Prediction of InSAR deformation time-series using a long short-term memory neural network

Chen, Yi; He, Yixin; Zhang, Lifeng; Chen, Youdong; Pu, Hongyu; Chen, Baoshan; Gao, Liya

doi:10.1080/01431161.2021.1947540

Cited by 65 publications

(24 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, while the ideal approach to assessing reliability involves comparing the studied methods with physical numerical models and empirical models, the accuracy of existing geological and hydrological data is often insu cient, lacks detail, or is con dential. Consequently, precise numerical or empirical modeling of ground subsidence remains challenging (Chen et al 2021). By utilizing singlevariable predictions for surface subsidence, we circumvent the limitations posed by inadequately informative data sources, achieving higher prediction accuracy even when subsidence-driving factor data is scarce.…”

Section: Comparison Of Machine Learning Methods For Urban Ground Subs...mentioning

confidence: 99%

Urban Ground Subsidence Monitoring and Prediction Using Time-Series InSAR and Machine Learning Approaches: A Case Study of Tianjin, China

Zhang,

Kou,

tao

et al. 2024

Preprint

View full text Add to dashboard Cite

Urban ground subsidence, a major geo-hazard threatening sustainable urban development, has been increasingly reported worldwide, yet comprehensive investigations integrating multi-temporal ground deformation monitoring and predictive modeling are still lacking. This study aims to characterize the spatial-temporal evolution of ground subsidence in Tianjin's Jinnan District from 2016 to 2023 using 193 Sentinel-1A ascending images and the advanced Synthetic Aperture Radar Interferometry (InSAR) techniques of SBAS-InSAR and PS-InSAR. The maximum cumulative subsidence reached − 326.92 mm, with an average subsidence rate of -0.39 mm/year concentrated in industrial, commercial, and residential areas with high population density. Further analysis revealed that subway construction, human engineering activities, and rainfall were the primary drivers of ground subsidence in this region. Simultaneously, this study compared the predictive capabilities of five machine learning methods, including Support Vector Machine, Gradient Boosting Decision Tree, Random Forest, Extremely Randomized Tree, and Long Short-Term Memory (LSTM) neural network, for future ground subsidence. The LSTM-based prediction model exhibited the highest accuracy, with a root mean square error of 2.11 mm. Subdomain predictions generally outperformed the overall prediction, highlighting the benefits of reducing spatial heterogeneity. These findings provide insights into the mechanisms and patterns of urban ground subsidence, facilitating sustainable urban planning and infrastructure development.

show abstract

Section: Comparison Of Machine Learning Methods For Urban Ground Subs...mentioning

confidence: 99%

Urban Ground Subsidence Monitoring and Prediction Using Time-Series InSAR and Machine Learning Approaches: A Case Study of Tianjin, China

Zhang,

Kou,

tao

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

“…Compared with the method based on mathematical statistics, the method based on deep learning can better learn the nonlinearity and randomness in data. Among the land subsidence prediction models based on deep learning, the models based on LSTM and its variants are most widely used (Chen et al, 2021, Kumar et al, 2021, Liu et al, 2021. In (Liu et al, 2021), considering the spatial heterogeneity of land subsidence, the authors first cluster the land subsidence data and then train the prediction model based on LSTM for each subclass.…”

Section: Land Subsidence Displacement Predictionmentioning

confidence: 99%

“…Although the implementation is relatively simple, the prediction accuracy is not high due to the single data and easy overfitting of the model. The method based on deep learning is mainly based on LSTM network (Chen et al, 2021, Kumar et al, 2021, Liu et al, 2021. Compared with previous methods, these methods improve the prediction accuracy to a certain extent by better simulating the nonlinear effects between land subsidence displacement and various influencing factors.…”

Section: Introductionmentioning

confidence: 99%

Land Subsidence Prediction Through Modeling of Temporal Attribute Prediction of Knowledge Graph

