Gencer Sumbul
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Gencer Sumbul has a total of 52 co-authors. Here are the top co-authors by number of shared publications:
- Begüm Demir
Berlin Institute for the Foundations of Learning and Data (BIFOLD)
19 shared publications - Devis Tuia
École Polytechnique Fédérale de Lausanne (EPFL)
4 shared publications Begum Demir
Technische Universität Berlin, Berlin, Germany
3 shared publications- Selim Aksoy
Department of Computer Engineering, Bilkent University, Ankara, Trkiye
3 shared publications - Barış Büyüktaş
Faculty of Electrical Engineering and Computer Science, Technische Universitt Berlin, Berlin, Germany
2 shared publications Jan Dirk Wegner
University of Zürich
2 shared publications
Latest Publications
- Open Access
- 12 November 2026
- Published by IOP Publishing in Environmental Research Letters
- Vol. 21 (21) , 214045
Towards high-resolution forecasting of tropical deforestation with deep learning
Reducing tropical deforestation effectively requires spatially explicit data on deforestation dynamics. While past deforestation is now detected and mapped with high accuracy, producing reliable future forecasts remains a challenge. We present a deep learning approach for high-resolution deforestation forecasting in the Brazilian Amazon (BA). We train a spatio-temporal Convolutional Long Short-Term Memory model on past deforestation maps and other covariates. Our model achieves an F 1 score of 0.47 when forecasting one year ahead, substantially improving over comparable prior work. For multi-year predictions, we obtain promising results within a 3 year horizon, while sensitivity to shifts in deforestation dynamics suggests limitations for longer-term forecasts. We generate forecasts for years 2016–2018 across the BA, and make them available through an interactive Google Earth Engine application. Our results demonstrate the potential for high-resolution, short- and medium-term forecasting that can support conservation efforts and help protect tropical forests.
- 12 November 2026
- Published by IOP Publishing in Environmental Research Letters
- Vol. 21 (21) , 214045
Open AccessTowards high-resolution forecasting of tropical deforestation with deep learning
Reducing tropical deforestation effectively requires spatially explicit data on deforestation dynamics. While past deforestation is now detected and mapped with high accuracy, producing reliable future forecasts remains a challenge. We present a deep learning approach for high-resolution deforestation forecasting in the Brazilian Amazon (BA). We train a spatio-temporal Convolutional Long Short-Term Memory model on past deforestation maps and other covariates. Our model achieves an F 1 score of 0.47 when forecasting one year ahead, substantially improving over comparable prior work. For multi-year predictions, we obtain promising results within a 3 year horizon, while sensitivity to shifts in deforestation dynamics suggests limitations for longer-term forecasts. We generate forecasts for years 2016–2018 across the BA, and make them available through an interactive Google Earth Engine application. Our results demonstrate the potential for high-resolution, short- and medium-term forecasting that can support conservation efforts and help protect tropical forests.
- Open Access
- 7 January 2026
- Published by Elsevier in ISPRS Journal of Photogrammetry and Remote Sensing
- Vol. 232, 578-593
Mapping land uses following tropical deforestation with location-aware deep learning
- 7 January 2026
- Published by Elsevier in ISPRS Journal of Photogrammetry and Remote Sensing
- Vol. 232, 578-593
Open AccessMapping land uses following tropical deforestation with location-aware deep learning
- Open Access
- 26 November 2025
- Published by Wiley in Methods in Ecology and Evolution
- Vol. 17 (1) , 188-206
MaskSDM with Shapley values to improve flexibility, robustness and explainability in species distribution modelling
Species distribution models (SDMs) play a vital role in biodiversity research, conservation planning and ecological niche modelling by predicting species distributions based on environmental conditions. The selection of predictors is crucial, strongly impacting both model accuracy and how well the predictions reflect ecological patterns. To ensure meaningful insights, input variables must be carefully chosen to match the study objectives and the ecological requirements of the target species. However, existing SDMs, including both traditional and deep learning‐based approaches, often lack key capabilities for variable selection: (i) flexibility to choose relevant predictors at inference without retraining; (ii) robustness to handle missing predictor values without compromising accuracy; and (iii) explainability to interpret and accurately quantify each predictor's contribution. To overcome these limitations, we build upon and extend MaskSDM, a novel deep learning‐based SDM that enables flexible predictor selection by employing a masked training strategy, in which input variables are randomly hidden during training to simulate missing or ignored predictors. This approach allows the model to make predictions with arbitrary subsets of input variables while remaining robust to missing data. It also provides a clearer understanding of how adding or removing a given predictor affects model performance and predictions. Furthermore, we introduce a new method for computing Shapley values with MaskSDM, enabling precise assessments of predictor contributions and improving upon traditional approximations. We conduct an extensive evaluation of MaskSDM on the global sPlotOpen dataset, modelling the distributions of 12,738 plant species. Our results show that MaskSDM outperforms imputation‐based methods and closely approximates models trained on specific subsets of variables, while also providing key local and global insights into predictor contributions through more accurate Shapley value estimation. These findings underscore MaskSDM's potential to increase the applicability and adoption of SDMs, laying the groundwork for developing foundation models in SDMs that can be readily applied to diverse ecological applications.
- 26 November 2025
- Published by Wiley in Methods in Ecology and Evolution
- Vol. 17 (1) , 188-206
Open AccessMaskSDM with Shapley values to improve flexibility, robustness and explainability in species distribution modelling
Species distribution models (SDMs) play a vital role in biodiversity research, conservation planning and ecological niche modelling by predicting species distributions based on environmental conditions. The selection of predictors is crucial, strongly impacting both model accuracy and how well the predictions reflect ecological patterns. To ensure meaningful insights, input variables must be carefully chosen to match the study objectives and the ecological requirements of the target species. However, existing SDMs, including both traditional and deep learning‐based approaches, often lack key capabilities for variable selection: (i) flexibility to choose relevant predictors at inference without retraining; (ii) robustness to handle missing predictor values without compromising accuracy; and (iii) explainability to interpret and accurately quantify each predictor's contribution. To overcome these limitations, we build upon and extend MaskSDM, a novel deep learning‐based SDM that enables flexible predictor selection by employing a masked training strategy, in which input variables are randomly hidden during training to simulate missing or ignored predictors. This approach allows the model to make predictions with arbitrary subsets of input variables while remaining robust to missing data. It also provides a clearer understanding of how adding or removing a given predictor affects model performance and predictions. Furthermore, we introduce a new method for computing Shapley values with MaskSDM, enabling precise assessments of predictor contributions and improving upon traditional approximations. We conduct an extensive evaluation of MaskSDM on the global sPlotOpen dataset, modelling the distributions of 12,738 plant species. Our results show that MaskSDM outperforms imputation‐based methods and closely approximates models trained on specific subsets of variables, while also providing key local and global insights into predictor contributions through more accurate Shapley value estimation. These findings underscore MaskSDM's potential to increase the applicability and adoption of SDMs, laying the groundwork for developing foundation models in SDMs that can be readily applied to diverse ecological applications.
- Open Access
- 28 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63 (01962892) , 1-11
A Multimodal Federated Learning Framework for Remote Sensing Image Classification
Federated learning (FL) enables the collaborative training of deep neural networks across decentralized data archives (i.e., clients) without sharing the local data of the clients. Most of the existing FL methods assume that the data distributed across all clients is associated with the same data modality. However, remote sensing (RS) images present in different clients can be associated with diverse data modalities. The joint use of the multi-modal RS data can significantly enhance classification performance. To effectively exploit decentralized and unshared multi-modal RS data, our paper introduces a novel multi-modal FL framework for RS image classification problems. The proposed framework comprises three modules: 1) multi-modal fusion (MF); 2) feature whitening (FW); and 3) mutual information maximization (MIM). The MF module employs iterative model averaging to facilitate learning without accessing multi-modal training data on clients. The FW module aims to address the limitations of training data heterogeneity by aligning data distributions across clients. The MIM module aims to model mutual information by maximizing the similarity between images from different modalities. For the experimental analyses, we focus our attention on multi-label classification and pixel-based classification tasks in RS. The results obtained using two benchmark archives show the effectiveness of the proposed framework when compared to state-of-the-art algorithms in the literature. The code of the proposed framework is publicly available at https://git.tu-berlin.de/rsim/multi-modal-FL.
