U.S. Geological Survey scientists, funded by the Climate and Land Use Change Research and Development Program, developed a dataset of 2006 and 2011 land use and land cover (LULC) information for selected 100-km2 sample blocks within 29 EPA Level 3 ecoregions across the conterminous United States. The data was collected for validation of new and existing national scale LULC datasets developed from remotely sensed data sources. The data can also be used with the previously published Land Cover Trends Dataset: 1973-2000 (http:// http://pubs.usgs.gov/ds/844/), to assess land-use/land-cover change in selected ecoregions over a 37-year study period. LULC data for 2006 and 2011 was manually delineated using the same sample block classification procedures as the previous Land Cover Trends project. The methodology is based on a statistical sampling approach, manual classification of land use and land cover, and post-classification comparisons of land cover across different dates. Landsat Thematic Mapper, and Enhanced Thematic Mapper Plus imagery was interpreted using a modified Anderson Level I classification scheme. Landsat data was acquired from the National Land Cover Database (NLCD) collection of images. For the 2006 and 2011 update, ecoregion specific alterations in the sampling density were made to expedite the completion of manual block interpretations. The data collection process started with the 2000 date from the previous assessment and any needed corrections were made before interpreting the next two dates of 2006 and 2011 imagery. The 2000 land cover was copied and any changes seen in the 2006 Landsat images were digitized into a new 2006 land cover image. Similarly, the 2011 land cover image was created after completing the 2006 delineation. Results from analysis of these data include ecoregion based statistical estimates of the amount of LULC change per time period, ranking of the most common types of conversions, rates of change, and percent composition. Overall estimated amount of change per ecoregion from 2001 to 2011 ranged from a low of 370 km2 in the Northern Basin and Range Ecoregion to a high of 78,782 km2 in the Southeastern Plains Ecoregion. The Southeastern Plains Ecoregion continues to encompass the most intense forest harvesting and regrowth in the country. Forest harvesting and regrowth rates in the southeastern U.S. and Pacific Northwest continued at late 20th century levels. The land use and land cover data collected by this study is ideally suited for training, validation, and regional assessments of land use and land cover change in the U.S. because it is collected using manual interpretation techniques of Landsat data aided by high resolution photography. The 2001-2011 Land Cover Trends Dataset is provided in an Albers Conical Equal Area projection using the NAD 1983 datum. The sample blocks have a 30-meter resolution and file names follow a specific naming convention that includes the number of the ecoregion containing the block, the block number, and the Landsat image date. The data files are organized by ecoregion, and are available in the ERDAS Imagine (.img) format. U.S. Geological Survey scientists, funded by the Climate and Land Use Change Research and Development Program, developed a dataset of 2006 and 2011 land use and land cover (LULC) information for selected 100-km2 sample blocks within 29 EPA Level 3 ecoregions across the conterminous United States. The data was collected for validation of new and existing national scale LULC datasets developed from remotely sensed data sources. The data can also be used with the previously published Land Cover Trends Dataset: 1973-2000 (http:// http://pubs.usgs.gov/ds/844/), to assess land-use/land-cover change in selected ecoregions over a 37-year study period. LULC data for 2006 and 2011 was manually delineated using the same sample block classification procedures as the previous Land Cover Trends project. The methodology is based on a statistical sampling approach, manual classification of land use and land cover, and post-classification comparisons of land cover across different dates. Landsat Thematic Mapper, and Enhanced Thematic Mapper Plus imagery was interpreted using a modified Anderson Level I classification scheme. Landsat data was acquired from the National Land Cover Database (NLCD) collection of images. For the 2006 and 2011 update, ecoregion specific alterations in the sampling density were made to expedite the completion of manual block interpretations. The data collection process started with the 2000 date from the previous assessment and any needed corrections were made before interpreting the next two dates of 2006 and 2011 imagery. The 2000 land cover was copied and any changes seen in the 2006 Landsat images were digitized into a new 2006 land cover image. Similarly, the 2011 land cover image was created after completing the 2006 delineation. Results from analysis of these data include ecoregion based statistical estimates of the amount of LULC change per time period, ranking of the most common types of conversions, rates of change, and percent composition. Overall estimated amount of change per ecoregion from 2001 to 2011 ranged from a low of 370 square km in the Northern Basin and Range Ecoregion to a high of 78,782 square km in the Southeastern Plains Ecoregion. The Southeastern Plains Ecoregion continues to encompass the most intense forest harvesting and regrowth in the country. Forest harvesting and regrowth rates in the southeastern U.S. and Pacific Northwest continued at late 20th century levels. The land use and land cover data collected by this study is ideally suited for training, validation, and regional assessments of land use and land cover change in the U.S. because it’s collected using manual interpretation techniques of Landsat data aided by high resolution photography. The 2001-2011 Land Cover Trends Dataset is provided in an Albers Conical Equal Area projection using the NAD 1983 datum. The sample blocks have a 30-meter resolution and file names follow a specific naming convention that includes the number of the ecoregion containing the block, the block number, and the Landsat image date. The data files are organized by ecoregion, and are available in the ERDAS Imagine (.img) format.

The U.S. Geological Survey (USGS), in partnership with several federal agencies, has developed and released five National Land Cover Database (NLCD) products over the past two decades: NLCD 1992, 2001, 2006, 2011, and 2016. The 2016 release saw landcover created for additional years of 2003, 2008, and 2013. These products provide spatially explicit and reliable information on the Nation’s land cover and land cover change. To continue the legacy of NLCD and further establish a long-term monitoring capability for the Nation’s land resources, the USGS has designed a new generation of NLCD products named NLCD 2019. The NLCD 2019 design aims to provide innovative, consistent, and robust methodologies for production of a multi-temporal land cover and land cover change database from 2001 to 2019 at 2–3-year intervals. Comprehensive research was conducted and resulted in developed strategies for NLCD 2019: continued integration between impervious surface and all landcover products with impervious surface being directly mapped as developed classes in the landcover, a streamlined compositing process for assembling and preprocessing based on Landsat imagery and geospatial ancillary datasets; a multi-source integrated training data development and decision-tree based land cover classifications; a temporally, spectrally, and spatially integrated land cover change analysis strategy; a hierarchical theme-based post-classification and integration protocol for generating land cover and change products; a continuous fields biophysical parameters modeling method; and an automated scripted operational system for the NLCD 2019 production. The performance of the developed strategies and methods were tested in twenty composite referenced areas throughout the conterminous U.S. An overall accuracy assessment from the 2016 publication give a 91% overall landcover accuracy, with the developed classes also showing a 91% accuracy in overall developed. Results from this study confirm the robustness of this comprehensive and highly automated procedure for NLCD 2019 operational mapping. Questions about the NLCD 2019 land cover product can be directed to the NLCD 2019 land cover mapping team at USGS EROS, Sioux Falls, SD (605) 594-6151 or mrlc@usgs.gov. See included spatial metadata for more details.

The Land Change Monitoring Assessment and Projection (LCMAP) raster dataset is a suite of five annual land surface change and five annual land cover (and land cover derivative) products. The LCMAP approach is the foundation for an integrated land change science framework led by the U.S. Geological Survey (USGS). The data were calculated using the Continuous Change Detection and Classification (CCDC) algorithm developed by Zhu and Woodcock (2014) and are derived from a time series of satellite imagery consisting of all available cloud- and shadow-free pixels in the USGS Landsat Analysis Ready Data (ARD) archive (Dwyer and others, 2018). The CCDC methodology supports the continuous tracking and characterization of changes in land cover, and condition enabling assessments of current, historical, and future processes of change. Landsat ARD, as the source data for LCMAP, are standardized Landsat data pre-processed to ensure the data meet a minimum set of requirements and are organized into a form that allows immediate analysis with a minimum of additional user effort. ARD data are provided as tiled, georegistered, surface reflectance products defined in a common equal area projection and tiled to a common grid. ARD observations must be transformed into time series vectors before further calculations using the CCDC methodology. The CCDC methodology, initially developed at Boston University (Zhu and Woodcock, 2014), has been adopted and modified by USGS for LCMAP. CCDC involves harmonic modeling that characterizes the seasonality, trends, and breaks from those trends based on the time series spectral reflectance data from multiple Landsat bands (i.e., green, red, near-infrared, short-wave infrared). The CCDC approach involves two major components: change detection and classification. The change detection component utilizes available high-quality surface reflectance data in a pixel-based time series to calculate a mathematical model for the spectral response of each pixel and to estimate the dates at which the spectral time series data diverge from past responses or patterns. The basis of change detection is the comparison of clear satellite observations with model predictions. 'Divergence' (referred to as a model 'break') often is identified as the result of an abrupt change (e.g. wildfire, logging, mining, and urban development) but may also result from a gradual shift (e.g., forest regrowth, insect infestation, disease) in the spectral signal over time. Breaks are detected by CCDC by applying a criterion based on the root mean square error of the harmonic modeling. Time periods for established models are referred to as 'model segments.' After a break is identified in the time series, a new model can be established following the break provided there are enough clear observations going forward in time. The classification component of CCDC involves using the coefficients of time series models as the inputs for land cover classification. The CCDC method has the capability to generate land cover for any date in the time series; the USGS has selected an annual time step for land cover classification. The suite of land cover and change products are nominally identified at a central point in the year, July 1. Classification is performed using a boosted decision tree method based on training data developed from 2001 NLCD land cover classes (Homer and others, 2007). The land cover legend for the Primary and Secondary Land Cover products is comparable to an Anderson level 1 classifcation scheme.

This is a dataset download, not a document. The Open button will start the download.This data layer is an element of the Oregon GIS Framework and has been clipped to the Oregon boundary and reprojected to Oregon Lambert (2992). The U.S. Geological Survey (USGS), in partnership with several federal agencies, has developed and released four National Land Cover Database (NLCD) products over the past two decades: NLCD 1992, 2001, 2006, and 2011. These products provide spatially explicit and reliable information on the Nation’s land cover and land cover change. To continue the legacy of NLCD and further establish a long-term monitoring capability for the Nation’s land resources, the USGS has designed a new generation of NLCD products named NLCD 2016. The NLCD 2016 design aims to provide innovative, consistent, and robust methodologies for production of a multi-temporal land cover and land cover change database from 2001 to 2016 at 2–3-year intervals. Comprehensive research was conducted and resulted in developed strategies for NLCD 2016: a streamlined process for assembling and preprocessing Landsat imagery and geospatial ancillary datasets; a multi-source integrated training data development and decision-tree based land cover classifications; a temporally, spectrally, and spatially integrated land cover change analysis strategy; a hierarchical theme-based post-classification and integration protocol for generating land cover and change products; a continuous fields biophysical parameters modeling method; and an automated scripted operational system for the NLCD 2016 production. The performance of the developed strategies and methods were tested in twenty World Reference System-2 path/row throughout the conterminous U.S. An overall agreement ranging from 71% to 97% between land cover classification and reference data was achieved for all tested area and all years. Results from this study confirm the robustness of this comprehensive and highly automated procedure for NLCD 2016 operational mapping. Questions about the NLCD 2016 land cover product can be directed to the NLCD 2016 land cover mapping team at USGS EROS, Sioux Falls, SD (605) 594-6151 or mrlc@usgs.gov. See included spatial metadata for more details.

This dataset provides two 30-m resolution time series products of annual land cover classifications over the Arctic Boreal Vulnerability Experiment (ABoVE) core domain for each year of the period 1984-2014. The data are the annual dominant plant functional type in a given 30-m pixel derived from Landsat surface reflectance, landcover training data mapped across the ABoVE domain (using Random Forests modeling, with clustering and interpretation of field photography) and very high resolution imagery to assign land cover classifications. One product has a 15-class land cover classification that breaks out forest and shrub types into several additional classes; the other product provides a simplified, 10-class approach. Classification accuracy assessment results are provided per year. Assessments were based on a probability-based random sample of reference data that supported statistically robust estimation of areas and uncertainties in mapped areas.

The "Sen-2 LULC Dataset" is a collection of 2,13,750+ pre-processed 10 m resolution images representing 7 distinct classes of Land Use Land Cover. The 7 classes are water, Dense forest, Sparse forest, Barren land, Built up, Agriculture land and Fallow land. Multiple classes are present in the single image of the dataset. The Sentinel-2 images of Central India are taken from USGS (United States Geological Survey) EXPLORER (https://earthexplorer.usgs.gov/) with cloud clover percentage ranging from 0 to 0.5. The images are combination of bands 3, 4 and 5 constituting the red, green and blue bands with spectral resolution of 10m. The images are taken within the months of March and April 2022. The images used in the dataset belongs to Sentinel-2 Level-2A product (https://sentinels.copernicus.eu/web/sentinel/user-guides/sentinel-2-msi/product-types/level-2a#:~:text=The%20Level%2D2A%20product%20provides,(UTM%2FWGS84%20projection).). The dataset contains equal number of mask images. The dataset contains 6 folders with train, test and validate images and train, test and validate masks. This dataset can be used for Land Use Land Cover Classification (LULC) of Indian region to build the deep learning models. This dataset is beneficial for LULC classification research.

Harmonized training samples based on LUCAS (Land Use and Coverage Area frame Survey) and CLC (CORINE Land Cover) Maps, according to the following classes:

• 111 - Urban fabric: the aggregated continuous and discontinuous urban fabric class is classified when urban structures are dominating the surface area. The impermeable features like buildings and artificially surfaced areas range from 30 to 100% land coverage.
• 122 - Road and rail networks and associated land: motorways and railways, including associated installations.
• 123 - Port areas: infrastructure of port areas, including quays, dockyards and marinas.
• 124 - Airports: airports installations as runways, buildings and associated land.
• 131 - Mineral extraction sites: areas of open-pit extraction of construction materials (sandpits, quarries) or other minerals (open-cast mines).
• 132 - Dump sites: public, industrial or mine dump sites.
• 133 - Construction sites: spaces under construction development, soil or bedrock excavations, earthworks.
• 141 - Green urban areas: areas with vegetation within urban fabric.
• 211 - Non-irrigated arable land: cultivated land parcels under rain-fed agricultural use for annually harvested non-permanent crops, normally under a crop rotation system.
• 212 - Permanently irrigated arable land: cultivated land parcels under agricultural use for arable crops that are permanently or periodically irrigated, using a permanent infrastructure (irrigation channels, drainage network and additional irrigation facilities).
• 213 - Rice fields: cultivated land parcels prepared for rice production, consisting of periodically flooded flat surfaces with irrigation channels.
• 221 - Vineyards: areas planted with vines.
• 222 - Fruit trees and berry plantations: cultivated parcels planted with fruit trees and shrubs, including nuts, intended for fruit production.
• 223 - Olive groves: cultivated areas planted with olive trees, including mixed occurrence of vines on the same parcel.
• 231 - Pastures: meadows where dispersed trees and shrubs may occupy up to 50% of surface. These meadows are characterized by rich floristic composition.
• 311 - Broad-leaved forest: vegetation formation composed principally of trees, including shrub and bush understorey, where broad-leaved species predominate.
• 312 - Coniferous forest: vegetation formation composed principally of trees, including shrub and bush understorey, where coniferous species predominate.
• 321 - Natural grasslands: grasslands under no or moderate human influence. Low productivity grasslands. Often situated in areas of rough, uneven ground, frequently include rocky areas, or patches of other (semi-)natural vegetation.
• 322 - Moors and heathland: vegetation with low and closed cover, dominated by bushes, shrubs (heather, briars, broom, gorse, laburnum, etc.
• 323 - Sclerophyllous vegetation: bushy sclerophyllous vegetation in a climax stage of development, including maquis, matorral and garrigue.
• 324 - Transitional woodland-shrub: transitional bushy and herbaceous vegetation with occasional scattered trees. Can represent either woodland degradation or forest regeneration / re-colonization
• 331 - Beaches, dunes, sands: unvegetated expanses of sand or pebble/gravel, in coastal or continental locations, like beaches, dunes, gravel pads.
• 332 - Bare rocks: scree, cliffs, rock outcrops, including areas of active erosion.
• 333 - Sparsely vegetated areas: Areas with sparse vegetation, covering 10-50% of the surface.
• 334 - Burnt areas: Areas affected by recent fires.
• 335 - Glaciers and perpetual snow: Land covered by ice or permanent snowfields
• 411 - Inland wetlands: low-lying land usually flooded in winter, and more or less saturated by water all year round; and wetlands with accumulation of considerable amount of decomposed moss and vegetation matter.
• 421 - Maritime wetlands: vegetated low-lying areas in the coastal zone, above the high-tide line, susceptible to flooding by seawater; salt-pans for extraction of salt from salt water; and coastal zones under tidal influence between open sea and land.
• 511 - Water courses: natural or artificial water courses serving as water drainage channels
• 512 - Water bodies: natural or artificial water surfaces covered by standing water most of the year
• 521 - Coastal lagoons: stretches of salt or brackish water in coastal areas which are separated from the sea by a tongue of land or other similar topography
• 522 - Estuaries: the mouth of a river under tidal influence within which the tide ebbs and flows.
• 523 - Sea and ocean: zone seaward of the lowest tide limit.

The samples were obtained from the geographic location of LUCAS (in-situ source) and the centroid of all polygons from CLC maps, harmonized according to the above classes and organized by year, where each unique combination of longitude, latitude and year was considered as a independent sample. Some specific CORINE samples (i.e. 111, 122, 131, 141, 211, 221, 222, 223, 231, 311, 312, 321, 411, 512) were filtered according to convergence with existing mapping products (OSM roads, railways and buildings; Copernicus High-Res. Layers - HRL), where, for example, “111 - Urban fabric” samples located in low density building areas (> 50% according to Copernicus-OSM building layer) were removed. The LUCAS points with a unique land-cover class received a confidence rating of 100%, while CORINE received 85%. Using these filtered samples a spacetime overlay was performed (check the code demonstration in eumap library) considering several raster layers for Continental Europe: four season quantiles for GLAD Landsat ARD (spectral bands and indices - Potapov, 2020), DTM-based elevation and slope (Hengl, 2020), VIIRS/SUOMI NPP night lights (Hillger, 2013), Global surface water frequency (Pekel, 2016), and geometric minimum and maximum temperature derived according to Kilibarda, 2014.

The provided samples has 5,362,229 rows and 263 columns, including the geographic location, the metadata of land cover harmonization, the mapped classes according to OSM and HRL, and the result of the spacetime overlay (178 covariates).

Use the following Python/R code to open the files lcv_landcover.hcl_lucas.corine_harm.samples.overlaid.*

import joblib
samples = joblib.load('lcv_landcover.hcl_lucas.corine_harm.samples.overlaid')

readRDS('lcv_landcover.hcl_lucas.corine_harm.samples.overlaid.rds')

These samples were used to train a spatiotemporal model, which predicts the land cover for continental Europe over 20 years (2000 - 2019). To access the predictions results (dominant class, probabilities and uncertainties) use the following services:

• Open Data Science Europe viewer: https://maps.opendatascience.eu
• S3 Cloud Object: https://medium.com/swlh/europe-from-above-space-time-machine-learning-reveals-our-changing-environment-1b05cb7be520

A publication describing, in detail, all processing steps, accuracy assessment and general analysis of land-cover changes in continental Europe is under preparation. To suggest any improvement/fix use https://gitlab.com/geoharmonizer_inea/spatial-layers/-/issues

Overcoming the obstacle of frequent cloud coverage in optical remote sensing data is essential for monitoring dynamic land surface processes from space. APiC, a novel adaptable pixel-based compositing and classification approach, is especially designed to use high resolution spatio-temporal space-borne data. Here, pixel-based compositing is used separately for training data and prediction data. First, cloud-free pixels covered by reference data are used within adapted composite periods to compile a training dataset. The compiled training dataset contains samples of spectral reflectances for respective land cover classes at each composite period. For land cover prediction, pixel-based compositing is then applied region-wide. Multiple prediction models are used based on temporal subsets of the compiled training dataset to dynamically account for cloud coverage at pixel level. Thus we present a data-driven classification approach which is applicable in regions with different weather conditions, species composition and phenology. The capability of our method is demonstrated by mapping 19 land cover classes across Germany for the year 2016 based on Sentinel-2A data. Since climatic conditions and thus plant phenology change on a large scale, the classification was carried out separately in six landscape regions of different biogeographical characteristics. The study drew on extensive ground validation data provided by the federal states of Germany. For each landscape region, composite periods of different lengths have been established, which differ regionally in their temporal arrangement as well as in their total number, emphasising the advantage of a flexible regionalised classification procedure. Using a random forest classifier and evaluating outcomes with independent reference data, an overall accuracy of 88% was achieved, with particularly high classification accuracy of around 90% for the major land cover types. We found that class imbalances have significant influence on classification accuracy. Based on multiple temporal subsets of the compiled training dataset, over 10,000 random forest models were calculated and their performance varied considerably across and within landscape regions. The calculated importance of composite periods show that a high temporal resolution of the compiled training dataset is necessary to better capture the different phenology of land cover types. In this study we demonstrate that APiC, due to its data-driven nature, is a very flexible compositing and classification approach making efficient use of dense satellite time series in areas with frequent cloud coverage. Hence, regionalisation can be given greater focus in future broad-scale classifications in order to facilitate better integration of small-scale biophysical conditions and achieve even better results in detailed land cover mapping.

Using 335,709 Landsat images on the Google Earth Engine, we built the first Landsat-derived annual land cover product of China (CLCD) from 1985 to 2019. We collected the training samples by combining stable samples extracted from China's Land-Use/Cover Datasets (CLUD), and visually-interpreted samples from satellite time-series data, Google Earth and Google Map. Several temporal metrics were constructed via all available Landsat data and fed to the random forest classifier to obtain classification results. A post-processing method incorporating spatial-temporal filtering and logical reasoning was further proposed to improve the spatial-temporal consistency of CLCD.

"*_albert.tif" are projected files via a proj4 string "+proj=aea +lat_1=25 +lat_2=47 +lat_0=0 +lon_0=105 +x_0=0 +y_0=0 +datum=WGS84 +units=m +no_defs".

CLCD in 2023 is now available.

1. Given that the USGS no longer maintains the Landsat Collection 1 data, we are now using the Collection 2 SR data to update the CLCD.

2. All files in this version have been exported as Cloud Optimized GeoTIFF for more efficient processing on the cloud. Please check here for more details.

3. Internal overviews and color tables are built into each file to speed up software loading and rendering.

MULTI-TEMPORAL REMOTE SENSING IMAGE CLASSIFICATION - A MULTI-VIEW APPROACH

VARUN CHANDOLA* AND RANGA RAJU VATSAVAI*

Abstract. Multispectral remote sensing images have been widely used for automated land use and land cover classification tasks. Often thematic classification is done using single date image, however in many instances a single date image is not informative enough to distinguish between different land cover types. In this paper we show how one can use multiple images, collected at different times of year (for example, during crop growing season), to learn a better classifier. We propose two approaches, an ensemble of classifiers approach and a co-training based approach, and show how both of these methods outperform a straightforward stacked vector approach often used in multi-temporal image classification. Additionally, the co-training based method addresses the challenge of limited labeled training data in supervised classification, as this classification scheme utilizes a large number of unlabeled samples (which comes for free) in conjunction with a small set of labeled training data.

Overview

A geospatial dataset of point geometries with a land use / land cover label and several remote-sensing derived predictor variables that can be used to train and test a land use / land cover classifier.

This dataset was generated with support from a Climate Change AI Innovation Grant and the Australian Centre for International Agricultural Research.

Each of the point geometries was assigned one of the following class labels:

1. water
2. mangrove
3. bare soil / rock
4. urban / impervious
5. cropland / agriculture
6. grassland
7. shrubland
8. trees

The class property associated with each POINT feature stores the point's class label.

Class definitions

The cropland / agriculture class is defined as any location where agricultural activities associated with cropping or livestock management were visible in high-resolution images. Land that is recently fallow, but where evidence of cropping or grazing activities is present, would be labelled as cropland. Grassland is defined as any low vegetation (e.g. below knee height) without a bush, shrub, or woody structure. Scrubland is defined as any vegetation that is below head height, does not form a closed canopy, and has a clearly visible bush, shrub, or woody structure. Trees are defined as any vegetation greater than head height forming a clear canopy.

Methods

Image interpretation and labelling points with a land cover class was undertaken within a custom Google Earth Engine application. Within a region of interest, a year’s worth of Sentinel-2 images was clustered into 15 classes using a k-means algorithm. A stratified random sample of points was generated for manual labelling using clusters as strata. Ground truth datasets ere generated in the Ba, Magodro, Rewa, Sigatoka, RakiRaki, Sigatoka, Suva, Suva (urban), Lautoka (urban), Noco, Vuya, Nadi, and Labasa regions.

To support image interpretation and labelling a point’s land cover using high-resolution images (Google satellite basemaps), ancillary datasets were used (e.g. Planet and Sentinel-2 images) in conjunction with field verification.

Two quality-checks were applied to the labelled land cover points. First, each point was manually screened and quality checked to ensure consistency in class labels. Second, using Planet NICFI basemaps and Sentinel-2 RGB composites 2019, 2020, and 2021, each of the labelled land cover points was screened for a change in land cover event occurring at any point during those three years. If a change in land cover was observed, the point was dropped from the dataset.

For each labelled point in 2019, 2020, and 2021 features were extracted comprising annual median cloud free spectral reflectance across Sentinel-2 wavebands, monthly NDVI composites, and annual median NDVI, NDBI, NDWI, and GCVI bands and elevation, slope, and aspect bands.

This resulted in a dataset of 13,914 labelled points across three years: 2019, 2020, and 2021. The difference in the number of points across years is due to cloud cover preventing features being generated in some years

Feature definitions:

• B_* - median annual cloud free spectral reflectance for Sentinel-2 wavebands
• ndvi - median annual cloud free NDVI computed from Sentinel-2
• gcvi - median annual cloud free GCVI computed from Sentinel-2
• ndwi - median annual cloud free NDWI computed from Sentinel-2
• ndbi - median annual cloud free NDBI computed from Sentinel-2
• ndvi_* - median monthly cloud free NDVI computed from Sentinel-2
• elevation - elevation computed from SRTM
• aspect - aspect computed from SRTM
• slope - slope computed from SRTM

The U.S. Geological Survey (USGS), in association with the Multi-Resolution Land Characteristics (MRLC) Consortium, produces the National Land Cover Database (NLCD) for the United States. The MRLC, a consortium of federal agencies who coordinate and generate consistent and relevant land cover information at the national scale for a wide variety of environmental, land management, and modeling applications, have been providing the scientific community with detailed land cover products for more than 30 years. Over that time, NLCD has been one of the most widely used geospatial datasets in the U.S., serving as a basis for understanding the Nation’s landscapes in thousands of studies and applications, trusted by scientists, land managers, students, city planners, and many more as a definitive source of U.S. land cover. NLCD land cover suite is created through the classification of Landsat imagery and uses partner data from the MRLC Consortium to help refine many of the land cover classes. The classification system used by NLCD is modified from the Anderson Land Cover Classification System. The NLCD Class Legend and Description is maintained at https://www.mrlc.gov/data/legends/national-land-cover-database-class-legend-and-description. The land cover theme includes two separate products. The first is a standard land cover product suite that provides 16 land cover classes for the conterminous United States and Alaska only land cover types and is available at https://www.mrlc.gov/data. The second product suite, NLCD Land Cover Science Products, provides additional discrimination and land cover classes differentiating grass and shrub and regenerating forest regime from grass and shrub and rangeland setting and is available at https://www.mrlc.gov/nlcd-2021-science-research-products. The latest release of NLCD land cover spans the timeframe from 2001 to 2021 in 2 to 3-year intervals. These new products use a streamlined compositing process for assembling and preprocessing Landsat imagery and geospatial ancillary datasets; a temporally, spectrally, and spatially integrated land cover change analysis strategy; a theme-based post-classification protocol for generating land cover and change products; a continuous fields biophysical parameters modeling method; and a scripted operational system. Unmasked Impervious - To produce the unmasked impervious layer a multilayered perceptron neural network (MLP) was deployed across CONUS. The MLP was trained to perform the regression task of predicting the 1-100 impervious fractional cover. To sample data to train the network, we broke CONUS into a grid comprised of 256x256 pixel regions of interest (ROIs) and sampled from that grid all ROIs with at least 40% impervious cover according to NLCD 2019 impervious fractional cover, which gave us samples from large impervious areas. From those ROIs, we then sampled 66 million training and 16 million validation data points with an even distribution across each impervious intensity (1-100). Those training points were then randomly split into 4 subsets, each corresponding to one of the following respective years: 2011, 2013, 2016, 2019. We used those points to query surface reflectance values from leaf-on composite and leaf-off synthetic imagery (see metadata for NLCD 2021 land cover), elevation data, and spatial urban intensity probabilities. The spatial urban intensity probabilities were generated by an ensemble of U-net models that were trained to predict the 4 urban intensity classes as defined by the NLCD product legend (open space, low intensity, medium intensity, high intensity). Two U-net models were trained using all ROIs in the CONUS 256x256 pixel grid. Inputs to these models included leaf-on composite and leaf-off synthetic imagery, and elevation data. To create the final training and validation datasets we randomly split the CONUS grid into to 2 equal sets: A and B. Using the ROIs from set A we queried the input features from the years 2011 and 2016 and from the ROIs in set B we queried input features from the years 2013 and 2019. These U-net models do not act as the final impervious predictors but instead as spatial feature generators. The spatial features learned by these convolutional neural networks were then fed into the pixel-based MLP, as spatial probabilities of urban intensity, to boost its predicting power. The U-nets were trained using categorical focal Jaccard loss and monitored with the Jaccard Index metric (IOU). The impervious fractional cover regression model (MLP) was trained using mean squared error as a loss function and monitored with mean absolute error as the metric. Initial impervious footprint - To generate an initial impervious footprint, three U-net models were trained on the multiclass-classification task of predicting “urban” and “roads”. The model was trained with 120,000 training and 40,000 validation 256X256 pixel Landsat image chips covering the entire extent of CONUS. The model inputs are consistent with what was used to generate the urban intensity U-net models; the only difference was the target mask the models were trained to predict. These models mapped all NLCD impervious footprint pixels to two classes (“urban” and “roads”); this was used to generate the impervious extent. Impervious Change Pixels - The initial 2021 impervious change pixels were created by comparing the 2021 urban footprint with the 2019 published urban descriptor and extracting the difference. These change pixels were manually edited for omission and commission errors. Ancillary data were then added to the change pixels to create the final 2021 impervious change pixels. These ancillary data consisted of solar installations, wind turbines, and roads. The solar installations dataset is an edited version of the Solar Photovoltaic Generating Units dataset produced by Kruitwagen et al (2021) (https://doi.org/10.5281/zenodo.5005867). The U.S. Wind Turbine Database from Hoen et al (2021) (https://doi.org/10.5066/F7TX3DN0) was used without edits. NavStreets road datasets were used in previous versions of NLCD but an updated version was not available to the USGS. New subdivision roads from the 2021 urban footprint and a small number of manually drawn roads were added to the 2021 impervious change pixels. 2021 impervious extent - The final impervious change pixels were added to that 2019 impervious descriptor file to create the new 2021 impervious descriptor file. This file maps the extent of all impervious for the 2021 NLCD. 2021 impervious product - The percent imperviousness values (1-100%) for the impervious change pixels were extracted from the unmasked impervious layer. Values for previously published urban remained the same except for areas that were 40% or more greater in value, in the unmasked impervious layer. 2021 impervious descriptor - The final impervious change pixels were mapped to the class legend for the NLCD 2019 published impervious descriptor. These pixels were then added to the NLCD 2019 impervious descriptor file to create the new 2021 impervious descriptor file.

