GLAD Landsat ARD

2. GLAD Landsat ARD

Multi-temporal metrics are a standard method of time-series data transformation. They simplify the analysis of land surface phenology, facilitate land cover mapping and change detection. The metrics approach was developed in the mid-1980s to extract phenology features from time-series of Normalized Difference Vegetation Index (NDVI) (Badwhar, 1984; Maligreau, 1986). At the same time, the idea of using NDVI time-series to extract spectral reflectance corresponding to certain phenological stages was proposed by Holben (1986). Later, both approaches were merged together by researchers from the Laboratory for Global Remote Sensing Studies at the University of Maryland (DeFries, Hansen, and Townsend, 1995). In their work, metrics were calculated by extracting spectral information for specific phenological stages defined by the NDVI annual dynamics. Later, the multi-temporal metrics were widely used for forest extent and structure monitoring at continental and global scales using MODIS (Hansen et al., 2010) and Landsat data (Hansen et al., 2013; Potapov et al., 2012; 2017; 2019;). The purpose of metrics is to create consistent inputs for annual vegetation mapping and change detection models and to overcome the inconsistency of clear-sky data availability that is typical for sensors with low observation frequency, like Landsat.

 

Implementation of the multi-temporal metric approach requires processing the entire archive of the Landsat observations. To simplify metric processing, the Global Land Analysis and Discovery team (GLAD) developed a Landsat Analysis Ready Data (ARD) product. The Landsat ARD produced by the GLAD team using an automated image processing system. The essence of the GLAD ARD approach is to convert individual Landsat images into a time-series of 16-day normalized surface reflectance composites with minimal atmospheric contamination. The Landsat data processing algorithms were prototyped by Hansen et al. (2008) and Potapov et al. (2012) and was significantly improved in the recent years (Potapov et al., 2019). The ARD serve as a data source to generate metrics suitable for annual land cover, land cover change, vegetation composition and structure mapping.

 

The following sections provide an overview of the Landsat processing methodology, explain ARD structure, and provide data download instructions.

    2.1. Landsat ARD Methodology

    2.2. Landsat ARD Structure

    2.3. Landsat ARD Download

    2.4. Topography Data Download

    2.5. License and Disclaimer

    2.6. User Registration

 

References

  1. Badwhar, G. D. (1984) Use of LANDSAT-derived profile features for spring small-grains classification, Int. J. Remote Sens. 5(5):783-797.
  2. Holben, B. N. (1986), Characteristics of maximum-value composite images from temporal AVHRR data, Int. J. Remote Sens. 12:1147-1163.
  3. Malingreau, J.-P. (1986) Global vegetation dynamics: satellite observations over Asia, Int. J. Remote Sens. 7:1121-1146.
  4. DeFries, R., Hansen, M., Townshend, J. (1995) Global discrimination of land cover types from metrics derived from AVHRR pathfinder data. Remote Sens. Environ. 54:209–222.
  5. Hansen, M., Stehman, S., Potapov, P. (2010) Quantification of global gross forest cover loss. Proc. Natl. Acad. Sci. USA. 107(19):8650–8655.
  6. Potapov, P.V., Turubanova, S.A., Hansen, M.C., Adusei, B., Broich, M., Altstatt, A., Mane, L., Justice, C.O. (2012) Quantifying forest cover loss in Democratic Republic of the Congo, 2000–2010, with Landsat ETM+ data. Remote Sens. Environ. 122:106-116.
  7. Hansen, M.C., Potapov, P.V., Moore, R., Hancher, M., Turubanova, S.A.A., Tyukavina, A., Thau, D., Stehman, S.V., Goetz, S.J., Loveland, T.R., Kommareddy, A. (2013) High-resolution global maps of 21st-century forest cover change. Science, 342(6160):850-853.
  8. Potapov, P. V., Turubanova, S.A., Tyukavina, A., Krylov, A.M., McCarty, J.L., Radeloff, V.C., Hansen, M.C. (2015) Eastern Europe’s forest cover dynamics from 1985 to 2012 quantified from the full Landsat archive. Remote Sens. Environ. 159:28–43.
  9. Potapov, P., Tyukavina, A., Turubanova, S., Talero, Y., Hernandez-Serna, A., Hansen, M.C., Saah, D., Tenneson, K., Poortinga, A., Aekakkararungroj, A. and Chishtie, F., 2019. Annual continuous fields of woody vegetation structure in the Lower Mekong region from 2000‐2017 Landsat time-series. Remote Sensing of Environment, 232, p.111278.