- 28 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63 (01962892) , 1-11
Open AccessA Multimodal Federated Learning Framework for Remote Sensing Image Classification
Federated learning (FL) enables the collaborative training of deep neural networks across decentralized data archives (i.e., clients) without sharing the local data of the clients. Most of the existing FL methods assume that the data distributed across all clients is associated with the same data modality. However, remote sensing (RS) images present in different clients can be associated with diverse data modalities. The joint use of the multi-modal RS data can significantly enhance classification performance. To effectively exploit decentralized and unshared multi-modal RS data, our paper introduces a novel multi-modal FL framework for RS image classification problems. The proposed framework comprises three modules: 1) multi-modal fusion (MF); 2) feature whitening (FW); and 3) mutual information maximization (MIM). The MF module employs iterative model averaging to facilitate learning without accessing multi-modal training data on clients. The FW module aims to address the limitations of training data heterogeneity by aligning data distributions across clients. The MIM module aims to model mutual information by maximizing the similarity between images from different modalities. For the experimental analyses, we focus our attention on multi-label classification and pixel-based classification tasks in RS. The results obtained using two benchmark archives show the effectiveness of the proposed framework when compared to state-of-the-art algorithms in the literature. The code of the proposed framework is publicly available at https://git.tu-berlin.de/rsim/multi-modal-FL.
- Open Access
Redundancy-Aware Pretraining of Vision-Language Foundation Models in Remote Sensing
The development of foundation models through pretraining of vision-language models (VLMs) has recently attracted great attention in remote sensing (RS). VLM pretraining aims to learn image and language alignments from a large number of image-text pairs. Each pretraining image is often associated with multiple captions containing redundant information due to repeated or semantically similar phrases, resulting in increased pretraining and inference time. To overcome this, we introduce a weighted feature aggregation (WFA) strategy for VLM pretraining in RS. Our strategy aims to extract and exploit complementary information from multiple captions per image while reducing redundancies through feature aggregation with importance weighting. To calculate adaptive importance weights for different captions of each image, we propose two techniques: (i) non-parametric uniqueness and (ii) learning-based attention. In the first technique, importance weights are calculated based on the bilingual evaluation understudy (BLEU) scores of the captions to emphasize unique sentences and reduce the influence of repetitive ones. In the second technique, importance weights are learned through an attention mechanism instead of relying on hand-crafted features. The effectiveness of the proposed WFA strategy with the two techniques is analyzed in terms of downstream performance on text-to-image retrieval in RS. Experimental results show that the proposed strategy enables efficient and effective pretraining of VLMs in RS. Based on the experimental analysis, we derive guidelines for selecting appropriate techniques depending on downstream task requirements and resource constraints. The code of this work is publicly available at https://git.tu-berlin.de/rsim/redundacy-aware-rs-vlm.
Open AccessRedundancy-Aware Pretraining of Vision-Language Foundation Models in Remote Sensing
The development of foundation models through pretraining of vision-language models (VLMs) has recently attracted great attention in remote sensing (RS). VLM pretraining aims to learn image and language alignments from a large number of image-text pairs. Each pretraining image is often associated with multiple captions containing redundant information due to repeated or semantically similar phrases, resulting in increased pretraining and inference time. To overcome this, we introduce a weighted feature aggregation (WFA) strategy for VLM pretraining in RS. Our strategy aims to extract and exploit complementary information from multiple captions per image while reducing redundancies through feature aggregation with importance weighting. To calculate adaptive importance weights for different captions of each image, we propose two techniques: (i) non-parametric uniqueness and (ii) learning-based attention. In the first technique, importance weights are calculated based on the bilingual evaluation understudy (BLEU) scores of the captions to emphasize unique sentences and reduce the influence of repetitive ones. In the second technique, importance weights are learned through an attention mechanism instead of relying on hand-crafted features. The effectiveness of the proposed WFA strategy with the two techniques is analyzed in terms of downstream performance on text-to-image retrieval in RS. Experimental results show that the proposed strategy enables efficient and effective pretraining of VLMs in RS. Based on the experimental analysis, we derive guidelines for selecting appropriate techniques depending on downstream task requirements and resource constraints. The code of this work is publicly available at https://git.tu-berlin.de/rsim/redundacy-aware-rs-vlm.
- 12 May 2025
- Published by Springer Nature
- p. 188-197
MaskSDM: Adaptive Species Distribution Modeling Through Data Masking
- 12 May 2025
- Published by Springer Nature
- p. 188-197
MaskSDM: Adaptive Species Distribution Modeling Through Data Masking
- 13 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63 (01962892) , 1-14
Exploring Masked Autoencoders for Sensor-Agnostic Image Retrieval in Remote Sensing
Self-supervised learning through masked autoencoders (MAEs) has recently attracted great attention for remote sensing (RS) image representation learning, and thus embodies a significant potential for content-based image retrieval (CBIR) from ever-growing RS image archives. However, the existing MAE based CBIR studies in RS assume that the considered RS images are acquired by a single image sensor, and thus are only suitable for uni-modal CBIR problems. The effectiveness of MAEs for cross-sensor CBIR, which aims to search semantically similar images across different image modalities, has not been explored yet. In this paper, we take the first step to explore the effectiveness of MAEs for sensor-agnostic CBIR in RS. To this end, we present a systematic overview on the possible adaptations of the vanilla MAE to exploit masked image modeling on multi-sensor RS image archives (denoted as cross-sensor masked autoencoders [CSMAEs]) in the context of CBIR. Based on different adjustments applied to the vanilla MAE, we introduce different CSMAE models. We also provide an extensive experimental analysis of these CSMAE models. We finally derive a guideline to exploit masked image modeling for uni-modal and cross-modal CBIR problems in RS. The code of this work is publicly available at https://github.com/jakhac/CSMAE.
