Learning Displacement-Robust Representations for Landslide Early Warning under Rainfall Forecast Uncertainty

Hamada Rizk; Hirozumi Yamaguchi; Ren Ozeki

arxiv: 2605.17419 · v1 · pith:BXRRBAG2new · submitted 2026-05-17 · 💻 cs.LG · cs.AI

Learning Displacement-Robust Representations for Landslide Early Warning under Rainfall Forecast Uncertainty

Ren Ozeki , Hamada Rizk , Hirozumi Yamaguchi This is my paper

Pith reviewed 2026-05-20 13:51 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords landslide early warningrainfall forecast uncertaintydisplacement robust representationscontrastive learningspatio-temporal environmental dataJapan regional landslide eventsshort-term rainfall prediction

0 comments

The pith

A contrastive learning method trains landslide models to stay accurate despite spatial shifts in rainfall forecasts.

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

The paper seeks to fix a practical failure in landslide early warning systems: short-term rainfall forecasts often displace rain fields slightly, which changes local totals and breaks predictions that assume perfect inputs. The authors introduce Rainfall-Motion-Aware Contrastive Learning to force the model to extract latent features from rainfall and terrain that remain consistent under such motion-like perturbations. These stable representations are then used for risk estimation. Experiments on two years of data across 19 Japanese regions show the approach raises precision by as much as 37 percent over current baselines. If the method generalizes, operational systems could issue more trustworthy near-future warnings even when forecasts contain typical spatial uncertainty.

Core claim

The central claim is that Rainfall-Motion-Aware Contrastive Learning, by introducing temporally correlated perturbations that emulate forecast-induced rainfall field displacements, produces latent representations from rainfall and terrain data that remain stable, thereby enabling reliable integration of uncertain short-term forecasts into landslide risk estimates.

What carries the argument

Rainfall-Motion-Aware Contrastive Learning (RMCL), which applies motion-aware perturbations to rainfall fields during training to enforce displacement-invariant representations for downstream risk prediction.

If this is right

Landslide early warning systems can integrate short-term forecasts without suffering large accuracy drops from typical spatial errors.
Precision improvements of up to 37 percent are observed on two years of rainfall and terrain data from 19 Japanese regions with recorded events.
More reliable near-future risk estimates become possible, supporting earlier and more confident evacuation decisions.
The same training strategy addresses the mismatch between assumed accurate inputs and the displaced fields present in operational forecast streams.

Where Pith is reading between the lines

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

The same perturbation approach could be tested on other spatial forecast tasks such as flood or wildfire risk where small location shifts also matter.
A direct comparison using live operational forecasts instead of synthetic perturbations would provide a stronger test of real-world utility.
Adding intensity or timing perturbations alongside spatial ones might reveal whether displacement is the dominant source of forecast error.

Load-bearing premise

That the simulated temporally correlated perturbations used in training are close enough to the actual spatial displacement patterns found in real operational rainfall forecasts.

What would settle it

Apply the trained model to a held-out set of real short-term rainfall forecasts paired with observed landslide outcomes and measure whether the reported precision gain disappears.

Figures

Figures reproduced from arXiv: 2605.17419 by Hamada Rizk, Hirozumi Yamaguchi, Ren Ozeki.

**Figure 2.** Figure 2: Rainfall Motion-Aware Contrastive Learning (RMCL) illustration. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 4.** Figure 4: Ablation study on RMCL D. Effect of Rainfall-Motion-Aware Contrastive Learning To evaluate the effectiveness of Rainfall-Motion-Aware Contrastive Learning (RMCL), we conducted an ablation study comparing three training strategies: (i) standard supervised learning using focal loss (End-to-end), (ii) supervised training using forecast rainfall as input (End-to-end w/ Forecast Rain), and (iii) the proposed RM… view at source ↗

