Our project is done in close collaboration with the Technical University of Munich. In particular with the TUM Data Science in Earth Observation (Sipeo) group. The complete list of associated publications might be also interesting for you and is available here.
Publications
Publications
2023 |
|
Xiangyu, Zhao; Jingliang, Hu; Lichao, Mou; Zhitong, Xiong; Xiang, Zhu Xiao Cross-city Landuse classification of remote sensing images via deep transfer learning Journal Article International Journal of Applied Earth Observation and Geoinformation, 122 , pp. 103358, 2023, ISSN: 1569-8432. Links | BibTeX | Tags: Cross-city classification, Deep learning, Domain adaptation, Local climate zone classification, Sentinel-1, Sentinel-2, Transfer learning @article{RN286, title = {Cross-city Landuse classification of remote sensing images via deep transfer learning}, author = {Zhao Xiangyu and Hu Jingliang and Mou Lichao and Xiong Zhitong and Zhu Xiao Xiang}, url = {https://www.sciencedirect.com/science/article/pii/S1569843223001826}, doi = {https://doi.org/10.1016/j.jag.2023.103358}, issn = {1569-8432}, year = {2023}, date = {2023-01-01}, journal = {International Journal of Applied Earth Observation and Geoinformation}, volume = {122}, pages = {103358}, keywords = {Cross-city classification, Deep learning, Domain adaptation, Local climate zone classification, Sentinel-1, Sentinel-2, Transfer learning}, pubstate = {published}, tppubtype = {article} } | |
Xin-Yi, Tong; Gui-Song, Xia; Xiang, Zhu Xiao Enabling country-scale land cover mapping with meter-resolution satellite imagery Journal Article ISPRS Journal of Photogrammetry and Remote Sensing, 196 , pp. 178-196, 2023, ISSN: 0924-2716. Links | BibTeX | Tags: Classification, Dataset, Deep learning, Domain adaptation, Gaofen-1, Gaofen-2, High-spatial resolution, Land cover mapping, Megacity, PlanetScope, Sentinel-2, Transfer learning @article{RN287, title = {Enabling country-scale land cover mapping with meter-resolution satellite imagery}, author = {Tong Xin-Yi and Xia Gui-Song and Zhu Xiao Xiang}, url = {https://www.sciencedirect.com/science/article/pii/S0924271622003264}, doi = {https://doi.org/10.1016/j.isprsjprs.2022.12.011}, issn = {0924-2716}, year = {2023}, date = {2023-01-01}, journal = {ISPRS Journal of Photogrammetry and Remote Sensing}, volume = {196}, pages = {178-196}, keywords = {Classification, Dataset, Deep learning, Domain adaptation, Gaofen-1, Gaofen-2, High-spatial resolution, Land cover mapping, Megacity, PlanetScope, Sentinel-2, Transfer learning}, pubstate = {published}, tppubtype = {article} } | |
Xiong, Zhitong; Huang, Wei; Hu, Jingtao; Zhu, Xiao THE Benchmark: Transferable Representation Learning for Monocular Height Estimation Journal Article IEEE Transactions on Geoscience and Remote Sensing, PP , pp. 1-1, 2023. @article{RN288, title = {THE Benchmark: Transferable Representation Learning for Monocular Height Estimation}, author = {Zhitong Xiong and Wei Huang and Jingtao Hu and Xiao Zhu}, doi = {10.1109/TGRS.2023.3311764}, year = {2023}, date = {2023-01-01}, journal = {IEEE Transactions on Geoscience and Remote Sensing}, volume = {PP}, pages = {1-1}, keywords = {}, pubstate = {published}, tppubtype = {article} } | |
Xu, Fang; Shi, Yilei; Ebel, Patrick; Yang, Wen; Zhu, Xiao High-Resolution Cloud Removal with Multi-Modal and Multi-Resolution Data Fusion: A New Baseline and Benchmark Journal Article 2023. BibTeX | Tags: @article{RN289, title = {High-Resolution Cloud Removal with Multi-Modal and Multi-Resolution Data Fusion: A New Baseline and Benchmark}, author = {Fang Xu and Yilei Shi and Patrick Ebel and Wen Yang and Xiao Zhu}, year = {2023}, date = {2023-01-01}, keywords = {}, pubstate = {published}, tppubtype = {article} } | |
Xu, Qingsong; Shi, Yilei; Guo, Jianhua; Ouyang, Chaojun; Zhu, Xiao UCDFormer: Unsupervised Change Detection Using a Transformer-driven Image Translation Journal Article Joint Urban Remote Sensing Event (JURSE), 2023 , 2023, ISBN: 978-1-6654-9374-1. @article{RN290, title = {UCDFormer: Unsupervised Change Detection Using a Transformer-driven Image Translation}, author = {Qingsong Xu and Yilei Shi and Jianhua Guo and Chaojun Ouyang and Xiao Zhu}, doi = {10.1109/JURSE57346.2023.10144178}, isbn = {978-1-6654-9374-1}, year = {2023}, date = {2023-01-01}, journal = {Joint Urban Remote Sensing Event (JURSE)}, volume = {2023}, keywords = {}, pubstate = {published}, tppubtype = {article} } | |
Xu, Qingsong; Shi, Yilei; Yuan, Xin; Zhu, Xiao Xiang Universal Domain Adaptation for Remote Sensing Image Scene Classification Journal Article IEEE Transactions on Geoscience and Remote Sensing, 61 , pp. 1-15, 2023, ( ). Links | BibTeX | Tags: Remote sensing Adaptation models Data models Uncertainty Image classification Entropy Earth Remote sensing image classification source data generation (SDG) transferable weight universal domain adaptation (DA) @article{RN291, title = {Universal Domain Adaptation for Remote Sensing Image Scene Classification}, author = {Qingsong Xu and Yilei Shi and Xin Yuan and Xiao Xiang Zhu}, doi = {10.1109/TGRS.2023.3235988}, year = {2023}, date = {2023-01-01}, journal = {IEEE Transactions on Geoscience and Remote Sensing}, volume = {61}, pages = {1-15}, note = { }, keywords = {Remote sensing Adaptation models Data models Uncertainty Image classification Entropy Earth Remote sensing image classification source data generation (SDG) transferable weight universal domain adaptation (DA)}, pubstate = {published}, tppubtype = {article} } | |
Zhang, Fahong; Shi, Yilei; Xiong, Zhitong; Huang, Wei; Zhu, Xiao Pseudo Features-Guided Self-Training for Domain Adaptive Semantic Segmentation of Satellite Images Journal Article IEEE Transactions on Geoscience and Remote Sensing, PP , pp. 1-1, 2023. @article{RN292, title = {Pseudo Features-Guided Self-Training for Domain Adaptive Semantic Segmentation of Satellite Images}, author = {Fahong Zhang and Yilei Shi and Zhitong Xiong and Wei Huang and Xiao Zhu}, doi = {10.1109/TGRS.2023.3281503}, year = {2023}, date = {2023-01-01}, journal = {IEEE Transactions on Geoscience and Remote Sensing}, volume = {PP}, pages = {1-1}, keywords = {}, pubstate = {published}, tppubtype = {article} } | |
Zhou, Fei; Sun, Xin; Sun, Chengze; Dong, Junyu; Zhu, Xiao Xiang Adaptive Morphology Filter: A Lightweight Module for Deep Hyperspectral Image Classification Journal Article IEEE Transactions on Geoscience and Remote Sensing, 61 , pp. 1-16, 2023. Links | BibTeX | Tags: Convolution Kernel Morphology Feature extraction Training Hyperspectral imaging Information filters Deep learning (DL) hyperspectral image (HSI) classification morphology filter structural reparameterization (SRP) @article{RN293, title = {Adaptive Morphology Filter: A Lightweight Module for Deep Hyperspectral Image Classification}, author = {Fei Zhou and Xin Sun and Chengze Sun and Junyu Dong and Xiao Xiang Zhu}, doi = {10.1109/TGRS.2023.3327418}, year = {2023}, date = {2023-01-01}, journal = {IEEE Transactions on Geoscience and Remote Sensing}, volume = {61}, pages = {1-16}, keywords = {Convolution Kernel Morphology Feature extraction Training Hyperspectral imaging Information filters Deep learning (DL) hyperspectral image (HSI) classification morphology filter structural reparameterization (SRP)}, pubstate = {published}, tppubtype = {article} } | |
Dong, Runmin; Mou, Lichao; Chen, Mengxuan; Li, Weijia; Tong, Xin-Yi; Yuan, Shuai; Zhang, Lixian; Zheng, Juepeng; Zhu, Xiao Xiang; Fu, Haohuan Large-Scale Land Cover Mapping with Fine-Grained Classes via Class-Aware Semi-Supervised Semantic Segmentation Miscellaneous Proceedings of ICCV 2023, Paris, France, 2023. @misc{RN37, title = {Large-Scale Land Cover Mapping with Fine-Grained Classes via Class-Aware Semi-Supervised Semantic Segmentation}, author = {Runmin Dong and Lichao Mou and Mengxuan Chen and Weijia Li and Xin-Yi Tong and Shuai Yuan and Lixian Zhang and Juepeng Zheng and Xiao Xiang Zhu and Haohuan Fu}, url = {https://doi.org/10.1109/ICCV51070.2023.01539}, doi = {10.1109/ICCV51070.2023.01539}, year = {2023}, date = {2023-01-01}, pages = {16737-16747}, howpublished = {Proceedings of ICCV 2023, Paris, France}, keywords = {}, pubstate = {published}, tppubtype = {misc} } | |
Ebel, Patrick; Garnot, Vivien Sainte Fare; Schmitt, Michael; Wegner, Jan; Zhu, Xiao UnCRtainTS: Uncertainty Quantification for Cloud Removal in Optical Satellite Time Series Book 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2023, ISBN: 979-8-3503-0250-9. @book{RN38, title = {UnCRtainTS: Uncertainty Quantification for Cloud Removal in Optical Satellite Time Series}, author = {Patrick Ebel and Vivien Sainte Fare Garnot and Michael Schmitt and Jan Wegner and Xiao Zhu}, doi = {10.1109/CVPRW59228.2023.00202}, isbn = {979-8-3503-0250-9}, year = {2023}, date = {2023-01-01}, pages = {2086-2096}, publisher = {2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)}, keywords = {}, pubstate = {published}, tppubtype = {book} } | |
2022 |
|
Bamber J. L., Oppenheimer Kopp Aspinall M R E W P; Cooke, R M Climate Processes Driving the Uncertainty in Projections of Future Sea Level Rise: Findings From a Structured Expert Judgement Approach Journal Article Earth's Future, 10 (10), 2022. Abstract | Links | BibTeX | Tags: . @article{Bamber2022, title = {Climate Processes Driving the Uncertainty in Projections of Future Sea Level Rise: Findings From a Structured Expert Judgement Approach}, author = {Bamber, J. L., Oppenheimer, M., Kopp, R. E., Aspinall, W. P., and Cooke, R. M. }, url = {https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022EF002772}, doi = {2022EF002772}, year = {2022}, date = {2022-10-03}, journal = {Earth's Future}, volume = {10}, number = {10}, abstract = {The ice sheets covering Antarctica and Greenland present the greatest uncertainty in, and largest potential contribution to, future sea level rise. The uncertainty arises from a paucity of suitable observations covering the full range of ice sheet behaviors, incomplete understanding of the influences of diverse processes, and limitations in defining key boundary conditions for the numerical models. To investigate the impact of these uncertainties on ice sheet projections we undertook a structured expert judgement study. Here, we interrogate the findings of that study to identify the dominant drivers of uncertainty in projections and their relative importance as a function of ice sheet and time. We find that for the 21st century, Greenland surface melting, in particular the role of surface albedo effects, and West Antarctic ice dynamics, specifically the role of ice shelf buttressing, dominate the uncertainty. The importance of these effects holds under both a high-end 5°C global warming scenario and another that limits global warming to 2°C. During the 22nd century the dominant drivers of uncertainty shift. Under the 5°C scenario, East Antarctic ice dynamics dominate the uncertainty in projections, driven by the possible role of ice flow instabilities. These dynamic effects only become dominant, however, for a temperature scenario above the Paris Agreement 2°C target and beyond 2100. Our findings identify key processes and factors that need to be addressed in future modeling and observational studies in order to reduce uncertainties in ice sheet projections.}, keywords = {.}, pubstate = {published}, tppubtype = {article} } The ice sheets covering Antarctica and Greenland present the greatest uncertainty in, and largest potential contribution to, future sea level rise. The uncertainty arises from a paucity of suitable observations covering the full range of ice sheet behaviors, incomplete understanding of the influences of diverse processes, and limitations in defining key boundary conditions for the numerical models. To investigate the impact of these uncertainties on ice sheet projections we undertook a structured expert judgement study. Here, we interrogate the findings of that study to identify the dominant drivers of uncertainty in projections and their relative importance as a function of ice sheet and time. We find that for the 21st century, Greenland surface melting, in particular the role of surface albedo effects, and West Antarctic ice dynamics, specifically the role of ice shelf buttressing, dominate the uncertainty. The importance of these effects holds under both a high-end 5°C global warming scenario and another that limits global warming to 2°C. During the 22nd century the dominant drivers of uncertainty shift. Under the 5°C scenario, East Antarctic ice dynamics dominate the uncertainty in projections, driven by the possible role of ice flow instabilities. These dynamic effects only become dominant, however, for a temperature scenario above the Paris Agreement 2°C target and beyond 2100. Our findings identify key processes and factors that need to be addressed in future modeling and observational studies in order to reduce uncertainties in ice sheet projections. | |
Yann Ziegler Bramha Dutt Vishwakarma, Aoibheann Brady Stephen Chuter Sam Royston Richard Westaway Jonathan Bamber M L Can GPS and GRACE data be used to separate past and present-day surface loading in a data-driven approach? Journal Article Geophysical Journal International, 232 (2), pp. 884-901, 2022. Abstract | Links | BibTeX | Tags: . @article{Ziegler2022, title = {Can GPS and GRACE data be used to separate past and present-day surface loading in a data-driven approach?}, author = {Yann Ziegler, Bramha Dutt Vishwakarma, Aoibheann Brady, Stephen Chuter, Sam Royston, Richard M Westaway, Jonathan L Bamber}, url = {https://doi.org/10.1093/gji/ggac365}, doi = {doi:10.1093/gji/ggac365}, year = {2022}, date = {2022-09-17}, journal = {Geophysical Journal International}, volume = {232}, number = {2}, pages = {884-901}, abstract = {Glacial isostatic adjustment (GIA) and the hydrological cycle are both associated with mass changes and vertical land motion (VLM), which are observed by GRACE and GPS, respectively. Hydrology-related VLM results from the instantaneous response of the elastic solid Earth to surface loading by freshwater, whereas GIA-related VLM reveals the long-term response of the viscoelastic Earth mantle to past ice loading history. Thus, observations of mass changes and VLM are interrelated, making GIA and hydrology difficult to quantify and study independently. In this work, we investigate the feasibility of separating these processes based on GRACE and GPS observations, in a fully data-driven and physically consistent approach. We take advantage of the differences in the spatio-temporal characteristics of the GIA and hydrology fields to estimate the respective contributions of each component using a Bayesian hierarchical modelling framework. A closed-loop synthetic test confirms that our method successfully solves this source separation problem. However, there are significant challenges when applying the same approach with actual observations and the answer to the main question of this study is more nuanced. In particular, in regions where GPS station coverage is sparse, the lack of informative data becomes a limiting factor.}, keywords = {.}, pubstate = {published}, tppubtype = {article} } Glacial isostatic adjustment (GIA) and the hydrological cycle are both associated with mass changes and vertical land motion (VLM), which are observed by GRACE and GPS, respectively. Hydrology-related VLM results from the instantaneous response of the elastic solid Earth to surface loading by freshwater, whereas GIA-related VLM reveals the long-term response of the viscoelastic Earth mantle to past ice loading history. Thus, observations of mass changes and VLM are interrelated, making GIA and hydrology difficult to quantify and study independently. In this work, we investigate the feasibility of separating these processes based on GRACE and GPS observations, in a fully data-driven and physically consistent approach. We take advantage of the differences in the spatio-temporal characteristics of the GIA and hydrology fields to estimate the respective contributions of each component using a Bayesian hierarchical modelling framework. A closed-loop synthetic test confirms that our method successfully solves this source separation problem. However, there are significant challenges when applying the same approach with actual observations and the answer to the main question of this study is more nuanced. In particular, in regions where GPS station coverage is sparse, the lack of informative data becomes a limiting factor. | |
Vishwakarma B. D., Ziegler Bamber Y J L; Royston, S Separating GIA signal from surface mass change using GPS and GRACE data Journal Article Geophysical Journal International, 232 (1), pp. 537-547, 2022. Abstract | Links | BibTeX | Tags: . @article{Vishwakarma2022, title = {Separating GIA signal from surface mass change using GPS and GRACE data}, author = {Vishwakarma, B. D., Ziegler, Y., Bamber, J. L., and Royston, S.}, url = {https://doi.org/10.1093/gji/ggac336}, doi = {doi:10.1093/gji/ggac336}, year = {2022}, date = {2022-08-23}, journal = {Geophysical Journal International}, volume = {232}, number = {1}, pages = {537-547}, abstract = {The visco-elastic response of the solid Earth to the past glacial cycles and the present-day surface mass change (PDSMC) are detected by the geodetic observation systems such as global navigation satellite system and satellite gravimetry. Majority of the contemporary PDSMC is driven by climate change and in order to better understand them using the aforementioned geodetic observations, glacial isostatic adjustment (GIA) signal should be accounted first. The default approach is to use forward GIA models that use uncertain ice-load history and approximate Earth rheology to predict GIA, yielding large uncertainties. The proliferation of contemporary, global, geodetic observations and their coverage have therefore enabled estimation of data-driven GIA solutions. A novel framework is presented that uses geophysical relations between the vertical land motion (VLM) and geopotential anomaly due to GIA and PDSMC to express GPS VLM trends and GRACE geopotential trends as a function of either GIA or PDSMC, which can be easily solved using least-squares regression. The GIA estimates are data-driven and differ significantly from forward models over Alaska and Greenland.}, keywords = {.}, pubstate = {published}, tppubtype = {article} } The visco-elastic response of the solid Earth to the past glacial cycles and the present-day surface mass change (PDSMC) are detected by the geodetic observation systems such as global navigation satellite system and satellite gravimetry. Majority of the contemporary PDSMC is driven by climate change and in order to better understand them using the aforementioned geodetic observations, glacial isostatic adjustment (GIA) signal should be accounted first. The default approach is to use forward GIA models that use uncertain ice-load history and approximate Earth rheology to predict GIA, yielding large uncertainties. The proliferation of contemporary, global, geodetic observations and their coverage have therefore enabled estimation of data-driven GIA solutions. A novel framework is presented that uses geophysical relations between the vertical land motion (VLM) and geopotential anomaly due to GIA and PDSMC to express GPS VLM trends and GRACE geopotential trends as a function of either GIA or PDSMC, which can be easily solved using least-squares regression. The GIA estimates are data-driven and differ significantly from forward models over Alaska and Greenland. | |
Fang Xu Yilei Shi, Patrick Ebel Lei Yu Gui-Song Xia Wen Yang ; Zhu, Xiao Xiang GLF-CR: SAR-Enhanced Cloud Removal with Global-Local Fusion Journal Article ISPRS Journal of Photogrammetry and Remote Sensing, 2022. @article{xu2022, title = {GLF-CR: SAR-Enhanced Cloud Removal with Global-Local Fusion}, author = {Fang Xu, Yilei Shi, Patrick Ebel,Lei Yu, Gui-Song Xia, Wen Yang, and Xiao Xiang Zhu}, year = {2022}, date = {2022-08-10}, journal = {ISPRS Journal of Photogrammetry and Remote Sensing}, keywords = {.}, pubstate = {published}, tppubtype = {article} } | |
Mrinalini Kochupillai Matthias Kahl, Michael Schmitt Hannes Taubenböck Xiao Xiang Zhu Artificial Intelligence for Earth Observation: Understanding Emerging Ethical Issues and Opportunities Journal Article IEEE Geoscience and Remote Sensing Magazine, 2022. @article{kochupillai2022, title = {Artificial Intelligence for Earth Observation: Understanding Emerging Ethical Issues and Opportunities}, author = {Mrinalini Kochupillai, Matthias Kahl, Michael Schmitt, Hannes Taubenböck, Xiao Xiang Zhu}, year = {2022}, date = {2022-08-10}, journal = {IEEE Geoscience and Remote Sensing Magazine}, keywords = {.}, pubstate = {published}, tppubtype = {article} } | |
Grinsted A., Bamber Bingham Buzzard Nias Ng J R S I K; Weeks, J The Transient Sea Level response to external forcing in CMIP6 models Journal Article Earth's Future, 10 (10), 2022. Abstract | Links | BibTeX | Tags: . @article{Grinsted2022, title = {The Transient Sea Level response to external forcing in CMIP6 models}, author = {Grinsted, A., Bamber, J., Bingham, R., Buzzard, S., Nias, I., Ng, K., and Weeks, J.}, url = {https://doi.org/10.1029/2022EF002696}, doi = {2022EF002696}, year = {2022}, date = {2022-08-02}, journal = {Earth's Future}, volume = {10}, number = {10}, abstract = {Earth is warming and sea levels are rising as land-based ice is lost to melt, and oceans expand due to accumulation of heat. The pace of ice loss and steric expansion is linked to the intensity of warming. How much faster sea level will rise as climate warms is, however, highly uncertain and difficult to model. Here, we quantify the transient sea level sensitivity of the sea level budget in both models and observations. Models show little change in sensitivity to warming between the first and second half of the twenty-first century for most contributors. The exception is glaciers and ice caps (GIC) that have a greater sensitivity pre-2050 (2.8 ± 0.4 mm/yr/K) compared to later (0.7 ± 0.1 mm/yr/K). We attribute this change to the short response time of glaciers and their changing area over time. Model sensitivities of steric expansion (1.5 ± 0.2 mm/yr/K), and Greenland Ice Sheet mass loss (0.8 ± 0.2 mm/yr/K) are greater than, but still compatible with, corresponding estimates from historical data (1.4 ± 0.5 and 0.4 ± 0.2 mm/yr/K). Antarctic Ice Sheet (AIS) models tends to show lower rates of sea level rise (SLR) with warming (−0.0 ± 0.3 mm/yr/K) in contrast to historical estimates (0.4 ± 0.2 mm/yr/K). This apparent low bias in AIS sensitivity is only partly able to account for a similar low bias identified in the sensitivity of global mean sea level excluding GIC (3.1 ± 0.4 vs. 2.3 ± 0.4 mm/yr/K). The balance temperature, where SLR is zero, lies close to the pre-industrial value, implying that SLR can only be mitigated by substantial global cooling.}, keywords = {.}, pubstate = {published}, tppubtype = {article} } Earth is warming and sea levels are rising as land-based ice is lost to melt, and oceans expand due to accumulation of heat. The pace of ice loss and steric expansion is linked to the intensity of warming. How much faster sea level will rise as climate warms is, however, highly uncertain and difficult to model. Here, we quantify the transient sea level sensitivity of the sea level budget in both models and observations. Models show little change in sensitivity to warming between the first and second half of the twenty-first century for most contributors. The exception is glaciers and ice caps (GIC) that have a greater sensitivity pre-2050 (2.8 ± 0.4 mm/yr/K) compared to later (0.7 ± 0.1 mm/yr/K). We attribute this change to the short response time of glaciers and their changing area over time. Model sensitivities of steric expansion (1.5 ± 0.2 mm/yr/K), and Greenland Ice Sheet mass loss (0.8 ± 0.2 mm/yr/K) are greater than, but still compatible with, corresponding estimates from historical data (1.4 ± 0.5 and 0.4 ± 0.2 mm/yr/K). Antarctic Ice Sheet (AIS) models tends to show lower rates of sea level rise (SLR) with warming (−0.0 ± 0.3 mm/yr/K) in contrast to historical estimates (0.4 ± 0.2 mm/yr/K). This apparent low bias in AIS sensitivity is only partly able to account for a similar low bias identified in the sensitivity of global mean sea level excluding GIC (3.1 ± 0.4 vs. 2.3 ± 0.4 mm/yr/K). The balance temperature, where SLR is zero, lies close to the pre-industrial value, implying that SLR can only be mitigated by substantial global cooling. | |
Vishwakarma BD Ramsankaran R, Azam MF Bolch Mandal Srivastava Kumar Sahu Navinkumar PJ Tanniru SR Javed Soheb Dimri AP Yadav Devaraju Chinnasamy Reddy MJ Murugesan GP Arora Jain SK Ojha CSP Harrison T A S P R A M M B P M S; J, Bamber Challenges in Understanding the Variability of the Cryosphere in the Himalaya and Its Impact on Regional Water Resources Journal Article Frontiers in Water, 4 , 2022. Abstract | Links | BibTeX | Tags: . @article{BD2022, title = {Challenges in Understanding the Variability of the Cryosphere in the Himalaya and Its Impact on Regional Water Resources}, author = {Vishwakarma BD, Ramsankaran R, Azam MF, Bolch T, Mandal A, Srivastava S, Kumar P, Sahu R, Navinkumar PJ, Tanniru SR, Javed A, Soheb M, Dimri AP, Yadav M, Devaraju B, Chinnasamy P, Reddy MJ, Murugesan GP, Arora M, Jain SK, Ojha CSP, Harrison S and Bamber J }, url = { https://doi.org/10.3389/frwa.2022.909246}, year = {2022}, date = {2022-07-28}, journal = {Frontiers in Water}, volume = {4}, abstract = {The Himalaya plays a vital role in regulating the freshwater availability for nearly a billion people living in the Indus, Ganga, and Brahmaputra River basins. Due to climate change and constantly evolving human-hydrosphere interactions, including land use/cover changes, groundwater extraction, reservoir or dam construction, water availability has undergone significant change, and is expected to change further in the future. Therefore, understanding the spatiotemporal evolution of the hydrological cycle over the Himalaya and its river basins has been one of the most critical exercises toward ensuring regional water security. However, due to the lack of extensive in-situ measurements, complex hydro-climatic environment, and limited collaborative efforts, large gaps in our understanding exist. Moreover, there are several significant issues with available studies, such as lack of consistent hydro-meteorological datasets, very few attempts at integrating different data types, limited spatiotemporal sampling of hydro-meteorological measurements, lack of open access to in-situ datasets, poorly accounted anthropogenic climate feedbacks, and limited understanding of the hydro-meteorological drivers over the region. These factors result in large uncertainties in our estimates of current and future water availability over the Himalaya, which constraints the development of sustainable water management strategies for its river catchments hampering our preparedness for the current and future changes in hydro-climate. To address these issues, a partnership development workshop entitled “Water sEcurity assessment in rIvers oriGinating from Himalaya (WEIGH),” was conducted between the 07th and 11th September 2020. Based on the intense discussions and deliberations among the participants, the most important and urgent research questions were identified. This white paper synthesizes the current understanding, highlights, and the most significant research gaps and research priorities for studying water availability in the Himalaya.}, keywords = {.}, pubstate = {published}, tppubtype = {article} } The Himalaya plays a vital role in regulating the freshwater availability for nearly a billion people living in the Indus, Ganga, and Brahmaputra River basins. Due to climate change and constantly evolving human-hydrosphere interactions, including land use/cover changes, groundwater extraction, reservoir or dam construction, water availability has undergone significant change, and is expected to change further in the future. Therefore, understanding the spatiotemporal evolution of the hydrological cycle over the Himalaya and its river basins has been one of the most critical exercises toward ensuring regional water security. However, due to the lack of extensive in-situ measurements, complex hydro-climatic environment, and limited collaborative efforts, large gaps in our understanding exist. Moreover, there are several significant issues with available studies, such as lack of consistent hydro-meteorological datasets, very few attempts at integrating different data types, limited spatiotemporal sampling of hydro-meteorological measurements, lack of open access to in-situ datasets, poorly accounted anthropogenic climate feedbacks, and limited understanding of the hydro-meteorological drivers over the region. These factors result in large uncertainties in our estimates of current and future water availability over the Himalaya, which constraints the development of sustainable water management strategies for its river catchments hampering our preparedness for the current and future changes in hydro-climate. To address these issues, a partnership development workshop entitled “Water sEcurity assessment in rIvers oriGinating from Himalaya (WEIGH),” was conducted between the 07th and 11th September 2020. Based on the intense discussions and deliberations among the participants, the most important and urgent research questions were identified. This white paper synthesizes the current understanding, highlights, and the most significant research gaps and research priorities for studying water availability in the Himalaya. | |