- 13 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 63 (01962892) , 1-14
Exploring Masked Autoencoders for Sensor-Agnostic Image Retrieval in Remote Sensing
Self-supervised learning through masked autoencoders (MAEs) has recently attracted great attention for remote sensing (RS) image representation learning, and thus embodies a significant potential for content-based image retrieval (CBIR) from ever-growing RS image archives. However, the existing MAE based CBIR studies in RS assume that the considered RS images are acquired by a single image sensor, and thus are only suitable for uni-modal CBIR problems. The effectiveness of MAEs for cross-sensor CBIR, which aims to search semantically similar images across different image modalities, has not been explored yet. In this paper, we take the first step to explore the effectiveness of MAEs for sensor-agnostic CBIR in RS. To this end, we present a systematic overview on the possible adaptations of the vanilla MAE to exploit masked image modeling on multi-sensor RS image archives (denoted as cross-sensor masked autoencoders [CSMAEs]) in the context of CBIR. Based on different adjustments applied to the vanilla MAE, we introduce different CSMAE models. We also provide an extensive experimental analysis of these CSMAE models. We finally derive a guideline to exploit masked image modeling for uni-modal and cross-modal CBIR problems in RS. The code of this work is publicly available at https://github.com/jakhac/CSMAE.
- 5 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 62 (01962892) , 1-11
Annotation Cost-Efficient Active Learning for Deep Metric Learning-Driven Remote Sensing Image Retrieval
Deep metric learning (DML) has shown to be effective for content-based image retrieval (CBIR) in remote sensing (RS). Most of DML methods for CBIR rely on a high number of annotated images to accurately learn model parameters of deep neural networks (DNNs). However, gathering such data is time-consuming and costly. To address this, we propose an annotation cost-efficient active learning (ANNEAL) method tailored to DML-driven CBIR in RS. ANNEAL aims to create a small but informative training set made up of similar and dissimilar image pairs to be utilized for accurately learning a metric space. The informativeness of image pairs is evaluated by combining uncertainty and diversity criteria. To assess the uncertainty of image pairs, we introduce two algorithms: 1) metric-guided uncertainty estimation (MGUE); and 2) binary classifier guided uncertainty estimation (BCGUE). MGUE algorithm automatically estimates a threshold value that acts as a boundary between similar and dissimilar image pairs based on the distances in the metric space. The closer the similarity between image pairs is to the estimated threshold value the higher their uncertainty. BCGUE algorithm estimates the uncertainty of the image pairs based on the confidence of the classifier in assigning correct similarity labels. The diversity criterion is assessed through a clustering-based strategy. ANNEAL combines either MGUE or BCGUE algorithm with the clustering-based strategy to select the most informative image pairs, which are then labelled by expert annotators as similar or dissimilar. This way of annotating images significantly reduces the annotation cost compared to annotating images with land-use land-cover class labels. Experimental results on two RS benchmark datasets demonstrate the effectiveness of our method. The code of this work is publicly available at https://git.tu-berlin.de/rsim/anneal_tgrs.
- 5 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 62 (01962892) , 1-11
Annotation Cost-Efficient Active Learning for Deep Metric Learning-Driven Remote Sensing Image Retrieval
Deep metric learning (DML) has shown to be effective for content-based image retrieval (CBIR) in remote sensing (RS). Most of DML methods for CBIR rely on a high number of annotated images to accurately learn model parameters of deep neural networks (DNNs). However, gathering such data is time-consuming and costly. To address this, we propose an annotation cost-efficient active learning (ANNEAL) method tailored to DML-driven CBIR in RS. ANNEAL aims to create a small but informative training set made up of similar and dissimilar image pairs to be utilized for accurately learning a metric space. The informativeness of image pairs is evaluated by combining uncertainty and diversity criteria. To assess the uncertainty of image pairs, we introduce two algorithms: 1) metric-guided uncertainty estimation (MGUE); and 2) binary classifier guided uncertainty estimation (BCGUE). MGUE algorithm automatically estimates a threshold value that acts as a boundary between similar and dissimilar image pairs based on the distances in the metric space. The closer the similarity between image pairs is to the estimated threshold value the higher their uncertainty. BCGUE algorithm estimates the uncertainty of the image pairs based on the confidence of the classifier in assigning correct similarity labels. The diversity criterion is assessed through a clustering-based strategy. ANNEAL combines either MGUE or BCGUE algorithm with the clustering-based strategy to select the most informative image pairs, which are then labelled by expert annotators as similar or dissimilar. This way of annotating images significantly reduces the annotation cost compared to annotating images with land-use land-cover class labels. Experimental results on two RS benchmark datasets demonstrate the effectiveness of our method. The code of this work is publicly available at https://git.tu-berlin.de/rsim/anneal_tgrs.
- 12 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Magazine
- Vol. 12 (3) , 64-80
Federated Learning Across Decentralized and Unshared Archives for Remote Sensing Image Classification: A review
Federated learning (FL) enables the collaboration of multiple deep learning (DL) models to learn from decentralized data archives (i.e., clients) without accessing data on the clients. Although FL offers ample opportunities in knowledge discovery from distributed image archives, it is seldom considered in remote sensing (RS). In this article, for the first time in RS, we present a comparative study of state-of-the-art FL algorithms for RS image classification problems. To this end, we initially provide a systematic review of the FL algorithms presented in the computer vision (CV) and machine learning (ML) communities. Then, we select several state-of-the-art FL algorithms based on their effectiveness with respect to training data heterogeneity across clients [known as not independent and identically distributed (non-IID) data]. After presenting an extensive overview of the selected algorithms, we conduct a theoretical comparison of them based on their 1) local training complexity, 2) aggregation complexity, 3) learning efficiency, 4) communication cost, and 5) scalability in terms of number of clients. After the theoretical comparison, experimental analyses are presented to compare the algorithms under different decentralization scenarios. For the experimental analyses, we focus our attention on multilabel image classification (MLC) problems in RS. Based on our comprehensive analyses, we finally derive a guideline for selecting suitable FL algorithms in RS. The code of this work is publicly available at https://git.tu-berlin.de/rsim/FL-RS .
- 12 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Magazine
- Vol. 12 (3) , 64-80
Federated Learning Across Decentralized and Unshared Archives for Remote Sensing Image Classification: A review
Federated learning (FL) enables the collaboration of multiple deep learning (DL) models to learn from decentralized data archives (i.e., clients) without accessing data on the clients. Although FL offers ample opportunities in knowledge discovery from distributed image archives, it is seldom considered in remote sensing (RS). In this article, for the first time in RS, we present a comparative study of state-of-the-art FL algorithms for RS image classification problems. To this end, we initially provide a systematic review of the FL algorithms presented in the computer vision (CV) and machine learning (ML) communities. Then, we select several state-of-the-art FL algorithms based on their effectiveness with respect to training data heterogeneity across clients [known as not independent and identically distributed (non-IID) data]. After presenting an extensive overview of the selected algorithms, we conduct a theoretical comparison of them based on their 1) local training complexity, 2) aggregation complexity, 3) learning efficiency, 4) communication cost, and 5) scalability in terms of number of clients. After the theoretical comparison, experimental analyses are presented to compare the algorithms under different decentralization scenarios. For the experimental analyses, we focus our attention on multilabel image classification (MLC) problems in RS. Based on our comprehensive analyses, we finally derive a guideline for selecting suitable FL algorithms in RS. The code of this work is publicly available at https://git.tu-berlin.de/rsim/FL-RS .