read the original abstract

Rainfall-induced landslides pose a growing risk worldwide as climate change intensifies extreme rainfall events. To provide sufficient evacuation time, landslide early warning systems (LEWS) for real-time disaster monitoring must estimate near-future landslide risk by integrating observed rainfall with short-term rainfall forecasts from spatio-temporal environmental data streams. Although recent landslide prediction methods have improved predictive performance using statistical and deep learning approaches, most assume accurate rainfall inputs. In operational settings, however, landslide prediction relies on rainfall forecasts, which often contain spatial displacement of rainfall fields due to forecasting uncertainties. Such displacement can alter local accumulated rainfall and degrade prediction accuracy. To address this challenge, we propose a novel LEWS robust to rainfall field displacement. The key idea is to learn latent representations from rainfall and terrain data that remain stable under displacement in rainfall field motion, enabling reliable geospatial data integration for landslide risk estimation. The landslide prediction model is trained using Rainfall-Motion-Aware Contrastive Learning (RMCL), which introduces temporally correlated rainfall field perturbations to emulate forecast-induced displacement in rainfall-driven spatio-temporal environmental data streams. Experiments were conducted using two years of rainfall and terrain data across Japan, covering 19 regions with landslide events. The proposed system achieved up to 37% higher precision than state-of-the-art baselines. These results demonstrate that modeling rainfall as a moving spatial field and addressing rainfall field displacement during learning significantly improve the reliability of short-term landslide prediction in operational early warning systems.

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 adapts contrastive learning with temporally correlated rainfall perturbations to handle forecast displacements in landslide prediction, but offers no direct check that those perturbations match real forecast error patterns.

read the letter

The main takeaway is that this work targets a practical gap: landslide early warning systems need to stay accurate when short-term rainfall forecasts shift the rainfall field spatially. They introduce Rainfall-Motion-Aware Contrastive Learning that adds those shifts during training so the model learns stable representations from rainfall and terrain data. That specific adaptation for moving rainfall fields is the clearest new element compared with earlier landslide models that treat inputs as fixed and accurate. The experiments run on two years of data from 19 Japanese regions with actual events, and they report up to 37% higher precision than the baselines they tested. That gives the claim some empirical footing in a real setting. The soft spot is the missing link between the training perturbations and actual forecast displacements. The abstract describes the perturbation mechanism but does not show any quantitative match, such as comparing displacement histograms or spatial correlation lengths against paired observed-versus-forecast fields from the same regions. Without that, it is unclear whether the learned invariance will hold when the model sees live forecasts. Details on baseline implementations, exact displacement ranges tested, and any statistical testing are also thin in the summary, which makes the size of the gain harder to judge. This is worth a serious referee for groups working on robust ML for environmental forecasting and disaster systems. A reader focused on handling input uncertainty in spatio-temporal prediction would get concrete ideas from the training procedure, even if the validation needs tightening before operational use. I would send it out for review rather than desk reject.

Referee Report

2 major / 2 minor

Summary. The paper proposes Rainfall-Motion-Aware Contrastive Learning (RMCL) to learn latent representations from rainfall and terrain data that remain stable under spatial displacements in rainfall fields. These displacements arise from uncertainties in short-term rainfall forecasts used for landslide early warning systems (LEWS). The method introduces temporally correlated rainfall field perturbations during training to emulate forecast-induced errors, enabling more reliable integration of geospatial data for near-future landslide risk estimation. Experiments on two years of rainfall and terrain data across 19 Japanese regions with landslide events report up to 37% higher precision than state-of-the-art baselines.

Significance. If the synthetic perturbations reproduce the spatial and temporal statistics of real operational forecast displacements, the approach could meaningfully improve the robustness and reliability of LEWS in settings where forecast errors degrade local accumulated rainfall estimates. The work targets a concrete operational gap between assumed-accurate rainfall inputs and actual forecast streams, and the empirical gains on a multi-region Japanese dataset provide a falsifiable starting point for further validation.

major comments (2)

[§3.2] §3.2 (RMCL perturbation mechanism): The claim that temporally correlated rainfall field perturbations emulate forecast-induced displacement rests on an unverified distributional match; no quantitative comparison (displacement magnitude histograms, spatial correlation lengths, or temporal autocorrelation functions) is reported between the synthetic perturbations and paired observed-vs-forecast rainfall fields over the 19 regions.
[§5] §5 (Experimental results): The headline 37% precision improvement is presented without specification of baseline implementations, the exact range of displacement magnitudes used in testing, statistical significance testing, or the cross-validation procedure, preventing verification that the reported gains are attributable to displacement robustness rather than other factors.