Sam Royston Rory J. Bingham, ; Bamber, Jonathan L Attributing decadal climate variability in coastal sea-level trends Journal Article Ocean Science, 18 (4), pp. 1093–1107, 2022. Abstract | Links | BibTeX | Tags: . @article{Royston2022, title = {Attributing decadal climate variability in coastal sea-level trends}, author = { Sam Royston, Rory J. Bingham, and Jonathan L. Bamber }, url = {https://doi.org/10.5194/os-18-1093-2022}, year = {2022}, date = {2022-07-27}, journal = {Ocean Science}, volume = {18}, number = {4}, pages = {1093–1107}, abstract = {Decadal sea-level variability masks longer-term changes due to natural and anthropogenic drivers in short-duration records and increases uncertainty in trend and acceleration estimates. When making regional coastal management and adaptation decisions, it is important to understand the drivers of these changes to account for periods of reduced or enhanced sea-level change. The variance in decadal sea-level trends about the global mean is quantified and mapped around the global coastlines of the Atlantic, Pacific, and Indian oceans from historical CMIP6 runs and a high-resolution ocean model forced by reanalysis data. We reconstruct coastal, sea-level trends via linear relationships with climate mode and oceanographic indices. Using this approach, more than one-third of the variability in decadal sea-level trends can be explained by climate indices at 24.6 % to 73.1 % of grid cells located within 25 km of a coast in the Atlantic, Pacific, and Indian oceans. At 10.9 % of the world's coastline, climate variability explains over two-thirds of the decadal sea-level trend. By investigating the steric, manometric, and gravitational components of sea-level trend independently, it is apparent that much of the coastal ocean variability is dominated by the manometric signal, the consequence of the open-ocean steric signal propagating onto the continental shelf. Additionally, decadal variability in the gravitational, rotational, and solid-Earth deformation (GRD) signal should not be ignored in the total. There are locations such as the Persian Gulf and African west coast where decadal sea-level variability is historically small that are susceptible to future changes in hydrology and/or ice mass changes that drive intensified regional GRD sea-level change above the global mean. The magnitude of variance explainable by climate modes quantified in this study indicates an enhanced uncertainty in projections of short- to mid-term regional sea-level trend.}, keywords = {.}, pubstate = {published}, tppubtype = {article} } Decadal sea-level variability masks longer-term changes due to natural and anthropogenic drivers in short-duration records and increases uncertainty in trend and acceleration estimates. When making regional coastal management and adaptation decisions, it is important to understand the drivers of these changes to account for periods of reduced or enhanced sea-level change. The variance in decadal sea-level trends about the global mean is quantified and mapped around the global coastlines of the Atlantic, Pacific, and Indian oceans from historical CMIP6 runs and a high-resolution ocean model forced by reanalysis data. We reconstruct coastal, sea-level trends via linear relationships with climate mode and oceanographic indices. Using this approach, more than one-third of the variability in decadal sea-level trends can be explained by climate indices at 24.6 % to 73.1 % of grid cells located within 25 km of a coast in the Atlantic, Pacific, and Indian oceans. At 10.9 % of the world's coastline, climate variability explains over two-thirds of the decadal sea-level trend. By investigating the steric, manometric, and gravitational components of sea-level trend independently, it is apparent that much of the coastal ocean variability is dominated by the manometric signal, the consequence of the open-ocean steric signal propagating onto the continental shelf. Additionally, decadal variability in the gravitational, rotational, and solid-Earth deformation (GRD) signal should not be ignored in the total. There are locations such as the Persian Gulf and African west coast where decadal sea-level variability is historically small that are susceptible to future changes in hydrology and/or ice mass changes that drive intensified regional GRD sea-level change above the global mean. The magnitude of variance explainable by climate modes quantified in this study indicates an enhanced uncertainty in projections of short- to mid-term regional sea-level trend. | |
van den Khan S. A., Colgan Neumann Broeke Brunt Noël Bamber Hassan W T A M R K M B J L J; Bjørk, A A Accelerating Ice Loss From Peripheral Glaciers in North Greenland Journal Article Geophysical Research Letters, 49 (12), 2022. Abstract | Links | BibTeX | Tags: . @article{Khan2022c, title = {Accelerating Ice Loss From Peripheral Glaciers in North Greenland}, author = {Khan, S. A., Colgan, W., Neumann, T. A., van den Broeke, M. R., Brunt, K. M., Noël, B., Bamber, J. L., Hassan, J., and Bjørk, A. A. }, url = {https://doi.org/10.1029/2022GL098915}, doi = {2022GL098915}, year = {2022}, date = {2022-06-16}, journal = {Geophysical Research Letters}, volume = {49}, number = {12}, abstract = {In recent decades, Greenland's peripheral glaciers have experienced large-scale mass loss, resulting in a substantial contribution to sea level rise. While their total area of Greenland ice cover is relatively small (4%), their mass loss is disproportionally large compared to the Greenland ice sheet. Satellite altimetry from Ice, Cloud, and land Elevation Satellite (ICESat) and ICESat-2 shows that mass loss from Greenland's peripheral glaciers increased from 27.2 ± 6.2 Gt/yr (February 2003–October 2009) to 42.3 ± 6.2 Gt/yr (October 2018–December 2021). These relatively small glaciers now constitute 11 ± 2% of Greenland's ice loss and contribute to global sea level rise. In the period October 2018–December 2021, mass loss increased by a factor of four for peripheral glaciers in North Greenland. While peripheral glacier mass loss is widespread, we also observe a complex regional pattern where increases in precipitation at high altitudes have partially counteracted increases in melt at low altitude.}, keywords = {.}, pubstate = {published}, tppubtype = {article} } In recent decades, Greenland's peripheral glaciers have experienced large-scale mass loss, resulting in a substantial contribution to sea level rise. While their total area of Greenland ice cover is relatively small (4%), their mass loss is disproportionally large compared to the Greenland ice sheet. Satellite altimetry from Ice, Cloud, and land Elevation Satellite (ICESat) and ICESat-2 shows that mass loss from Greenland's peripheral glaciers increased from 27.2 ± 6.2 Gt/yr (February 2003–October 2009) to 42.3 ± 6.2 Gt/yr (October 2018–December 2021). These relatively small glaciers now constitute 11 ± 2% of Greenland's ice loss and contribute to global sea level rise. In the period October 2018–December 2021, mass loss increased by a factor of four for peripheral glaciers in North Greenland. While peripheral glacier mass loss is widespread, we also observe a complex regional pattern where increases in precipitation at high altitudes have partially counteracted increases in melt at low altitude. | |
Rosa, Laura Elena Cué La; Oliveira, Dário Augusto Borges Learning from Label Proportions with Prototypical Contrastive Clustering Journal Article Proceedings of the AAAI Conference on Artificial Intelligence, 36 (2), pp. 2153-2161, 2022. @article{Rosa_Oliveira_2022, title = {Learning from Label Proportions with Prototypical Contrastive Clustering}, author = {Laura Elena Cué La Rosa and Dário Augusto Borges Oliveira}, url = {https://ojs.aaai.org/index.php/AAAI/article/view/20112}, doi = {10.1609/aaai.v36i2.20112}, year = {2022}, date = {2022-06-01}, journal = {Proceedings of the AAAI Conference on Artificial Intelligence}, volume = {36}, number = {2}, pages = {2153-2161}, keywords = {}, pubstate = {published}, tppubtype = {article} } | |
Chuter S. J., Zammit-Mangion Rougier Dawson A J G; Bamber, J L Mass evolution of the Antarctic Peninsula over the last 2 decades from a joint Bayesian inversion Journal Article The Cryosphere, 16 (4), pp. 1349-1367, 2022. Abstract | Links | BibTeX | Tags: . @article{Chuter2022, title = {Mass evolution of the Antarctic Peninsula over the last 2 decades from a joint Bayesian inversion}, author = {Chuter, S. J., Zammit-Mangion, A., Rougier, J., Dawson, G., and Bamber, J. L. }, url = {https://tc.copernicus.org/articles/16/1349/2022/}, doi = {tc-16-1349-2022}, year = {2022}, date = {2022-04-12}, journal = {The Cryosphere}, volume = {16}, number = {4}, pages = {1349-1367}, abstract = {The Antarctic Peninsula has become an increasingly important component of the Antarctic Ice Sheet mass budget over the last 2 decades, with mass losses generally increasing. However, due to the challenges presented by the topography and geometry of the region, there remain large variations in mass balance estimates from conventional approaches and in assessing the relative contribution of individual ice sheet processes. Here, we use a regionally optimized Bayesian hierarchical model joint inversion approach that combines data from multiple altimetry studies (ENVISAT, ICESat, CryoSat-2 swath), gravimetry (GRACE and GRACE-FO), and localized DEM differencing observations to solve for annual mass trends and their attribution to individual driving processes for the period 2003–2019. This is first time that such localized observations have been assimilated directly to estimate mass balance as part of a wider-scale regional assessment. The region experienced a mass imbalance rate of Gt yr−1 between 2003 and 2019, predominantly driven by accelerations in ice dynamic mass losses in the first decade and sustained thereafter. Inter-annual variability is driven by surface processes, particularly in 2016 due to increased precipitation driven by an extreme El Niño, which temporarily returned the sector back to a state of positive mass balance. In the West Palmer Land and the English Coast regions, surface processes are a greater contributor to mass loss than ice dynamics in the early part of the 2010s. Our results show good agreement with conventional and other combination approaches, improving confidence in the robustness of mass trend estimates, and in turn, understanding of the region's response to changes in external forcing.}, keywords = {.}, pubstate = {published}, tppubtype = {article} } The Antarctic Peninsula has become an increasingly important component of the Antarctic Ice Sheet mass budget over the last 2 decades, with mass losses generally increasing. However, due to the challenges presented by the topography and geometry of the region, there remain large variations in mass balance estimates from conventional approaches and in assessing the relative contribution of individual ice sheet processes. Here, we use a regionally optimized Bayesian hierarchical model joint inversion approach that combines data from multiple altimetry studies (ENVISAT, ICESat, CryoSat-2 swath), gravimetry (GRACE and GRACE-FO), and localized DEM differencing observations to solve for annual mass trends and their attribution to individual driving processes for the period 2003–2019. This is first time that such localized observations have been assimilated directly to estimate mass balance as part of a wider-scale regional assessment. The region experienced a mass imbalance rate of Gt yr−1 between 2003 and 2019, predominantly driven by accelerations in ice dynamic mass losses in the first decade and sustained thereafter. Inter-annual variability is driven by surface processes, particularly in 2016 due to increased precipitation driven by an extreme El Niño, which temporarily returned the sector back to a state of positive mass balance. In the West Palmer Land and the English Coast regions, surface processes are a greater contributor to mass loss than ice dynamics in the early part of the 2010s. Our results show good agreement with conventional and other combination approaches, improving confidence in the robustness of mass trend estimates, and in turn, understanding of the region's response to changes in external forcing. | |
