Global forest change что это

Обновлено: 18.05.2024

Results from time-series analysis of Landsat images in characterizing global forest extent and change from 2000 through 2020. For additional information about these results, please see the associated journal article (Hansen et al., Science 2013).

Web-based visualizations of these results are also available at our main site:

Please use that URL when linking to this dataset.

We anticipate releasing updated versions of this dataset. To keep up to date with the latest updates, and to help us better understand how these data are used, please register as a user. Thanks!

Global Forest Change: первая детальная карта изменений лесного покрова мира

15 ноября 2013 года в журнале Science опубликована статья группы ученых из Мэрилендского университета и ряда других научно-исследовательских организаций о первой детальной карте изменений лесного покрова мира в XXI веке (с 2000 по 2012 г.г.) - High-Resolution Global Maps of 21st-Century Forest Cover Change.

Сама карта размещена в открытом доступе в сети Интернет:

Пространственное разрешение карты - 30 метров, основой ее являются снимки спутников Landsat. Карта отражает фактическое изменение покрытых лесом площадей, включая как потери лесного покрова в результате вырубок, пожаров, ветровалов, расчисток и т.д., так и его увеличение в результате зарастания вырубок, гарей, брошенных сельхозугодий, лесовосстановления, лесоразведения и т.д. В общей сложности за рассматриваемый период площадь, на которой лесной покров исчез (на время или навсегда), составила 230 миллионов гектаров, а площадь, на которой он появился или восстановился - 80 миллионов гектаров.

Эта карта делалась коллективом авторов в течение нескольких лет, и полученный результат явно оправдал затраты труда и времени. Получился важнейший для понимания динамики лесного покрова мира информационный продукт, который, скорее всего, станет основой для большого количества разнообразных других исследований и оценок.

Usage Notes

The Global Land Analysis and Discovery (GLAD) laboratory at the University of Maryland, in partnership with Global Forest Watch (GFW), provides annually updated global-scale forest loss data, derived using Landsat time-series imagery. These data, available here, are a relative indicator of spatiotemporal trends in forest loss dynamics globally. However, inconsistencies exist due to the following factors:

  1. Differences in Landsat sensor technology, whether Thematic Mapper, Enhanced Thematic Mapper Plus, or Operational Land Image data. For example, the Operational Land Imager (2013-onward) onboard the Landsat 8 spacecraft employs a pushbroom sensor technology that increases per observation dwell time compared to past whiskbroom systems. The result is a signal to noise ratio that is a magnitude greater than that of Landsat 7’s Enhanced Thematic Mapper Plus sensor. The increased signal enables better detection capabilities in mapping land change.
  2. Data richness, or the number of viable land observations available as inputs to analysis. The global acquisition strategy has improved over time, with acquisitions increasing from under 150k per year in the early 2000s to over 250k per year in recent years. Additionally, Landsat 7 was the only input for the 2001-2012 initial product, and is affected by the scan-line corrector malfunction of the Enhanced Thematic Mapper from 2002 onward, where nearly a quarter of the footprint of each scene is not collected. Also, the gap between the decommissioning of Landsat 5 in 2011 and the launch of Landsat 8 in 2013 resulted in a total 2012 global collection of less than 100k Landsat 7 images.
  3. Algorithm adjustments, including modifications of training data and input image feature space. For example, the original 2001-2012 forest loss map was made using a single algorithm run, compared to subsequent years that have been added individually. Additionally, models have been iterated to improve performance in the 2012-forward period. Such changes in the mapping method can result in year to year inconsistencies.

While the resulting map data are a largely viable relative indicator of trends, care must be taken when comparing change across any interval. Applying a temporal filter, for example a 3-year moving average, is often useful in discerning trends. However, definitive area estimation should not be made using pixels counts from the forest loss layers.

The Intergovernmental Panel on Climate Change (IPCC) provides guidance on reporting areal extent and change of land cover and land use, requiring the use of estimators that neither over or underestimate dynamics to the degree possible, and that have known uncertainties. The maps provided by GLAD do not have these properties. However, the maps can be leveraged to facilitate appropriate probability-based statistical methods in deriving statistically valid areas of forest extent and change. Specifically, the maps may be used as a stratifier in targeting forest extent and/or change by a probability sample. The team at GLAD has demonstrated such approaches using the GLAD forest loss data in sample-based area estimation (Tyukavina et al., ERL, 2018, Turubanova et al., ERL, 2019, and Potapov et al., RSE, 2019, among others).

User Notes for Version 1.8 Update

This update of gross forest cover loss includes new 2020 loss-year and multispectral imagery layers. Relative to the version 1.0 product our method has been modified in numerous ways, and the new update should be seen as part of a transition to a future version 2.0 that is more consistent over the entire 2000-onward period. Key changes include:

  1. The use of Landsat 8 OLI data for 2013 onward,
  2. The reprocessing of data from 2011 onward in measuring loss,
  3. Improved training data for calibrating the loss model,
  4. Improved per sensor quality assessment models to filter input data, and
  5. Improved input spectral features for building and applying the loss model.

Some examples of improved change detection in the 2011–2020 update include the following:

  1. Improved detection of boreal forest loss due to fire.
  2. Improved detection of smallholder rotation agricultural clearing in dry and humid tropical forests.
  3. Improved detection of selective logging.
  4. Improved detection of the clearing of short cycle plantations in sub-tropical and tropical ecozones.

These are examples of dynamics that may be differentially mapped over the 2001-2020 period in Version 1.8. A version 2.0 reprocessing of the 2000-onward record is planned, but no delivery date is yet confirmed.

The original version 1.0 data is also still available for download here.

In addition, to reduce confusion, beginning with version 1.4 we are no longer releasing loss as a separate layer from lossyear . Loss as previously releaed corresponds to nonzero values of loss year.

A Guide to GFW’s Forest Change Data

By Mikaela Weisse Forest change data, which shows how tree cover changes in the world’s forests, is the backbone of our work at Global Forest Watch, and we’re constantly striving to provide better, more up-to-date data for our users. But, as we add more data, it can become harder to understand how data sets differ, and know which ones to use for your purposes. Below, we answer some common user questions about the forest change data currently on GFW, and new data that’s coming soon.

Where does all this data come from?

All of the “forest change data” on Global Forest Watch comes from satellite images (other types of data, like “forest use” and “concessions” might be derived from other sources like government surveys, field measurements, and models). Satellite images are a good way to detect forest change because they cover large areas, are consistent across time and space, and can serve as a source of information about the past. Our data mainly comes from two NASA satellite-based sensors: Landsat and MODIS, because all of the images from these two systems are free. Scientists use a variety of techniques to detect forest changes from these images, including manually identifying the changes, statistical modelling, and “machine learning” (computers are shown samples images of forest change and develop pattern recognition). The techniques used in each of the data sets are too complicated to explore in depth here, but you can read about the methods of creating each data set in its information window on GFW (accessed by clicking the icon).

What is resolution?

Satellite images, like all pictures, are made up of pixels. The resolution of an image is how large of an area each pixel captures—for example, a 30-meter resolution image has pixels that each capture a 30 by 30-meter land area or roughly the size of a baseball diamond. Higher resolution data allows us to see changes at a much finer scale. Most of the data on GFW is at either 30-meter or 250-meter resolution, because those are the maximum resolutions of Landsat and MODIS images, respectively. Generally speaking, satellites with higher resolution images cover a smaller area of land each day. Landsat sensors currently take 8 days to cover the Earth, while MODIS sensors now pass every spot on Earth twice a day. Because of the differences in resolution and frequency, there is a tradeoff in our data sets. Many of our lower resolution data sets can be updated more often, but don’t have as much detail as some of the higher-resolution data sets.

Why are there so many data sets?

Global Forest Watch is committed to providing the best information available on forest change. That gets tricky, however, because different data sets have advantages for different areas or purposes. Some of the data sets are best used for measuring trends and areas of deforestation (e.g. GLAD tree cover loss, PRODES), while others are better suited for monitoring forest changes that are happening now (e.g. FORMA, Terra-i, GLAD alerts). Some of the data sets cover the entire world, or most of it, while others are calibrated for a particular area and may be more accurate. Some of the data on GFW is “official” data from countries (e.g. PRODES, MINAM) and thus analysis with that data will be more respected by those governments. The table below explains the best purpose for each data set.

What’s the difference between alert and non-alert data?

Many of the “alert” products like Terra-i and FORMA, are intended to help identify areas where recent clearing is likely to have occurred—calculated based on changes in the vegetation as seen from satellites. They are often coarse, but updated as frequently as daily. Their pixels of loss are more warnings than confirmation of wholesale clearing like the annual GLAD tree cover loss. Some of the upcoming data products, such as GLAD alerts (distinct from the annual global data) and FORMA 250 (currently, FORMA is at 500-m resolution), offer more reliable estimates of loss, but are still intended to flag new places of potential loss rather than the exact location of loss. Ideally, these alerts will help users prioritize areas for further investigation using verification methods, such as field visits, scrutinizing up-to-date satellite imagery, or flying drones to confirm recent changes.

Which data are right?

There may be cases where different data sets show different locations of forest change. This is inevitable—all data sets use different methods (some of them slight differences, some of them major), may measure different things, and all have some amount of inaccuracy. It is impossible to say which data is “right” and which data is “wrong” in every case. A conservative approach is to focus on areas where multiple alert products agree—these areas are very likely to have real forest changes. You can also always reach out to the GFW discussion forum to see if other users have suggestions for how to address your needs. Here’s a quick guide to the various forest change data sets currently on Global Forest Watch, and those coming soon. If you have questions on these data sets, please reach out to our community through the GFW discussion forum.