Lei

Song

Fan

et al. 2022

ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

Abstract. Land subsidence is a geological disaster. It will lead to the decline of land elevation, resulting in the potential safety hazards of urban facilities. Thus, the prediction of land subsidence displacement is significant. Among the existing prediction methods, the methods based on the time-series prediction model only analyze the settlement series data without considering the settlement mechanism, so they are easy to apply. However, they less consider the influence of other factors on land subsidence. Besides, they independently input displacement time-series data from different monitoring points without considering their relationship. To solve these problems, we take the monitoring point as the entity and take the DEM, soil type, building height, and land subsidence displacement sequence at the corresponding position of the monitoring point as the attributes to construct the knowledge graph. And then, we propose a framework Graph-TAP for modeling temporal attribute prediction of the knowledge graph’s entity. This framework learns the representation of events with the specific entity at first. Then it captures the temporal dependency between historical events using GRU. Finally, it predicts the entity’s displacement attribute. We randomly selected 61 subsidence monitoring points in Shenzhen, China. We used the land subsidence displacement InSAR time-series data (12-day time interval) and other attribute data from June 22, 2015, to April 5, 2016, for model training, validation, and testing. The experimental results show that our method is better than the time-series prediction based on the LSTM model and the DARNET (a knowledge graph temporal attribute prediction framework).

show abstract

“…Nukala et al [ 42 ] proposed a new method based on recurrent neural networks (RNN), applying Sentinel-1 to predict time-series deformation maps and achieved good prediction performance. LSTM (Long Short-Term Memory) networks address the limitations of gradient explosion and disappearance that older RNN variants may suffer when learning long-term dependencies of data, improving time-series InSAR deformation prediction models [ 43 , 44 , 45 ]. Wang et al proposed an innovative InSAR deformation prediction integrated algorithm based on transformer models to accurately predict time-series deformation surrounding Salt Lake [ 46 ].…”

Section: Introductionmentioning

confidence: 99%

Evaluating Permafrost Degradation in the Tuotuo River Basin by MT-InSAR and LSTM Methods

Ping

Liu

Zhang

et al. 2023

Sensors

View full text Add to dashboard Cite

Permafrost degradation can significantly affect vegetation, infrastructure, and sustainable development on the Qinghai-Tibet Plateau (QTP). The permafrost on the QTP faces a risk of widespread degradation due to climate change and ecosystem disturbances; thus, monitoring its changes is critical. In this study, we conducted a permafrost surface deformation prediction over the Tuotuo River tributary watershed in the southwestern part of the QTP using the Long Short-Term Memory model (LSTM). The LSTM model was applied to the deformation information derived from a time series of Multi-Temporal Interferometry Synthetic Aperture Radar (MT-InSAR). First, we designed a quadtree segmentation-based Small BAseline Subset (SBAS) to monitor the seasonal permafrost deformation from March 2017 to April 2022. Then, the types of frozen soil were classified using the spatio-temporal deformation information and the temperature at the top of the permafrost. Finally, the time-series deformation trends of different types of permafrost were predicted using the LSTM model. The results showed that the deformation rates in the Tuotuo River Basin ranged between −80 to 60 mm/yr. Permafrost, seasonally frozen ground, and potentially degraded permafrost covered 7572.23, 900.87, and 921.70 km2, respectively. The LSTM model achieved high precision for frozen soil deformation prediction at the point scale, with a root mean square error of 4.457 mm and mean absolute error of 3.421 mm. The results demonstrated that deformation monitoring and prediction using MT-InSAR technology integrated with the LSTM model can be used to accurately identify types of permafrost over a large region and quantitatively evaluate its degradation trends.

show abstract

Prediction of InSAR deformation time-series using a long short-term memory neural network

Cited by 65 publications

References 31 publications

Urban Ground Subsidence Monitoring and Prediction Using Time-Series InSAR and Machine Learning Approaches: A Case Study of Tianjin, China

Urban Ground Subsidence Monitoring and Prediction Using Time-Series InSAR and Machine Learning Approaches: A Case Study of Tianjin, China

Land Subsidence Prediction Through Modeling of Temporal Attribute Prediction of Knowledge Graph

Evaluating Permafrost Degradation in the Tuotuo River Basin by MT-InSAR and LSTM Methods

Contact Info

Product

Resources

About