Machine learning for monitoring groundwater resources over Europe
Machine learning for monitoring groundwater resources over Europe
View/Type of Scholarly Publication
DissertationDate of Exam
04.05.2022Date of Publication
30.08.2022Advisor
Kollet, Stefan J.Co-Referee
Simmer, ClemensMetadata
Show full item record
Citable Links 
Abstract
Groundwater (GW) is an important natural resource for Europe and the world, and has been affected by extreme weather and climate, e.g., summer heat waves and droughts, and human overexploitation. As climate change and human interventions increase, extreme events and GW depletion are expected to become more frequent and severe in many parts of Europe in the future, aggravating the vulnerability of GW systems. This emphasizes the necessity of GW monitoring in GW management. Up to date, however, it is still challenging to monitor GW at the large, continental scale, mainly due to the lack of water table depth (wtd) observations.
In order to address the challenge, the PhD work proposes an indirect, generic methodology based on advanced machine learning (ML) techniques, that are Long Short-Term Memory (LSTM) networks and transfer learning (TL), to produce reliable monthly wtd anomaly (wtda) estimates at the continental scale. The methodology is named LSTM-TL. While in this work, LSTM-TL has been implemented over Europe, it is transferable to other regions in the world. The methodology relies on the close connection between GW and other atmospheric and terrestrial compartments in the water cycle, using precipitation and soil moisture anomalies (pra and θa) as input, which have data available at large scales from, e.g., remotely sensed observations. Several steps were involved in the development of LSTM-TL for GW monitoring.
In the first step, LSTM networks were applied in combination with spatiotemporally continuous pra and wtda data from uncalibrated integrated hydrologic simulation results (named the TSMP-G2A data set) over Europe to capture the time-varying and time-lagged relationship between pra and wtda in order to obtain reliable networks to estimate wtda at the individual pixel level assuming that pra is a useful proxy for wtda. In most European regions, LSTM networks showed good skill with respect to the TSMP-G2A data set in predicting wtda with pra as input. The results indicated that the local factors, that are yearly averaged wtd, evapotranspiration (ET), soil moisture (θ), and snow water equivalent (SWE), had a significant impact on the performance of the LSTM networks. Moreover, the decrease in the network test performance at some pixels was attributed to a change in the temporal TSMP-G2A pra-wtda pattern during the study period.
In the second step, a number of input hydrometeorological variables, in addition to pra, were included in the construction of LSTM networks to arrive at improved wtda estimates at individual pixels over Europe in various experiments. All input and target data were derived from the TSMP-G2A data set. Improved LSTM networks were found with pra and θa as input. Considering θa strongly increased the network test performance particularly in the areas with wtd ≤ 3 m (i.e., the major wtd category of Europe), suggesting the substantial contribution of θa to the estimation of wtda over Europe. The results highlight the importance to combine θ information with precipitation information in quantifying and predicting wtda.
In the final step, LSTM-TL was proposed for real-world applications. In LSTM-TL, LSTM networks were first trained on TSMP-G2A anomalies, and then, without additional training, utilized to estimate wtda with pra and θa from common observational datasets as input, thus, transferring knowledge from simulation results (i.e., the TSMP-G2A data set) to the observation-based estimation of wtda. Applying TL addressed the issue of scarce wtda observations (wtda,o) to train LSTM networks at the European scale. The implementation of LSTM-TL was based on two assumptions, that are i) the modeled relationship between wtda and input hydrometeorological variables (i.e., pra and θa) agreed well with the observed; and ii) the internal LSTM networks successfully captured the modeled relationship. The obtained wtda estimates were evaluated with collated in-situ wtda measurements from approximately 2,600 European GW monitoring wells, which demonstrated the good skill of LSTM-TL in estimating wtda. LSTM-TL was used for reconstructing monthly wtda from the early 1980s to the near present over Europe. The reconstructed wtda data exhibited seasonal wtda trends in different European regions in the past, contributing significantly to the understanding of historical GW dynamics at the continental scale over Europe, which has not been possible before.
