LCMAP and NLCD: Complementary Data for Understanding the Geography of the United States

Science Center Objects

Land Change Monitoring, Assessment, and Projection (LCMAP) Collection 1 science products provide unprecedented monitoring of past changes occurring in land cover and condition across the conterminous U.S. over more than 30 years at an annual timestep.

Color image of land cover map of Conterminous United States

Map of Primary Land Cover for the Conterminous United States (CONUS) from the USGS Land Change Monitoring, Assessment, and Projection initiative. (click image for details)

The methods underlying LCMAP can capture abrupt and gradual land surface change by harnessing the depth of the Landsat archive. The approach adopted by LCMAP produces annual data on land surface change and land cover, providing essential context for understanding how the land surface changes. However, LCMAP is just the most recent effort providing land cover information for the United States at a 30-m spatial resolution using a Landsat time series.

USGS produces another well-known and well-cited geospatial land cover database for the United States called the National Land Cover Database (NLCD), the flagship land cover product published by the USGS. NLCD was launched in the 1990s and the earliest published dataset characterized land cover for 1992. Since the early development of NLCD, multiple eras of land cover and other associated datasets about land surface characteristics have been produced. The newest version of NLCD, released in 2019, provides land cover data for the following time periods; 2001, 2004, 2006, 2008, 2011, 2013, and 2016.

Similarities, Interdependence

While there are some key differences between the LCMAP and NLCD land cover products and associated technical approaches, there are also similarities and interdependencies. For example, they both utilize Landsat satellite data as input and have a nominal 30-m spatial resolution. They both present geospatial information on land cover and can be used to explore how it changes through time. The LCMAP change detection method relies on a harmonic modeling process based on a long time series of Landsat observations starting with 1982 (from TM, ETM+, and OLI instruments) that has been reconditioned into Landsat Analysis Ready Data (ARD). Specifically, the Continuous Change Detection and Classification (CCDC) method was first developed to estimate change in forested landscapes but is now being used to track the land surface across a variety of land cover types.

The mapping methods used by NLCD include data preparation, land cover change detection and classification, theme-based postprocessing, and integration of various classification layers. In NLCD, change detection relies on Landsat spectral and temporal information and knowledge-based trajectory analysis. Initial overall accuracy of the 2016 version is reported at 82% at a similar thematic classification to LCMAP (86% for the finer detailed classes).

One of the LCMAP annual products is Primary Land Cover (LCPRI). These data help us to understand the annual change that is detected through implementing the CCDC methodology. The LCPRI data are represented using a general land cover scheme with eight land cover classes. NLCD provides more detailed thematic land cover information with 16 land cover classes for the conterminous U.S.

Land Cover Animation of Palm Springs, CA area

Land cover animation of urban growth in the Palm Springs, CA area. (click animation for details)

LCMAP has some interdependency with NLCD, and Collection 1 has adopted NLCD for the year 2001 (2011 publication date) as a source for training data for classification. Early testing suggested that NLCD for 2001 supported the best classification accuracy.

As we might assume, in the years where LCMAP and NLCD overlap, LCMAP LCPRI has a high percentage match (on average of 89% calculated at an ARD tile basis) to NLCD, when NLCD classes are translated into the LCMAP LCPRI legend.

During 2020, NLCD team has been working on a new release of land cover data for the year 2019. They have recently updated their process for getting access to “clean” Landsat image composites for their seasons of interest by utilizing LCMAP code that accesses the Landsat ARD. Access to parallel source imagery and tools for analysis is a recent example of integration between LCMAP and NLCD.

Differences

Unlike NLCD, LCMAP CCDC products are derived from an end-to-end automated change detection and classification methodology (CCDC) using machine learning and based on a time series of every good-quality Landsat observation. The LCMAP product suite consists of five land surface change products and five land cover products. The creation and production of all LCMAP products is integrated and they inform and complement each other. The land cover is produced as a partner to the change data that provide information on when and to what extent change has occurred on the land surface.

LCMAP has also recently published a reference dataset covering approximately 25,000 Landsat pixel-level plots interpreted in collaboration with the U.S. Forest Service. These data have been used to validate the LCMAP land cover. Collection 1 LCMAP primary land cover has an overall accuracy of 82.5% across the record.

NLCD provides additional geospatial information for the country and involves partner products such as tree canopy, urban imperviousness, and western U.S. shrub and grassland area at varying periodicities. NLCD data layers are not produced at an annual time step, but land cover is produced at a two- or three-year interval.

User Considerations

NLCD 2016 Land Cover map of the conterminous United States

National Land Cover Database land cover for the conterminous United States, represented as 16 land cover classes. (click image for details)

In choosing between LCMAP or NLCD land cover products, researchers should consider the different thematic detail of the land cover legends, the level of accuracy required, the period of record needed, and the temporal repeat or frequency of coverage.

LCMAP has a lower level of thematic detail, with 8 thematic classes vs. 16 for NLCD (e.g., NLCD has three forest classes and the LCMAP has one class called “Tree Cover”). For some applications, the higher level of thematic detail found in NLCD is needed. The land cover in NLCD is also rendered to represent change at a patch level, whereas the LCMAP represents pixel-level cover and change without patch-level rendering.

LCMAP Collection 1 currently extends across a 33-year historical period and new years of land cover data will soon be added . For some modeling applications, extending the period of record over which change is depicted improves modeling accuracy, and there is evidence that characterization over a longer time period provides a better perspective on complex land cover dynamics.

The annual time step of the LCMAP data provides a higher level of temporal precision on land surface changes. One annual product in the LCMAP suite, the timing of spectral change (SCTIME) provides an estimate of the day of year when change was detected. Further analysis of SCTIME integrated with LCPRI, for example, can show what month or season that disturbance or change in the land surface occurred. In combination with other LCMAP products such as spectral stability (SCSTAB) or change magnitude (SCMAG), the suite of annual LCMAP products is a powerful tool to assess change at finer time scales, or more subtle changes in landscape condition that may not be represented in the thematic land cover product (LCPRI).

National Land Cover Database impervious descriptor layer for Dallas-Ft. Worth area

National Land Cover Database 2016 Impervious Descriptor layer for an area on the western side of Dallas/Ft. Worth, Texas. (click image for details)

NLCD also provides unique products not represented in LCMAP data that may be useful for specific user applications. Impervious surface layers (currently provided at nominal 5-year intervals) provide information valuable for hydrologic applications, for example. Fractional vegetation products (e.g, tree canopy or shrub percentage) may be useful for applications related to biodiversity or carbon dynamics.

Because the harmonic modeling method is the basis for both change detection and land cover classification, LCMAP provides land cover information that is directly related to the land surface change information. This is inherently a different method than land cover change that is a result of comparing one land cover map to another.

LCMAP and NLCD – Comprehensive Land Change Database

While both LCMAP and NLCD provide thematic land cover products (with varying temporal and thematic resolutions), other products produced by each project have little duplication. In combination, LCMAP and NLCD provide a comprehensive suite of land change data datasets, capable of characterizing changes in land cover (thematic land cover classes), as well as more subtle changes related to land cover condition, with each product providing a slightly different piece of information on the land surface state or condition. We encourage users to explore the joint utilization of both suites of data products to address their particular application or area of study.

Looking to the Future

It’s difficult to quantify the many applications that need geospatial information on land cover and land surface change and anticipate future needs. Topics run the gamut from tracking urban development to studying forest harvest cycles or forest pests and disease, from estimating hurricane damage to mapping wildfire scars, from monitoring agricultural patterns to ecosystem health. The user community has become more sophisticated in recent years. Users are constantly improving their capabilities to ingest higher volumes of geospatial data, at higher spatial, temporal, and thematic detail. Many also require data produced with lower latency and expressing an even bigger variety of characteristics related to land surface change, cover, use, and condition.

While the combination of LCMAP and NLCD provides a comprehensive land change database covering the conterminous U.S., we are looking to the future with a focus on 1) reducing latency in delivery of data, providing information at “the speed of decision,” 2) improving efficiencies, utilizing state-of-the-art technologies to improve reproducibility, increase accuracy, and develop tighter synergies between the LCMAP and NLCD projects, and 3) expanding the breadth of land change information provided by USGS, with a focus on designing new operational products that are best able to address broad stakeholder and partner needs.