Qian, Kun; Wang, Yuanyuan; Shi, Yilei; Zhu, Xiao Xiang γ-Net: Superresolving SAR Tomographic Inversion via Deep Learning Journal Article IEEE Transactions on Geoscience and Remote Sensing, 2022. @article{, title = {γ-Net: Superresolving SAR Tomographic Inversion via Deep Learning}, author = {Kun Qian and Yuanyuan Wang and Yilei Shi and Xiao Xiang Zhu}, year = {2022}, date = {2022-03-30}, booktitle = {IEEE Transactions on Geoscience and Remote Sensing}, journal = {IEEE Transactions on Geoscience and Remote Sensing}, abstract = {(This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.) Synthetic aperture radar tomography (TomoSAR) has been extensively employed in 3-D reconstruction in dense urban areas using high-resolution SAR acquisitions. Compressive sensing (CS)-based algorithms are generally considered as the state of the art in super-resolving TomoSAR, in particular in the single look case. This superior performance comes at the cost of extra computational burdens, because of the sparse reconstruction, which cannot be solved analytically and we need to employ computationally expensive iterative solvers. In this paper, we propose a novel deep learning-based super-resolving TomoSAR inversion approach, γ-Net, to tackle this challenge. γ-Net adopts advanced complex-valued learned iterative shrinkage thresholding algorithm (CV-LISTA) to mimic the iterative optimization step in sparse reconstruction. Simulations show the height estimate from a well-trained γ-Net approaches the Cramér-Rao lower bound while improving the computational efficiency by 1 to 2 orders of magnitude comparing to the first-order CS-based methods. It also shows no degradation in the super-resolution power comparing to the state-of-the-art second-order TomoSAR solvers, which are much more computationally expensive than the first-order methods. Specifically, γ-Net reaches more than 90% detection rate in moderate super-resolving cases at 25 measurements at 6dB SNR. Moreover, simulation at limited baselines demonstrates that the proposed algorithm outperforms the second-order CS-based method by a fair margin. Test on real TerraSAR-X data with just 6 interferograms also shows high-quality 3-D reconstruction with high-density detected double scatterers.}, keywords = {}, pubstate = {published}, tppubtype = {article} } (This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.) Synthetic aperture radar tomography (TomoSAR) has been extensively employed in 3-D reconstruction in dense urban areas using high-resolution SAR acquisitions. Compressive sensing (CS)-based algorithms are generally considered as the state of the art in super-resolving TomoSAR, in particular in the single look case. This superior performance comes at the cost of extra computational burdens, because of the sparse reconstruction, which cannot be solved analytically and we need to employ computationally expensive iterative solvers. In this paper, we propose a novel deep learning-based super-resolving TomoSAR inversion approach, γ-Net, to tackle this challenge. γ-Net adopts advanced complex-valued learned iterative shrinkage thresholding algorithm (CV-LISTA) to mimic the iterative optimization step in sparse reconstruction. Simulations show the height estimate from a well-trained γ-Net approaches the Cramér-Rao lower bound while improving the computational efficiency by 1 to 2 orders of magnitude comparing to the first-order CS-based methods. It also shows no degradation in the super-resolution power comparing to the state-of-the-art second-order TomoSAR solvers, which are much more computationally expensive than the first-order methods. Specifically, γ-Net reaches more than 90% detection rate in moderate super-resolving cases at 25 measurements at 6dB SNR. Moreover, simulation at limited baselines demonstrates that the proposed algorithm outperforms the second-order CS-based method by a fair margin. Test on real TerraSAR-X data with just 6 interferograms also shows high-quality 3-D reconstruction with high-density detected double scatterers. | |
van den Shfaqat A. Khan Jonathan L. Bamber, Eric Rignot Veit Helm Andy Aschwanden David Holland Michiel Broeke Michalea King Brice Noël Martin Truffer Angelika Humbert William Colgan Saurabh Vijay Peter Kuipers Munneke M Greenland Mass Trends From Airborne and Satellite Altimetry During 2011–2020 Journal Article Journal of Geophysical Research: Earth Surface, 127 (4), 2022. Abstract | Links | BibTeX | Tags: . @article{Khan2022b, title = {Greenland Mass Trends From Airborne and Satellite Altimetry During 2011–2020}, author = {Shfaqat A. Khan, Jonathan L. Bamber, Eric Rignot, Veit Helm, Andy Aschwanden, David M. Holland, Michiel van den Broeke, Michalea King, Brice Noël, Martin Truffer, Angelika Humbert, William Colgan, Saurabh Vijay, Peter Kuipers Munneke }, url = {https://doi.org/10.1029/2021JF006505}, doi = {e2021JF006505}, year = {2022}, date = {2022-03-21}, journal = {Journal of Geophysical Research: Earth Surface}, volume = {127}, number = {4}, abstract = {We use satellite and airborne altimetry to estimate annual mass changes of the Greenland Ice Sheet. We estimate ice loss corresponding to a sea-level rise of 6.9 ± 0.4 mm from April 2011 to April 2020, with a highest annual ice loss rate of 1.4 mm/yr sea-level equivalent from April 2019 to April 2020. On a regional scale, our annual mass loss timeseries reveals 10–15 m/yr dynamic thickening at the terminus of Jakobshavn Isbræ from April 2016 to April 2018, followed by a return to dynamic thinning. We observe contrasting patterns of mass loss acceleration in different basins across the ice sheet and suggest that these spatiotemporal trends could be useful for calibrating and validating prognostic ice sheet models. In addition to resolving the spatial and temporal fingerprint of Greenland's recent ice loss, these mass loss grids are key for partitioning contemporary elastic vertical land motion from longer-term glacial isostatic adjustment (GIA) trends at GPS stations around the ice sheet. Our ice-loss product results in a significantly different GIA interpretation from a previous ice-loss product.}, keywords = {.}, pubstate = {published}, tppubtype = {article} } We use satellite and airborne altimetry to estimate annual mass changes of the Greenland Ice Sheet. We estimate ice loss corresponding to a sea-level rise of 6.9 ± 0.4 mm from April 2011 to April 2020, with a highest annual ice loss rate of 1.4 mm/yr sea-level equivalent from April 2019 to April 2020. On a regional scale, our annual mass loss timeseries reveals 10–15 m/yr dynamic thickening at the terminus of Jakobshavn Isbræ from April 2016 to April 2018, followed by a return to dynamic thinning. We observe contrasting patterns of mass loss acceleration in different basins across the ice sheet and suggest that these spatiotemporal trends could be useful for calibrating and validating prognostic ice sheet models. In addition to resolving the spatial and temporal fingerprint of Greenland's recent ice loss, these mass loss grids are key for partitioning contemporary elastic vertical land motion from longer-term glacial isostatic adjustment (GIA) trends at GPS stations around the ice sheet. Our ice-loss product results in a significantly different GIA interpretation from a previous ice-loss product. | |
Tom Mitcham G. Hilmar Gudmundsson, ; Bamber, Jonathan L The instantaneous impact of calving and thinning on the Larsen C Ice Shelf Journal Article The Cryosphere, 16 (3), pp. 883–901, 2022. Abstract | Links | BibTeX | Tags: . @article{Mitcham2022, title = {The instantaneous impact of calving and thinning on the Larsen C Ice Shelf}, author = {Tom Mitcham, G. Hilmar Gudmundsson, and Jonathan L. Bamber }, url = {https://doi.org/10.5194/tc-16-883-2022}, doi = {tc-16-883-2022}, year = {2022}, date = {2022-03-11}, journal = {The Cryosphere}, volume = {16}, number = {3}, pages = {883–901}, abstract = {The Antarctic Peninsula has seen rapid and widespread changes in the extent of its ice shelves in recent decades, including the collapse of the Larsen A and B ice shelves in 1995 and 2002, respectively. In 2017 the Larsen C Ice Shelf (LCIS) lost around 10 % of its area by calving one of the largest icebergs ever recorded (A68). This has raised questions about the structural integrity of the shelf and the impact of any changes in its extent on the flow of its tributary glaciers. In this work, we used an ice flow model to study the instantaneous impact of changes in the thickness and extent of the LCIS on ice dynamics and in particular on changes in the grounding line flux (GLF). We initialised the model to a pre-A68 calving state and first replicated the calving of the A68 iceberg. We found that there was a limited instantaneous impact on upstream flow – with speeds increasing by less than 10 % across almost all of the shelf – and a 0.28 % increase in GLF. This result is supported by observations of ice velocity made before and after the calving event. We then perturbed the ice-shelf geometry through a series of instantaneous, idealised calving and thinning experiments of increasing magnitude. We found that significant changes to the geometry of the ice shelf, through both calving and thinning, resulted in limited instantaneous changes in GLF. For example, to produce a doubling of GLF from calving, the new calving front needed to be moved to 5 km from the grounding line, removing almost the entire ice shelf. For thinning, over 200 m of the ice-shelf thickness had to be removed across the whole shelf to produce a doubling of GLF. Calculating the instantaneous increase in GLF (607 %) after removing the entire ice shelf allowed us to quantify the total amount of buttressing provided by the LCIS. From this, we identified that the region of the ice shelf in the first 5 km downstream of the grounding line provided over 80 % of the buttressing capacity of the shelf. This is due to the large resistive stresses generated in the narrow, local embayments downstream of the largest tributary glaciers.}, keywords = {.}, pubstate = {published}, tppubtype = {article} } The Antarctic Peninsula has seen rapid and widespread changes in the extent of its ice shelves in recent decades, including the collapse of the Larsen A and B ice shelves in 1995 and 2002, respectively. In 2017 the Larsen C Ice Shelf (LCIS) lost around 10 % of its area by calving one of the largest icebergs ever recorded (A68). This has raised questions about the structural integrity of the shelf and the impact of any changes in its extent on the flow of its tributary glaciers. In this work, we used an ice flow model to study the instantaneous impact of changes in the thickness and extent of the LCIS on ice dynamics and in particular on changes in the grounding line flux (GLF). We initialised the model to a pre-A68 calving state and first replicated the calving of the A68 iceberg. We found that there was a limited instantaneous impact on upstream flow – with speeds increasing by less than 10 % across almost all of the shelf – and a 0.28 % increase in GLF. This result is supported by observations of ice velocity made before and after the calving event. We then perturbed the ice-shelf geometry through a series of instantaneous, idealised calving and thinning experiments of increasing magnitude. We found that significant changes to the geometry of the ice shelf, through both calving and thinning, resulted in limited instantaneous changes in GLF. For example, to produce a doubling of GLF from calving, the new calving front needed to be moved to 5 km from the grounding line, removing almost the entire ice shelf. For thinning, over 200 m of the ice-shelf thickness had to be removed across the whole shelf to produce a doubling of GLF. Calculating the instantaneous increase in GLF (607 %) after removing the entire ice shelf allowed us to quantify the total amount of buttressing provided by the LCIS. From this, we identified that the region of the ice shelf in the first 5 km downstream of the grounding line provided over 80 % of the buttressing capacity of the shelf. This is due to the large resistive stresses generated in the narrow, local embayments downstream of the largest tributary glaciers. | |