The proposed LSTM-TL has three salient features. First, the methodology does not rely on wtda,o to estimate wtda, which enables its usage over large regions even without wtd observations. Second, the methodology can be used to generate wtda estimates beyond the time period of the TSMP-G2A data set utilized for training, which is useful for reconstructing historical wtda and predicting future wtda at the continental scale. Third, once the internal LSTM networks are successfully trained, the methodology can be directly implemented without additional training, and thus, requires low computational cost in comparison to physically-based numerical simulation systems to generate new wtda estimates.
This PhD work presents a novel approach in the field of ML to estimate wtda in the absence of wtd observations, which advances significantly GW monitoring capacities at large scales. Since the TSMP-G2A data set provides a near-natural representation of the terrestrial water and energy cycles, the current implementation of LSTM-TL does not account for anthropogenetic impacts on GW dynamics. Nevertheless, LSTM-TL has been shown to produce reliable wtda estimates over Europe and can serve as an alternative methodology to in-situ wtda measurements. In addition to data reconstruction, the methodology can be employed for online GW monitoring and predictions, which is useful to GW management in Europe and beyond.
In order to address the challenge, the PhD work proposes an indirect, generic methodology based on advanced machine learning (ML) techniques, that are Long Short-Term Memory (LSTM) networks and transfer learning (TL), to produce reliable monthly wtd anomaly (wtda) estimates at the continental scale. The methodology is named LSTM-TL. While in this work, LSTM-TL has been implemented over Europe, it is transferable to other regions in the world. The methodology relies on the close connection between GW and other atmospheric and terrestrial compartments in the water cycle, using precipitation and soil moisture anomalies (pra and θa) as input, which have data available at large scales from, e.g., remotely sensed observations. Several steps were involved in the development of LSTM-TL for GW monitoring.
In the first step, LSTM networks were applied in combination with spatiotemporally continuous pra and wtda data from uncalibrated integrated hydrologic simulation results (named the TSMP-G2A data set) over Europe to capture the time-varying and time-lagged relationship between pra and wtda in order to obtain reliable networks to estimate wtda at the individual pixel level assuming that pra is a useful proxy for wtda. In most European regions, LSTM networks showed good skill with respect to the TSMP-G2A data set in predicting wtda with pra as input. The results indicated that the local factors, that are yearly averaged wtd, evapotranspiration (ET), soil moisture (θ), and snow water equivalent (SWE), had a significant impact on the performance of the LSTM networks. Moreover, the decrease in the network test performance at some pixels was attributed to a change in the temporal TSMP-G2A pra-wtda pattern during the study period.
In the second step, a number of input hydrometeorological variables, in addition to pra, were included in the construction of LSTM networks to arrive at improved wtda estimates at individual pixels over Europe in various experiments. All input and target data were derived from the TSMP-G2A data set. Improved LSTM networks were found with pra and θa as input. Considering θa strongly increased the network test performance particularly in the areas with wtd ≤ 3 m (i.e., the major wtd category of Europe), suggesting the substantial contribution of θa to the estimation of wtda over Europe. The results highlight the importance to combine θ information with precipitation information in quantifying and predicting wtda.
In the final step, LSTM-TL was proposed for real-world applications. In LSTM-TL, LSTM networks were first trained on TSMP-G2A anomalies, and then, without additional training, utilized to estimate wtda with pra and θa from common observational datasets as input, thus, transferring knowledge from simulation results (i.e., the TSMP-G2A data set) to the observation-based estimation of wtda. Applying TL addressed the issue of scarce wtda observations (wtda,o) to train LSTM networks at the European scale. The implementation of LSTM-TL was based on two assumptions, that are i) the modeled relationship between wtda and input hydrometeorological variables (i.e., pra and θa) agreed well with the observed; and ii) the internal LSTM networks successfully captured the modeled relationship. The obtained wtda estimates were evaluated with collated in-situ wtda measurements from approximately 2,600 European GW monitoring wells, which demonstrated the good skill of LSTM-TL in estimating wtda. LSTM-TL was used for reconstructing monthly wtda from the early 1980s to the near present over Europe. The reconstructed wtda data exhibited seasonal wtda trends in different European regions in the past, contributing significantly to the understanding of historical GW dynamics at the continental scale over Europe, which has not been possible before.
