U.S. Geological Survey
20200604
Remote Sensing Shrub/Grass National Land Cover Database (NLCD) Back-in-Time (BIT) Products for the Western U.S., 1985 - 2018
Raster Digital Data Set
Earth Resources Observation and Science Center, Sioux Falls, SD
U.S. Geological Survey
Rigge, M., C. Homer, H. Shi, and D. Meyer. 2019. Validating a Landsat time-series of fractional component cover across western U.S. rangelands. Remote Sensing 11, 3009.
RuleQuest Research. Cubist, version 2.05; Rule-Quest Pty, St Ives: New South Wales, Australia, 2008.
Shi, H.; Homer, C.; Rigge, M.; Postma, K.; Xian, G. 2020. Assessing fractional component change in a shrubland ecosystem with both long-term field observations and a Landsat time-series in Wyoming USA. Ecospher. in review.
Wylie, B. K., N. J. Pastick, J. J. Picotte, and C. A. Deering.2018. Geospatial data mining for digital raster mapping. GIScience & Remote Sensing 56:406–429.
https://doi.org/10.5066/P9C9O66W
https://www.mrlc.gov/data-services-page
https://www.mrlc.gov/data
Matthew Rigge
Hua Shi
Collin Homer
Patrick Danielson
Brian Granneman
20190422
Long-term trajectories of fractional component change in the Northern Great Basin, USA
Publication (Journal)
Ecosphere, Volume 10, Issue 6, June 2019
https://doi.org/10.1002/ecs2.2762
The need to monitor change in sagebrush steppe is urgent due to the increasing impacts of climate change, shifting fire regimes, and management practices on ecosystem health. Remote sensing provides a cost-effective and reliable method for monitoring change through time and attributing changes to drivers. We report an automated method of mapping rangeland fractional component cover over a large portion of the Northern Great Basin, USA, from 1986 to 2016 using a dense Landsat imagery time series. 2012 was excluded from the time-series due to a lack of quality imagery. Our method improved upon the traditional change vector method by considering the legacy of change at each pixel. We evaluate cover trends stratified by climate bin and assess spatial and temporal relationships with climate variables. Finally, we statistically evaluate the minimum time density needed to accurately characterize temporal patterns and relationships with climate drivers. Over the 30-yr period, shrub cover declined and bare ground increased. While few pixels had >10% cover change, a large majority had at least some change. All fractional components had significant spatial relationships with water year precipitation (WYPRCP), maximum temperature (WYTMAX), and minimum temperature (WYTMIN) in all years. Shrub and sagebrush cover in particular respond positively to warming WYTMIN, resulting from the largest increases in WYTMIN being in the coolest and wettest areas, and respond negatively to warming WYTMAX because the largest increases in WYTMAX are in the warmest and driest areas. These data can be used to answer critical questions regarding the influence of climate change and the suitability of management practices. Component products can be downloaded from www.mrlc.gov.
The goal of this project is to provide a dense Landsat imagery time series of shrubland fractional components. An inventory of land cover validated products and estimates of precision across the Western U.S from 1985 - 2018. 2012 was excluded from the time-series due to a lack of quality imagery. Climate change, shifting fire regimes, and management practices are increasingly impacting the health of the ecosystem. This dataset will fill the need for improved monitoring change within the large shrubland habitat.
Although this Federal Geographic Data Committee-compliant metadata file is intended to document the data set in nonproprietary form, as well as in Esri format, this metadata file may include some Esri-specific terminology.
19850501
20181031
ground condition
As needed
-130.2380
-99.6688
52.7905
26.2039
ISO 19115 Topic Category
biota
environment
geoscientificInformation
imageryBaseMapsEarthCover
USGS Thesaurus
shrubland ecosystems
terrestrial ecosystems
Alexandria Digital Library Feature Type Thesaurus
shrublands
time series
back-in-time
trends
grassland change
shrubland change
vegetation change
climate change
rangeland management
None
shrub
sagebrush
big sagebrush
herbaceous
annual herbaceous
litter
grass
vegetation
bare ground
rangeland
shrubland
Common geographic areas
United States
Washington
Oregon
Montana
North Dakota
South Dakota
Wyoming
Idaho
Nebraska
Nevada
Utah
California
Colorado
Arizona
Texas
New Mexico
None
NM
TX
AZ
CO
CA
UT
NV
NE
ID
WY
SD
ND
MT
OR
WA
Great Basin
Arizona Plateau
Black Hills
Blue Mountains
Chihuahuan Desert
Colorado Plateau
Columbia Plateau
Grand Canyon
Middle Rockies
Rocky Mountains
Gunnison
Sonoran Desert
Southwest Tablelands
Three Forks
Wasatch
Western US
Yellowstone
Northern Mountainous
Any downloading and use of these data signifies a user's agreement to comprehension and compliance of the USGS Standard Disclaimer. Ensure all portions of metadata are read and clearly understood before using these data in order to protect both user and USGS interests.
There is no guarantee of warranty concerning the accuracy of the data. Users should be aware that these data were developed from models which can contain some local error. Users should not use these data for critical applications without a full awareness of their limitations. Acknowledgement of the originating agencies would be appreciated in products derived from these data. Any user who modifies these data is obligated to describe the types of modifications they perform. User specifically agrees not to misrepresent the data, nor to imply that changes made were approved or endorsed by the U.S. Geological Survey. Please refer to https://www.usgs.gov/privacy.html for the USGS disclaimer.
U.S. Geological Survey
Customer Services Representative
mailing and physical
47914 252nd Street
Sioux Falls
SD
57198-0001
U.S.
605-594-6151
605-594-6589
custserv@usgs.gov
U.S. Geological Survey (USGS)
Bureau of Land Management (BLM)
Environment as of Metadata Creation: Microsoft Windows 7 Version 6.1 (Build 7601) Service Pack 1; Esri ArcGIS 10.5.1 (Build 7333) Service Pack N/A (Build N/A)
Two robust validation approaches have been applied used to evaluate the accuracy of BIT products. First, an approach capitalizing on field data collected concurrently with high-resolution satellite (HRS) images over multiple locations (n = 42) and years (Rigge et al. 2020). HRS sites spanned a broad range of vegetation, biophysical, climatic, and disturbance regimes in Wyoming, Nevada, and Montana. Field observations were used to train regression tree models, predicting the component cover across each HRS image. We evaluated the spatial and temporal relationships between HRS and BIT component cover and compare spatio-temporal climate responses. For each HRS site-year (n = 77) we averaged both the HRS and BIT predictions within each site separately and regressed the averages to quantify the temporal accuracy. Next, we regressed individual pixel values of corresponding HRS and BIT predictions to quantify the spatio-temporal accuracy. Results showed strong temporal correlations with an average R2 of 0.63 and Root Mean Square Error (RMSE) of 5.47% as well as strong spatio-temporal correlations with an average R2 of 0.52 and RMSE of 7.89% across components.
The second major validation approach for BIT is comparison of results to data from two long term monitoring sites in southwest Wyoming (Shi et al. In Review). The field data included ten years of consistent observation at 126 plots over the period of 2006-2008. Field observations and Landsat times-series predictions generally responded similarly to interannual variation in weather, chiefly driven by precipitation. The spatial-temporal correlation across all 126 plots showed robust correlations for all components, with an R2 of 0.69 for bare ground and 0.40 for shrub cover and averaging 0.46 across components. Changes observed in both the field and BIT data were typically gradual, within-state, changes which are most difficult to resolve, which were successfully captured.
The methods employed to map fractional vegetation components in shrubland ecosystems in the Western U.S. include: modelling shrublands as a series of independent continuous field components from 1985 – 2018 (2012 not included), and acquiring multiple seasons (summer and fall) of 172 path/rows of 30-meter Landsat imagery for large-area modelling using regression tree classification technology that optimizes data mining of multiple image dates, indices, and bands with ancillary data.
This fractional estimation of six shrubland habitat variables ranging from 1985 - 2018, in the Western U.S. is the version dated 20200424. Data set is considered complete for the information presented, as described in the abstract. Users are advised to read the rest of the metadata record carefully for additional details.
A formal accuracy assessment of the horizontal positional information in the data set has not been conducted.
A formal accuracy assessment of the vertical positional information in the data set has either not been conducted, or is not applicable.
The NLCD Back in Time (BIT) dataset quantified the percent cover of rangeland components across the western U.S. using Landsat imagery from 1985-2018. The BIT product suite consists of six fractional components: annual herbaceous, bare ground, herbaceous, litter, sagebrush and shrub, and the temporal trends of each. The four primary components bare ground, shrub, litter, and herbaceous sum to 100% in each pixel, the only deviation is when there are forested areas <40% coverage. The secondary components of annual herbaceous and sagebrush are subsets of the primary components of herbaceous and shrub. 2012 was excluded from the time-series due to a lack of quality imagery. The base products developed in October of 2019 were the baseline and the training data source of these data.
20200601
Image processing--
Landsat Collection 1 Level-2: Landsat C1 Level-2 Albers (U.S.)
The Higher-Level Albers Scene Bundle consists of Level-2 products at 30-meter resolution in the Albers Equal Area (AEA) map projection. Products include Top of Atmosphere (TOA)Reflectance, and Quality Assessment (QA), which are needed for BIT modeling.
For Landsat 7 ETM+ and Landsat 4-5 TM data, the processing flow for the creation of Level-2 Albers products starts with Collection 1 Tier 1 scenes processed through the Level-1 Product Generation System (LPGS) directly into the AEA map projection.
Landsat 8 products are likewise built from Collection 1 Tier 2 scenes. Science algorithms are then applied in the Level-2 Product Generation System (L2PGS), delivering Level-2 Albers products in the same WRS-2 path/row grid as Level 1 products, but in Albers projection.
Summer and fall Landsat 5,7 (SLC-on only), and 8 images were selected based on a balance of several competing criteria. First, that the date (but more importantly the phenology) of the summer and fall images of a given target year correspond to that of the base year summer and fall images. Second, that cloud, shadow, and snow cover, and active burns within the analysis extent of the image are minimized. Scenes were selected from 1985 – 2018 (no imagery for 2012). A total of 344 Landsat scenes were used for the mapping region covering the Western U.S.
20200211
Ancillary Data--
In addition to Landsat imagery, we included several ancillary variables in the regression modelling of fractional components. 1. Data related to topographic features, as independent variables, such as slope, aspect and a position index. 2. For each selected Landsat image, three spectral indices are calculated, the Soil Adjusted Vegetation Index (SAVI), Normalized Difference Built-up Index (NDBI), and Normalized Difference Water Index (NDWI). 3. Water bodies occurring at any point during the time series are detected using the NDWI. Areas identified as agriculture or water at any point in the time series are applied to all data throughout the training selection process and were included in the land cover mask. 4. The QA band for each selected image was converted to an fmask equivalent. The summer and fall fmask equivalents were stacked across all target years. 5. Monitoring Trends in Burn Severity (MTBS) and GeoMac burn data were used to produce a suite of burned area products.
20200331
Land Cover Masking
Products were masked using the base land cover mask (Rigge et al. 2020) based on pixels identified as non-rangeland in the circa 2016 base year. The base year land cover mask included forested areas with greater than 40% tree canopy cover in the NLCD 2016 fractional tree canopy cover product and additional forested pixels identified using NDVI or MSAVI thresholds (Rigge et al. 2020). Next, urban areas, major roads, snow, and ice were masked according to NLCD 2016 land cover classes. Third, cultivated crop and pasture/hay fields were masked using a combination of the 2013 Cropland Data Layer from the National Agricultural Statistics Service and the NLCD 2016 agricultural classes. Fourth, open water areas were identified using Normalized Difference Water Index (NDWI) thresholds. Supplemental hand edits were then applied where issues needed correction.
Some pixels identified as rangeland in the base year were previously (or later) non-rangeland land covers. We developed a series of models to detect and mask out pixels that were inundated with water or were actively farmed at any point in the time-series. To exclude water, we calculated the maximum NDWI through the time series. Pixels with zero topographic slope and a high NDWI value were added to the land cover mask. Next, we calculated a variety of spatial and temporal indicators based on summer and fall SAVI data. Specifically, we highlighted pixels with both a high maximum SAVI value and with a high amount of average variance between summer and fall as potential agriculture. We used elevation, land ownership, and ecoregion data to remove commission from pixels identified as agriculture. The water and agriculture masks were added to the base mask and applied throughout the process for each path row and to the final components.
20200331
Normalization--
An approach was developed to normalize the individual scenes in the same season to the base year image. Linear regression models were fit for the spatial relationship between band x of a given summer/fall target year image vs band x of the summer/fall image from the base year. Pixels with cloud, shadow, or snow cover and those that were recently burned were removed from the normalization regressions. This procedure is completed for all 6 spectral bands and for both seasons. The regression parameters of slope (a), y-intercept (b), and correlation (R²) are used to adjust the spectral band values of the target year image, so that spatial mean matched that of the base year. If the R² value of a given seasonal pair regression is less than 0.30, the scene is flagged, and a new image is selected. Normalization serves to minimize differences related to phenological variability, sun angle, and atmospheric contamination. Additionally, it helps to stabilize any difference related to Landsat sensor. The normalized imagery was only used for identifying change pixels, with non-normalized imagery reserved for the component predictions.
20200331
Change detection --
Change vector analysis –
Each normalized target and base image pair are compared using a Change Vector (CV) approach, finding differences in spectral values by band. These are summed across bands to produce a change vector magnitude value for each image pair. If a pixel has a change vector magnitude beyond a standard deviation threshold in a given target year image for a given NLCD 2016 land cover class, it is flagged as changed (Rigge et al. 2019). Two out of two seasons agreement is needed for a pixel to be flagged as changed between a target and base year. The requirement for two season agreements is critical to remove cloud, shadow, and haze influence missed by the QA bands.
Change fraction –
We use an approach which considers the range of CV values through time at each pixel. The CV magnitude range represents the extent of variability through time at each pixel, those pixels with stable conditions will have a low value, while those with disturbance will have a high value. We determine how each yearly CV magnitude compares to the range, allowing us to define the degree of confidence in that change. We define this metric as Change Fraction (CF). The CF considers three weighted attributes in change detection. First, the seasonal CV magnitude score for a given image in relation to the CV magnitude range per pixel. This gives a high value to pixels with a high amount of spectral change relative to the total range of variation in the time-series. Second, the CV magnitude range per pixel, relative to the maximum in each path-row. This gives a high value to pixels with a high total variation in the time-series relative to those pixels with lower variation through time. Third, the CV magnitude per pixel relative to the maximum CV magnitude in each path-row, each year. This gives a high value to pixels with a high amount of spectral change relative to the maximum change in that given image. The total score for each season is calculated and the maximum of the two seasons is given as the final CF. Values of CF range from 0 (high confidence unchanged) to 100 (high confidence changed).
Final change area –
We used a convergence of evidence approach to produce the final change area in each target year, considering both the CV and CF. The combination of CV and CF scenarios that best captured known change and minimized false change were selected based on testing multiple combinations in areas known to have changed and not changed.
Specifically, the criteria for change were as follows:
If CV = 1 and CF >34 or if CV = 0 and CF >80, (CV of 1 = change detected and 0 = no change detected). If either scenario is satisfied, then the pixel is flagged as change in that target year. Determination of change in pixels with cloud cover in one season is based on the CV and CF values from the non-cloudy season, if both seasons are cloudy, no change can be detected in that target year. This procedure produces a time-series of changed/unchanged areas which allow changed areas to be labeled with the appropriate fractional component cover and unchanged areas to be used to train target year predictions.
20200331
Training--
In larger patches (>5 pixels) of unchanged areas we can be more confident that the pixels are indeed unchanged, and therefore the base year predictions are appropriate to use as training. Within this potential training area, we randomly sample 200,000 pixels for every target year. To these points, we extract the dependent and independent variables, in addition to the SAVI temporal mean per pixel. Independent variables are slope, aspect, topographic position index, non-normalized target year Landsat images, spectral indices SAVI, NDWI, and NDBI, and summer thermal band of each target year. Next, we append all yearly sampled data files per components into a combined spatio-temporal data base. We apply two procedures to the training database to improve dynamic range through space and time. First, to improve the dynamic range through space, we randomly remove half the observations with 1 stand deviation of the mean for each component dependent variable (Wylie et al. 2018). Next to improve temporal dynamic range, we compare the ratio of the yearly summer SAVI conditions relative to the long-term mean. Wet years will have values over 100, dry years lower than 100. We again remove half of the observations within 1 standard deviation of the mean. The overall objective being to stretch out component histograms through space and time. Finally, we placed 30k points in recently burned areas, (those burned in the target year and preceding four years) to which we assigned a dependent variable value of 0% cover. We then appended those observations to train burned areas in the shrub and sagebrush components.
20200331
Base Adjustment Model --
We leverage the time-series data to improve the quality of the base year training data. Using training database for each component we make initial cubist predictions of component cover in the year preceding and following the base. We then take the mean of the predictions from the base year and two adjacent years. If there is no change in either adjacent year, and there have been no fires in the base and preceding 5 years, then the original base is replaced with the mean value, otherwise the original base is maintained. The final output of this procedure was to update the base year prediction for all components. We then repeated the procedures described in the prior section, extracting the adjusted base year training values.
20200331
Component Prediction—
Using the spatio-temporal training pool with the adjusted base values we developed Cubist regression tree models (RuleQuest Research 2008) for each component, using the component values as the dependent variable, and the full set of independent variables. We developed the Cubist models for the spatio-temporal training dataset but applied the model to data from each target year to produce component cover predictions. A second set of predictions was run for shrub and sagebrush, using training data that included the burned area observations. This burned area prediction for shrub and sagebrush was inserted in pixels that burned in the target year or preceding four years, while the standard prediction was kept elsewhere. In pixels with cloud or shadow cover in either season of the target year we filled the target year prediction with that of the prediction from the following target year.
20200331
Change Labeling --
In pixels meeting the criteria of change, described in the section above, we apply the target year prediction. If the pixels are identified as non-changed, the base year prediction is maintained. Next we limited component cover change in four situations. First, if component change between a given target and base year was greater than defined thresholds described in (Rigge et al. 2019) we lowered it to the limit. Next, we summed the primary fractional components (bare ground, litter, herbaceous, and shrub) in each pixel, which should fall within a range between 90 and 110%. We identified pixels with a summation over 110% or less than 90% and adjusted the bare ground value to sum to 90 and 110%, respectively. Third, we identified areas in which most pixels in a 5 by 5 moving window had +/- 1% component cover change between the base and target year. Similarly, we identified areas in which most pixels had no change, and less than 20% had +/- 1% component cover change. Since the change in both above scenarios is likely to be noise, we inserted the base year fractional component cover in these changed pixels, while the target year estimate was maintained in the remaining area. Fourth, we determined the total number of fractional components with change in each target year. If only 1 out of 6 fractional components in a pixel had change, the base year fractional component cover was maintained for all fractional components. The logic is that if one fractional component changes, then other component(s) should change in response. Finally, we ensured that fractional component relationships are intact (shrub > sagebrush, herbaceous > annual herbaceous). If not, we forced the relationships to be intact.
20200331
Post Processing –
We created a series of models to ensure accurate post-burn temporal trajectories in shrub and sagebrush. First, we created a time since most recent fire layer which resets to zero with every new burn and accrues with subsequent years without a burn. Pixels are treated as unburned until the time of first fire. We created second-order polynomial regressions of shrub and sagebrush cover limits (Rigge et al. 2019) by time since fire based on the literature and expert knowledge which allow more change than in the unburned area.
To correct for noise in the time series we developed a model to detect and remove outliers. We defined noise under two conditions, first, if target year x is unchanged relative to the base, but the previous and following target years are changed. We replaced the predicted value of the target year with the mean of the previous and following years. Second, we compared each target year prediction to the prediction of the previous year, to determine if change is beyond limits set by expert knowledge. The limits for increase between years was 100 (bare), >5 (shrub/sage), 100 (litter), and >15% (herb/annual herb). The limits of decrease between years was <-20 (bare), -100 (shrub/sage), <-10 (litter), and -100% (herb/annual herb). If change was beyond these limits, we replaced the predicted value with the limit. These noise removal procedures were implemented only in pixels that never burned.
20200331
Final Path Row Products
172 mapping Landsat path rows were mapped through this process. Any errors found were addressed before we began our mosaicking process.
20200402
Mosaicking
The final step was mosaicking all 172 Landsat path row components together. Using ERDAS mosaicking functionality we “stitched” all path rows together to create the final products. We generated automated seamlines to place them in overlap areas with the lowest value component value difference. We used a feathering process of 990m on either side of the cutline. After the ERDAS processing we did a visual inspection and addressed any seam lines or data voids within the mapping region. The land cover mask was also applied to smooth any issues between the path rows. The pixel counts were reconciled for all components for the mask value of 101 and the area outside the mapping region with a value of 102. The secondary components were reconciled to the primary components for sagebrush and shrub, and annual herbaceous to herbaceous. Annual herbaceous could not have a higher density than herbaceous, and sagebrush a higher density than shrub. This process resulted in a wall-to-wall prediction from the mid U.S. to the western edge of the U.S. encompassing the largest area of shrublands.
20200424
Mosaic Trends
We calculated temporal trends in each component across the time-series using linear models applied to the mosaicked products. For each pixel, we calculated the slope, t-score, average change, and standard errors.
20200429
Raster
Grid Cell
87666
71204
1
Albers Conical Equal Area
29.5
45.5
-96.0
23.0
0.0
0.0
row and column
30.0
30.0
meters
WGS_1984
WGS 84
6378137.0
298.257223563
Shrub BIT Attribute Table
Table containing attribute information associated with each of the Shrub BIT data sets.
Producer Defined
Value
Column for value indicating per-pixel percent for bare ground, shrub, herbaceous, litter, sagebrush, and annual herbaceous, component range from 0 to 100, a value of 101 for the masked areas, and a value of 102 for outside mapping data areas.
Producer defined
0
102
Percent
Count
All raster attribute tables include a column for count describing a nominal integer value that designates the number of pixels that have each value in the file; histogram column in ERDAS Imagine raster attributes table.
Producer defined
0
2919282288
Integer
Red
Red color code for RGB. The value is arbitrarily assigned by the display software package, unless defined by user.
Producer defined
0
1
Percent
Green
Green color code for RGB. The value is arbitrarily assigned by the display software package, unless defined by user.
Producer defined
0
1
Percent
Blue
Blue color code for RGB. The value is arbitrarily assigned by the display software package, unless defined by user.
Producer defined
0
1
Percent
Opacity
A measure of how opaque, or solid, a color is displayed in a layer.
Producer defined
0
1
Percent
The entity and attribute information provided here describes the tabular data associated with the data set. Please review the detailed descriptions that are provided (the individual attribute descriptions) for information on the values that appear as fields/table entries of the data set.
The entity and attribute information was generated by the individual and/or agency identified as the originator of the data set. Please review the rest of the metadata record for additional details and information.
U.S. Geological Survey
GS ScienceBase
mailing address
Denver Federal Center, Building 810, Mail Stop 302
Denver
CO
80225
United States
1-888-275-8747
sciencebase@usgs.gov
Unless otherwise stated, all data, metadata and related materials are considered to satisfy the quality standards relative to the purpose for which the data were collected. Although these data and associated metadata have been reviewed for accuracy and completeness and approved for release by the U.S. Geological Survey (USGS), no warranty expressed or implied is made regarding the display or utility of the data for other purposes, nor on all computer systems, nor shall the act of distribution constitute any such warranty.
Digital Data
https://doi.org/10.5066/P9C9O66W
None
20200603
U.S. Geological Survey
Customer Services Representative
mailing and physical
47914 252nd Street
Sioux Falls
SD
57198-0001
U.S.
605-594-6151
605-594-6589
custserv@usgs.gov
FGDC Content Standard for Digital Geospatial Metadata
FGDC-STD-001-1998