minor comments (2)

[Abstract] The abstract and §4 do not state the precise two-year time window or the public data sources for rainfall and terrain, which would aid reproducibility.
[§3] Notation for the contrastive loss in RMCL is introduced without an explicit equation or algorithm box, making the training objective harder to follow.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive feedback and for recognizing the operational relevance of our work on displacement-robust representations for landslide early warning. We address each major comment below and indicate the revisions we will undertake.

read point-by-point responses

Referee: [§3.2] §3.2 (RMCL perturbation mechanism): The claim that temporally correlated rainfall field perturbations emulate forecast-induced displacement rests on an unverified distributional match; no quantitative comparison (displacement magnitude histograms, spatial correlation lengths, or temporal autocorrelation functions) is reported between the synthetic perturbations and paired observed-vs-forecast rainfall fields over the 19 regions.

Authors: We agree that a direct distributional comparison would strengthen the justification for the perturbation design. The temporally correlated perturbations in RMCL are constructed to reflect typical spatial displacement patterns documented in short-term rainfall forecast literature, with correlation lengths chosen to approximate forecast error scales. However, our experimental dataset comprises observed rainfall and terrain fields without paired forecast realizations for the 19 regions, so the requested quantitative match cannot be computed from available data. In the revision we will expand §3.2 to explicitly describe the perturbation generation process, cite the meteorological references motivating the correlation parameters, and state the limitation regarding direct validation against operational forecasts. revision: partial
Referee: [§5] §5 (Experimental results): The headline 37% precision improvement is presented without specification of baseline implementations, the exact range of displacement magnitudes used in testing, statistical significance testing, or the cross-validation procedure, preventing verification that the reported gains are attributable to displacement robustness rather than other factors.

Authors: We accept that the current presentation of results in §5 lacks sufficient detail for full reproducibility and attribution. The 37% figure is the maximum observed precision gain across the 19 regions relative to the strongest baseline under the tested displacement conditions. In the revised manuscript we will augment §5 with: explicit descriptions and hyper-parameter settings for all baselines; the precise spatial displacement magnitudes (in km) applied during both training and testing; results of statistical significance tests (paired t-tests across regions with reported p-values); and the cross-validation scheme (temporal hold-out with region-wise stratification). These additions will clarify that the gains arise from the displacement-aware contrastive objective rather than other modeling choices. revision: yes

standing simulated objections not resolved

Direct quantitative comparison of synthetic perturbations against paired observed-vs-forecast rainfall fields over the 19 regions, because the dataset contains only observed rainfall and does not include corresponding forecast fields.

Circularity Check

0 steps flagged

No significant circularity; empirical training procedure with independent evaluation.

full rationale

The paper describes an empirical method: RMCL introduces temporally correlated rainfall field perturbations during training to learn displacement-robust representations, then reports precision gains on two years of real rainfall/terrain data across 19 Japanese regions. No equations, derivations, or self-citations are presented that reduce the 37% improvement to a fitted parameter, self-definition, or input by construction. The evaluation uses held-out operational-style data, making the central claim falsifiable outside the training perturbations themselves. This is a standard non-circular empirical ML setup.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The approach rests on standard contrastive learning assumptions plus the untested modeling choice that synthetic perturbations match real forecast displacement statistics.

axioms (1)

domain assumption Contrastive learning with appropriate augmentations produces representations invariant to the chosen perturbations.
Invoked when stating that RMCL learns stable representations under rainfall field motion.

invented entities (1)

Rainfall-Motion-Aware Contrastive Learning (RMCL) no independent evidence
purpose: Training objective that introduces temporally correlated perturbations to emulate forecast displacement.
New training procedure introduced in the paper.

pith-pipeline@v0.9.0 · 5791 in / 1254 out tokens · 36618 ms · 2026-05-20T13:51:19.849257+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages

[1]

The global costs of extreme weather that are attributable to climate change,

R. Newman and I. Noy, “The global costs of extreme weather that are attributable to climate change,”Nature Communications, vol. 14, no. 1, p. 6103, 2023

work page 2023
[2]