The proposed LSTM-TL has three salient features. First, the methodology does not rely on wtda,o to estimate wtda, which enables its usage over large regions even without wtd observations. Second, the methodology can be used to generate wtda estimates beyond the time period of the TSMP-G2A data set utilized for training, which is useful for reconstructing historical wtda and predicting future wtda at the continental scale. Third, once the internal LSTM networks are successfully trained, the methodology can be directly implemented without additional training, and thus, requires low computational cost in comparison to physically-based numerical simulation systems to generate new wtda estimates.
This PhD work presents a novel approach in the field of ML to estimate wtda in the absence of wtd observations, which advances significantly GW monitoring capacities at large scales. Since the TSMP-G2A data set provides a near-natural representation of the terrestrial water and energy cycles, the current implementation of LSTM-TL does not account for anthropogenetic impacts on GW dynamics. Nevertheless, LSTM-TL has been shown to produce reliable wtda estimates over Europe and can serve as an alternative methodology to in-situ wtda measurements. In addition to data reconstruction, the methodology can be employed for online GW monitoring and predictions, which is useful to GW management in Europe and beyond.
Subjects
deep learning, LSTM-TL, Long Short-Term Memory networks, transfer learning, groundwater, drought monitoring, Europe
Classification (DDC)
550 GeowissenschaftenMa, Yueling: Machine learning for monitoring groundwater resources over Europe. - Bonn, 2022. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5-67321
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5-67321
@phdthesis{handle:20.500.11811/10198,
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5-67321,
author = {{Yueling Ma}},
title = {Machine learning for monitoring groundwater resources over Europe},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2022,
month = aug,
note = {Groundwater (GW) is an important natural resource for Europe and the world, and has been affected by extreme weather and climate, e.g., summer heat waves and droughts, and human overexploitation. As climate change and human interventions increase, extreme events and GW depletion are expected to become more frequent and severe in many parts of Europe in the future, aggravating the vulnerability of GW systems. This emphasizes the necessity of GW monitoring in GW management. Up to date, however, it is still challenging to monitor GW at the large, continental scale, mainly due to the lack of water table depth (wtd) observations.
In order to address the challenge, the PhD work proposes an indirect, generic methodology based on advanced machine learning (ML) techniques, that are Long Short-Term Memory (LSTM) networks and transfer learning (TL), to produce reliable monthly wtd anomaly (wtda) estimates at the continental scale. The methodology is named LSTM-TL. While in this work, LSTM-TL has been implemented over Europe, it is transferable to other regions in the world. The methodology relies on the close connection between GW and other atmospheric and terrestrial compartments in the water cycle, using precipitation and soil moisture anomalies (pra and θa) as input, which have data available at large scales from, e.g., remotely sensed observations. Several steps were involved in the development of LSTM-TL for GW monitoring.
In the first step, LSTM networks were applied in combination with spatiotemporally continuous pra and wtda data from uncalibrated integrated hydrologic simulation results (named the TSMP-G2A data set) over Europe to capture the time-varying and time-lagged relationship between pra and wtda in order to obtain reliable networks to estimate wtda at the individual pixel level assuming that pra is a useful proxy for wtda. In most European regions, LSTM networks showed good skill with respect to the TSMP-G2A data set in predicting wtda with pra as input. The results indicated that the local factors, that are yearly averaged wtd, evapotranspiration (ET), soil moisture (θ), and snow water equivalent (SWE), had a significant impact on the performance of the LSTM networks. Moreover, the decrease in the network test performance at some pixels was attributed to a change in the temporal TSMP-G2A pra-wtda pattern during the study period.
