이 데이터 세트는 2000년부터 2022년까지 30m 공간 해상도로 초지(경작 및 자연/반자연)의 전 세계 연간 우세 클래스 지도를 제공합니다.
Land & Carbon Lab Global Pasture Watch 이니셔티브에서 제작한 매핑된 초지 범위에는 건조하거나 습한 저지대 식물이 30% 이상 포함되고, 풀과 초본 (3미터 미만)이 우세하며 다음을 포함하는 모든 토지 피복 유형이 포함됩니다.
최대 50% 의 수관 (5미터 초과)
기타 목본 식물 (관목 및 개방형 관목지)의 최대 70%
농경지 및 기타 식생의 모자이크 경관에서 활성 농경지 피복이 최대 50% 입니다.
초지 면적은 두 가지 클래스로 분류됩니다.
- 경작된 초지: 풀과 기타 사료 식물이 의도적으로 심어지고 관리되는 지역과 가축의 지향성 방목과 같은 특정 인간 지향적 용도로 적극적이고 집중적인 관리가 명확하게 이루어지는 토착 초지 유형 식물 지역.
- 자연/반자연 초지: 과거에 다양한 수준의 인간 활동을 경험한 지역뿐만 아니라 스텝, 툰드라와 같이 비교적 방해받지 않은 토착 초지/단신 식물로, 역사적 토지 이용 및 자연적 과정으로 인해 토착종과 도입종이 혼합되어 있을 수 있습니다.
일반적으로 다양한 식물의 자연스러운 패턴과 지형 전체에 걸쳐 명확하게 정렬된 수문학적 관계를 보여줍니다.
구현된 방법론에서는 GLAD Landsat ARD-2 이미지
(구름이 없는 격월 집계로 처리됨, Consoli et al, 2024 참고)를 기후, 지형, 근접성 공변량, 시공간 머신러닝 (클래스별 랜덤 포레스트) 및 230만 개 이상의 참조 샘플 (매우 높은 해상도 이미지에서 시각적으로 해석됨)과 함께 고려했습니다. 5겹 공간 교차 검증과 균형 잡힌 정밀도 및 재현율 값을 기반으로 하는 맞춤 확률 기준값을 사용하여 우세한 클래스 지도를 도출했습니다. 경작지 및 자연/반자연 초지 확률 기준값은 각각 0.32와 0.42입니다.
제한사항: 아프리카 남동부 (짐바브웨 및 모잠비크)와 오스트레일리아 동부 (멀가 생태 지역의 관목지 및 삼림)에서 초지 면적이 부분적으로 과소 예측됩니다. 북아프리카, 아라비아 반도, 서오스트레일리아, 뉴질랜드, 볼리비아 중부, 마토그로소주 (브라질) 일부 지역에서 농경지가 초지로 잘못 분류되었습니다. Landsat 7 SLC 오류로 인해 특히 2012년에 필지 수준에서 초지 확률의 규칙적인 스트라이프가 표시됩니다. 더 낮은 해상도 레이어 (접근성 지도 및 MODIS 제품)를 사용하면 우루과이, 아르헨티나 남서부, 앙골라 남부, 아프리카 사헬 지역에 곡선형 거시적 오류 (3차 스플라인 기반 다운스케일링 전략으로 인해)가 발생했습니다. 사용자는 초기 예측 단계에서 지도를 적절하게 사용하기 위해 제한사항과 알려진 문제를 인지하고 이를 신중하게 고려해야 합니다. GPW는 Geo-Wiki 플랫폼을 통해 체계적인 의견을 수집하고, 현재 버전을 검증하고, 향후 버전의 데이터 세트를 개선하기 위해 적극적으로 노력하고 있습니다.