Dataset Details

Tree canopy cover for year 2000 ( treecover2000 ) Tree cover in the year 2000, defined as canopy closure for all vegetation taller than 5m in height. Encoded as a percentage per output grid cell, in the range 0–100. Global forest cover gain 2000–2012 ( gain ) Forest gain during the period 2000–2012, defined as the inverse of loss, or a non-forest to forest change entirely within the study period. Encoded as either 1 (gain) or 0 (no gain). Year of gross forest cover loss event ( lossyear ) Forest loss during the period 2000–2020, defined as a stand-replacement disturbance, or a change from a forest to non-forest state. Encoded as either 0 (no loss) or else a value in the range 1–20, representing loss detected primarily in the year 2001–2020, respectively. Data mask ( datamask ) Three values representing areas of no data (0), mapped land surface (1), and permanent water bodies (2). Circa year 2000 Landsat 7 cloud-free image composite ( first ) Reference multispectral imagery from the first available year, typically 2000. If no cloud-free observations were available for year 2000, imagery was taken from the closest year with cloud-free data, within the range 1999–2012. Circa year 2020 Landsat cloud-free image composite ( last ) Reference multispectral imagery from the last available year, typically 2020. If no cloud-free observations were available for year 2020, imagery was taken from the closest year with cloud-free data.

The g factor was chosen independently for each band to preserve the band-specific dynamic range, as shown in the following table:

Landsat Bandg
Band 3 (red)508
Band 4 (NIR)254
Band 5 (SWIR)363
Band 7 (SWIR)423

Hansen Global Forest Change v1.8 (2000-2020)

UMD/hansen/global_forest_change_2020_v1_8

Earth Engine is free to use for research, education, and nonprofit use.

To access this dataset in Earth Engine, please sign up for Earth Engine then return to this page.

Description

Results from time-series analysis of Landsat images in characterizing global forest extent and change.

Please see the User Notes for this Version 1.8 update, as well as the associated journal article: Hansen, Potapov, Moore, Hancher et al. “High-resolution global maps of 21st-century forest cover change.” Science 342.6160 (2013): 850-853.

Bands

Resolution
30.92 meters

Tree canopy cover for year 2000, defined as canopy closure for all vegetation taller than 5m in height.

Forest loss during the study period, defined as a stand-replacement disturbance (a change from a forest to non-forest state).

Bitmask for loss

  • Bit 0: Forest loss during the study period.
    • 0: Not loss
    • 1: Loss

    Forest gain during the period 2000–2012, defined as the inverse of loss (a non-forest to forest change entirely within the study period). Note that this has not been updated in subsequent versions.

    Bitmask for gain

    • Bit 0: Forest gain during the period 2000–2012.
      • 0: No gain
      • 1: Gain

      Landsat 7 band 3 (red) cloud-free image composite. Reference multispectral imagery from the first available year, typically 2000.

      Landsat 7 band 4 (NIR) cloud-free image composite. Reference multispectral imagery from the first available year, typically 2000.

      Landsat 7 band 5 (SWIR) cloud-free image composite. Reference multispectral imagery from the first available year, typically 2000.

      Landsat 7 band 7 (SWIR) cloud-free image composite. Reference multispectral imagery from the first available year, typically 2000.

      Landsat 7 band 3 (red) cloud-free image composite. Reference multispectral imagery from the last available year, typically the last year of the study period.

      Landsat 7 band 4 (NIR) cloud-free image composite. Reference multispectral imagery from the last available year, typically the last year of the study period.

      Landsat 7 band 5 (SWIR) cloud-free image composite. Reference multispectral imagery from the last available year, typically the last year of the study period.

      Landsat 7 band 7 (SWIR) cloud-free image composite. Reference multispectral imagery from the last available year, typically the last year of the study period.

      Three values representing areas of no data, mapped land surface, and permanent water bodies.

      Bitmask for datamask

      • Bits 0-1: Three values representing areas of no data, mapped land surface, and permanent water bodies.
        • 0: No data
        • 1: Mapped land surface
        • 2: Permanent water bodies

        Year of gross forest cover loss event. Forest loss during the study period, defined as a stand-replacement disturbance, or a change from a forest to non-forest state. Encoded as either 0 (no loss) or else a value in the range 1–20, representing loss detected primarily in the year 2001–2020, respectively.

        Terms of Use

        Terms of Use

        Licensed under the Creative Commons Attribution 4.0 International License.

        Citations

        Explore in Earth Engine

        Sign up for Earth Engine

        Earth Engine is free to use for research, education, and nonprofit use.

        To access this dataset in Earth Engine, please sign up for Earth Engine then return to this page.

        License and Attribution

        This work is licensed under a Creative Commons Attribution 4.0 International License. You are free to copy and redistribute the material in any medium or format, and to transform and build upon the material for any purpose, even commercially. You must give appropriate credit, provide a link to the license, and indicate if changes were made.

        Use the following credit when these data are displayed:

        Use the following credit when these data are cited:

        Download Instructions

        To download individual 10x10 degree granules, click on a region on the map below and then click on the URLs underneath it.

        Читайте также: