Automatic Extraction and Filtering of OpenStreetMap Data to Generate Training Datasets for Land Use Land Cover Classification

Fonte, Cidália C.; Patriarca, Joaquim; Jesus, Ismael; Duarte, Diogo

doi:10.3390/rs12203428

Cited by 14 publications

(7 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Disagreements between the auxiliary datasets resulted in excluding a few areas from the automatic training, potentially impacting on classification. Alternative generation of training data can be considered to improve their quality, such as refining the filtering criteria used and adding more auxiliary data [58]. Other reasons limiting classification accuracy may include the length of the time series (one year), classification features, classification method, and so forth.…”

Section: Multi-stage Approachmentioning

confidence: 99%

Spatially Stratified and Multi-Stage Approach for National Land Cover Mapping Based on Sentinel-2 Data and Expert Knowledge

et al. 2022

View full text Add to dashboard Cite

Portugal is building a land cover monitoring system to deliver land cover products annually for its mainland territory. This paper presents the methodology developed to produce a prototype relative to 2018 as the first land cover map of the future annual map series (COSsim). A total of thirteen land cover classes are represented, including the most important tree species in Portugal. The mapping approach developed includes two levels of spatial stratification based on landscape dynamics. Strata are analysed independently at the higher level, while nested sublevels can share data and procedures. Multiple stages of analysis are implemented in which subsequent stages improve the outputs of precedent stages. The goal is to adjust mapping to the local landscape and tackle specific problems or divide complex mapping tasks in several parts. Supervised classification of Sentinel-2 time series and post-classification analysis with expert knowledge were performed throughout four stages. The overall accuracy of the map is estimated at 81.3% (±2.1) at the 95% confidence level. Higher thematic accuracy was achieved in southern Portugal, and expert knowledge significantly improved the quality of the map.

show abstract

Section: Multi-stage Approachmentioning

confidence: 99%

Spatially Stratified and Multi-Stage Approach for National Land Cover Mapping Based on Sentinel-2 Data and Expert Knowledge

et al. 2022

View full text Add to dashboard Cite

show abstract

“…In the case of large-scale supervised classification, both the quantity and quality of samples are important (Foody and Arora, 1997). ISA is a highly variable object, and its attributes in the Sentinel-2 multispectral images are related to materials, viewing angles, and atmospheric conditions, while its response to the Sentinel-1 SAR instrument depends on dielectric properties, geometry, and surface roughness.…”

Section: Sample Collectionmentioning

confidence: 99%

“…From the perspective of the global ISA mapping methods, supervised classification has been widely employed (Table 1). The quality of the training samples is the major factor affecting the classification results (Foody, 2009). Visual interpretation and automatic extraction from the existing datasets are two common ways to generate training samples.…”

Section: Introductionmentioning

confidence: 99%

Mapping 10 m global impervious surface area (GISA-10m) using multi-source geospatial data

Huang

Yang

Wang

et al. 2022

Earth Syst. Sci. Data

View full text Add to dashboard Cite

Abstract. Artificial impervious surface area (ISA) documents the human footprint. Accurate, timely, and detailed ISA datasets are therefore essential for global climate change studies and urban planning. However, due to the lack of sufficient training samples and operational mapping methods, global ISA datasets at a 10 m resolution are still lacking. To this end, we proposed a global ISA mapping method leveraging multi-source geospatial data. Based on the existing satellite-derived ISA maps and crowdsourced OpenStreetMap (OSM) data, 58 million training samples were extracted via a series of temporal, spatial, spectral, and geometric rules. We then produced a 10 m resolution global ISA dataset (GISA-10m) from over 2.7 million Sentinel optical and radar images on the Google Earth Engine platform. Based on test samples that are independent of the training set, GISA-10m achieves an overall accuracy of greater than 86 %. In addition, the GISA-10m dataset was comprehensively compared with the existing global ISA datasets, and the superiority of GISA-10m was confirmed. The global road area was further investigated, courtesy of this 10 m dataset. It was found that China and the US have the largest areas of ISA and road. The global rural ISA was found to be 2.2 times that of urban while the rural road area was found to be 1.5 times larger than that of the urban regions. The global road area accounts for 14.2 % of the global ISA, 57.9 % of which is located in the top 10 countries. Generally speaking, the produced GISA-10m dataset and the proposed sampling and mapping method are able to achieve rapid and efficient global mapping, and have the potential for detecting other land covers. It is also shown that global ISA mapping can be improved by incorporating OSM data. The GISA-10m dataset could be used as a fundamental parameter for Earth system science, and will provide valuable support for urban planning and water cycle study. The GISA-10m can be freely downloaded from https://doi.org/10.5281/zenodo.5791855 (Huang et al., 2021a).

show abstract

“…A natural language processing (NLP)-based approach is used to classify point-of-interest (POI), land use, and roads data extracted from Baidu Maps that can infer building types 49 . Similar methods are used for correcting OSM building annotations 50 , street labels predictions 51 , autonomous robot navigation 52 , 3D building models 53 , and land cover classification 54 , 55 . However, these approaches either use region-specific or proprietary datasets that are hard to obtain for applying the models in different places.…”

Section: Related Workmentioning

confidence: 99%

Predicting building types using OpenStreetMap

Atwal

Anderson

Pfoser

et al. 2022

Sci Rep

View full text Add to dashboard Cite

Having accurate building information is paramount for a plethora of applications, including humanitarian efforts, city planning, scientific studies, and navigation systems. While volunteered geographic information from sources such as OpenStreetMap (OSM) has good building geometry coverage, descriptive attributes such as the type of a building are sparse. To fill this gap, this study proposes a supervised learning-based approach to provide meaningful, semantic information for OSM data without manual intervention. We present a basic demonstration of our approach that classifies buildings into either residential or non-residential types for three study areas: Fairfax County in Virginia (VA), Mecklenburg County in North Carolina (NC), and the City of Boulder in Colorado (CO). The model leverages (i) available OSM tags capturing non-spatial attributes, (ii) geometric and topological properties of the building footprints including adjacent types of roads, proximity to parking lots, and building size. The model is trained and tested using ground truth data available for the three study areas. The results show that our approach achieves high accuracy in predicting building types for the selected areas. Additionally, a trained model is transferable with high accuracy to other regions where ground truth data is unavailable. The OSM and data science community are invited to build upon our approach to further enrich the volunteered geographic information in an automated manner.

show abstract

Automatic Extraction and Filtering of OpenStreetMap Data to Generate Training Datasets for Land Use Land Cover Classification

Cited by 14 publications

References 39 publications

Spatially Stratified and Multi-Stage Approach for National Land Cover Mapping Based on Sentinel-2 Data and Expert Knowledge

Spatially Stratified and Multi-Stage Approach for National Land Cover Mapping Based on Sentinel-2 Data and Expert Knowledge

Mapping 10 m global impervious surface area (GISA-10m) using multi-source geospatial data

Predicting building types using OpenStreetMap

Contact Info

Product

Resources

About