Parente, L., Sloat, L., Mesquita, V., et al. (2024)
Global Pasture Watch - 30m 공간 해상도의 연간 초지 등급 및 범위 지도 (2000~2022년) (버전 v1) [데이터 세트]. Zenodo
doi:https://doi.org/10.5281/zenodo.13890401
이 데이터 세트는 2000년부터 2022년까지 30m 공간 해상도로 초지 (경작 및 자연/반자연)의 전 세계 연간 우세 클래스 지도를 제공합니다. Land & Carbon Lab Global Pasture Watch 이니셔티브에서 제작한 매핑된 초지 범위에는 건조하거나 습한 저지대 초지가 30% 이상 포함된 모든 토지 피복 유형이 포함됩니다.
[null,null,[],[[["\u003cp\u003eThis dataset provides annual maps of global grassland types (cultivated and natural/semi-natural) at 30-meter resolution from 2000 to 2022.\u003c/p\u003e\n"],["\u003cp\u003eDeveloped by the Land & Carbon Lab Global Pasture Watch initiative, it identifies areas with at least 30% low vegetation dominated by grasses and forbs, with specific tree and shrubland cover limitations.\u003c/p\u003e\n"],["\u003cp\u003eThe mapping methodology uses Landsat imagery, environmental covariates, and machine learning, validated with over 2.3 million reference samples.\u003c/p\u003e\n"],["\u003cp\u003eKnown limitations include potential under-prediction in certain regions and misclassification of cropland as grassland in others.\u003c/p\u003e\n"],["\u003cp\u003eThe dataset is available under a CC-BY-4.0 license and users are encouraged to consider the limitations for appropriate application.\u003c/p\u003e\n"]]],["The dataset offers annual global grassland maps from 2000-2022, at 30-meter resolution, classifying grasslands into cultivated and natural/semi-natural types. Created by the Land & Carbon Lab Global Pasture Watch, the maps identify areas with at least 30% low vegetation, with limitations on tree cover, woody vegetation, and cropland. The data utilizes GLAD Landsat ARD-2 images, machine learning, and over 2.3 million reference samples. Users can access the data through Google Earth Engine using the provided code and should be aware of the documented limitations.\n"],null,["# GPW Annual Dominant Class of Grasslands v1\n\ninfo\n\n\nThis dataset is part of a Publisher Catalog, and not managed by Google Earth Engine.\n\nContact [Land \\& Carbon Lab](https://landcarbonlab.org/subscribe)\n\nfor bugs or [view more datasets](https://developers.google.com/earth-engine/datasets/publisher/global-pasture-watch)\nfrom the Global Pasture Watch Catalog. [Learn more about Publisher datasets](/earth-engine/datasets/publisher). \n[](https://landcarbonlab.org/data/global-grassland-and-livestock-monitoring) \n\nCatalog Owner\n: Global Pasture Watch\n\nDataset Availability\n: 2000-01-01T00:00:00Z--2023-01-01T00:00:00Z\n\nDataset Provider\n:\n\n\n [Land and Carbon Lab Global Pasture Watch](https://landcarbonlab.org/data/global-grassland-and-livestock-monitoring)\n\nContact\n: [Land \\& Carbon Lab](https://landcarbonlab.org/subscribe)\n\nCadence\n: 1 Year\n\nTags\n:\n[global](/earth-engine/datasets/tags/global) [global-pasture-watch](/earth-engine/datasets/tags/global-pasture-watch) [land](/earth-engine/datasets/tags/land) [landcover](/earth-engine/datasets/tags/landcover) [landuse](/earth-engine/datasets/tags/landuse) [landuse-landcover](/earth-engine/datasets/tags/landuse-landcover) [pasture](/earth-engine/datasets/tags/pasture) [publisher-dataset](/earth-engine/datasets/tags/publisher-dataset) [rangeland](/earth-engine/datasets/tags/rangeland) [vegetation](/earth-engine/datasets/tags/vegetation) \n\n#### Description\n\nThis dataset provides global annual dominant class maps of grasslands\n(cultivated and natural/semi-natural) from 2000 to 2022 at 30-m spatial\nresolution.\nProduced by Land \\&\nCarbon Lab Global Pasture Watch initiative, the mapped grassland extent\nincludes any land cover type, which contains at least 30% of dry or wet\nlow vegetation, dominated by grasses and forbs (less than 3 meters)\nand a:\n\n- maximum of 50% tree canopy cover (greater than 5 meters),\n- maximum of 70% of other woody vegetation (scrubs and open shrubland), and\n- maximum of 50% active cropland cover in mosaic landscapes of cropland \\& other vegetation.\n\nThe grassland extent is classified into two classes:\n- **Cultivated grassland** : Areas where grasses and other forage plants have\nbeen intentionally planted and managed, as well as areas of native\ngrassland-type vegetation where they clearly exhibit active and\nheavy management for specific human-directed uses, such as directed\ngrazing of livestock.\n- **Natural/Semi-natural grassland**: Relatively undisturbed native\ngrasslands/short-height vegetation, such as steppes and tundra,\nas well as areas that have experienced varying degrees of human\nactivity in the past, which may contain a mix of native and\nintroduced species due to historical land use and natural processes.\nIn general, they exhibit natural-looking patterns of varied vegetation\nand clearly ordered hydrological relationships throughout the landscape.\n\nThe implemented methodology considered [GLAD Landsat ARD-2 images](https://glad.umd.edu/ard) (processed into cloud-free bi-monthly\naggregates, see [Consoli et al, 2024](https://doi.org/10.7717/peerj.18585)\n), accompanied by climatic, landform and proximity covariates,\nspatiotemporal machine learning (per-class Random Forest) and over\n2.3 million reference samples (visually interpreted in Very High\nResolution imagery). Custom probability thresholds (based on five-fold\nspatial cross-validation and balanced precision and recall values)\nwere used to derive dominant class maps, 0.32 and 0.42 for\ncultivated and natural/semi-natural grassland probability thresholds, respectively.\n\n**Limitations:** Grassland extent is partly under-predicted in southeastern\nAfrica (Zimbabwe and Mozambique) and in eastern Australia (shrublands and\nwoodlands of the Mulga ecoregion). Cropland is misclassified as grassland\nin parts of northern Africa, the Arabian Peninsula, Western Australia,\nNew Zealand, the center of Bolivia, and Mato Grosso state (Brazil). Due\nto the Landsat 7 SLC failure, regular stripes of grassland probabilities\nare visible at parcel-level, particularly in the year 2012. The usage of\ncoarser resolution layers (accessibility maps and MODIS products)\nintroduced curvilinear macroscopic errors (due to the downscaling\nstrategy based on cubicspline) in Uruguay, Southwest Argentina, South\nof Angola and in the Sahel region in Africa. Users need to be aware\nof the limitations and known issues; whilst considering them\ncarefully to ensure appropriate use of maps at this initial prediction\nstage. GPW is working actively to collect systematic feedback via the [Geo-Wiki\nplatform](https://www.geo-wiki.org), validate the current version\nand improve future versions of the dataset.\n\n**For more information see [Parente et. al, 2024](http://doi.org/10.1038/s41597-024-04139-6),\n[Zenodo](https://zenodo.org/records/13890401) and\n\u003chttps://github.com/wri/global-pasture-watch\u003e**\n\n### Bands\n\n**Bands**\n\n| Name | Min | Max | Pixel Size | Description |\n|------------------|-----|-----|------------|--------------------------------------------------------------------|\n| `dominant_class` | 0 | 2 | 30 meters | Dominant class derived through Random Forest and probability maps. |\n\n**dominant_class Class Table**\n\n| Value | Color | Description |\n|-------|---------|--------------------------------|\n| 0 | #ffffff | Other |\n| 1 | #ffcd73 | Cultivated grassland |\n| 2 | #ff9916 | Natural/Semi-natural grassland |\n\n### Image Properties\n\n**Image Properties**\n\n| Name | Type | Description |\n|---------|------|-----------------|\n| version | INT | Product version |\n\n### Terms of Use\n\n**Terms of Use**\n\n[CC-BY-4.0](https://spdx.org/licenses/CC-BY-4.0.html)\n\n### Citations\n\nCitations:\n\n- Parente, L., Sloat, L., Mesquita, V., et al. (2024)\n Global Pasture Watch - Annual grassland class and extent\n maps at 30-m spatial resolution (2000---2022) (Version v1)\n \\[Data set\\]. Zenodo\n [doi:https://doi.org/10.5281/zenodo.13890401](https://doi.org/10.5281/zenodo.13890401)\n- Parente, L., Sloat, L., Mesquita, V., et al. (2024).\n Annual 30-m maps of global grassland class and extent (2000--2022)\n based on spatiotemporal Machine Learning, Scientific Data.\n [doi: http://doi.org/10.1038/s41597-024-04139-6](http://doi.org/10.1038/s41597-024-04139-6)\n\n### DOIs\n\n- \u003chttps://doi.org/10.1038/s41597-024-04139-6\u003e\n- \u003chttps://doi.org/10.5281/zenodo.13890401\u003e\n\n### Explore with Earth Engine\n\n| **Important:** Earth Engine is a platform for petabyte-scale scientific analysis and visualization of geospatial datasets, both for public benefit and for business and government users. Earth Engine is free to use for research, education, and nonprofit use. To get started, please [register for Earth Engine access.](https://console.cloud.google.com/earth-engine)\n\n### Code Editor (JavaScript)\n\n```javascript\nMap.setCenter(-49.265188, -16.602052, 4);\n\nvar domi_grassland = ee.ImageCollection(\n \"projects/global-pasture-watch/assets/ggc-30m/v1/grassland_c\"\n)\nvar visParams = {\"opacity\":1, \"min\":1,\"max\":2,\"palette\":[\"ffcd73\",\"ff9916\"]};\n\nvar domi_grassland_2022 = domi_grassland.filterDate('2022-01-01', '2023-01-01').first();\nMap.addLayer(\n domi_grassland_2022.selfMask(), \n visParams, 'Dominant grassland class (2022)'\n);\n\nvar domi_grassland_2000 = domi_grassland.filterDate('2000-01-01', '2001-01-01').first();\nMap.addLayer(\n domi_grassland_2000.selfMask(), \n visParams, 'Dominant grassland class (2000)'\n);\n```\n[Open in Code Editor](https://code.earthengine.google.com/?scriptPath=Examples:Datasets/global-pasture-watch/projects_global-pasture-watch_assets_ggc-30m_v1_grassland_c) \n[GPW Annual Dominant Class of Grasslands v1](/earth-engine/datasets/catalog/projects_global-pasture-watch_assets_ggc-30m_v1_grassland_c) \nThis dataset provides global annual dominant class maps of grasslands (cultivated and natural/semi-natural) from 2000 to 2022 at 30-m spatial resolution. Produced by Land \\& Carbon Lab Global Pasture Watch initiative, the mapped grassland extent includes any land cover type, which contains at least 30% of dry or wet low ... \nprojects/global-pasture-watch/assets/ggc-30m/v1/grassland_c, global,global-pasture-watch,land,landcover,landuse,landuse-landcover,pasture,publisher-dataset,rangeland,vegetation \n2000-01-01T00:00:00Z/2023-01-01T00:00:00Z \n-90 -180 90 180 \nGoogle Earth Engine \nhttps://developers.google.com/earth-engine/datasets\n\n- [https://doi.org/10.5281/zenodo.13890401](https://doi.org/https://landcarbonlab.org/data/global-grassland-and-livestock-monitoring)\n- [https://doi.org/10.5281/zenodo.13890401](https://doi.org/https://developers.google.com/earth-engine/datasets/catalog/projects_global-pasture-watch_assets_ggc-30m_v1_grassland_c)"]]