In the second step, a number of input hydrometeorological variables, in addition to pra, were included in the construction of LSTM networks to arrive at improved wtda estimates at individual pixels over Europe in various experiments. All input and target data were derived from the TSMP-G2A data set. Improved LSTM networks were found with pra and θa as input. Considering θa strongly increased the network test performance particularly in the areas with wtd ≤ 3 m (i.e., the major wtd category of Europe), suggesting the substantial contribution of θa to the estimation of wtda over Europe. The results highlight the importance to combine θ information with precipitation information in quantifying and predicting wtda.
In the final step, LSTM-TL was proposed for real-world applications. In LSTM-TL, LSTM networks were first trained on TSMP-G2A anomalies, and then, without additional training, utilized to estimate wtda with pra and θa from common observational datasets as input, thus, transferring knowledge from simulation results (i.e., the TSMP-G2A data set) to the observation-based estimation of wtda. Applying TL addressed the issue of scarce wtda observations (wtda,o) to train LSTM networks at the European scale. The implementation of LSTM-TL was based on two assumptions, that are i) the modeled relationship between wtda and input hydrometeorological variables (i.e., pra and θa) agreed well with the observed; and ii) the internal LSTM networks successfully captured the modeled relationship. The obtained wtda estimates were evaluated with collated in-situ wtda measurements from approximately 2,600 European GW monitoring wells, which demonstrated the good skill of LSTM-TL in estimating wtda. LSTM-TL was used for reconstructing monthly wtda from the early 1980s to the near present over Europe. The reconstructed wtda data exhibited seasonal wtda trends in different European regions in the past, contributing significantly to the understanding of historical GW dynamics at the continental scale over Europe, which has not been possible before.
The proposed LSTM-TL has three salient features. First, the methodology does not rely on wtda,o to estimate wtda, which enables its usage over large regions even without wtd observations. Second, the methodology can be used to generate wtda estimates beyond the time period of the TSMP-G2A data set utilized for training, which is useful for reconstructing historical wtda and predicting future wtda at the continental scale. Third, once the internal LSTM networks are successfully trained, the methodology can be directly implemented without additional training, and thus, requires low computational cost in comparison to physically-based numerical simulation systems to generate new wtda estimates.
This PhD work presents a novel approach in the field of ML to estimate wtda in the absence of wtd observations, which advances significantly GW monitoring capacities at large scales. Since the TSMP-G2A data set provides a near-natural representation of the terrestrial water and energy cycles, the current implementation of LSTM-TL does not account for anthropogenetic impacts on GW dynamics. Nevertheless, LSTM-TL has been shown to produce reliable wtda estimates over Europe and can serve as an alternative methodology to in-situ wtda measurements. In addition to data reconstruction, the methodology can be employed for online GW monitoring and predictions, which is useful to GW management in Europe and beyond.},
url = {https://hdl.handle.net/20.500.11811/10198}
}
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5-67321,
author = {{Yueling Ma}},
title = {Machine learning for monitoring groundwater resources over Europe},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2022,
month = aug,
note = {Groundwater (GW) is an important natural resource for Europe and the world, and has been affected by extreme weather and climate, e.g., summer heat waves and droughts, and human overexploitation. As climate change and human interventions increase, extreme events and GW depletion are expected to become more frequent and severe in many parts of Europe in the future, aggravating the vulnerability of GW systems. This emphasizes the necessity of GW monitoring in GW management. Up to date, however, it is still challenging to monitor GW at the large, continental scale, mainly due to the lack of water table depth (wtd) observations.
In order to address the challenge, the PhD work proposes an indirect, generic methodology based on advanced machine learning (ML) techniques, that are Long Short-Term Memory (LSTM) networks and transfer learning (TL), to produce reliable monthly wtd anomaly (wtda) estimates at the continental scale. The methodology is named LSTM-TL. While in this work, LSTM-TL has been implemented over Europe, it is transferable to other regions in the world. The methodology relies on the close connection between GW and other atmospheric and terrestrial compartments in the water cycle, using precipitation and soil moisture anomalies (pra and θa) as input, which have data available at large scales from, e.g., remotely sensed observations. Several steps were involved in the development of LSTM-TL for GW monitoring.
In the first step, LSTM networks were applied in combination with spatiotemporally continuous pra and wtda data from uncalibrated integrated hydrologic simulation results (named the TSMP-G2A data set) over Europe to capture the time-varying and time-lagged relationship between pra and wtda in order to obtain reliable networks to estimate wtda at the individual pixel level assuming that pra is a useful proxy for wtda. In most European regions, LSTM networks showed good skill with respect to the TSMP-G2A data set in predicting wtda with pra as input. The results indicated that the local factors, that are yearly averaged wtd, evapotranspiration (ET), soil moisture (θ), and snow water equivalent (SWE), had a significant impact on the performance of the LSTM networks. Moreover, the decrease in the network test performance at some pixels was attributed to a change in the temporal TSMP-G2A pra-wtda pattern during the study period.
In the second step, a number of input hydrometeorological variables, in addition to pra, were included in the construction of LSTM networks to arrive at improved wtda estimates at individual pixels over Europe in various experiments. All input and target data were derived from the TSMP-G2A data set. Improved LSTM networks were found with pra and θa as input. Considering θa strongly increased the network test performance particularly in the areas with wtd ≤ 3 m (i.e., the major wtd category of Europe), suggesting the substantial contribution of θa to the estimation of wtda over Europe. The results highlight the importance to combine θ information with precipitation information in quantifying and predicting wtda.
In the final step, LSTM-TL was proposed for real-world applications. In LSTM-TL, LSTM networks were first trained on TSMP-G2A anomalies, and then, without additional training, utilized to estimate wtda with pra and θa from common observational datasets as input, thus, transferring knowledge from simulation results (i.e., the TSMP-G2A data set) to the observation-based estimation of wtda. Applying TL addressed the issue of scarce wtda observations (wtda,o) to train LSTM networks at the European scale. The implementation of LSTM-TL was based on two assumptions, that are i) the modeled relationship between wtda and input hydrometeorological variables (i.e., pra and θa) agreed well with the observed; and ii) the internal LSTM networks successfully captured the modeled relationship. The obtained wtda estimates were evaluated with collated in-situ wtda measurements from approximately 2,600 European GW monitoring wells, which demonstrated the good skill of LSTM-TL in estimating wtda. LSTM-TL was used for reconstructing monthly wtda from the early 1980s to the near present over Europe. The reconstructed wtda data exhibited seasonal wtda trends in different European regions in the past, contributing significantly to the understanding of historical GW dynamics at the continental scale over Europe, which has not been possible before.
The proposed LSTM-TL has three salient features. First, the methodology does not rely on wtda,o to estimate wtda, which enables its usage over large regions even without wtd observations. Second, the methodology can be used to generate wtda estimates beyond the time period of the TSMP-G2A data set utilized for training, which is useful for reconstructing historical wtda and predicting future wtda at the continental scale. Third, once the internal LSTM networks are successfully trained, the methodology can be directly implemented without additional training, and thus, requires low computational cost in comparison to physically-based numerical simulation systems to generate new wtda estimates.
This PhD work presents a novel approach in the field of ML to estimate wtda in the absence of wtd observations, which advances significantly GW monitoring capacities at large scales. Since the TSMP-G2A data set provides a near-natural representation of the terrestrial water and energy cycles, the current implementation of LSTM-TL does not account for anthropogenetic impacts on GW dynamics. Nevertheless, LSTM-TL has been shown to produce reliable wtda estimates over Europe and can serve as an alternative methodology to in-situ wtda measurements. In addition to data reconstruction, the methodology can be employed for online GW monitoring and predictions, which is useful to GW management in Europe and beyond.},
url = {https://hdl.handle.net/20.500.11811/10198}
}
The following license files are associated with this item:
