Thematic accuracy assessment of the NLCD 2019 land cover for the conterminous United States

James Wickham; Stephen V. Stehman; Daniel G. Sorenson; Leila Gass; Jon Dewitz

doi:10.1080/15481603.2023.2181143

Home/Archive/USGS

USGS · 70241085

Thematic accuracy assessment of the NLCD 2019 land cover for the conterminous United States

James Wickham·Stephen V. Stehman·Daniel G. Sorenson·Leila Gass·Jon Dewitz

View original source ↗Export metadata ↓

Abstract

The National Land Cover Database (NLCD), a product suite produced through the MultiResolution Land Characteristics (MRLC) consortium, is an operational land cover monitoring program. Starting from a base year of 2001, NLCD releases a land cover database every 2–3-years. The recent release of NLCD2019 extends the database to 18 years. We implemented a stratified random sample to collect land cover reference data for the 2016 and 2019 components of the NLCD2019 database at Level II and Level I of the classification hierarchy. For both dates, Level II land cover overall accuracies (OA) were 77.5% ± 1% (± value is the standard error) when agreement was defined as a match between the map label and primary reference label only, and increased to 87.1% ± 0.7% when agreement was defined as a match between the map label and either the primary or alternate reference label. At Level I of the classification hierarchy, land cover OA was 83.1% ± 0.9% for both 2016 and 2019 when agreement was defined as a match between the map label and primary reference label only, and increased to 90.3% ± 0.7% when agreement also included the alternate reference label. The Level II and Level I OA for the 2016 land cover in the NLCD2019 database were 5% higher compared to the 2016 land cover component of the NLCD2016 database when agreement was defined as a match between the map label and primary reference label only. No improvement was realized by the NLCD2019 database when agreement also included the alternate reference label. User’s accuracies (UA) for forest loss and grass gain were>70% when agreement included either the primary or alternate label, and UA was generally<50% for all other change themes. Producer’s accuracies (PA) were>70% for grass loss and gain and water gain and generally<50% for the other change themes. We conducted a post-analysis review for map-reference agreement to identify patterns of disagreement, and these findings are discussed in the context of potential adjustments to mapping and reference data collection procedures that may lead to improved map accuracy going forward.

Explore related subjects

GIScience & Remote Sensing Article

Source-reported bounding extent: 25.08° to 49.38905° latitude; -124.68721° to -66.96466° longitude. This indicates report coverage, not an exact sampling location. View area on OpenStreetMap.

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

James Wickham, Stephen V. Stehman, Daniel G. Sorenson, Leila Gass, Jon Dewitz. 2023. Thematic accuracy assessment of the NLCD 2019 land cover for the conterminous United States. https://doi.org/10.1080/15481603.2023.2181143

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related USGS reports

USGS2026

Thematic accuracy assessment of the National Land Cover Database (NLCD2021) for the conterminous United States

The MultiResolution Land Characteristics (MRLC) consortium’s National Land Cover Database (NLCD) was established as an operational, stakeholder-oriented land cover monitoring program with the release of the NLCD2001 database. Both land cover and land cover change data became available with the release of NLCD2006. Here, we report the land cover and land cover change accuracies for NLCD2021, documenting Level II and I accuracies for the 2016, 2019, and 2021 land cover datasets and 2016–2021 land cover change. An additional objective, not addressed in previous NLCD assessments, is the estimation of the total variance of area estimates, where total variance includes sampling variance and variability in reference label assignment. We estimate the total variance of the area estimates for the stable and change classes targeted by the strata used in the sampling design. For the 16 Level II classes, the overall accuracy (OA) was 73% ± 1% when agreement was defined using only the primary reference class label and 89% ± 0.7% when agreement was defined as a match between the map label and either the primary or alternate reference label (± standard error, SE). For the eight Level I classes, the corresponding OA estimates were 81% ± 0.9% and 89% ± 0.7%. Level II and Level I OA tended to be higher for 2016 land cover and lower and more similar for 2019 and 2021 land cover. Land cover change user and producer accuracies (UA and PA) tended to be <50%. When the alternate reference label was included in the definition of agreement, exceptions of high class-specific accuracies (≥75%) were more prevalent for PA than UA. The consistency of reference label interpretations was greater for no-change classes than for change classes. The estimates of the total variance incorporating interpreter variability were often smaller than the standard variance estimates, indicating the possibility that the total variance estimator is highly unstable. Further study is needed to improve the utility of this total variance estimator for practical applications.

conterminous United States↗

USGS2022

Trends analysis of Rangeland Condition Monitoring Assessment and Projection (RCMAP) fractional component time series (1985–2020)

Rangelands have a dynamic response to climate change, fire, and other anthropogenic disturbances. The Rangeland Condition, Monitoring, Assessment, and Projection (RCMAP) product aims to capture this response by quantifying the percent cover of eight rangeland components, associated error, and trends across the western United States using Landsat from 1985 to 2020. The current generation of RCMAP has been improved with more training data, regional-scale Landsat composites, and more robust change detection. We assess the temporal patterns in each component with a linear model and a structural change method that determines break points using an 8-year temporal moving window. The linear and structural change methods generally agreed on patterns of change, but the latter found breaks more often, with at least one break point in most pixels. The structural change model provides more robust statistics on the significant minority of pixels with non-monotonic trends, while detrending some interannual signal potentially superfluous from a long-term perspective. Although break point density within one year of fire and vegetation treatments was ~10× and ~4× that of unburned areas, respectively, break point detection in the correct year of fire was only moderately accurate. Climate responses in break points proved more robust, with strong spatiotemporal relation in break point density with both aridity index values and aridity index change. Break point density strongly responds to both increased and decreased aridity and is reflective of ecosystem resilience. Data provide spatiotemporal information on the occurrence of breaks, but even more importantly, attribute those change events to specific component(s).

Arizona, California, Colorado, Idaho, Kansas, Mont↗

USGS2022

Multiple agricultural cropland products of South Asia developed using Landsat-8 30 m and MODIS 250 m data using machine learning on the Google Earth Engine (GEE) cloud and spectral matching techniques (SMTs) in support of food and water security

Cropland products are of great importance in water and food security assessments, especially in South Asia, which is home to nearly 2 billion people and 230 million hectares of net cropland area. In South Asia, croplands account for about 90% of all human water use. Cropland extent, cropping intensity, crop watering methods, and crop types are important factors that have a bearing on the quantity, quality, and location of production. Currently, cropland products are produced using mainly coarse-resolution (250–1000 m) remote sensing data. As multiple cropland products are needed to address food and water security challenges, our study was aimed at producing three distinct products that would be useful overall in South Asia. The first of these, Product 1, was meant to assess irrigated versus rainfed croplands in South Asia using Landsat 30 m data on the Google Earth Engine (GEE) platform. The second, Product 2, was tailored for major crop types using Moderate Resolution Imaging Spectroradiometer (MODIS) 250 m data. The third, Product 3, was designed for cropping intensity (single, double, and triple cropping) using MODIS 250 m data. For the kharif season (the main cropping season in South Asia, Jun–Oct), 10 major crops (5 irrigated crops: rice, soybean, maize, sugarcane, cotton; and 5 rainfed crops: pulses, rice, sorghum, millet, groundnut) were mapped. For the rabi season (post-rainy season, Nov–Feb), five major crops (three irrigated crops: rice, wheat, maize; and two rainfed crops: chickpea, pulses) were mapped. The irrigated versus rainfed 30 m product showed an overall accuracy of 79.8% with the irrigated cropland class providing a producer’s accuracy of 79% and the rainfed cropland class 74%. The overall accuracy demonstrated by the cropping intensity product was 85.3% with the producer’s accuracies of 88%, 85%, and 67% for single, double, and triple cropping, respectively. Crop types were mapped to accuracy levels ranging from 72% to 97%. A comparison of the crop-type area statistics with national statistics explained 63–98% variability. The study produced multiple-cropland products that are crucial for food and water security assessments, modeling, mapping, and monitoring using multiple-satellite sensor big-data, and Random Forest (RF) machine learning algorithms by coding, processing, and computing on the GEE cloud.

GIScience & Remote Sensing↗