 

Landsat ARD Methodology

2.1. Landsat ARD Methodology

The Landsat Analysis Ready Data (ARD) developed by the Global Land Analysis and Discovery team (GLAD) provides globally consistent inputs for land cover mapping and change detection. The entire archive of the Landsat Collection 1 TM/ETM+/OLI data available from the United States Geological Survey National Center for Earth Resources Observation and Science (USGS EROS) was automatically processed to derive ARD time-series. The following section documents the Landsat data processing and temporal aggregation. For all questions and comment contact Peter Potapov (potapov@umd.edu). The following publication summarizes data processing steps: Potapov et al., 2019 (Potapov_RSE_2019.pdf)

 

The Landsat data processing includes four steps: (1) conversion to radiometric quantity, (2) observation quality assessment, (3) reflectance normalization, and (4) temporal integration into 16-day composites.


1. Conversion to radiometric quantity

In the first step, all data is converted to top-of-atmosphere reflectance (Chander et al., 2009) for reflective bands and brightness temperature for the emissive band. Spectral reflectance (value range from zero to one) are rescaled to the range from 1 to 40,000; temperature is recorded as degrees C multiplied by 100 to preserve measurement precision. Only spectral bands with matching wavelengths between TM, ETM+ and OLI/TIRS sensors are processed (Table 1).

 

Table 1. Landsat spectral bands used for the ARD and corresponding MODIS spectral bands

Band name

Wavelength, nm

Landsat 5 TM

Landsat 7 ETM+

Landsat 8 OLI/TIRS

MODIS

Blue

450-520

441-514

452-512

459-479

Green

520-600

519-601

533-590

545-565

Red

630-690

631-692

636-673

620-670

NIR

760-900

772-898

851-879

841-876

SWIR1

1,550-1,750

1,547-1,749

1,566-1,651

1,628-1,652

SWIR2

2,080-2,350

2,064-2,345

2,107-2,294

2,105-2,155

Thermal

10,410-12,500

10,310-12,360

10,600-11,190

10,780-11,280

 

The following equations summarize data conversion and scaling. The value of PI defined as 3.1415926535897932384626433832. All other coefficients are from Landsat Collection 1 metadata and lookup tables from Chander et al., 2009. DN stands for the source raster value.

 

Landsat 8 OLI

TOA reflectance = ((0.00002*DN-0.1)/sin(SUNELEV*PI/180))*40000

 

From the image metadata:

SUNELEV = "SUN_ELEVATION"

 

Landsat 8 TIRS

Brightness temperature = (K2/log(K1/(0.0003342*DN+0.1)+1))*100+0.5

 

From the image metadata:

K1 = “K1_CONSTANT_BAND_10”

K2 = “K2_CONSTANT_BAND_10”

 

Landsat 4/5 TM and 7 ETM+

TOA reflectance = ((PI*D^2*(G*DN+B))/(ESUN*sin(SUNELEV *PI/180)))*40000

 

From the image metadata:

SUNELEV = “SUN_ELEVATION”

 

From the lookup tables (Chander et al., 2009):

G – gain (sensor- and band-specific)

B – bias (sensor- and band-specific)

ESUN – exoatmospheric irradiance (sensor- and band-specific)

D – Sun-Earth distance

 

Brightness temperature = (K2/log(K1/(G*DN+B)+1))*100+0.5

 

From the lookup tables (Chander et al., 2009):

K1 = Thermal band calibration constant 1 (sensor-specific)

K2 = Thermal band calibration constant 2 (sensor-specific)


2. Observation quality assessment

During the second step, we determine per-pixel observation quality, i.e. a probability of an image pixel to be collected during clear sky conditions. The GLAD quality assessment model developed by our team represents a set of regionally adapted decision trees to map probability of a pixel to represent cloud, cloud shadow, heavy haze, and, for clear-sky observations, land, water, or snow/ice. The Landsat Collection 1 data include observation quality layers based on the globally consistent CFMask cloud and cloud shadow detection algorithm (Foga et al., 2017). Since our primary goal is to reduce the presence of clouds and shadows in the time-series data, we merge both the CFMask product (high-probability clouds and shadows) with the GLAD algorithm output. Through these outputs, cloud, shadow, haze, water, and land masks are created for each Landsat image. The masks are subsequently aggregated into a single quality flag that highlights cloud/shadow contaminated observations, indicates confidence of cloud/shadow detection, separates topography shadows from likely cloud shadows, and specifies the proximity to high-confidence clouds and cloud shadows. Table 2 summarize possible observation quality flags. For the subsequent analysis, quality flags 1, 2, 5, 6, 11, 12, and 14 are treated as clear-sky observations, while codes 3, 4, 7, 8, 9, and 10 are considered contaminated by clouds and shadows. Codes 5, 6, 11, 12, and 14 are used for 16-day composition to prioritize observations with minimal atmospheric contamination.

