Spatio-temporal data fusion for fine-resolution subsidence estimation
Introduction
Many regions of the earth face land subsidence crises due to over-exploitation of groundwater for agriculture and other anthropogenic activities (Hu et al., 2004; Galloway and Burbey, 2011; Minderhoud et al., 2017; Mahmoudpour et al., 2016). Extraction of large volumes of groundwater increases the effective stress of the aquifer system resulting in compaction and subsequent land subsidence (Galloway and Burbey, 2011; Minderhoud et al., 2017). A spatiotemporal model has been developed an accurate representations of annual land subsidence changes from groundwater exploitation (Ali et al., 2020). However, the high spatio-temporal resolution of subsidence monitoring data can help us summarize the processes and patterns of land subsidence.
Subsidence monitoring, which aims to characterize the spatial distribution of land subsidence, generally includes leveling surveys, borehole monitoring data, GPS, and InSAR (Hung et al., 2010). However, the sensors and surveys used in these investigations produce measurements with different spatial and temporal resolutions. Data with high spatial and temporal characteristics are difficult to acquire because of the tradeoff in sensor and survey designs, which balances spatial resolution and temporal coverage (Zhang et al., 2013). For instance, sensor data can be applied at fine spatial resolutions but can rarely be used to capture the temporal changes due to their low temporal resolutions. The available subsidence leveling data still cannot satisfy our requirement of high-frequency land subsidence change measurements caused by groundwater extraction from highly heterogeneous aquifer systems. Leveling data are usually measured at low frequencies due to the high cost of obtaining this information. Leveling can deliver pointwise vertical displacements that are accurate to few mm. Compaction measurements are typically measured at monthly intervals. Compaction monitoring wells can provide depth-dependent measurements of compaction at various intervals within a single borehole by noting the depth of individual magnetic rings that are emplaced at various depths throughout the depth of the well (Hung et al., 2010). The measurement of compaction monitoring wells provide monthly subsidence information. However, the depth of the compaction monitoring well is limited due to the installation cost and geological conditions.
High temporal resolution data are provided by sensors but cannot reach the requirements of interpretation and visualization due to their low spatial resolution. Therefore, spatiotemporal data fusion represents a feasible solution for the aforementioned problem (Zhu et al., 2018). Spatiotemporal data fusion was originally designed for blending reflectance bandwidths from Landsat and MODIS data to produce daily Landsat-like surface reflectance (Chen et al., 2015). Spatiotemporal data fusion is performed to fuse satellite images from two independent sensors, namely, a sensor with very high frequency but coarse spatial resolution such as MODIS, with very high spatial resolution but low frequency sensor such as Landsat (Zhang et al., 2013; Chen et al., 2015). In this study, spatiotemporal data fusion can generate high spatial temporal resolution subsidence combining leveling data and compaction monitoring well data. Typical spatiotemporal data fusion methods are categorized into five groups by using the specific techniques used to link coarse and fine data sets, namely, unmixing-based, weight function-based, Bayesian-based, learning-based, and hybrid-based methods (Zhu et al., 2018). Weight function-based methods are the most popular approach. The weight function is empirical and provides estimates at a fine pixel scale by combining information from all input images using weight functions. The non-spatial weighting assumption is invalid for heterogeneous aquifer systems. In this study, the data fusion method is spatial regression-based but also uses kernel weights (Fotheringham et al., 2003). Our study assumes that the relation between leveling data and subsidence monitoring wells is invariant with time. Thus, on the basis of the spatial relations between the leveling data and subsidence information, the high spatial temporal resolution fused data can be determined. In addition, few studies have shown the spatio-temporal subsidence changes using multiple sensors (Tangdamrongsub et al., 2019). Our fused subsidence data are expected to show where and when subsidence hotspots will occur.
This study aims to provide high spatio-temporal resolution land subsidence estimates by using a data fusion approach between annual leveling data and monthly compaction observations. We hypothesize that data fusion will maximize the spatio-temporal resolution of observations. The subsidence information between annual leveling data and compaction observations are constant with time. Their temporal transformation is a spatial function e.g. location-based linear transformation. Moreover, we will provide interpolated subsidence maps obtained from data fusion. The data fusion approach proposed here will provide a reliable representation of the seasonal subsidence in spite of the temporal and spatial heterogeneity associated with subsidence. In addition, the feasibility of the subsidence hotspot movement detection from the fused data is presented.
Section snippets
Definition of subsidence data fusion
represents the subsidence obtained from leveling data (primary variable) with location set and time set . In this investigation, two sets of leveling data are used (May 2014 and May 2015). is the subsidence from compaction monitoring wells (auxiliary variable) with location set and time set . These data are collected monthly. . . and are the maximum number of temporal and spatial observations (i.e., and ), respectively. The estimated can
Leveling data
Annual subsidence rates from 671 leveling points (locations in Fig. 2b) are used in this investigation and are shown in Fig. 2d. A leveling network with over 1000 km in length is used to calculate subsidence for every 1.5 km interval along the leveling routes. Leveling specifications satisfy a loop closure of less than 3 mm, where K is the length of the leveling circuit in kilometers. The vertical accuracy of the leveling data is generally within 1 cm (Hung et al., 2010). Leveling has a high
Monthly subsidence from monitoring wells
Subsidence between leveling points and subsidence monitoring wells are found to be highly correlated. The correlation coefficient for the annual subsidence between leveling points and subsidence monitoring wells with collocated locations is approximately 0.80. Fig. 4 shows the monthly subsidence from the 31 subsidence monitoring wells. The results imply that the subsidence varies with season. The average subsidence is less than 1.5 cm at the first eight months, but increases dramatically from
Discussion
Subsidence data provided by only one sensor are greatly limited in usefulness due to the tradeoff in sensor designs that balance spatial resolutions and temporal coverage (Chen et al., 2015). As a result, integrated use of data from multiple sensors has become important for understanding subsidence trends from human activities and precipitation patterns. Fused data with high spatial resolution is critical for monitoring subsidence dynamics in heterogeneous aquifer systems, climate, and human
Conclusions
Subsidence data fusion, which integrates both leveling data and monthly subsidence information from compaction monitoring wells, is used to provide better understanding of subsidence patterns influenced by human activities and natural factors. Leveling data contain high spatial resolution but low temporal resolution, and compaction monitoring well data have high temporal resolution but low spatial resolution. However, together these two data sets provide a powerful way to analyze subsidence.
The
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
We acknowledge the financial support from the Ministry of Science and Technology (MOST) and the data support from Water Resources Agency in Taiwan. This research was funded by MOST, grant number 105-2621-M-006 -011 -. We also thank Dr. Bo Huang for scientific advisors. The authors would like to thank the editors and anonymous reviewers for providing suggestions of paper improvement.
References (20)
- et al.
Review on current status and challenging issues of land subsidence in China
Eng. Geol.
(2004) - et al.
Modeling aquifer-system compaction and predicting land subsidence in central Taiwan
Eng. Geol.
(2012) - et al.
Numerical simulation and prediction of regional land subsidence caused by groundwater exploitation in the southwest plain of Tehran, Iran
Eng. Geol.
(2016) - et al.
Quantifying water storage change and land subsidence induced by reservoir impoundment using GRACE, Landsat, and GPS data
Rem. Sens. Environ.
(2019) - et al.
Identifying poromechanism and spatially varying parameters of aquifer compaction in Choushui River alluvial fan, Taiwan
Eng. Geol.
(2018) - et al.
Understanding space–time patterns of groundwater system by empirical orthogonal functions: a case study in the Choshui River alluvial fan, Taiwan
J. Hydrol.
(2010) - et al.
Mapping and predicting subsidence from spatio-temporal regression models of groundwater-drawdown and subsidence observations
Hydrogeol. J.
(2020) - et al.
Comparison of spatiotemporal fusion models: a review
Rem. Sens.
(2015) - et al.
Spatiotemporal analysis of extracted groundwater volumes estimated from electricity consumption
Groundwater
(2020) - et al.
Geographically Weighted Regression: the Analysis of Spatially Varying Relationships
(2003)
Cited by (9)
Parallel finite layer method for land subsidence and its homotopy parameter inversion
2024, Environmental Modelling and SoftwareRegional groundwater sequential forecasting using global and local LSTM models
2023, Journal of Hydrology: Regional StudiesDevelopment of spatially varying groundwater-drawdown functions for land subsidence estimation
2021, Journal of Hydrology: Regional StudiesCitation Excerpt :Currently, the rate of compaction in the coastal area is declining. However, the subsidence bowl is moving inland due to over-pumping in this region (Hwang et al., 2008; Chu et al., 2021). Interferometric Synthetic Aperture Radar (InSAR) offers a powerful and useful method for identifying regional land subsidence (Bürgmann et al., 2000; Bonì et al., 2016; Chaussard et al., 2014b; Galloway et al., 1998b; Hoffmann et al., 2001).
Spatiotemporal subsidence feature decomposition and hotspot identification
2024, Environmental Earth SciencesAn Efficient and Fully Refined Deformation Extraction Method for Deriving Mining-Induced Subsidence by the Joint of Probability Integral Method and SBAS-InSAR
2023, IEEE Transactions on Geoscience and Remote Sensing