To evaluate land use and land cover (LULC) maps an independent and representative test dataset is required. Here, a test dataset was generated via stratified random sampling approach across all areas in Fiji not used to generate training data (i.e. all Tikinas which did not contain a training data point were valid for sampling to generate the test dataset). Following equation 13 in Olofsson et al. (2014), the sample size of the test dataset was 834. This was based on a desired standard error of the overall accuracy score of 0.01 and a user's accuracy of 0.75 for all classes. The strata for sampling test samples were the eight LULC classes: water, mangrove, bare soil, urban, agriculture, grassland, shrubland, and trees.

There are different strategies for allocating samples to strata for evaluating LULC maps, as discussed by Olofsson et al. (2014). Equal allocation of samples to strata ensures coverage of rarely occurring classes and minimise the standard error of estimators of user's accuracy. However, equal allocation does not optimise the standard error of the estimator of overall accuracy. Proportional allocation of samples to strata, based on the proportion of the strata in the overall dataset, can result in rarely occurring classes being underrepresented in the test dataset. Optimal allocation of samples to strata is challenging to implement when there are multiple evaluation objectives. Olofsson et al. (2014) recommend a "simple" allocation procedure where 50 to 100 samples are allocated to rare classes and proportional allocation is used to allocate samples to the remaining majority classes. The number of samples to allocate to rare classes can be determined by iterating over different allocations and computing estimated standard errors for performance metrics. Here, the 2021 all-Fiji LULC map, minus the Tikinas used for generating training samples, was used to estimate the proportional areal coverage of each LULC class. The LULC map from 2021 was used to permit comparison with other LULC products with a 2021 layer, notably the ESA WorldCover 10m v200 2021 product.

The 2021 LULC map was dominated by the tree class (74\% of the area classified) and the remaining classes had less than 10\% coverage each. Therefore, a "simple" allocation of 100 samples to the seven minority classes and an allocation of 133 samples to the tree class was used. This ensured all the minority classes had sufficient coverage in the test set while balancing the requirement to minimise standard errors for the estimate of overall accuracy. The allocated number of test dataset points were randomly sampled within each strata and were manually labelled using 2021 annual median RGB composites from Sentinel-2 and Planet NICFI and high-resolution Google Satellite Basemaps.

Data format

The Fiji LULC test data is available in GeoJSON format in the file fiji-lulc-test-data.geojson. Each point feature has two attributes: ref_class (the LULC class manually labelled and quality checked) and strata (the strata the sampled point belongs to derived from the 2021 all-Fiji LULC map). The following integers correspond to the ref_class and strata labels:

1. water
2. mangrove
3. bare earth / rock
4. urban / impervious
5. agriculture
6. grassland
7. shrubland
8. tree

Use

When evaluating LULC maps using test data derived from a stratified sample, the nature of the stratified sampling needs to be accounted for when estimating performance metrics such as overall accuracy, user's accuracy, and producer's accuracy. This is particulary so if the strata do not match the map classes (i.e. when comparing different LULC products). Stehman (2014) provide formulas for estimating performance metrics and their standard errors when using test data with a stratified sampling structure.

To support LULC accuracy assessment a Python package has been developed which provides implementations of Stehman's (2014) formulas. The package can be installed via:

pip install lulc-validation

with documentation and examples here.

In order to compute performance metrics accounting for the stratified nature of the sample the total number of points / pixels available to be sampled in each strata must be known. For this dataset that is:

1. 1779768,
2. 3549325,
3. 541204,
4. 687659,
5. 14279258,
6. 15115599,
7. 4972515,
8. 116131948

Acknowledgements

This dataset was generated with support from a Climate Change AI Innovation Grant.

Download linkSizeType2019 NLCD2.28 GBapplication/zipThe U.S. Geological Survey (USGS), in partnership with several federal agencies, has developed and released five National Land Cover Database (NLCD) products over the past two decades: NLCD 1992, 2001, 2006, 2011 and 2016. The 2016 release saw land cover created for additional years of 2003, 2008, and 2013. These products provide spatially explicit and reliable information on the Nation’s land cover and land cover change. To continue the legacy of NLCD and further establish a long-term monitoring capability for the Nation’s land resources, the USGS has designed a new generation of NLCD products named NLCD 2019.The NLCD 2019 design aims to provide innovative, consistent, and robust methodologies for production of a multi-temporal land cover and land cover change database from 2001 to 2019 at 2–3-year intervals. Comprehensive research was conducted and resulted in developed strategies for NLCD 2019: continued integration between impervious surface and all landcover products with impervious surface being directly mapped as developed classes in the landcover, a streamlined compositing process for assembling and preprocessing based on Landsat imagery and geospatial ancillary datasets; a multi-source integrated training data development and decision-tree based land cover classifications; a temporally, spectrally, and spatially integrated land cover change analysis strategy; a hierarchical theme-based post-classification and integration protocol for generating land cover and change products; a continuous fields biophysical parameters modeling method; and an automated scripted operational system for the NLCD 2019 production. The performance of the developed strategies and methods were tested in twenty composite referenced areas throughout the conterminous U.S. An overall accuracy assessment from the 2016 publication give a 91% overall landcover accuracy, with the developed classes also showing a 91% accuracy in overall developed. Results from this study confirm the robustness of this comprehensive and highly automated procedure for NLCD 2019 operational mapping. Questions about the NLCD 2019 land cover product can be directed to the NLCD 2019 land cover mapping team at USGS EROS, Sioux Falls, SD (605) 594-6151 or mrlc@usgs.gov. See included spatial metadata for more details.National Land Cover Database (NLCD) 2019 Impervious ProductsNational Land Cover Database (NLCD) 2019 Land Cover Products

Annual land cover mapping for continental Europe based on Ensemble Machine Learning (EML), samples obtained from LUCAS (Land Use and Coverage Area frame Survey) and CLC (CORINE Land Cover) Maps, and several harmonized raster layers (e.g. GLAD Landsat ARD imagery and Continental EU DTM). The EML predicted the dominant land cover, probabilities and uncertainties for 33 classes compatible with CLC over 20 years (2000–2019), and was implemented in R and Python (eumap library).

The raster layers were mainly composed by the GLAD Landsat ARD imagery, which were downloaded for the years 1999 to 2020 considering the Continental Europe extent (land mask area and tiling system), screened to reduce cloud cover (GLAD quality assessment band), aggregated by season according with three different quantiles (i.e. 25th, 50th and 75th), and gap-filled using the Temporal Moving Window Median approach available in the eumap library. The images for each season were selected using the same calendar dates for all period:

• Winter: December 2 of previous year until March 20 of current year
• Spring: March 21 until June 24 of current year
• Summer: June 25 until September 12 of current year
• Fall: September 13 until December 1 of current year

In addition to Landsat spectral data, the EML considered night lights (VIIRS/SUOMI NPP), Global surface water frequency, Continental EU DTM, Landsat spectral indices (SAVI, NDVI, NBR, NBR2, REI and NDWI) and the max/min. monthly geometric temperature, estimated on a pixel basis and for each month.

The training data were obtained from the geographic location of LUCAS (in-situ source) and the centroid of all polygons of CORINE (supplementary source), harmonized according to the 33 CLC and organized by year, where each unique combination of longitude, latitude and year was treated as a independent sample with the following classes (the class descriptions are here):

• 111: Urban fabric
• 122: Road and rail networks and associated land
• 123: Port areas
• 124: Airports
• 131: Mineral extraction sites
• 132: Dump sites
• 133: Construction sites
• 141: Green urban areas
• 211: Non-irrigated arable land
• 212: Permanently irrigated arable land
• 213: Rice fields
• 221: Vineyards
• 222: Fruit trees and berry plantations
• 223: Olive groves
• 231: Pastures
• 311: Broad-leaved forest
• 312: Coniferous forest
• 321: Natural grasslands
• 322: Moors and heathland
• 323: Sclerophyllous vegetation
• 324: Transitional woodland-shrub
• 331: Beaches, dunes, sands
• 332: Bare rocks
• 333: Sparsely vegetated areas
• 334: Burnt areas
• 335: Glaciers and perpetual snow
• 411: Inland wetlands
• 421: Maritime wetlands
• 511: Water courses
• 512: Water bodies
• 521: Coastal lagoons
• 522: Estuaries
• 523: Sea and ocean

The LUCAS points with a unique land cover class received a confidence rating of 100%, while CORINE points received 85%, values which were considered by EML as sample weight in the training phase. The points were used in a spacetime overlay approach, which considered the location and the year to retrieve the pixel values of all rasters. Some specific land cover samples (i.e. 111, 122, 131, 141, 211, 221, 222, 223, 231, 311, 312, 321, 411, 512) were screened according to convergence with pre-existing mapping products (OSM roads, OSM railways and Copernicus-OSM buildings; Copernicus high resolution layers), where, for example, “111: Urban fabric” samples located in low density building areas (> 50% according to Copernicus-OSM building layer) were removed from the final training data ( ~5.3 million samples and 178 covariates/features).

Using this training data, three ML models were trained to predict probabilities (i.e. Random Forest, XGBoost, Artificial Neural Network), which served as input to train a linear meta-model (i.e. Logistic regression classifier), responsable for predicting the final land cover probabilities of all classes. The hyperparameter optimization was conducted using a 5-fold spatial cross validation, based on a 30x30km tilling system. The uncertainties were calculated for all classes according to the standard deviation of the three predicted probabilities for each pixel, and the highest probability was selected as the dominant land cover class, resulting in 20 annual maps for continental Europe.

The training samples, covariates/features and fitted models are available through lcv_landcover.hcl_lucas.corine.eml_p_landmapper_full.lz4, a LandMapper class instance that can be loaded by eumap library (check the code demonstration). The production code used to generate the current version of the annual land cover maps is available in the spatial layer repository and considered a lighter LandMapper class instance (lcv_landcover.hcl_lucas.corine.eml_p_landmapper_light.lz4,), which not includes the training samples.

Only the dominant land cover classes are provided here. To access the probabilities and uncertainties use:

• Open Data Science Europe viewer: https://maps.opendatascience.eu
• S3 Cloud Object Service: https://medium.com/swlh/europe-from-above-space-time-machine-learning-reveals-our-changing-environment-1b05cb7be520

A publication describing, in detail, all processing steps, accuracy assessment and general analysis of land-cover changes in continental Europe is under preparation. To suggest any improvement/fix use https://gitlab.com/geoharmonizer_inea/spatial-layers/-/issues

This layer displays a global map of land use/land cover (LULC) derived from ESA Sentinel-2 imagery at 10m resolution. Each year is generated from Impact Observatory’s deep learning AI land classification model used a massive training dataset of billions of human-labeled image pixels developed by the National Geographic Society. The global maps were produced by applying this model to the Sentinel-2 scene collection on Microsoft’s Planetary Computer, processing over 400,000 Earth observations per year.The algorithm generates LULC predictions for 10 classes, described in detail below. The year 2017 has a land cover class assigned for every pixel, but its class is based upon fewer images than the other years. The years 2018-2021 are based upon a more complete set of imagery. For this reason, the year 2017 may have less accurate land cover class assignments than the years 2018-2021.Variable mapped: Land use/land cover in 2017, 2018, 2019, 2020, 2021Data Projection: Universal Transverse Mercator (UTM)Mosaic Projection: WGS84Extent: GlobalSource imagery: Sentinel-2Cell Size: 10m (0.00008983152098239751 degrees)Type: ThematicSource: Esri Inc.Publication date: January 2022What can you do with this layer?Global land use/land cover maps provide information on conservation planning, food security, and hydrologic modeling, among other things. This dataset can be used to visualize land use/land cover anywhere on Earth. It should be noted that since land use focus does not provide the spatial detail of a land cover map for the built area classification – yards, parks, small groves will appear as built area rather than trees or rangeland classes This layer can also be used in analyses that require land use/land cover input. For example, the Zonal Statistics tools allow a user to understand the composition of a specified area by reporting the total estimates for each of the classes. Land Cover processingThis map was produced by a deep learning model trained using over 5 billion hand-labeled Sentinel-2 pixels, sampled from over 20,000 sites distributed across all major biomes of the world.The underlying deep learning model uses 6 bands of Sentinel-2 surface reflectance data: visible blue, green, red, near infrared, and two shortwave infrared bands. To create the final map, the model is run on multiple dates of imagery throughout the year, and the outputs are composited into a final representative map for each year.Processing platformSentinel-2 L2A/B data was accessed via Microsoft’s Planetary Computer and scaled using Microsoft Azure Batch.Class definitions1. WaterAreas where water was predominantly present throughout the year; may not cover areas with sporadic or ephemeral water; contains little to no sparse vegetation, no rock outcrop nor built up features like docks; examples: rivers, ponds, lakes, oceans, flooded salt plains.2. TreesAny significant clustering of tall (~15 feet or higher) dense vegetation, typically with a closed or dense canopy; examples: wooded vegetation, clusters of dense tall vegetation within savannas, plantations, swamp or mangroves (dense/tall vegetation with ephemeral water or canopy too thick to detect water underneath).4. Flooded vegetationAreas of any type of vegetation with obvious intermixing of water throughout a majority of the year; seasonally flooded area that is a mix of grass/shrub/trees/bare ground; examples: flooded mangroves, emergent vegetation, rice paddies and other heavily irrigated and inundated agriculture.5. CropsHuman planted/plotted cereals, grasses, and crops not at tree height; examples: corn, wheat, soy, fallow plots of structured land.7. Built AreaHuman made structures; major road and rail networks; large homogenous impervious surfaces including parking structures, office buildings and residential housing; examples: houses, dense villages / towns / cities, paved roads, asphalt.8. Bare groundAreas of rock or soil with very sparse to no vegetation for the entire year; large areas of sand and deserts with no to little vegetation; examples: exposed rock or soil, desert and sand dunes, dry salt flats/pans, dried lake beds, mines.9. Snow/IceLarge homogenous areas of permanent snow or ice, typically only in mountain areas or highest latitudes; examples: glaciers, permanent snowpack, snow fields. 10. CloudsNo land cover information due to persistent cloud cover.11. RangelandOpen areas covered in homogenous grasses with little to no taller vegetation; wild cereals and grasses with no obvious human plotting (i.e., not a plotted field); examples: natural meadows and fields with sparse to no tree cover, open savanna with few to no trees, parks/golf courses/lawns, pastures. Mix of small clusters of plants or single plants dispersed on a landscape that shows exposed soil or rock; scrub-filled clearings within dense forests that are clearly not taller than trees; examples: moderate to sparse cover of bushes, shrubs and tufts of grass, savannas with very sparse grasses, trees or other plants.CitationKarra, Kontgis, et al. “Global land use/land cover with Sentinel-2 and deep learning.” IGARSS 2021-2021 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2021.AcknowledgementsTraining data for this project makes use of the National Geographic Society Dynamic World training dataset, produced for the Dynamic World Project by National Geographic Society in partnership with Google and the World Resources Institute.For questions please email environment@esri.com

The U.S. Geological Survey (USGS), in partnership with several federal agencies, has developed and released five National Land Cover Database (NLCD) products over the past two decades: NLCD 1992, 2001, 2006, 2011, and 2016. The 2016 release saw landcover created for additional years of 2003, 2008, and 2013. These products provide spatially explicit and reliable information on the Nation’s land cover and land cover change. To continue the legacy of NLCD and further establish a long-term monitoring capability for the Nation’s land resources, the USGS has designed a new generation of NLCD products named NLCD 2019. The NLCD 2019 design aims to provide innovative, consistent, and robust methodologies for production of a multi-temporal land cover and land cover change database from 2001 to 2019 at 2–3-year intervals. Comprehensive research was conducted and resulted in developed strategies for NLCD 2019: continued integration between impervious surface and all landcover products with impervious surface being directly mapped as developed classes in the landcover, a streamlined compositing process for assembling and preprocessing based on Landsat imagery and geospatial ancillary datasets; a multi-source integrated training data development and decision-tree based land cover classifications; a temporally, spectrally, and spatially integrated land cover change analysis strategy; a hierarchical theme-based post-classification and integration protocol for generating land cover and change products; a continuous fields biophysical parameters modeling method; and an automated scripted operational system for the NLCD 2019 production. The performance of the developed strategies and methods were tested in twenty composite referenced areas throughout the conterminous U.S. An overall accuracy assessment from the 2016 publication give a 91% overall landcover accuracy, with the developed classes also showing a 91% accuracy in overall developed. Results from this study confirm the robustness of this comprehensive and highly automated procedure for NLCD 2019 operational mapping. Questions about the NLCD 2019 land cover product can be directed to the NLCD 2019 land cover mapping team at USGS EROS, Sioux Falls, SD (605) 594-6151 or mrlc@usgs.gov. See included spatial metadata for more details.

The U.S. Geological Survey (USGS), in partnership with several federal agencies, has developed and released four National Land Cover Database (NLCD) products over the past two decades: NLCD 1992, 2001, 2006, and 2011. These products provide spatially explicit and reliable information on the Nation’s land cover and land cover change. To continue the legacy of NLCD and further establish a long-term monitoring capability for the Nation’s land resources, the USGS has designed a new generation of NLCD products named NLCD 2016. The NLCD 2016 design aims to provide innovative, consistent, and robust methodologies for production of a multi-temporal land cover and land cover change database from 2001 to 2016 at 2–3-year intervals. Comprehensive research was conducted and resulted in developed strategies for NLCD 2016: a streamlined process for assembling and preprocessing Landsat imagery and geospatial ancillary datasets; a multi-source integrated training data development and decision-tree based land cover classifications; a temporally, spectrally, and spatially integrated land cover change analysis strategy; a hierarchical theme-based post-classification and integration protocol for generating land cover and change products; a continuous fields biophysical parameters modeling method; and an automated scripted operational system for the NLCD 2016 production. The performance of the developed strategies and methods were tested in twenty World Reference System-2 path/row throughout the conterminous U.S. An overall agreement ranging from 71% to 97% between land cover classification and reference data was achieved for all tested area and all years. Results from this study confirm the robustness of this comprehensive and highly automated procedure for NLCD 2016 operational mapping. Questions about the NLCD 2016 land cover product can be directed to the NLCD 2016 land cover mapping team at USGS EROS, Sioux Falls, SD (605) 594-6151 or mrlc@usgs.gov. See included spatial metadata for more details.