 

Table 2. Per-pixel observation quality flags

Flag

Quality

 Definition

1

Land

Clear-sky land observation.

2

Water

Clear-sky water observation.

3

Cloud

Cloud detected by either GLAD or CFMask algorithm.

4

Cloud shadow

Shadow detected by either GLAD or CFMask algorithm. The pixels located within the projection of a detected cloud. Cloud projection defined using the solar elevation and azimuth and limited to 9 km distance from the cloud.

5

Topographic shadow

Shadow detected by either GLAD or CFMask algorithm. The pixel located outside cloud projections and within estimated topographic shadow (defined using SRTM DEM and the observation solar elevation and azimuth).

6

Snow

Clear-sky observation classified as snow by the GLAD algorithm

7

Haze

Classified as haze (dense semi-transparent cloud) by the GLAD algorithm.

8

Cloud proximity

Aggregation (OR) of two rules:

(i) 1-pixel buffer around detected clouds.

(ii) Above-zero cloud likelihood (estimated by GLAD cloud detection model) within 3-pixel buffer around detected clouds.

9

Shadow proximity

Shadow likelihood (estimated by GLAD shadow detection model) above 10% for pixels either (i) located within the projection of a detected cloud; OR (ii) within 3 pixels of a detected cloud or cloud shadow.

10

Other shadows

Shadow detected by either GLAD or CFMask algorithm. The pixel located outside (a) projection of a detected cloud and (b) estimated topographic shadow.

11

Additional cloud proximity over land

Clear-sky land pixels located closer than 7 pixels of detected clouds

12

Additional cloud proximity over water

Clear-sky water pixels located closer than 7 pixels of detected clouds

14

Additional shadow proximity over land

Clear-sky land pixels located closer than 7 pixels of detected cloud shadows

 


3. Reflectance normalization

The third step consists of reflectance and brightness temperature normalization to reduce the effects of atmospheric scattering and surface anisotropy. The purpose of relative normalization is to simplify extrapolation of classification models in space and time by ensuring spectral similarity of the same land-cover types. Relative normalization is not computationally heavy and does not require synchronously collected or historical data on atmospheric properties and land-cover specific anisotropy. The normalized surface reflectance is not equal to surface reflectance and should not be used as a source dataset for the analysis of reflectance properties and dynamic. The purpose of the ARD product is solely to facilitate land cover and land cover change mapping.

 

The Landsat image normalization consists of four steps: (A) Production of the normalization target dataset; (B) selection of pseudo-invariant target pixels to derive the normalization model; (C) model parametrization; and (D) model application.

 

A. Normalization target

The normalization target data were collected from the MODIS 44C surface reflectance product. We used MODIS bands with a similar wavelength to the selected Landsat bands (see Table 1). MODIS surface reflectance and brightness temperature were scaled using the same coefficients as for the processed Landsat data. To ensure spatial consistency of reflectance target dataset, we used all 16-day global MODIS observation composites from the year 2000 to 2011. To reduce the effects of atmospheric contamination and surface anisotropy, only near-nadir clear-sky and low aerosol observations were retained in the time-series. From all selected observations, a single reflectance composite was created for each spectral band. We calculated the normalization target composite value as the mean reflectance of observations with normalized difference vegetation index (NDVI) above the 75% percentile.

 

B. Selection of pseudo-invariant target pixels

We define the pseudo-invariant target pixels as clear-sky land observations that represent the same land cover type and phenology stage in Landsat image and MODIS normalization target composite. To check for the land cover type and condition, we calculate the absolute difference between Landsat and MODIS spectral reflectance for red and shortwave infrared bands and select pixels with difference below 0.1 reflectance value for both spectral bands. Bright objects (with red band reflectance above 0.5) are excluded from the mask. Landsat images with less than 10,000 selected pseudo-invariant target pixels are discarded from the processing chain.

 

C. Model parametrization

To parametrize the reflectance normalization model, we calculate the median bias between MODIS and Landsat reflectance of pseudo-invariant pixels for each reflective band for each 10 km interval of distance from the Landsat ground track. The set of median values is used to parametrize a per-band linear regression model using least squares fitting method. For each image and each spectral band, we derive Gain and Bias coefficients to predict the reflectance bias as a function of the distance from the ground track:

 

[Reflectance Bias] = Gain * [Distance from ground track] + Bias

 

For images where only a small portion (less than 1/16 of the image) is covered by land, a mean reflectance bias is calculated instead (Gain = 0). Such conditions are usually found in coastal regions or over small islands. For the brightness temperature band, we calculate a single mean bias value for all pseudo-invariant target pixels within the image.

 
D. Model application

The model is then applied to all pixels within the image to estimate the reflectance (brightness temperature) bias. We subtract the estimated reflectance bias from the Landsat top-of-atmosphere reflectance for each spectral band and from brightness temperature values. The model has been shown to effectively normalize land surface reflectance and reduce effects of atmospheric scattering and surface anisotropy (Potapov et al., 2012)

 


4. Temporal integration into 16-day composites

The final step of Landsat ARD processing is temporal aggregation of individual images into 16-day composites. The range of dates for each composite is provided in the Table 3.

 

Table 3. 16-day composite intervals

Interval ID

DOY start

DOY end

Date start

Date end

1

1

16

1-Jan

16-Jan

2

17

32

17-Jan

1-Feb

3

33

48

2-Feb

17-Feb

4

49

64

18-Feb

4-Mar

5

65

80

5-Mar

20-Mar

6

81

96

21-Mar

5-Apr

7

97

112

6-Apr

21-Apr

8

113

128

22-Apr

7-May

9

129

144

8-May

23-May

10

145

160

24-May

8-Jun

11

161

176

9-Jun

24-Jun

12

177

192

25-Jun

10-Jul

13

193

208

11-Jul

26-Jul

14

209

224

27-Jul

11-Aug

15

225

240

12-Aug

27-Aug

16

241

256

28-Aug

12-Sep

17

257

272

13-Sep

28-Sep

18

273

288

29-Sep

14-Oct

19

289

304

15-Oct

30-Oct

20

305

320

31-Oct

15-Nov

21

321

336

16-Nov

1-Dec

22

337

352

2-Dec

17-Dec

23

353

366

18-Dec

31-Dec

 

Temporal compositing is done per-pixel using all overlapping observations within the 16-day interval. From all available observations we retain the one with the highest observation quality. The proximity to clouds/shadows is used as a criterion to select the least contaminated observation. If several clear-sky observations were available, the per-band mean reflectance value is retained in the composite. Each 16-day composite consists of normalized surface reflectance for six spectral bands (blue, green, red, NIR, and two SWIR), normalized brightness temperature, and the quality flag as separate raster layers. The 16-day interval composites are stored in the geographic coordinates (+proj=longlat +ellps=WGS84 +datum=WGS84 +no_defs) as raster tiles of 1.0005 by 1.0005 degree size and pixel size of 0.00025 degree. The tile system features a 2-pixel overlap to simplify parallelization of the focal average computation. For detailed information about the Landsat ARD see Landsat ARD Structure section.

 
Important notes on data limitations:
  • Images that were not processed to Tier 1 standard, as well as images with high cloud cover and Landsat 5 TM sensor issues were automatically detected and excluded from processing.
  • Due to low sun azimuth and similarity between snow cover and clouds during winter season in temperate and boreal climates, the GLAD Landsat ARD algorithm is not suitable for winter time image processing above 30N and below 45S Latitude. Using the MODIS snow and ice product and NDVI time-series we defined dates with high probability of the seasonal snow cover for each WRS2 path/row. We are not processing images for these dates, and the resulting 16-day intervals have no data. Some of the images (and resulting 16-day composites) may still include snow-covered observations. We suggest to further filter such observations using “snow/ice” quality flag or by removing certain 16-day intervals from data processing.
  • The GLAD team is committed to the annual ARD update. Due to the delay of Tier 1 data processing by USGS EROS (between 14 and 26 days after the data acquisition) and the time required to download and process new data by GLAD, the current data processing system is not designed for real-time ARD delivery.
  • The users should be aware that the image normalization method is not design to deal with specular reflectance and thus introduces bias over the water surfaces.

Landsat ARD Structure

2.2. Landsat ARD Structure

Tile system

The global Landsat ARD product is provided as a set of 1x1 degree tiles. Tile names are derived from the tile center, and refer to their integer value of the tile center degrees. E.g., the name of a tile with center 17.5E and 52.5N is 017E_52N. Please use the shapefile glad_landsat_tiles.zip to select tiles for your area of interest.