Li T., Dawson Chuter G J S J; Bamber, J L A high-resolution Antarctic grounding zone product from ICESat-2 laser altimetry Journal Article Earth System Science Data, 14 (2), pp. 535–557, 2022. Abstract | Links | BibTeX | Tags: . @article{Li2022c, title = {A high-resolution Antarctic grounding zone product from ICESat-2 laser altimetry}, author = {Li, T., Dawson, G. J., Chuter, S. J., and Bamber, J. L. }, url = {https://doi.org/10.5194/essd-14-535-2022}, doi = {essd-14-535-2022}, year = {2022}, date = {2022-02-08}, journal = {Earth System Science Data}, volume = {14}, number = {2}, pages = {535–557}, abstract = {The Antarctic grounding zone, which is the transition between the fully grounded ice sheet to freely floating ice shelf, plays a critical role in ice sheet stability, mass budget calculations, and ice sheet model projections. It is therefore important to continuously monitor its location and migration over time. Here we present the first ICESat-2-derived high-resolution grounding zone product of the Antarctic Ice Sheet, including three important boundaries: the inland limit of tidal flexure (Point F), inshore limit of hydrostatic equilibrium (Point H), and the break in slope (Point Ib). This dataset was derived from automated techniques developed in this study, using ICESat-2 laser altimetry repeat tracks between 30 March 2019 and 30 September 2020. The new grounding zone product has a near-complete coverage of the Antarctic Ice Sheet with a total of 21 346 Point F, 18 149 Point H, and 36 765 Point Ib locations identified, including the difficult-to-survey grounding zones, such as the fast-flowing glaciers draining into the Amundsen Sea embayment. The locations of newly derived ICESat-2 landward limit of tidal flexure agree well with the most recent differential synthetic aperture radar interferometry (DInSAR) observations in 2018, with a mean absolute separation and standard deviation of 0.02 and 0.02 km, respectively. By comparing the ICESat-2-derived grounding zone with the previous grounding zone products, we find a grounding line retreat of up to 15 km on the Crary Ice Rise of Ross Ice Shelf and a pervasive landward grounding line migration along the Amundsen Sea embayment during the past 2 decades. We also identify the presence of ice plains on the Filchner–Ronne Ice Shelf and the influence of oscillating ocean tides on grounding zone migration. The product derived from this study is available at https://doi.org/10.5523/bris.bnqqyngt89eo26qk8keckglww (Li et al., 2021) and is archived and maintained at the National Snow and Ice Data Center.}, keywords = {.}, pubstate = {published}, tppubtype = {article} } The Antarctic grounding zone, which is the transition between the fully grounded ice sheet to freely floating ice shelf, plays a critical role in ice sheet stability, mass budget calculations, and ice sheet model projections. It is therefore important to continuously monitor its location and migration over time. Here we present the first ICESat-2-derived high-resolution grounding zone product of the Antarctic Ice Sheet, including three important boundaries: the inland limit of tidal flexure (Point F), inshore limit of hydrostatic equilibrium (Point H), and the break in slope (Point Ib). This dataset was derived from automated techniques developed in this study, using ICESat-2 laser altimetry repeat tracks between 30 March 2019 and 30 September 2020. The new grounding zone product has a near-complete coverage of the Antarctic Ice Sheet with a total of 21 346 Point F, 18 149 Point H, and 36 765 Point Ib locations identified, including the difficult-to-survey grounding zones, such as the fast-flowing glaciers draining into the Amundsen Sea embayment. The locations of newly derived ICESat-2 landward limit of tidal flexure agree well with the most recent differential synthetic aperture radar interferometry (DInSAR) observations in 2018, with a mean absolute separation and standard deviation of 0.02 and 0.02 km, respectively. By comparing the ICESat-2-derived grounding zone with the previous grounding zone products, we find a grounding line retreat of up to 15 km on the Crary Ice Rise of Ross Ice Shelf and a pervasive landward grounding line migration along the Amundsen Sea embayment during the past 2 decades. We also identify the presence of ice plains on the Filchner–Ronne Ice Shelf and the influence of oscillating ocean tides on grounding zone migration. The product derived from this study is available at https://doi.org/10.5523/bris.bnqqyngt89eo26qk8keckglww (Li et al., 2021) and is archived and maintained at the National Snow and Ice Data Center. | |
Li, Tian; Dawson, Geoffrey J; Chuter, Stephen J; Bamber, Jonathan L A high-resolution Antarctic grounding zone product from ICESat-2 laser altimetry Journal Article Earth System Science Data, 14 , pp. 535-557, 2022, ISSN: 1866-3516. Abstract | Links | BibTeX | Tags: @article{Li2022, title = {A high-resolution Antarctic grounding zone product from ICESat-2 laser altimetry}, author = {Tian Li and Geoffrey J Dawson and Stephen J Chuter and Jonathan L Bamber}, url = {https://essd.copernicus.org/articles/14/535/2022/}, doi = {10.5194/essd-14-535-2022}, issn = {1866-3516}, year = {2022}, date = {2022-01-01}, journal = {Earth System Science Data}, volume = {14}, pages = {535-557}, abstract = { }, <p><![CDATA[Abstract. The Antarctic grounding zone, which is the transition between the fully grounded ice sheet to freely floating ice shelf, plays a critical role in ice sheet stability, mass budget calculations, and ice sheet model projections. It is therefore important to continuously monitor its location and migration over time. Here we present the first ICESat-2-derived high-resolution grounding zone product of the Antarctic Ice Sheet, including three important boundaries: the inland limit of tidal flexure (Point F), inshore limit of hydrostatic equilibrium (Point H), and the break in slope (Point Ib). This dataset was derived from automated techniques developed in this study, using ICESat-2 laser altimetry repeat tracks between 30 March 2019 and 30 September 2020. The new grounding zone product has a near-complete coverage of the Antarctic Ice Sheet with a total of 21 346 Point F, 18 149 Point H, and 36 765 Point Ib locations identified, including the difficult-to-survey grounding zones, such as the fast-flowing glaciers draining into the Amundsen Sea embayment. The locations of newly derived ICESat-2 landward limit of tidal flexure agree well with the most recent differential synthetic aperture radar interferometry (DInSAR) observations in 2018, with a mean absolute separation and standard deviation of 0.02 and 0.02 km, respectively. By comparing the ICESat-2-derived grounding zone with the previous grounding zone products, we find a grounding line retreat of up to 15 km on the Crary Ice Rise of Ross Ice Shelf and a pervasive landward grounding line migration along the Amundsen Sea embayment during the past 2 decades. We also identify the presence of ice plains on the Filchner–Ronne Ice Shelf and the influence of oscillating ocean tides on grounding zone migration. The product derived from this study is available at https://doi.org/10.5523/bris.bnqqyngt89eo26qk8keckglww (Li et al., 2021) and is archived and maintained at the National Snow and Ice Data Center.]]></p> | |
Lehmann, Fanny; Vishwakarma, Bramha Dutt; Bamber, Jonathan How well are we able to close the water budget at the global scale? Journal Article Hydrology and Earth System Sciences, 26 , pp. 35-54, 2022, ISSN: 1607-7938. Abstract | Links | BibTeX | Tags: @article{Lehmann2022, title = {How well are we able to close the water budget at the global scale?}, author = {Fanny Lehmann and Bramha Dutt Vishwakarma and Jonathan Bamber}, url = {https://hess.copernicus.org/articles/26/35/2022/}, doi = {10.5194/hess-26-35-2022}, issn = {1607-7938}, year = {2022}, date = {2022-01-01}, journal = {Hydrology and Earth System Sciences}, volume = {26}, pages = {35-54}, abstract = { Abstract. The water budget equation describes the exchange of water between the land, ocean, and atmosphere. Being able to adequately close the water budget gives confidence in our ability to model and/or observe the spatio-temporal variations in the water cycle and its components. Due to advances in observation techniques, satellite sensors, and modelling, a number of data products are available that represent the components of water budget in both space and time. Despite these advances, closure of the water budget at the global scale has been elusive. In this study, we attempt to close the global water budget using precipitation, evapotranspiration, and runoff data at the catchment scale. The large number of recent state-of-the-art datasets provides a new evaluation of well-used datasets. These estimates are compared to terrestrial water storage (TWS) changes as measured by the Gravity Recovery And Climate Experiment (GRACE) satellite mission. We investigated 189 river basins covering more than 90 % of the continental land area. TWS changes derived from the water balance equation were compared against GRACE data using two metrics: the Nash–Sutcliffe efficiency (NSE) and the cyclostationary NSE. These metrics were used to assess the performance of more than 1600 combinations of the various datasets considered. We found a positive NSE and cyclostationary NSE in 99 % and 62 % of the basins examined respectively. This means that TWS changes reconstructed from the water balance equation were more accurate than the long-term (NSE) and monthly (cyclostationary NSE) mean of GRACE time series in the corresponding basins. By analysing different combinations of the datasets that make up the water balance, we identified data products that performed well in certain regions based on, for example, climatic zone. We identified that some of the good results were obtained due to the cancellation of errors in poor estimates of water budget components. Therefore, we used coefficients of variation to determine the relative quality of a data product, which helped us to identify bad combinations giving us good results. In general, water budget components from ERA5-Land and the Catchment Land Surface Model (CLSM) performed better than other products for most climatic zones. Conversely, the latest version of CLSM, v2.2, performed poorly for evapotranspiration in snow-dominated catchments compared, for example, with its predecessor and other datasets available. Thus, the nature of the catchment dynamics and balance between components affects the optimum combination of datasets. For regional studies, the combination of datasets that provides the most realistic TWS for a basin will depend on its climatic conditions and factors that cannot be determined a priori. We believe that the results of this study provide a road map for studying the water budget at catchment scale. },keywords = {}, pubstate = {published}, tppubtype = {article} } <p>Abstract. The water budget equation describes the exchange of water between the land, ocean, and atmosphere. Being able to adequately close the water budget gives confidence in our ability to model and/or observe the spatio-temporal variations in the water cycle and its components. Due to advances in observation techniques, satellite sensors, and modelling, a number of data products are available that represent the components of water budget in both space and time. Despite these advances, closure of the water budget at the global scale has been elusive. In this study, we attempt to close the global water budget using precipitation, evapotranspiration, and runoff data at the catchment scale. The large number of recent state-of-the-art datasets provides a new evaluation of well-used datasets. These estimates are compared to terrestrial water storage (TWS) changes as measured by the Gravity Recovery And Climate Experiment (GRACE) satellite mission. We investigated 189 river basins covering more than 90 % of the continental land area. TWS changes derived from the water balance equation were compared against GRACE data using two metrics: the Nash–Sutcliffe efficiency (NSE) and the cyclostationary NSE. These metrics were used to assess the performance of more than 1600 combinations of the various datasets considered. We found a positive NSE and cyclostationary NSE in 99 % and 62 % of the basins examined respectively. This means that TWS changes reconstructed from the water balance equation were more accurate than the long-term (NSE) and monthly (cyclostationary NSE) mean of GRACE time series in the corresponding basins. By analysing different combinations of the datasets that make up the water balance, we identified data products that performed well in certain regions based on, for example, climatic zone. We identified that some of the good results were obtained due to the cancellation of errors in poor estimates of water budget components. Therefore, we used coefficients of variation to determine the relative quality of a data product, which helped us to identify bad combinations giving us good results. In general, water budget components from ERA5-Land and the Catchment Land Surface Model (CLSM) performed better than other products for most climatic zones. Conversely, the latest version of CLSM, v2.2, performed poorly for evapotranspiration in snow-dominated catchments compared, for example, with its predecessor and other datasets available. Thus, the nature of the catchment dynamics and balance between components affects the optimum combination of datasets. For regional studies, the combination of datasets that provides the most realistic TWS for a basin will depend on its climatic conditions and factors that cannot be determined a priori. We believe that the results of this study provide a road map for studying the water budget at catchment scale.</p> | |
Gawlikowski, Jakob; Saha, Sudipan; Kruspe, Anna; Zhu, Xiao Xiang An Advanced Dirichlet Prior Network for Out-of-Distribution Detection in Remote Sensing Journal Article IEEE Transactions on Geoscience and Remote Sensing, 2022. BibTeX | Tags: @article{Gawlikowski2022, title = {An Advanced Dirichlet Prior Network for Out-of-Distribution Detection in Remote Sensing}, author = {Jakob Gawlikowski and Sudipan Saha and Anna Kruspe and Xiao Xiang Zhu}, year = {2022}, date = {2022-01-01}, journal = {IEEE Transactions on Geoscience and Remote Sensing}, publisher = {IEEE}, keywords = {}, pubstate = {published}, tppubtype = {article} } | |
Saha, Sudipan; Zhao, Shan; Zhu, Xiao Xiang Multi-target domain adaptation for remote sensing classification using graph neural network Journal Article IEEE Geoscience and Remote Sensing Letters, 2022. BibTeX | Tags: @article{Saha2022, title = {Multi-target domain adaptation for remote sensing classification using graph neural network}, author = {Sudipan Saha and Shan Zhao and Xiao Xiang Zhu}, year = {2022}, date = {2022-01-01}, journal = {IEEE Geoscience and Remote Sensing Letters}, publisher = {IEEE}, keywords = {}, pubstate = {published}, tppubtype = {article} } | |
Roschlaub, Robert; Glock, Clemens; Möst, Karin; Hümmer, Frank; Li, Qingyu; Auer, Stefan; Kruspe, Anna; Zhu, Xiao Xiang Implementierung einer KI-Infrastruktur zur automatisierten Erkennung von landesweiten Gebäudeveränderungen aus Luftbildern Journal Article ZFV - Zeitschrift für Geodasie, Geoinformation und Landmanagement, 2022. Abstract | Links | BibTeX | Tags: @article{Roschlaub2022, title = {Implementierung einer KI-Infrastruktur zur automatisierten Erkennung von landesweiten Gebäudeveränderungen aus Luftbildern}, author = {Robert Roschlaub and Clemens Glock and Karin Möst and Frank Hümmer and Qingyu Li and Stefan Auer and Anna Kruspe and Xiao Xiang Zhu}, url = {https://elib.dlr.de/185326/}, year = {2022}, date = {2022-01-01}, journal = {ZFV - Zeitschrift für Geodasie, Geoinformation und Landmanagement}, publisher = {Wißner-Verlag}, abstract = {Aufbauend auf einer vorangegangenen Studie über den Einsatz von Künstlicher Intelligenz (KI) zur Detektion von Gebäudeveränderungen im amtlichen Liegenschaftskataster, wird in diesem Beitrag die Transformation von der Projektphase in einen Produktivbetrieb vorgestellt, es werden verschiedene KI-Architekturen angesprochen, miteinander verglichen und eine Auswahlentscheidung gegeben. Die erzielten Ergebnisse einer landesweiten Gebäudedetektion werden gemeinsam mit Untersuchungen zu den Trainingsläufen, zur Performance und den von einem Vermessungsamt validierten KI-Ergebnissen präsentiert.}, keywords = {}, pubstate = {published}, tppubtype = {article} } Aufbauend auf einer vorangegangenen Studie über den Einsatz von Künstlicher Intelligenz (KI) zur Detektion von Gebäudeveränderungen im amtlichen Liegenschaftskataster, wird in diesem Beitrag die Transformation von der Projektphase in einen Produktivbetrieb vorgestellt, es werden verschiedene KI-Architekturen angesprochen, miteinander verglichen und eine Auswahlentscheidung gegeben. Die erzielten Ergebnisse einer landesweiten Gebäudedetektion werden gemeinsam mit Untersuchungen zu den Trainingsläufen, zur Performance und den von einem Vermessungsamt validierten KI-Ergebnissen präsentiert. | |
Stadtler, Scarlet; Betancourt, Clara; Roscher, Ribana Explainable Machine Learning Reveals Capabilities, Redundancy, and Limitations of a Geospatial Air Quality Benchmark Dataset Journal Article Machine Learning and Knowledge Extraction, 4 , pp. 150-171, 2022, ISSN: 2504-4990. Abstract | Links | BibTeX | Tags: @article{Stadtler2022, title = {Explainable Machine Learning Reveals Capabilities, Redundancy, and Limitations of a Geospatial Air Quality Benchmark Dataset}, author = {Scarlet Stadtler and Clara Betancourt and Ribana Roscher}, url = {https://www.mdpi.com/2504-4990/4/1/8}, doi = {10.3390/make4010008}, issn = {2504-4990}, year = {2022}, date = {2022-01-01}, journal = {Machine Learning and Knowledge Extraction}, volume = {4}, pages = {150-171}, publisher = {Multidisciplinary Digital Publishing Institute}, abstract = { Air quality is relevant to society because it poses environmental risks to humans and nature. We use explainable machine learning in air quality research by analyzing model predictions in relation to the underlying training data. The data originate from worldwide ozone observations, paired with geospatial data. We use two different architectures: a neural network and a random forest trained on various geospatial data to predict multi-year averages of the air pollutant ozone. To understand how both models function, we explain how they represent the training data and derive their predictions. By focusing on inaccurate predictions and explaining why these predictions fail, we can (i) identify underrepresented samples, (ii) flag unexpected inaccurate predictions, and (iii) point to training samples irrelevant for predictions on the test set. Based on the underrepresented samples, we suggest where to build new measurement stations. We also show which training samples do not substantially contribute to the model performance. This study demonstrates the application of explainable machine learning beyond simply explaining the trained model. },keywords = {}, pubstate = {published}, tppubtype = {article} } <p>Air quality is relevant to society because it poses environmental risks to humans and nature. We use explainable machine learning in air quality research by analyzing model predictions in relation to the underlying training data. The data originate from worldwide ozone observations, paired with geospatial data. We use two different architectures: a neural network and a random forest trained on various geospatial data to predict multi-year averages of the air pollutant ozone. To understand how both models function, we explain how they represent the training data and derive their predictions. By focusing on inaccurate predictions and explaining why these predictions fail, we can (i) identify underrepresented samples, (ii) flag unexpected inaccurate predictions, and (iii) point to training samples irrelevant for predictions on the test set. Based on the underrepresented samples, we suggest where to build new measurement stations. We also show which training samples do not substantially contribute to the model performance. This study demonstrates the application of explainable machine learning beyond simply explaining the trained model.</p> | |
Betancourt, C; Stomberg, T T; Edrich, A -K; Patnala, A; Schultz, M G; Roscher, R; Kowalski, J; Stadtler, S Global, high-resolution mapping of tropospheric ozone – explainable machine learning and impact of uncertainties Journal Article Geoscientific Model Development Discussions, 2022 , pp. 1-36, 2022. @article{Betancourt2022, title = {Global, high-resolution mapping of tropospheric ozone – explainable machine learning and impact of uncertainties}, author = {C Betancourt and T T Stomberg and A -K Edrich and A Patnala and M G Schultz and R Roscher and J Kowalski and S Stadtler}, url = {https://gmd.copernicus.org/preprints/gmd-2022-2/}, doi = {10.5194/gmd-2022-2}, year = {2022}, date = {2022-01-01}, journal = {Geoscientific Model Development Discussions}, volume = {2022}, pages = {1-36}, keywords = {}, pubstate = {published}, tppubtype = {article} } | |
Li, Tian; Dawson, Geoffrey J; Chuter, Stephen J; Bamber, Jonathan L A high-resolution Antarctic grounding zone product from ICESat-2 laser altimetry Journal Article Earth System Science Data, 14 , pp. 535–557, 2022, ISSN: 1866-3516. Abstract | Links | BibTeX | Tags: @article{Li2022b, title = {A high-resolution Antarctic grounding zone product from ICESat-2 laser altimetry}, author = {Tian Li and Geoffrey J Dawson and Stephen J Chuter and Jonathan L Bamber}, url = {https://essd.copernicus.org/articles/14/535/2022/}, doi = {10.5194/essd-14-535-2022}, issn = {1866-3516}, year = {2022}, date = {2022-01-01}, journal = {Earth System Science Data}, volume = {14}, pages = {535--557}, abstract = { }, <p><![CDATA[Abstract. The Antarctic grounding zone, which is the transition between the fully grounded ice sheet to freely floating ice shelf, plays a critical role in ice sheet stability, mass budget calculations, and ice sheet model projections. It is therefore important to continuously monitor its location and migration over time. Here we present the first ICESat-2-derived high-resolution grounding zone product of the Antarctic Ice Sheet, including three important boundaries: the inland limit of tidal flexure (Point F), inshore limit of hydrostatic equilibrium (Point H), and the break in slope (Point Ib). This dataset was derived from automated techniques developed in this study, using ICESat-2 laser altimetry repeat tracks between 30 March 2019 and 30 September 2020. The new grounding zone product has a near-complete coverage of the Antarctic Ice Sheet with a total of 21 346 Point F, 18 149 Point H, and 36 765 Point Ib locations identified, including the difficult-to-survey grounding zones, such as the fast-flowing glaciers draining into the Amundsen Sea embayment. The locations of newly derived ICESat-2 landward limit of tidal flexure agree well with the most recent differential synthetic aperture radar interferometry (DInSAR) observations in 2018, with a mean absolute separation and standard deviation of 0.02 and 0.02 km, respectively. By comparing the ICESat-2-derived grounding zone with the previous grounding zone products, we find a grounding line retreat of up to 15 km on the Crary Ice Rise of Ross Ice Shelf and a pervasive landward grounding line migration along the Amundsen Sea embayment during the past 2 decades. We also identify the presence of ice plains on the Filchner–Ronne Ice Shelf and the influence of oscillating ocean tides on grounding zone migration. The product derived from this study is available at https://doi.org/10.5523/bris.bnqqyngt89eo26qk8keckglww (Li et al., 2021) and is archived and maintained at the National Snow and Ice Data Center.]]></p> | |
Gawlikowski, Jakob; Saha, Sudipan; Kruspe, Anna; Zhu, Xiao Xiang An Advanced Dirichlet Prior Network for Out-of-Distribution Detection in Remote Sensing Journal Article IEEE Transactions on Geoscience and Remote Sensing, 2022. BibTeX | Tags: @article{Gawlikowski2022b, title = {An Advanced Dirichlet Prior Network for Out-of-Distribution Detection in Remote Sensing}, author = {Jakob Gawlikowski and Sudipan Saha and Anna Kruspe and Xiao Xiang Zhu}, year = {2022}, date = {2022-01-01}, journal = {IEEE Transactions on Geoscience and Remote Sensing}, publisher = {IEEE}, keywords = {}, pubstate = {published}, tppubtype = {article} } | |
Roschlaub, Robert; Glock, Clemens; Möst, Karin; Hümmer, Frank; Li, Qingyu; Auer, Stefan; Kruspe, Anna; Zhu, Xiao Xiang Implementierung einer KI-Infrastruktur zur automatisierten Erkennung von landesweiten Gebäudeveränderungen aus Luftbildern Journal Article ZFV - Zeitschrift für Geodasie, Geoinformation und Landmanagement, 2022. Abstract | Links | BibTeX | Tags: @article{Roschlaub2022b, title = {Implementierung einer KI-Infrastruktur zur automatisierten Erkennung von landesweiten Gebäudeveränderungen aus Luftbildern}, author = {Robert Roschlaub and Clemens Glock and Karin Möst and Frank Hümmer and Qingyu Li and Stefan Auer and Anna Kruspe and Xiao Xiang Zhu}, url = {https://elib.dlr.de/185326/}, year = {2022}, date = {2022-01-01}, journal = {ZFV - Zeitschrift für Geodasie, Geoinformation und Landmanagement}, publisher = {Wißner-Verlag}, abstract = {Aufbauend auf einer vorangegangenen Studie über den Einsatz von Künstlicher Intelligenz (KI) zur Detektion von Gebäudeveränderungen im amtlichen Liegenschaftskataster, wird in diesem Beitrag die Transformation von der Projektphase in einen Produktivbetrieb vorgestellt, es werden verschiedene KI-Architekturen angesprochen, miteinander verglichen und eine Auswahlentscheidung gegeben. Die erzielten Ergebnisse einer landesweiten Gebäudedetektion werden gemeinsam mit Untersuchungen zu den Trainingsläufen, zur Performance und den von einem Vermessungsamt validierten KI-Ergebnissen präsentiert.}, keywords = {}, pubstate = {published}, tppubtype = {article} } Aufbauend auf einer vorangegangenen Studie über den Einsatz von Künstlicher Intelligenz (KI) zur Detektion von Gebäudeveränderungen im amtlichen Liegenschaftskataster, wird in diesem Beitrag die Transformation von der Projektphase in einen Produktivbetrieb vorgestellt, es werden verschiedene KI-Architekturen angesprochen, miteinander verglichen und eine Auswahlentscheidung gegeben. Die erzielten Ergebnisse einer landesweiten Gebäudedetektion werden gemeinsam mit Untersuchungen zu den Trainingsläufen, zur Performance und den von einem Vermessungsamt validierten KI-Ergebnissen präsentiert. | |
Khan, Shfaqat A; Bamber, Jonathan L; Rignot, Eric; Helm, Veit; Aschwanden, Andy; Holland, David M; van den Broeke, Michiel; King, Michalea; Noël, Brice; Truffer, Martin; Humbert, Angelika; Colgan, William; Vijay, Saurabh; Munneke, Peter Kuipers Greenland Mass Trends From Airborne and Satellite Altimetry During 2011–2020 Journal Article Journal of Geophysical Research: Earth Surface, 127 , pp. e2021JF006505, 2022, ISSN: 2169-9011. @article{Khan2022, title = {Greenland Mass Trends From Airborne and Satellite Altimetry During 2011–2020}, author = {Shfaqat A Khan and Jonathan L Bamber and Eric Rignot and Veit Helm and Andy Aschwanden and David M Holland and Michiel van den Broeke and Michalea King and Brice Noël and Martin Truffer and Angelika Humbert and William Colgan and Saurabh Vijay and Peter Kuipers Munneke}, url = {https://onlinelibrary.wiley.com/doi/full/10.1029/2021JF006505 https://onlinelibrary.wiley.com/doi/abs/10.1029/2021JF006505 https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2021JF006505}, doi = {10.1029/2021JF006505}, issn = {2169-9011}, year = {2022}, date = {2022-01-01}, journal = {Journal of Geophysical Research: Earth Surface}, volume = {127}, pages = {e2021JF006505}, publisher = {John Wiley & Sons, Ltd}, keywords = {}, pubstate = {published}, tppubtype = {article} } | |
