deep learning architecture for flood segmentation and mapping through multi-sensor data fusion,

M. Fawakherji and L. Hashemi-Beni, “Multi-head encoder-decoder deep learning architecture for flood segmentation and mapping through multi-sensor data fusion,” inIGARSS 2024-2024 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2024, pp. 1191– 1195

2024
[7]

Physics-informed neural networks for solving the two-dimensional shallow water equations with terrain topography and rainfall source terms,

Y . Tian, S. Ding, L. Huang, G. Su, and J. Chen, “Physics-informed neural networks for solving the two-dimensional shallow water equations with terrain topography and rainfall source terms,”Water Resources Research, vol. 61, no. 9, p. e2025WR040052, 2025

2025
[8]

Ai-integrated traffic information system: A synergistic approach of physics informed 16 neural network and gpt-4 for traffic estimation and real-time assistance,

T. S. Gebre, L. Beni, E. T. Wasehun, and F. E. Dorbu, “Ai-integrated traffic information system: A synergistic approach of physics informed 16 neural network and gpt-4 for traffic estimation and real-time assistance,” IEEE Access, vol. 12, pp. 65 869–65 882, 2024

2024
[9]

Real-time traffic insights with physics-informed neural networks: Integrating the aw- rascle model and llms,

T. S. Gebre, S. E. Ashebir, J. Blay, and L. H. Beni, “Real-time traffic insights with physics-informed neural networks: Integrating the aw- rascle model and llms,”IEEE Access, vol. PP, no. 99, pp. 1–1, 2025

2025
[10]

An integrated framework of gpt-4 and pinn for dynamic traffic estimation and support,

T. S. Gebre and L. Hashemi-Beni, “An integrated framework of gpt-4 and pinn for dynamic traffic estimation and support,” inIGARSS 2024- 2024 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2024, pp. 5457–5460

2024
[11]

A change detection approach to flood mapping in urban areas using terrasar-x,

L. Giustarini, R. Hostache, P. Matgen, G. J.-P. Schumann, P. D. Bates, and D. C. Mason, “A change detection approach to flood mapping in urban areas using terrasar-x,”IEEE Transactions on Geoscience and Remote Sensing, vol. 51, no. 4, p. 2417–2430, Apr. 2013. [Online]. Available: http://dx.doi.org/10.1109/TGRS.2012.2210901

work page doi:10.1109/tgrs.2012.2210901 2013
[12]

Towards operational near real-time flood detection using a split-based automatic thresholding procedure on high resolution terrasar-x data,

S. Martinis, A. Twele, and S. V oigt, “Towards operational near real-time flood detection using a split-based automatic thresholding procedure on high resolution terrasar-x data,”Natural Hazards and Earth System Sciences, vol. 9, no. 2, p. 303–314, Mar. 2009. [Online]. Available: http://dx.doi.org/10.5194/nhess-9-303-2009

work page doi:10.5194/nhess-9-303-2009 2009
[13]

Multichannel deep learning-based architecture for flood detection and mapping,

M. Fawakherji and L. H. Beni, “Multichannel deep learning-based architecture for flood detection and mapping,” inAGU Fall Meeting Abstracts, vol. 2023, no. 722, 2023, pp. NH41B–0722

2023
[14]

Flood impact risk map- ping in settlement areas from a 3d perspective: A case study of hurricane matthew,

J. Blay, M. Fawakherji, and L. Hashemi-Beni, “Flood impact risk map- ping in settlement areas from a 3d perspective: A case study of hurricane matthew,” inIGARSS 2024-2024 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2024, pp. 3939–3942

2024
[15]

Ronneberger, P

O. Ronneberger, P. Fischer, and T. Brox,U-Net: Convolutional Networks for Biomedical Image Segmentation. Springer International Publishing, 2015, p. 234–241. [Online]. Available: http://dx.doi.org/10. 1007/978-3-319-24574-4\ 28

2015
[16]

Rtformer: Efficient design for real-time semantic segmentation with transformer,

E. Ding, H. Feng, C. Gou, J. Han, J. Wang, J. Wang, and Q. Wu, “Rtformer: Efficient design for real-time semantic segmentation with transformer,” inAdvances in Neural Information Processing Systems 35, ser. NeurIPS 2022. Neural Information Processing Systems Foundation, Inc. (NeurIPS), 2022, p. 7423–7436. [Online]. Available: http://dx.doi.org/10.52202/06...

work page doi:10.52202/068431-0539 2022
[17]

Sen1floods11: a georeferenced dataset to train and test deep learning flood algorithms for sentinel-1,

D. Bonafilia, B. Tellman, T. Anderson, and E. Issenberg, “Sen1floods11: a georeferenced dataset to train and test deep learning flood algorithms for sentinel-1,” in2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). IEEE, Jun. 2020, p. 835–845. [Online]. Available: http://dx.doi.org/10.1109/CVPRW50498. 2020.00113

work page doi:10.1109/cvprw50498 2020
[18]

Flood detection in urban areas using terrasar-x,

D. Mason, R. Speck, B. Devereux, G.-P. Schumann, J. Neal, and P. Bates, “Flood detection in urban areas using terrasar-x,”IEEE Transactions on Geoscience and Remote Sensing, vol. 48, no. 2, p. 882–894, Feb. 2010. [Online]. Available: http://dx.doi.org/10.1109/TGRS.2009.2029236

work page doi:10.1109/tgrs.2009.2029236 2010
[19]

Nature Reviews Physics , author =

G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, “Physics-informed machine learning,”Nature Reviews Physics, vol. 3, no. 6, p. 422–440, May 2021. [Online]. Available: http://dx.doi.org/10.1038/s42254-021-00314-5

work page doi:10.1038/s42254-021-00314-5 2021
[20]

Smart traffic solutions: Enhanced uav-based traffic mon- itoring, object detection and ai-enhanced guidance system,

T. S. Gebre, “Smart traffic solutions: Enhanced uav-based traffic mon- itoring, object detection and ai-enhanced guidance system,” Ph.D. dis- sertation, North Carolina Agricultural and Technical State University, 2025

2025
[21]

Physics-informed neural networks: A deep learn- ing framework for solving forward and inverse problems involving nonlinear partial differential equations

M. Raissi, P. Perdikaris, and G. Karniadakis, “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,”Journal of Computational Physics, vol. 378, p. 686–707, Feb. 2019. [Online]. Available: http://dx.doi.org/10.1016/j.jcp.2018.10.045

work page doi:10.1016/j.jcp.2018.10.045 2019
[22]

Physics-informed neural networks for high-speed flows,

Z. Mao, A. D. Jagtap, and G. E. Karniadakis, “Physics-informed neural networks for high-speed flows,”Computer Methods in Applied Mechanics and Engineering, vol. 360, p. 112789, Mar. 2020. [Online]. Available: http://dx.doi.org/10.1016/j.cma.2019.112789

work page doi:10.1016/j.cma.2019.112789 2020
[23]

Understanding and mitigating gradient flow pathologies in physics-informed neural networks,

S. Wang, Y . Teng, and P. Perdikaris, “Understanding and mitigating gradient flow pathologies in physics-informed neural networks,”SIAM Journal on Scientific Computing, vol. 43, no. 5, p. A3055–A3081, Jan
[24]

Available: http://dx.doi.org/10.1137/20M1318043

[Online]. Available: http://dx.doi.org/10.1137/20M1318043

work page doi:10.1137/20m1318043
[25]

When and why PINNs fail to train: A neural tangent kernel perspective.Journal of Computational Physics, 449:110768, 2022

S. Wang, X. Yu, and P. Perdikaris, “When and why pinns fail to train: A neural tangent kernel perspective,”Journal of Computational Physics, vol. 449, p. 110768, Jan. 2022. [Online]. Available: http://dx.doi.org/10.1016/j.jcp.2021.110768

work page doi:10.1016/j.jcp.2021.110768 2022
[26]

Be confident in what you know: Bayesian parameter efficient fine-tuning of vision foundation models,

D. Pandey, S. Pyakurel, and Q. Yu, “Be confident in what you know: Bayesian parameter efficient fine-tuning of vision foundation models,” inAdvances in Neural Information Processing Systems 37, ser. NeurIPS 2024. Neural Information Processing Systems Foundation, Inc. (NeurIPS), 2024, p. 44814–44844. [Online]. Available: http://dx.doi.org/10.52202/079017-1424

work page doi:10.52202/079017-1424 2024
[27]

Evidential fuzzy rule-based machine learning to quantify classification uncertainty,

H. Shiraishi, H. Ishibuchi, and M. Nakata, “Evidential fuzzy rule-based machine learning to quantify classification uncertainty,” Jun. 2025. [Online]. Available: http://dx.doi.org/10.36227/techrxiv. 174900814.43483543/v1 Tewodros Syum Gebreholds a B.S. in Hydraulic Engineering from Arba Minch University, Ethiopia, in 2009, an M.S. in Civil (Road and Trans...

work page doi:10.36227/techrxiv 2025

[1] [1]

Residual wave vision u-net for flood mapping using dual polarization sentinel-1 sar imagery,

A. Jamali, S. K. Roy, L. H. Beni, B. Pradhan, J. Li, and P. Ghamisi, “Residual wave vision u-net for flood mapping using dual polarization sentinel-1 sar imagery,”International Journal of Applied Earth Obser- vation and Geoinformation, vol. 127, p. 103662, 2024

2024

[2] [2]

Deepflood for inundated vegetation high-resolution dataset for accurate flood mapping and segmentation,

M. Fawakherji, J. Blay, M. Anokye, L. Hashemi-Beni, and J. Dorton, “Deepflood for inundated vegetation high-resolution dataset for accurate flood mapping and segmentation,”Scientific Data, vol. 12, no. 1, p. 271, 2025

2025

[3] [3]

Flood detection with sar: A review of techniques and datasets,

D. Amitrano, G. Di Martino, A. Di Simone, and P. Imperatore, “Flood detection with sar: A review of techniques and datasets,”Remote Sensing, vol. 16, no. 4, p. 656, 2024

2024

[4] [4]

Inundated vegetation mapping using sar data: A comparison of polarization configurations of uavsar l-band and sentinel c-band,

A. Salem and L. Hashemi-Beni, “Inundated vegetation mapping using sar data: A comparison of polarization configurations of uavsar l-band and sentinel c-band,”Remote Sensing, vol. 14, no. 24, p. 6374, 2022

2022

[5] [5]

Flood extent mapping: An integrated method using deep learning and region growing using uav optical data,

L. Hashemi-Beni and A. A. Gebrehiwot, “Flood extent mapping: An integrated method using deep learning and region growing using uav optical data,”IEEE Journal of Selected Topics in Applied Earth Obser- vations and Remote Sensing, vol. 14, pp. 2127–2135, 2021

2021

[6] [6]

Multi-head encoder-decoder deep learning architecture for flood segmentation and mapping through multi-sensor data fusion,

M. Fawakherji and L. Hashemi-Beni, “Multi-head encoder-decoder deep learning architecture for flood segmentation and mapping through multi-sensor data fusion,” inIGARSS 2024-2024 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2024, pp. 1191– 1195

2024

[7] [7]

Physics-informed neural networks for solving the two-dimensional shallow water equations with terrain topography and rainfall source terms,

Y . Tian, S. Ding, L. Huang, G. Su, and J. Chen, “Physics-informed neural networks for solving the two-dimensional shallow water equations with terrain topography and rainfall source terms,”Water Resources Research, vol. 61, no. 9, p. e2025WR040052, 2025

2025

[8] [8]

Ai-integrated traffic information system: A synergistic approach of physics informed 16 neural network and gpt-4 for traffic estimation and real-time assistance,

T. S. Gebre, L. Beni, E. T. Wasehun, and F. E. Dorbu, “Ai-integrated traffic information system: A synergistic approach of physics informed 16 neural network and gpt-4 for traffic estimation and real-time assistance,” IEEE Access, vol. 12, pp. 65 869–65 882, 2024

2024

[9] [9]

Real-time traffic insights with physics-informed neural networks: Integrating the aw- rascle model and llms,

T. S. Gebre, S. E. Ashebir, J. Blay, and L. H. Beni, “Real-time traffic insights with physics-informed neural networks: Integrating the aw- rascle model and llms,”IEEE Access, vol. PP, no. 99, pp. 1–1, 2025

2025

[10] [10]

An integrated framework of gpt-4 and pinn for dynamic traffic estimation and support,

T. S. Gebre and L. Hashemi-Beni, “An integrated framework of gpt-4 and pinn for dynamic traffic estimation and support,” inIGARSS 2024- 2024 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2024, pp. 5457–5460

2024

[11] [11]

A change detection approach to flood mapping in urban areas using terrasar-x,

L. Giustarini, R. Hostache, P. Matgen, G. J.-P. Schumann, P. D. Bates, and D. C. Mason, “A change detection approach to flood mapping in urban areas using terrasar-x,”IEEE Transactions on Geoscience and Remote Sensing, vol. 51, no. 4, p. 2417–2430, Apr. 2013. [Online]. Available: http://dx.doi.org/10.1109/TGRS.2012.2210901

work page doi:10.1109/tgrs.2012.2210901 2013

[12] [12]

Towards operational near real-time flood detection using a split-based automatic thresholding procedure on high resolution terrasar-x data,

S. Martinis, A. Twele, and S. V oigt, “Towards operational near real-time flood detection using a split-based automatic thresholding procedure on high resolution terrasar-x data,”Natural Hazards and Earth System Sciences, vol. 9, no. 2, p. 303–314, Mar. 2009. [Online]. Available: http://dx.doi.org/10.5194/nhess-9-303-2009

work page doi:10.5194/nhess-9-303-2009 2009

[13] [13]

Multichannel deep learning-based architecture for flood detection and mapping,

M. Fawakherji and L. H. Beni, “Multichannel deep learning-based architecture for flood detection and mapping,” inAGU Fall Meeting Abstracts, vol. 2023, no. 722, 2023, pp. NH41B–0722

2023

[14] [14]

Flood impact risk map- ping in settlement areas from a 3d perspective: A case study of hurricane matthew,

J. Blay, M. Fawakherji, and L. Hashemi-Beni, “Flood impact risk map- ping in settlement areas from a 3d perspective: A case study of hurricane matthew,” inIGARSS 2024-2024 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2024, pp. 3939–3942

2024

[15] [15]

Ronneberger, P

O. Ronneberger, P. Fischer, and T. Brox,U-Net: Convolutional Networks for Biomedical Image Segmentation. Springer International Publishing, 2015, p. 234–241. [Online]. Available: http://dx.doi.org/10. 1007/978-3-319-24574-4\ 28

2015

[16] [16]

Rtformer: Efficient design for real-time semantic segmentation with transformer,

E. Ding, H. Feng, C. Gou, J. Han, J. Wang, J. Wang, and Q. Wu, “Rtformer: Efficient design for real-time semantic segmentation with transformer,” inAdvances in Neural Information Processing Systems 35, ser. NeurIPS 2022. Neural Information Processing Systems Foundation, Inc. (NeurIPS), 2022, p. 7423–7436. [Online]. Available: http://dx.doi.org/10.52202/06...

work page doi:10.52202/068431-0539 2022

[17] [17]

Sen1floods11: a georeferenced dataset to train and test deep learning flood algorithms for sentinel-1,

D. Bonafilia, B. Tellman, T. Anderson, and E. Issenberg, “Sen1floods11: a georeferenced dataset to train and test deep learning flood algorithms for sentinel-1,” in2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). IEEE, Jun. 2020, p. 835–845. [Online]. Available: http://dx.doi.org/10.1109/CVPRW50498. 2020.00113

work page doi:10.1109/cvprw50498 2020

[18] [18]

Flood detection in urban areas using terrasar-x,

D. Mason, R. Speck, B. Devereux, G.-P. Schumann, J. Neal, and P. Bates, “Flood detection in urban areas using terrasar-x,”IEEE Transactions on Geoscience and Remote Sensing, vol. 48, no. 2, p. 882–894, Feb. 2010. [Online]. Available: http://dx.doi.org/10.1109/TGRS.2009.2029236

work page doi:10.1109/tgrs.2009.2029236 2010

[19] [19]

Nature Reviews Physics , author =

G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, “Physics-informed machine learning,”Nature Reviews Physics, vol. 3, no. 6, p. 422–440, May 2021. [Online]. Available: http://dx.doi.org/10.1038/s42254-021-00314-5

work page doi:10.1038/s42254-021-00314-5 2021

[20] [20]

Smart traffic solutions: Enhanced uav-based traffic mon- itoring, object detection and ai-enhanced guidance system,

T. S. Gebre, “Smart traffic solutions: Enhanced uav-based traffic mon- itoring, object detection and ai-enhanced guidance system,” Ph.D. dis- sertation, North Carolina Agricultural and Technical State University, 2025

2025

[21] [21]

Physics-informed neural networks: A deep learn- ing framework for solving forward and inverse problems involving nonlinear partial differential equations

M. Raissi, P. Perdikaris, and G. Karniadakis, “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,”Journal of Computational Physics, vol. 378, p. 686–707, Feb. 2019. [Online]. Available: http://dx.doi.org/10.1016/j.jcp.2018.10.045

work page doi:10.1016/j.jcp.2018.10.045 2019

[22] [22]

Physics-informed neural networks for high-speed flows,

Z. Mao, A. D. Jagtap, and G. E. Karniadakis, “Physics-informed neural networks for high-speed flows,”Computer Methods in Applied Mechanics and Engineering, vol. 360, p. 112789, Mar. 2020. [Online]. Available: http://dx.doi.org/10.1016/j.cma.2019.112789

work page doi:10.1016/j.cma.2019.112789 2020

[23] [23]

Understanding and mitigating gradient flow pathologies in physics-informed neural networks,

S. Wang, Y . Teng, and P. Perdikaris, “Understanding and mitigating gradient flow pathologies in physics-informed neural networks,”SIAM Journal on Scientific Computing, vol. 43, no. 5, p. A3055–A3081, Jan

[24] [24]

Available: http://dx.doi.org/10.1137/20M1318043

[Online]. Available: http://dx.doi.org/10.1137/20M1318043

work page doi:10.1137/20m1318043

[25] [25]

When and why PINNs fail to train: A neural tangent kernel perspective.Journal of Computational Physics, 449:110768, 2022

S. Wang, X. Yu, and P. Perdikaris, “When and why pinns fail to train: A neural tangent kernel perspective,”Journal of Computational Physics, vol. 449, p. 110768, Jan. 2022. [Online]. Available: http://dx.doi.org/10.1016/j.jcp.2021.110768

work page doi:10.1016/j.jcp.2021.110768 2022

[26] [26]

Be confident in what you know: Bayesian parameter efficient fine-tuning of vision foundation models,

D. Pandey, S. Pyakurel, and Q. Yu, “Be confident in what you know: Bayesian parameter efficient fine-tuning of vision foundation models,” inAdvances in Neural Information Processing Systems 37, ser. NeurIPS 2024. Neural Information Processing Systems Foundation, Inc. (NeurIPS), 2024, p. 44814–44844. [Online]. Available: http://dx.doi.org/10.52202/079017-1424

work page doi:10.52202/079017-1424 2024

[27] [27]

Evidential fuzzy rule-based machine learning to quantify classification uncertainty,

H. Shiraishi, H. Ishibuchi, and M. Nakata, “Evidential fuzzy rule-based machine learning to quantify classification uncertainty,” Jun. 2025. [Online]. Available: http://dx.doi.org/10.36227/techrxiv. 174900814.43483543/v1 Tewodros Syum Gebreholds a B.S. in Hydraulic Engineering from Arba Minch University, Ethiopia, in 2009, an M.S. in Civil (Road and Trans...

work page doi:10.36227/techrxiv 2025