 

16-day intervals

Each data granule (GeoTIFF file) contains observation data collected for a single 16-day interval. There are 23 intervals per year, see Table 2 (Landsat ARD Methodology) for interval dates. Each interval has a unique numeric ID, starting from the first interval of the year 1980. Use the 16-day interval ID table (16d_intervals.xlsx) to select intervals for your analysis.

 

Raster data

Each 16-day interval data stored as a single 8-band LZW-compressed GeoTIFF file in geographic coordinates (+proj=longlat +ellps=WGS84 +datum=WGS84 +no_defs), with the size of 4004x4004 pixels, an extent of 1.0005 by 1.0005 degrees, and a pixel size of 0.00025 degree. The tile system features a 2-pixel overlap to simplify parallelization of the focal average computation.

 

All raster layers are 16-bit unsigned. The list of layers are provided in Table 1 (see Landsat ARD Methodology for band wavelengths). The last layer (band 8) consists of a quality flag (QA) that reflects the quality of observation used to create the composite. Table 2 provides QA flag metadata. QA values 1, 15, and 2 are always considered as clear-sky observations. QA values 5 and 6 indicate seasonal quality issues (topographic shadows and snow cover). These observations may be removed from data processing if the number of clear-sky observations is sufficient. QA values 11-14 and 16-17 are considered as cloud-free data with an indication of potential cloud or cloud shadow contamination. QA values 15-17 are added to simplify intermittent water analysis (e.g., floods detection). QA values 3, 4, and 7-10 are considered contaminated by clouds and shadows and are usually excluded from data processing.

 

Table 1. 16-day composite image layers

Band number

Data

Units, Format

1

Normalized surface reflectance of blue band

Normalized surface reflectance scaled to the range from 1 to 40,000, UInt16

2

Normalized surface reflectance of green band

3

Normalized surface reflectance of red band

4

Normalized surface reflectance of NIR band

5

Normalized surface reflectance of SWIR1 band

6

Normalized surface reflectance of SWIR2 band

7

Normalized brightness temperature

degrees C * 100, UInt16

8

Observation quality code (QA). See 16d_intervals.xlsx for QA code description.

code, UInt16

 

Table 2. Per-pixel observation quality flag (QA)

QA code

Description

Quality

1

Land

clear-sky

2

Water

clear-sky

3

Cloud

Cloud contaminated

4

Cloud shadow

Shadow contaminated

5

Hillshade

clear-sky

6

Snow

clear-sky

7

Haze

Cloud contaminated

8

Cloud buffer

Cloud contaminated

9

Shadow buffer

Shadow contaminated

10

Shadow high likelihood

Shadow contaminated

11

Additional cloud buffer over land

clear-sky

12

Additional cloud buffer over water

clear-sky

14

Additional shadow buffer over land

clear-sky

15

Land, water detected but not used

clear-sky

16

Additional cloud buffer over land, water detected but not used

clear-sky

17

Additional shadow buffer over land, water detected but not used

clear-sky

 

Note for ARD v1.0 users:

The v1.0 16-day data was produced using the same methodology that the V1.1 data. The differences include the use of Landsat pre-collection data, incomplete coverage of Landsat 5 data (exclude recently added to the USGS archive), and different QA code (see below). The ARD v1.1 data is currently available east of 30W longitude. The data west of 30W longitude is only available in the v1.0 standard and is being reprocessed to the v1.1.

 

The QA flag is combined with the number of observations used to calculate the value for the 16-day interval. The output value is constructed as:

[QA ARD v1.0] = [Observation quality code * 100] + [number of observations]

In the v1.1 the number of observations is deprecated and the QA value consists of the observation quality code.

 

 

Landsat ARD Download

2.3. Landsat ARD Download

To download the Landsat ARD data, follow the steps below:

  • Create/define local folder for data download
    • If you are downloading data for the first time or for a new project, create a new folder where data will be stored. Make sure you have enough disk space. The data volume for a single 1x1 degree tile, one year of data, is between 4 and 5 GB on average.
    • If you are adding data to the existing project, use name of the folder with previously downloaded 16-day composites.
  • Select tiles to download from the tile database (glad_landsat_tiles.zip). Make a text file that contains only the list of selected tiles (single column, tile names only – see example tiles.txt).
  • Select dates from the 16-day interval IDs (16d_intervals.xlsx).
  • For a new user: register to the service and receive your username and password (User Registration)
  • To download a single interval for a single tile, use the following command:

curl -u <username>:<password> -X GET https://glad.umd.edu/dataset/landsat_v1.1/<lat>/<tile>/<interval>.tif -o <outfolder>/<interval>.tif

 

Required parameters:

<username> and <password> - User login information

<tile> - lat/long 1x1 degree tile name (e.g. 105E_13N)

<lat> - tile latitude, second half of the tile name (e.g., for the 105E_13N, <lat> is 13N)

<interval> - unique 16-day interval ID (16d_intervals.xlsx).

<outfolder> - output folder (make sure that each tile is stored in a separate folder)

 

To download multiple tiles, use data download code (download_V1.1.pl)

  • Make sure that the ActivePERL is installed on your computer (see GLAD Tools Setup for details).
  • Make list of tiles as a text file (single column, tile names only – see example tiles.txt).

Usage: > perl download_V1.1.pl <username> <password> <tile list file> <start interval> <end interval> <output folder>

 

Required parameters:

<username> and <password> - User login information

<tile list file> - list of lat/long 1x1 degree tiles (in text format)

<start interval> <end interval> - a range of 16-day intervals (16d_intervals.xlsx).

<output folder> - output folder

 

 

Topography Data Download

2.4. Topography Data Download

Topography metrics (elevation, slope, and aspect) are frequently used as additional source data for image classification along with multi-temporal spectral metrics. To facilitate the image classification applications, the GLAD team provides selected topography metrics in the ARD tile format. The source data was extracted from the NASA Making Earth System Data Records for Use in Research Environments (MEaSUREs) version of the Shuttle Radar Topography Mission (SRTM) global elevation dataset. The source product (SRTMGL1v003) has a spatial resolution of 1 arc second (~30 meters) and is freely distributed by NASA LPDAAC. The dataset has global coverage between 60°N to 56°S.

 

The GLAD team re-projected and subset the SRTM elevation data to the standard ARD tile format. The dem.tif files contain SRTM elevation in meters, data format Int16 (signed). The No Data value is -32768. Two other topography metrics (slope and aspect) were derived from the global DEM dataset using the following procedure: (1) the DEM was resampled into a local UTM map projection with pixel size of 30m; (2) aspect and slope were calculated using GDAL tools; and (3) slope was scaled by multiplying by 500, and aspect by multiplying by 100, and both variables were stored as UInt16 images (slope.tif and aspect.tif).

 

To download multiple tiles, use data download code (download_V1.1.pl)

  • Make sure that the ActivePERL is installed on your computer (see GLAD Tools Setup for details).
  • Make list of tiles as a text file (single column, tile names only – see example tiles.txt).

Usage: > perl download_SRTM.pl <username> <password> <tile list file> <output folder>

 

Required parameters:

<username> and <password> - User login information

<tile list file> - list of lat/long 1x1 degree tiles (in text format)

<output folder> - output folder

 

License and Disclaimer

2.5. License and Disclaimer

The GLAD Landsat Analysis Ready Data (ARD) data is available online, with no charges for access and no restrictions on subsequent redistribution or use, as long as the proper citation is provided as specified by the Creative Commons Attribution License (CC BY).

 

Copyright © Global Land Analysis and Discovery Team, University of Maryland

 

Suggested citation

Global Land Analysis and Discovery team (GLAD). Landsat Analysis Ready Data. Downloaded from https://glad.umd.edu/ on 07/07/2020.

 

Suggested citation for ARD methodology

Potapov, P., Hansen, M.C., Kommareddy, I., Kommareddy, A., Turubanova, S., Pickens, A., Adusei, B., Tyukavina A., and Ying, Q., 2020. Landsat analysis ready data for global land cover and land cover change mapping. Remote Sens. 2020, 12, 426; doi:10.3390/rs12030426 (Potapov_RS_2020.pdf)

 

Disclaimer

While the GLAD makes every effort to ensure completeness and consistency of the Landsat ARD product, we acknowledge that it may contain faults and unreadable data. We ask that you notify us immediately of any problems in our data. We will make every effort to correct them.

 

With respect to the Landsat ARD data available from this service, the GLAD team does not make any warranty, express or implied, including the warranties of merchantability and fitness for a particular purpose; nor assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of the data products.

 

Although the GLAD team is committed to operational product update and to maintain open data access, our work is contingent on adequate funding and resources. Therefore, this service may be interrupted or canceled at any time without notice.

 

User Registration