Hua, Yuansheng; Mou, Lichao; Jin, Pu; Zhu, Xiao Xiang MultiScene: A Large-Scale Dataset and Benchmark for Multiscene Recognition in Single Aerial Images Journal Article IEEE Transactions on Geoscience and Remote Sensing, 60 , 2022, ISSN: 15580644. Abstract | Links | BibTeX | Tags: @article{Hua2022, title = {MultiScene: A Large-Scale Dataset and Benchmark for Multiscene Recognition in Single Aerial Images}, author = {Yuansheng Hua and Lichao Mou and Pu Jin and Xiao Xiang Zhu}, doi = {10.1109/TGRS.2021.3110314}, issn = {15580644}, year = {2022}, date = {2022-01-01}, journal = {IEEE Transactions on Geoscience and Remote Sensing}, volume = {60}, publisher = {Institute of Electrical and Electronics Engineers Inc.}, abstract = {Aerial scene recognition is a fundamental research problem in interpreting high-resolution aerial imagery. Over the past few years, most studies focus on classifying an image into one scene category, while in real-world scenarios, it is more often that a single image contains multiple scenes. Therefore, in this article, we investigate a more practical yet underexplored task—multiscene recognition in single images. To this end, we create a large-scale dataset, called MultiScene, composed of 100 000 unconstrained high-resolution aerial images. Considering that manually labeling such images is extremely arduous, we resort to low-cost annotations from crowdsourcing platforms, e.g., OpenStreetMap (OSM). However, OSM data might suffer from incompleteness and incorrectness, which introduce noise into image labels. To address this issue, we visually inspect 14 000 images and correct their scene labels, yielding a subset of cleanly annotated images, named MultiScene-Clean. With it, we can develop and evaluate deep networks for multiscene recognition using clean data. Moreover, we provide crowdsourced annotations of all images for the purpose of studying network learning with noisy labels. We conduct experiments with extensive baseline models on both MultiScene-Clean and MultiScene to offer benchmarks for multiscene recognition in single images and learning from noisy labels for this task, respectively. To facilitate progress, we make our dataset and trained models available on https://gitlab.lrz.de/ai4eo/reasoning/multiscene.}, keywords = {}, pubstate = {published}, tppubtype = {article} } Aerial scene recognition is a fundamental research problem in interpreting high-resolution aerial imagery. Over the past few years, most studies focus on classifying an image into one scene category, while in real-world scenarios, it is more often that a single image contains multiple scenes. Therefore, in this article, we investigate a more practical yet underexplored task—multiscene recognition in single images. To this end, we create a large-scale dataset, called MultiScene, composed of 100 000 unconstrained high-resolution aerial images. Considering that manually labeling such images is extremely arduous, we resort to low-cost annotations from crowdsourcing platforms, e.g., OpenStreetMap (OSM). However, OSM data might suffer from incompleteness and incorrectness, which introduce noise into image labels. To address this issue, we visually inspect 14 000 images and correct their scene labels, yielding a subset of cleanly annotated images, named MultiScene-Clean. With it, we can develop and evaluate deep networks for multiscene recognition using clean data. Moreover, we provide crowdsourced annotations of all images for the purpose of studying network learning with noisy labels. We conduct experiments with extensive baseline models on both MultiScene-Clean and MultiScene to offer benchmarks for multiscene recognition in single images and learning from noisy labels for this task, respectively. To facilitate progress, we make our dataset and trained models available on https://gitlab.lrz.de/ai4eo/reasoning/multiscene. | |
Cummings, Sol; Kondmann, Lukas; Zhu, Xiaoxiang Siamese Attention U-Net for Multi-Class Change Detection. Inproceedings 2022. BibTeX | Tags: @inproceedings{Cummings2022, title = {Siamese Attention U-Net for Multi-Class Change Detection.}, author = {Sol Cummings and Lukas Kondmann and Xiaoxiang Zhu}, year = {2022}, date = {2022-01-01}, journal = {2022 IEEE International Geoscience and Remote Sensing Symposium IGARSS}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } | |
EPFL, Jonathan Sauder; Genzel, Martin; Berlin, Peter TU Jung Gradient-Based Learning of Discrete Structured Measurement Operators for Signal Recovery Miscellaneous 2022. Abstract | Links | BibTeX | Tags: @misc{JonathanSauder2022, title = {Gradient-Based Learning of Discrete Structured Measurement Operators for Signal Recovery}, author = {Jonathan Sauder EPFL and Martin Genzel and Peter TU Jung Berlin}, url = {https://arxiv.org/abs/2202.03391v1}, doi = {10.48550/arxiv.2202.03391}, year = {2022}, date = {2022-01-01}, abstract = {Countless signal processing applications include the reconstruction of signals from few indirect linear measurements. The design of effective measurement operators is typically constrained by the underlying hardware and physics, posing a challenging and often even discrete optimization task. While the potential of gradient-based learning via the unrolling of iterative recovery algorithms has been demonstrated, it has remained unclear how to leverage this technique when the set of admissible measurement operators is structured and discrete. We tackle this problem by combining unrolled optimization with Gumbel reparametrizations, which enable the computation of low-variance gradient estimates of categorical random variables. Our approach is formalized by GLODISMO (Gradient-based Learning of DIscrete Structured Measurement Operators). This novel method is easy-to-implement, computationally efficient, and extendable due to its compatibility with automatic differentiation. We empirically demonstrate the performance and flexibility of GLODISMO in several prototypical signal recovery applications, verifying that the learned measurement matrices outperform conventional designs based on randomization as well as discrete optimization baselines.}, keywords = {}, pubstate = {published}, tppubtype = {misc} } Countless signal processing applications include the reconstruction of signals from few indirect linear measurements. The design of effective measurement operators is typically constrained by the underlying hardware and physics, posing a challenging and often even discrete optimization task. While the potential of gradient-based learning via the unrolling of iterative recovery algorithms has been demonstrated, it has remained unclear how to leverage this technique when the set of admissible measurement operators is structured and discrete. We tackle this problem by combining unrolled optimization with Gumbel reparametrizations, which enable the computation of low-variance gradient estimates of categorical random variables. Our approach is formalized by GLODISMO (Gradient-based Learning of DIscrete Structured Measurement Operators). This novel method is easy-to-implement, computationally efficient, and extendable due to its compatibility with automatic differentiation. We empirically demonstrate the performance and flexibility of GLODISMO in several prototypical signal recovery applications, verifying that the learned measurement matrices outperform conventional designs based on randomization as well as discrete optimization baselines. | |
Lehmann, Fanny; Vishwakarma, Bramha Dutt; Bamber, Jonathan How well are we able to close the water budget at the global scale? Journal Article Hydrology and Earth System Sciences, 26 , pp. 35-54, 2022, ISSN: 1607-7938. Abstract | Links | BibTeX | Tags: @article{Lehmann2022b, title = {How well are we able to close the water budget at the global scale?}, author = {Fanny Lehmann and Bramha Dutt Vishwakarma and Jonathan Bamber}, url = {https://hess.copernicus.org/articles/26/35/2022/}, doi = {10.5194/hess-26-35-2022}, issn = {1607-7938}, year = {2022}, date = {2022-01-01}, journal = {Hydrology and Earth System Sciences}, volume = {26}, pages = {35-54}, publisher = {Copernicus GmbH}, abstract = { Abstract. The water budget equation describes the exchange of water between the land, ocean, and atmosphere. Being able to adequately close the water budget gives confidence in our ability to model and/or observe the spatio-temporal variations in the water cycle and its components. Due to advances in observation techniques, satellite sensors, and modelling, a number of data products are available that represent the components of water budget in both space and time. Despite these advances, closure of the water budget at the global scale has been elusive. In this study, we attempt to close the global water budget using precipitation, evapotranspiration, and runoff data at the catchment scale. The large number of recent state-of-the-art datasets provides a new evaluation of well-used datasets. These estimates are compared to terrestrial water storage (TWS) changes as measured by the Gravity Recovery And Climate Experiment (GRACE) satellite mission. We investigated 189 river basins covering more than 90 % of the continental land area. TWS changes derived from the water balance equation were compared against GRACE data using two metrics: the Nash–Sutcliffe efficiency (NSE) and the cyclostationary NSE. These metrics were used to assess the performance of more than 1600 combinations of the various datasets considered. We found a positive NSE and cyclostationary NSE in 99 % and 62 % of the basins examined respectively. This means that TWS changes reconstructed from the water balance equation were more accurate than the long-term (NSE) and monthly (cyclostationary NSE) mean of GRACE time series in the corresponding basins. By analysing different combinations of the datasets that make up the water balance, we identified data products that performed well in certain regions based on, for example, climatic zone. We identified that some of the good results were obtained due to the cancellation of errors in poor estimates of water budget components. Therefore, we used coefficients of variation to determine the relative quality of a data product, which helped us to identify bad combinations giving us good results. In general, water budget components from ERA5-Land and the Catchment Land Surface Model (CLSM) performed better than other products for most climatic zones. Conversely, the latest version of CLSM, v2.2, performed poorly for evapotranspiration in snow-dominated catchments compared, for example, with its predecessor and other datasets available. Thus, the nature of the catchment dynamics and balance between components affects the optimum combination of datasets. For regional studies, the combination of datasets that provides the most realistic TWS for a basin will depend on its climatic conditions and factors that cannot be determined a priori. We believe that the results of this study provide a road map for studying the water budget at catchment scale. },keywords = {}, pubstate = {published}, tppubtype = {article} } <p>Abstract. The water budget equation describes the exchange of water between the land, ocean, and atmosphere. Being able to adequately close the water budget gives confidence in our ability to model and/or observe the spatio-temporal variations in the water cycle and its components. Due to advances in observation techniques, satellite sensors, and modelling, a number of data products are available that represent the components of water budget in both space and time. Despite these advances, closure of the water budget at the global scale has been elusive. In this study, we attempt to close the global water budget using precipitation, evapotranspiration, and runoff data at the catchment scale. The large number of recent state-of-the-art datasets provides a new evaluation of well-used datasets. These estimates are compared to terrestrial water storage (TWS) changes as measured by the Gravity Recovery And Climate Experiment (GRACE) satellite mission. We investigated 189 river basins covering more than 90 % of the continental land area. TWS changes derived from the water balance equation were compared against GRACE data using two metrics: the Nash–Sutcliffe efficiency (NSE) and the cyclostationary NSE. These metrics were used to assess the performance of more than 1600 combinations of the various datasets considered. We found a positive NSE and cyclostationary NSE in 99 % and 62 % of the basins examined respectively. This means that TWS changes reconstructed from the water balance equation were more accurate than the long-term (NSE) and monthly (cyclostationary NSE) mean of GRACE time series in the corresponding basins. By analysing different combinations of the datasets that make up the water balance, we identified data products that performed well in certain regions based on, for example, climatic zone. We identified that some of the good results were obtained due to the cancellation of errors in poor estimates of water budget components. Therefore, we used coefficients of variation to determine the relative quality of a data product, which helped us to identify bad combinations giving us good results. In general, water budget components from ERA5-Land and the Catchment Land Surface Model (CLSM) performed better than other products for most climatic zones. Conversely, the latest version of CLSM, v2.2, performed poorly for evapotranspiration in snow-dominated catchments compared, for example, with its predecessor and other datasets available. Thus, the nature of the catchment dynamics and balance between components affects the optimum combination of datasets. For regional studies, the combination of datasets that provides the most realistic TWS for a basin will depend on its climatic conditions and factors that cannot be determined a priori. We believe that the results of this study provide a road map for studying the water budget at catchment scale.</p> | |
Saha, Sudipan; Zhao, Shan; Zhu, Xiao Xiang Multi-target domain adaptation for remote sensing classification using graph neural network Journal Article IEEE Geoscience and Remote Sensing Letters, 2022. BibTeX | Tags: @article{Saha2022b, title = {Multi-target domain adaptation for remote sensing classification using graph neural network}, author = {Sudipan Saha and Shan Zhao and Xiao Xiang Zhu}, year = {2022}, date = {2022-01-01}, journal = {IEEE Geoscience and Remote Sensing Letters}, publisher = {IEEE}, keywords = {}, pubstate = {published}, tppubtype = {article} } | |
Stadtler, Scarlet; Betancourt, Clara; Roscher, Ribana Explainable Machine Learning Reveals Capabilities, Redundancy, and Limitations of a Geospatial Air Quality Benchmark Dataset Journal Article Machine Learning and Knowledge Extraction, 4 , pp. 150-171, 2022, ISSN: 2504-4990. Abstract | Links | BibTeX | Tags: @article{Stadtler2022b, title = {Explainable Machine Learning Reveals Capabilities, Redundancy, and Limitations of a Geospatial Air Quality Benchmark Dataset}, author = {Scarlet Stadtler and Clara Betancourt and Ribana Roscher}, url = {https://www.mdpi.com/2504-4990/4/1/8}, doi = {10.3390/make4010008}, issn = {2504-4990}, year = {2022}, date = {2022-01-01}, journal = {Machine Learning and Knowledge Extraction}, volume = {4}, pages = {150-171}, publisher = {Multidisciplinary Digital Publishing Institute}, abstract = { Air quality is relevant to society because it poses environmental risks to humans and nature. We use explainable machine learning in air quality research by analyzing model predictions in relation to the underlying training data. The data originate from worldwide ozone observations, paired with geospatial data. We use two different architectures: a neural network and a random forest trained on various geospatial data to predict multi-year averages of the air pollutant ozone. To understand how both models function, we explain how they represent the training data and derive their predictions. By focusing on inaccurate predictions and explaining why these predictions fail, we can (i) identify underrepresented samples, (ii) flag unexpected inaccurate predictions, and (iii) point to training samples irrelevant for predictions on the test set. Based on the underrepresented samples, we suggest where to build new measurement stations. We also show which training samples do not substantially contribute to the model performance. This study demonstrates the application of explainable machine learning beyond simply explaining the trained model. },keywords = {}, pubstate = {published}, tppubtype = {article} } <p>Air quality is relevant to society because it poses environmental risks to humans and nature. We use explainable machine learning in air quality research by analyzing model predictions in relation to the underlying training data. The data originate from worldwide ozone observations, paired with geospatial data. We use two different architectures: a neural network and a random forest trained on various geospatial data to predict multi-year averages of the air pollutant ozone. To understand how both models function, we explain how they represent the training data and derive their predictions. By focusing on inaccurate predictions and explaining why these predictions fail, we can (i) identify underrepresented samples, (ii) flag unexpected inaccurate predictions, and (iii) point to training samples irrelevant for predictions on the test set. Based on the underrepresented samples, we suggest where to build new measurement stations. We also show which training samples do not substantially contribute to the model performance. This study demonstrates the application of explainable machine learning beyond simply explaining the trained model.</p> | |
Mou, Lichao; Saha, Sudipan; Hua, Yuansheng; Bovolo, Francesca; Bruzzone, Lorenzo; Zhu, Xiao Xiang Deep Reinforcement Learning for Band Selection in Hyperspectral Image Classification Journal Article IEEE Transactions on Geoscience and Remote Sensing, 60 , 2022, ISSN: 15580644. Abstract | Links | BibTeX | Tags: @article{Mou2022, title = {Deep Reinforcement Learning for Band Selection in Hyperspectral Image Classification}, author = {Lichao Mou and Sudipan Saha and Yuansheng Hua and Francesca Bovolo and Lorenzo Bruzzone and Xiao Xiang Zhu}, doi = {10.1109/TGRS.2021.3067096}, issn = {15580644}, year = {2022}, date = {2022-01-01}, journal = {IEEE Transactions on Geoscience and Remote Sensing}, volume = {60}, publisher = {Institute of Electrical and Electronics Engineers Inc.}, abstract = {Band selection refers to the process of choosing the most relevant bands in a hyperspectral image. By selecting a limited number of optimal bands, we aim at speeding up model training, improving accuracy, or both. It reduces redundancy among spectral bands while trying to preserve the original information of the image. By now, many efforts have been made to develop unsupervised band selection approaches, of which the majorities are heuristic algorithms devised by trial and error. In this article, we are interested in training an intelligent agent that, given a hyperspectral image, is capable of automatically learning policy to select an optimal band subset without any hand-engineered reasoning. To this end, we frame the problem of unsupervised band selection as a Markov decision process, propose an effective method to parameterize it, and finally solve the problem by deep reinforcement learning. Once the agent is trained, it learns a band-selection policy that guides the agent to sequentially select bands by fully exploiting the hyperspectral image and previously picked bands. Furthermore, we propose two different reward schemes for the environment simulation of deep reinforcement learning and compare them in experiments. This, to the best of our knowledge, is the first study that explores a deep reinforcement learning model for hyperspectral image analysis, thus opening a new door for future research and showcasing the great potential of deep reinforcement learning in remote sensing applications. Extensive experiments are carried out on four hyperspectral data sets, and experimental results demonstrate the effectiveness of the proposed method. The code is publicly available.}, keywords = {}, pubstate = {published}, tppubtype = {article} } Band selection refers to the process of choosing the most relevant bands in a hyperspectral image. By selecting a limited number of optimal bands, we aim at speeding up model training, improving accuracy, or both. It reduces redundancy among spectral bands while trying to preserve the original information of the image. By now, many efforts have been made to develop unsupervised band selection approaches, of which the majorities are heuristic algorithms devised by trial and error. In this article, we are interested in training an intelligent agent that, given a hyperspectral image, is capable of automatically learning policy to select an optimal band subset without any hand-engineered reasoning. To this end, we frame the problem of unsupervised band selection as a Markov decision process, propose an effective method to parameterize it, and finally solve the problem by deep reinforcement learning. Once the agent is trained, it learns a band-selection policy that guides the agent to sequentially select bands by fully exploiting the hyperspectral image and previously picked bands. Furthermore, we propose two different reward schemes for the environment simulation of deep reinforcement learning and compare them in experiments. This, to the best of our knowledge, is the first study that explores a deep reinforcement learning model for hyperspectral image analysis, thus opening a new door for future research and showcasing the great potential of deep reinforcement learning in remote sensing applications. Extensive experiments are carried out on four hyperspectral data sets, and experimental results demonstrate the effectiveness of the proposed method. The code is publicly available. | |
Saha, Sudipan; Ebel, Patrick; Zhu, Xiao Xiang Self-Supervised Multisensor Change Detection Journal Article IEEE Transactions on Geoscience and Remote Sensing, 60 , 2022, ISSN: 15580644. Abstract | Links | BibTeX | Tags: @article{Saha2022c, title = {Self-Supervised Multisensor Change Detection}, author = {Sudipan Saha and Patrick Ebel and Xiao Xiang Zhu}, doi = {10.1109/TGRS.2021.3109957}, issn = {15580644}, year = {2022}, date = {2022-01-01}, journal = {IEEE Transactions on Geoscience and Remote Sensing}, volume = {60}, publisher = {Institute of Electrical and Electronics Engineers Inc.}, abstract = {Most change detection (CD) methods assume that prechange and postchange images are acquired by the same sensor. However, in many real-life scenarios, e.g., natural disasters, it is more practical to use the latest available images before and after the occurrence of incidence, which may be acquired using different sensors. In particular, we are interested in the combination of the images acquired by optical and synthetic aperture radar (SAR) sensors. SAR images appear vastly different from the optical images even when capturing the same scene. Adding to this, CD methods are often constrained to use only target image-pair, no labeled data, and no additional unlabeled data. Such constraints limit the scope of traditional supervised machine learning and unsupervised generative approaches for multisensor CD. The recent rapid development of self-supervised learning methods has shown that some of them can even work with only few images. Motivated by this, in this work, we propose a method for multisensor CD using only the unlabeled target bitemporal images that are used for training a network in a self-supervised fashion by using deep clustering and contrastive learning. The proposed method is evaluated on four multimodal bitemporal scenes showing change, and the benefits of our self-supervised approach are demonstrated. Code is available at https://gitlab.lrz.de/ai4eo/cd/-/tree/main/sarOpticalMultisensorTgrs2021.}, keywords = {}, pubstate = {published}, tppubtype = {article} } Most change detection (CD) methods assume that prechange and postchange images are acquired by the same sensor. However, in many real-life scenarios, e.g., natural disasters, it is more practical to use the latest available images before and after the occurrence of incidence, which may be acquired using different sensors. In particular, we are interested in the combination of the images acquired by optical and synthetic aperture radar (SAR) sensors. SAR images appear vastly different from the optical images even when capturing the same scene. Adding to this, CD methods are often constrained to use only target image-pair, no labeled data, and no additional unlabeled data. Such constraints limit the scope of traditional supervised machine learning and unsupervised generative approaches for multisensor CD. The recent rapid development of self-supervised learning methods has shown that some of them can even work with only few images. Motivated by this, in this work, we propose a method for multisensor CD using only the unlabeled target bitemporal images that are used for training a network in a self-supervised fashion by using deep clustering and contrastive learning. The proposed method is evaluated on four multimodal bitemporal scenes showing change, and the benefits of our self-supervised approach are demonstrated. Code is available at https://gitlab.lrz.de/ai4eo/cd/-/tree/main/sarOpticalMultisensorTgrs2021. | |
Betancourt, C; Stomberg, T T; Edrich, A -K; Patnala, A; Schultz, M G; Roscher, R; Kowalski, J; Stadtler, S GMDD - Global, high-resolution mapping of tropospheric ozone – explainable machine learning and impact of uncertainties Journal Article Geosci. Model Dev., pp. 1-36, 2022. @article{Betancourt2022b, title = {GMDD - Global, high-resolution mapping of tropospheric ozone – explainable machine learning and impact of uncertainties}, author = {C Betancourt and T T Stomberg and A -K Edrich and A Patnala and M G Schultz and R Roscher and J Kowalski and S Stadtler}, url = {https://gmd.copernicus.org/preprints/gmd-2022-2/}, doi = {10.5194/gmd-2022-2}, year = {2022}, date = {2022-01-01}, journal = {Geosci. Model Dev.}, pages = {1-36}, keywords = {}, pubstate = {published}, tppubtype = {article} } | |
Song, Qian; Albrecht, Conrad M; Xiong, Zhitong; Zhu, Xiao Xiang Towards Global Forest Biomass estimators from tree height data Inproceedings 2022. BibTeX | Tags: @inproceedings{Song2022, title = {Towards Global Forest Biomass estimators from tree height data}, author = {Qian Song and Conrad M Albrecht and Zhitong Xiong and Xiao Xiang Zhu}, year = {2022}, date = {2022-01-01}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } | |
L., Wang Zhu Yuan Mou Q X Z From Easy to Hard: Learning Language-guided Curriculum for Visual Question Answering on Remote Sensing Data Journal Article IEEE Transactions on Geoscience and Remote Sensing, 2022, (in press). BibTeX | Tags: @article{Yuan2022, title = {From Easy to Hard: Learning Language-guided Curriculum for Visual Question Answering on Remote Sensing Data}, author = {Wang Zhu Q X Yuan Z. Mou L.}, year = {2022}, date = {2022-01-01}, journal = {IEEE Transactions on Geoscience and Remote Sensing}, note = {in press}, keywords = {}, pubstate = {published}, tppubtype = {article} } | |
Heidler, Konrad; Mou, Lichao; Loebel, Erik; Scheinert, Mirko; `e, Sébastien" Lef; Zhu, Xiao Xiang Deep Active Contour Models for Delineating Glacier Calving Fronts Inproceedings IGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium, 2022. BibTeX | Tags: @inproceedings{igarss2022_3131, title = {Deep Active Contour Models for Delineating Glacier Calving Fronts}, author = {Konrad Heidler and Lichao Mou and Erik Loebel and Mirko Scheinert and Sébastien" Lef{`e}vre and Xiao Xiang Zhu}, year = {2022}, date = {2022-01-01}, booktitle = {IGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } | |
Gütter, Jonas; Niebling, Julia; Zhu, Xiao Xiang Analysing the interactions between training dataset size, label noise and model performance in remote sensing data Inproceedings 2022 IEEE International Geoscience and Remote Sensing Symposium IGARSS, IEEE 2022. BibTeX | Tags: @inproceedings{gütter2022, title = {Analysing the interactions between training dataset size, label noise and model performance in remote sensing data}, author = {Jonas Gütter and Julia Niebling and Xiao Xiang Zhu}, year = {2022}, date = {2022-01-01}, booktitle = {2022 IEEE International Geoscience and Remote Sensing Symposium IGARSS}, organization = {IEEE}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } | |
Wang, Yi; Albrecht, Conrad M; Zhu, Xiao Xiang Self-supervised Vision Transformers for Joint SAR-optical Representation Learning Journal Article 2022. BibTeX | Tags: @article{wang2022self, title = {Self-supervised Vision Transformers for Joint SAR-optical Representation Learning}, author = {Yi Wang and Conrad M Albrecht and Xiao Xiang Zhu}, year = {2022}, date = {2022-01-01}, booktitle = {2022 IEEE International Geoscience and Remote Sensing Symposium}, keywords = {}, pubstate = {published}, tppubtype = {article} } |