State of the global climate 2021,

W. M. Associationet al., “State of the global climate 2021,” 2022

work page 2021
[3]

A study on a method for determining non-linear critical line of slope failures during heavy rainfall based on rbf network,

K. KURAMOTO, H. TETSUGA, N. HIGASHI, M. ARAKAW A, H. NAKAY AMA, and K. FURUKAW A, “A study on a method for determining non-linear critical line of slope failures during heavy rainfall based on rbf network,”Doboku Gakkai Ronbunshu, vol. 2001, no. 672, pp. 117–132, 2001

work page 2001
[4]

Deep learning forecast of rainfall-induced shallow landslides,

A. C. Mondini, F. Guzzetti, and M. Melillo, “Deep learning forecast of rainfall-induced shallow landslides,”Nature communications, vol. 14, no. 1, p. 2466, 2023

work page 2023
[5]

Landslide detection, monitoring and prediction with remote-sensing techniques,

N. Casagli, E. Intrieri, V . Tofani, G. Gigli, and F. Raspini, “Landslide detection, monitoring and prediction with remote-sensing techniques,” Nature Reviews Earth & Environment, vol. 4, no. 1, pp. 51–64, 2023

work page 2023
[6]

Advection-free convolutional neural network for convective rainfall nowcasting,

J. Ritvanen, B. Harnist, M. Aldana, T. M ¨akinen, and S. Pulkkinen, “Advection-free convolutional neural network for convective rainfall nowcasting,”IEEE Journal of Selected Topics in Applied Earth Ob- servations and Remote Sensing, vol. 16, pp. 1654–1667, 2023

work page 2023
[7]

Determination of rainfall thresholds for landslide prediction using an algorithm-based approach: Case study in the darjeeling himalayas, india,

T. S. Teja, A. Dikshit, and N. Satyam, “Determination of rainfall thresholds for landslide prediction using an algorithm-based approach: Case study in the darjeeling himalayas, india,”Geosciences, vol. 9, no. 7, 2019

work page 2019
[8]

Performance evaluation of a physically based model for shallow landslide prediction,

J.-Y . Ho and K. T. Lee, “Performance evaluation of a physically based model for shallow landslide prediction,”Landslides, vol. 14, no. 3, 2017

work page 2017
[9]

Development of a novel hybrid intelligence approach for landslide spatial prediction,

P. T. Nguyen, T. T. Tuyen, A. Shirzadi, B. T. Pham, H. Shahabi, E. Omidvar, A. Amini, H. Entezami, I. Prakash, T. V . Phonget al., “Development of a novel hybrid intelligence approach for landslide spatial prediction,”Applied Sciences, vol. 9, no. 14, p. 2824, 2019

work page 2019
[10]

Algorithms for intelligent prediction of landslide displacements,

Z. Liu, D. Guo, S. Lacasse, J.-h. Li, B. Yang, and J. C. Choi, “Algorithms for intelligent prediction of landslide displacements,” 2020

work page 2020
[11]

Litetransnet: An interpretable approach for landslide displacement prediction using transformer model with attention mechanism,

Q. Ge, J. Li, X. Wang, Y . Deng, K. Zhang, and H. Sun, “Litetransnet: An interpretable approach for landslide displacement prediction using transformer model with attention mechanism,”Engineering Geology, vol. 331, p. 107446, 2024

work page 2024
[12]

Landslide susceptibility mapping considering landslide local-global features based on cnn and transformer,

Z. Zhao, T. Chen, J. Dou, G. Liu, and A. Plaza, “Landslide susceptibility mapping considering landslide local-global features based on cnn and transformer,”IEEE Journal of Selected Topics in Applied Earth Obser- vations and Remote Sensing, vol. 17, pp. 7475–7489, 2024

work page 2024
[13]

Investigating the potential of a global precipitation forecast to inform landslide prediction,

S. Khan, D. Kirschbaum, and T. Stanley, “Investigating the potential of a global precipitation forecast to inform landslide prediction,”Weather and Climate Extremes, vol. 33, p. 100364, 2021

work page 2021
[14]

From spatio-temporal landslide susceptibility to landslide risk forecast,

T. Wang, A. Dahal, Z. Fang, C. van Westen, K. Yin, and L. Lombardo, “From spatio-temporal landslide susceptibility to landslide risk forecast,” Geoscience Frontiers, vol. 15, no. 2, p. 101765, 2024

work page 2024
[15]

An attention-based teacher-student model for multivariate short-term landslide displacement prediction incorporating weather forecast data,

J. Chen, W. Hu, Y . Zhang, H. Qiu, and R. Wang, “An attention-based teacher-student model for multivariate short-term landslide displacement prediction incorporating weather forecast data,”Journal of Mountain Science, vol. 21, no. 8, pp. 2739–2753, 2024

work page 2024
[16]

Predicting and understanding landslide events with explainable ai,

E. Collini, L. I. Palesi, P. Nesi, G. Pantaleo, N. Nocentini, and A. Rosi, “Predicting and understanding landslide events with explainable ai,” IEEE Access, vol. 10, pp. 31 175–31 189, 2022

work page 2022
[17]

The application of long short-term mem- ory (lstm) method on displacement prediction of multifactor-induced landslides,

P. Xie, A. Zhou, and B. Chai, “The application of long short-term mem- ory (lstm) method on displacement prediction of multifactor-induced landslides,”IEEE Access, vol. 7, pp. 54 305–54 311, 2019

work page 2019
[18]

Resam- pling techniques for materials informatics: limitations in crystal point groups classification,

A. A. Alsaui, Y . A. Alghofaili, M. Alghadeer, and F. H. Alharbi, “Resam- pling techniques for materials informatics: limitations in crystal point groups classification,”Journal of Chemical Information and Modeling, vol. 62, no. 15, pp. 3514–3523, 2022

work page 2022
[19]

Credit card fraud detection under extreme imbalanced data: a comparative study of data-level algo- rithms,

A. Singh, R. K. Ranjan, and A. Tiwari, “Credit card fraud detection under extreme imbalanced data: a comparative study of data-level algo- rithms,”Journal of Experimental & Theoretical Artificial Intelligence, vol. 34, no. 4, pp. 571–598, 2022

work page 2022
[20]

Under- sampling with support vectors for multi-class imbalanced data classi- fication,

B. Krawczyk, C. Bellinger, R. Corizzo, and N. Japkowicz, “Under- sampling with support vectors for multi-class imbalanced data classi- fication,” in2021 International Joint Conference on Neural Networks (IJCNN), 2021, pp. 1–7

work page 2021
[21]

Smote: synthetic minority over-sampling technique,

N. V . Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “Smote: synthetic minority over-sampling technique,”Journal of artificial intel- ligence research, vol. 16, pp. 321–357, 2002

work page 2002
[22]

Deepsmote: Fusing deep learning and smote for imbalanced data,

D. Dablain, B. Krawczyk, and N. V . Chawla, “Deepsmote: Fusing deep learning and smote for imbalanced data,”IEEE Transactions on Neural Networks and Learning Systems, 2022

work page 2022
[23]

Using cost-sensitive learning and feature selection algorithms to improve the performance of imbalanced classification,

F. Feng, K.-C. Li, J. Shen, Q. Zhou, and X. Yang, “Using cost-sensitive learning and feature selection algorithms to improve the performance of imbalanced classification,”IEEE Access, vol. 8, 2020

work page 2020
[24]

Training deep neural networks on imbalanced data sets,

S. Wang, W. Liu, J. Wu, L. Cao, Q. Meng, and P. J. Kennedy, “Training deep neural networks on imbalanced data sets,” in2016 International Joint Conference on Neural Networks (IJCNN), 2016, pp. 4368–4374

work page 2016
[25]

Focal loss for dense object detection,

T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Doll ´ar, “Focal loss for dense object detection,” inProceedings of the IEEE international conference on computer vision, 2017, pp. 2980–2988

work page 2017
[26]

Addressing the class imbalance problem in twitter spam detection using ensemble learning,

S. Liu, Y . Wang, J. Zhang, C. Chen, and Y . Xiang, “Addressing the class imbalance problem in twitter spam detection using ensemble learning,” Computers&Security, vol. 69, pp. 35–49, 2017

work page 2017
[27]

Random balance: Ensembles of variable priors classifiers for imbal- anced data,

J. F. D ´ıez-Pastor, J. J. Rodr´ıguez, C. Garc´ıa-Osorio, and L. I. Kuncheva, “Random balance: Ensembles of variable priors classifiers for imbal- anced data,”Knowledge-Based Systems, vol. 85, pp. 96–111, 2015

work page 2015
[28]

Pyramidal implementation of the affine lucas kanade feature tracker description of the algorithm,

J.-Y . Bouguetet al., “Pyramidal implementation of the affine lucas kanade feature tracker description of the algorithm,”Intel corporation, vol. 5, no. 1-10, p. 4, 2001

work page 2001
[29]

Pysteps: An open-source python library for probabilistic precipitation nowcasting (v1. 0),

S. Pulkkinen, D. Nerini, A. A. P ´erez Hortal, C. Velasco-Forero, A. Seed, U. Germann, and L. Foresti, “Pysteps: An open-source python library for probabilistic precipitation nowcasting (v1. 0),”Geoscientific Model Development, vol. 12, no. 10, pp. 4185–4219, 2019

work page 2019
[30]

Supervised contrastive learn- ing,

P. Khosla, P. Teterwak, C. Wang, A. Sarna, Y . Tian, P. Isola, A. Maschinot, C. Liu, and D. Krishnan, “Supervised contrastive learn- ing,”Advances in neural information processing systems, vol. 33, pp. 18 661–18 673, 2020

work page 2020
[31]

Quantitative precipitation estimation and quantitative precip- itation forecasting by the japan meteorological agency,

K. Nagata, “Quantitative precipitation estimation and quantitative precip- itation forecasting by the japan meteorological agency,”RSMC Tokyo– Typhoon Center Technical Review, vol. 13, pp. 37–50, 2011

work page 2011

[1] [1]

The global costs of extreme weather that are attributable to climate change,

R. Newman and I. Noy, “The global costs of extreme weather that are attributable to climate change,”Nature Communications, vol. 14, no. 1, p. 6103, 2023

work page 2023

[2] [2]

State of the global climate 2021,

W. M. Associationet al., “State of the global climate 2021,” 2022

work page 2021

[3] [3]

A study on a method for determining non-linear critical line of slope failures during heavy rainfall based on rbf network,

K. KURAMOTO, H. TETSUGA, N. HIGASHI, M. ARAKAW A, H. NAKAY AMA, and K. FURUKAW A, “A study on a method for determining non-linear critical line of slope failures during heavy rainfall based on rbf network,”Doboku Gakkai Ronbunshu, vol. 2001, no. 672, pp. 117–132, 2001

work page 2001

[4] [4]

Deep learning forecast of rainfall-induced shallow landslides,

A. C. Mondini, F. Guzzetti, and M. Melillo, “Deep learning forecast of rainfall-induced shallow landslides,”Nature communications, vol. 14, no. 1, p. 2466, 2023

work page 2023

[5] [5]

Landslide detection, monitoring and prediction with remote-sensing techniques,

N. Casagli, E. Intrieri, V . Tofani, G. Gigli, and F. Raspini, “Landslide detection, monitoring and prediction with remote-sensing techniques,” Nature Reviews Earth & Environment, vol. 4, no. 1, pp. 51–64, 2023

work page 2023

[6] [6]

Advection-free convolutional neural network for convective rainfall nowcasting,

J. Ritvanen, B. Harnist, M. Aldana, T. M ¨akinen, and S. Pulkkinen, “Advection-free convolutional neural network for convective rainfall nowcasting,”IEEE Journal of Selected Topics in Applied Earth Ob- servations and Remote Sensing, vol. 16, pp. 1654–1667, 2023

work page 2023

[7] [7]

Determination of rainfall thresholds for landslide prediction using an algorithm-based approach: Case study in the darjeeling himalayas, india,

T. S. Teja, A. Dikshit, and N. Satyam, “Determination of rainfall thresholds for landslide prediction using an algorithm-based approach: Case study in the darjeeling himalayas, india,”Geosciences, vol. 9, no. 7, 2019

work page 2019

[8] [8]

Performance evaluation of a physically based model for shallow landslide prediction,

J.-Y . Ho and K. T. Lee, “Performance evaluation of a physically based model for shallow landslide prediction,”Landslides, vol. 14, no. 3, 2017

work page 2017

[9] [9]

Development of a novel hybrid intelligence approach for landslide spatial prediction,

P. T. Nguyen, T. T. Tuyen, A. Shirzadi, B. T. Pham, H. Shahabi, E. Omidvar, A. Amini, H. Entezami, I. Prakash, T. V . Phonget al., “Development of a novel hybrid intelligence approach for landslide spatial prediction,”Applied Sciences, vol. 9, no. 14, p. 2824, 2019

work page 2019

[10] [10]

Algorithms for intelligent prediction of landslide displacements,

Z. Liu, D. Guo, S. Lacasse, J.-h. Li, B. Yang, and J. C. Choi, “Algorithms for intelligent prediction of landslide displacements,” 2020

work page 2020

[11] [11]

Litetransnet: An interpretable approach for landslide displacement prediction using transformer model with attention mechanism,

Q. Ge, J. Li, X. Wang, Y . Deng, K. Zhang, and H. Sun, “Litetransnet: An interpretable approach for landslide displacement prediction using transformer model with attention mechanism,”Engineering Geology, vol. 331, p. 107446, 2024

work page 2024

[12] [12]

Landslide susceptibility mapping considering landslide local-global features based on cnn and transformer,

Z. Zhao, T. Chen, J. Dou, G. Liu, and A. Plaza, “Landslide susceptibility mapping considering landslide local-global features based on cnn and transformer,”IEEE Journal of Selected Topics in Applied Earth Obser- vations and Remote Sensing, vol. 17, pp. 7475–7489, 2024

work page 2024

[13] [13]

Investigating the potential of a global precipitation forecast to inform landslide prediction,

S. Khan, D. Kirschbaum, and T. Stanley, “Investigating the potential of a global precipitation forecast to inform landslide prediction,”Weather and Climate Extremes, vol. 33, p. 100364, 2021

work page 2021

[14] [14]

From spatio-temporal landslide susceptibility to landslide risk forecast,

T. Wang, A. Dahal, Z. Fang, C. van Westen, K. Yin, and L. Lombardo, “From spatio-temporal landslide susceptibility to landslide risk forecast,” Geoscience Frontiers, vol. 15, no. 2, p. 101765, 2024

work page 2024

[15] [15]

An attention-based teacher-student model for multivariate short-term landslide displacement prediction incorporating weather forecast data,

J. Chen, W. Hu, Y . Zhang, H. Qiu, and R. Wang, “An attention-based teacher-student model for multivariate short-term landslide displacement prediction incorporating weather forecast data,”Journal of Mountain Science, vol. 21, no. 8, pp. 2739–2753, 2024

work page 2024

[16] [16]

Predicting and understanding landslide events with explainable ai,

E. Collini, L. I. Palesi, P. Nesi, G. Pantaleo, N. Nocentini, and A. Rosi, “Predicting and understanding landslide events with explainable ai,” IEEE Access, vol. 10, pp. 31 175–31 189, 2022

work page 2022

[17] [17]

The application of long short-term mem- ory (lstm) method on displacement prediction of multifactor-induced landslides,

P. Xie, A. Zhou, and B. Chai, “The application of long short-term mem- ory (lstm) method on displacement prediction of multifactor-induced landslides,”IEEE Access, vol. 7, pp. 54 305–54 311, 2019

work page 2019

[18] [18]

Resam- pling techniques for materials informatics: limitations in crystal point groups classification,

A. A. Alsaui, Y . A. Alghofaili, M. Alghadeer, and F. H. Alharbi, “Resam- pling techniques for materials informatics: limitations in crystal point groups classification,”Journal of Chemical Information and Modeling, vol. 62, no. 15, pp. 3514–3523, 2022

work page 2022

[19] [19]

Credit card fraud detection under extreme imbalanced data: a comparative study of data-level algo- rithms,

A. Singh, R. K. Ranjan, and A. Tiwari, “Credit card fraud detection under extreme imbalanced data: a comparative study of data-level algo- rithms,”Journal of Experimental & Theoretical Artificial Intelligence, vol. 34, no. 4, pp. 571–598, 2022

work page 2022

[20] [20]

Under- sampling with support vectors for multi-class imbalanced data classi- fication,

B. Krawczyk, C. Bellinger, R. Corizzo, and N. Japkowicz, “Under- sampling with support vectors for multi-class imbalanced data classi- fication,” in2021 International Joint Conference on Neural Networks (IJCNN), 2021, pp. 1–7

work page 2021

[21] [21]

Smote: synthetic minority over-sampling technique,

N. V . Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “Smote: synthetic minority over-sampling technique,”Journal of artificial intel- ligence research, vol. 16, pp. 321–357, 2002

work page 2002

[22] [22]

Deepsmote: Fusing deep learning and smote for imbalanced data,

D. Dablain, B. Krawczyk, and N. V . Chawla, “Deepsmote: Fusing deep learning and smote for imbalanced data,”IEEE Transactions on Neural Networks and Learning Systems, 2022

work page 2022

[23] [23]

Using cost-sensitive learning and feature selection algorithms to improve the performance of imbalanced classification,

F. Feng, K.-C. Li, J. Shen, Q. Zhou, and X. Yang, “Using cost-sensitive learning and feature selection algorithms to improve the performance of imbalanced classification,”IEEE Access, vol. 8, 2020

work page 2020

[24] [24]

Training deep neural networks on imbalanced data sets,

S. Wang, W. Liu, J. Wu, L. Cao, Q. Meng, and P. J. Kennedy, “Training deep neural networks on imbalanced data sets,” in2016 International Joint Conference on Neural Networks (IJCNN), 2016, pp. 4368–4374

work page 2016

[25] [25]

Focal loss for dense object detection,

T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Doll ´ar, “Focal loss for dense object detection,” inProceedings of the IEEE international conference on computer vision, 2017, pp. 2980–2988

work page 2017

[26] [26]

Addressing the class imbalance problem in twitter spam detection using ensemble learning,

S. Liu, Y . Wang, J. Zhang, C. Chen, and Y . Xiang, “Addressing the class imbalance problem in twitter spam detection using ensemble learning,” Computers&Security, vol. 69, pp. 35–49, 2017

work page 2017

[27] [27]

Random balance: Ensembles of variable priors classifiers for imbal- anced data,

J. F. D ´ıez-Pastor, J. J. Rodr´ıguez, C. Garc´ıa-Osorio, and L. I. Kuncheva, “Random balance: Ensembles of variable priors classifiers for imbal- anced data,”Knowledge-Based Systems, vol. 85, pp. 96–111, 2015

work page 2015

[28] [28]

Pyramidal implementation of the affine lucas kanade feature tracker description of the algorithm,

J.-Y . Bouguetet al., “Pyramidal implementation of the affine lucas kanade feature tracker description of the algorithm,”Intel corporation, vol. 5, no. 1-10, p. 4, 2001

work page 2001

[29] [29]

Pysteps: An open-source python library for probabilistic precipitation nowcasting (v1. 0),

S. Pulkkinen, D. Nerini, A. A. P ´erez Hortal, C. Velasco-Forero, A. Seed, U. Germann, and L. Foresti, “Pysteps: An open-source python library for probabilistic precipitation nowcasting (v1. 0),”Geoscientific Model Development, vol. 12, no. 10, pp. 4185–4219, 2019

work page 2019

[30] [30]

Supervised contrastive learn- ing,

P. Khosla, P. Teterwak, C. Wang, A. Sarna, Y . Tian, P. Isola, A. Maschinot, C. Liu, and D. Krishnan, “Supervised contrastive learn- ing,”Advances in neural information processing systems, vol. 33, pp. 18 661–18 673, 2020

work page 2020

[31] [31]

Quantitative precipitation estimation and quantitative precip- itation forecasting by the japan meteorological agency,

K. Nagata, “Quantitative precipitation estimation and quantitative precip- itation forecasting by the japan meteorological agency,”RSMC Tokyo– Typhoon Center Technical Review, vol. 13, pp. 37–50, 2011

work page 2011