- Open Access
- 9 December 2023
- Published by Springer Nature
- p. 21-42
Analyzing and Improving the Quality and Fitness for Purpose of OpenStreetMap as Labels in Remote Sensing Applications
OpenStreetMap (OSM) is a well-known example of volunteered geographic information. It has evolved to one of the most used geographic databases. As data quality of OSM is heterogeneous both in space and across different thematic domains, data quality assessment is of high importance for potential users of OSM data. As use cases differ with respect to their requirements, it is not data quality per se that is of interest for the user but fitness for purpose. We investigate the fitness for purpose of OSM to derive land-use and land-cover labels for remote sensing-based classification models. Therefore, we evaluated OSM land-use and land-cover information by two approaches: (1) assessment of OSM fitness for purpose for samples in relation to intrinsic data quality indicators at the scale of individual OSM objects and (2) assessment of OSM-derived multi-labels at the scale of remote sensing patches ( $$1.22 \times 1.22$$ km) in combination with deep learning approaches. The first approach was applied to 1000 randomly selected relevant OSM objects. The quality score for each OSM object in the samples was combined with a large set of intrinsic quality indicators (such as the experience of the mapper, the number of mappers in a region, and the number of edits made to the object) and auxiliary information about the location of the OSM object (such as the continent or the ecozone). Intrinsic indicators were derived by a newly developed tool based on the OSHDB (OpenStreetMap History DataBase). Afterward, supervised and unsupervised shallow learning approaches were used to identify relationships between the indicators and the quality score. Overall, investigated OSM land-use objects were of high quality: both geometry and attribute information were mostly accurate. However, areas without any land-use information in OSM existed even in well-mapped areas such as Germany. The regression analysis at the level of the individual OSM objects revealed associations between intrinsic indicators, but also a strong variability. Even if more experienced mappers tend to produce higher quality and objects which underwent multiple edits tend to be of higher quality, an inexperienced mapper might map a perfect land-use polygon. This result indicates that it is hard to predict data quality of individual land-use objects purely on intrinsic data quality indicators. The second approach employed a label-noise robust deep learning method on remote sensing data with OSM labels. As the quality of the OSM labels was manually assessed beforehand, it was possible to control the amount of noise in the dataset during the experiment. The addition of artificial noise allowed for an even more fine-grained analysis on the effect of noise on prediction quality. The noise-tolerant deep learning method was capable to identify correct multi-labels even for situations with significant levels of noise added. The method was also used to identify areas where input labels were likely wrong. Thereby, it is possible to provide feedback to the OSM community as areas of concern can be flagged.- 9 December 2023
- Published by Springer Nature
- p. 21-42
Open AccessAnalyzing and Improving the Quality and Fitness for Purpose of OpenStreetMap as Labels in Remote Sensing Applications
OpenStreetMap (OSM) is a well-known example of volunteered geographic information. It has evolved to one of the most used geographic databases. As data quality of OSM is heterogeneous both in space and across different thematic domains, data quality assessment is of high importance for potential users of OSM data. As use cases differ with respect to their requirements, it is not data quality per se that is of interest for the user but fitness for purpose. We investigate the fitness for purpose of OSM to derive land-use and land-cover labels for remote sensing-based classification models. Therefore, we evaluated OSM land-use and land-cover information by two approaches: (1) assessment of OSM fitness for purpose for samples in relation to intrinsic data quality indicators at the scale of individual OSM objects and (2) assessment of OSM-derived multi-labels at the scale of remote sensing patches ( $$1.22 \times 1.22$$ km) in combination with deep learning approaches. The first approach was applied to 1000 randomly selected relevant OSM objects. The quality score for each OSM object in the samples was combined with a large set of intrinsic quality indicators (such as the experience of the mapper, the number of mappers in a region, and the number of edits made to the object) and auxiliary information about the location of the OSM object (such as the continent or the ecozone). Intrinsic indicators were derived by a newly developed tool based on the OSHDB (OpenStreetMap History DataBase). Afterward, supervised and unsupervised shallow learning approaches were used to identify relationships between the indicators and the quality score. Overall, investigated OSM land-use objects were of high quality: both geometry and attribute information were mostly accurate. However, areas without any land-use information in OSM existed even in well-mapped areas such as Germany. The regression analysis at the level of the individual OSM objects revealed associations between intrinsic indicators, but also a strong variability. Even if more experienced mappers tend to produce higher quality and objects which underwent multiple edits tend to be of higher quality, an inexperienced mapper might map a perfect land-use polygon. This result indicates that it is hard to predict data quality of individual land-use objects purely on intrinsic data quality indicators. The second approach employed a label-noise robust deep learning method on remote sensing data with OSM labels. As the quality of the OSM labels was manually assessed beforehand, it was possible to control the amount of noise in the dataset during the experiment. The addition of artificial noise allowed for an even more fine-grained analysis on the effect of noise on prediction quality. The noise-tolerant deep learning method was capable to identify correct multi-labels even for situations with significant levels of noise added. The method was also used to identify areas where input labels were likely wrong. Thereby, it is possible to provide feedback to the OSM community as areas of concern can be flagged.