A Landsat Data Tiling and Compositing Approach Optimized for Change Detection in the Conterminous United States

Nelson, Kurtis J.; Steinwand, Daniel R.

doi:10.14358/pers.81.7.573

Cited by 15 publications

(11 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Temporal variability is often a challenge of wetland classification since some classes are changing dynamically their condition [14,[51][52][53], e.g., through flooding, pronounced dry phases, or cropping cycles. Several image compositing approaches were developed in the past years [54][55][56][57][58][59][60][61][62]. The general idea of those methods is to select on a per-pixel basis the most suitable observations from a predefined selection of images that fulfill different quality criteria and represent spectral reflectance of one particular day of year in the best way.…”

Section: Satellite Data and Pre-processingmentioning

confidence: 99%

Long-Term Land Use/Land Cover Change Assessment of the Kilombero Catchment in Tanzania Using Random Forest Classification and Robust Change Vector Analysis

Thonfeld

Steinbach²,

Muro

et al. 2020

Remote Sensing

View full text Add to dashboard Cite

Information about land use/land cover (LULC) and their changes is useful for different stakeholders to assess future pathways of sustainable land use for food production as well as for nature conservation. In this study, we assess LULC changes in the Kilombero catchment in Tanzania, an important area of recent development in East Africa. LULC change is assessed in two ways: first, post-classification comparison (PCC) which allows us to directly assess changes from one LULC class to another, and second, spectral change detection. We perform LULC classification by applying random forests (RF) on sets of multitemporal metrics that account for seasonal within-class dynamics. For the spectral change detection, we make use of the robust change vector analysis (RCVA) and determine those changes that do not necessarily lead to another class. The combination of the two approaches enables us to distinguish areas that show (a) only PCC changes, (b) only spectral changes that do not affect the classification of a pixel, (c) both types of change, or (d) no changes at all. Our results reveal that only one-quarter of the catchment has not experienced any change. One-third shows both, spectral changes and LULC conversion. Changes detected with both methods predominantly occur in two major regions, one in the West of the catchment, one in the Kilombero floodplain. Both regions are important areas of food production and economic development in Tanzania. The Kilombero floodplain is a Ramsar protected area, half of which was converted to agricultural land in the past decades. Therefore, LULC monitoring is required to support sustainable land management. Relatively poor classification performances revealed several challenges during the classification process. The combined approach of PCC and RCVA allows us to detect spatial patterns of LULC change at distinct dimensions and intensities. With the assessment of additional classifier output, namely class-specific per-pixel classification probabilities and derived parameters, we account for classification uncertainty across space. We overlay the LULC change results and the spatial assessment of classification reliability to provide a thorough picture of the LULC changes taking place in the Kilombero catchment.

show abstract

Section: Satellite Data and Pre-processingmentioning

confidence: 99%

Long-Term Land Use/Land Cover Change Assessment of the Kilombero Catchment in Tanzania Using Random Forest Classification and Robust Change Vector Analysis

Thonfeld

Steinbach²,

Muro

et al. 2020

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…Landsat data were corrected to surface reflectance [17,18] using the USGS Earth Resources Observation and Science (EROS) Center Science Processing Architecture (ESPA) [19] processing system (https://espa.cr.usgs.gov/, accessed 12 June 2019), reprojected to USA Albers Equal Area Conic, and resampled to 30 m. Using high-performance computing architecture, multiple Landsat scenes were stacked and ranked pixel by pixel to produce a cloud-free image composite. This process, known as best-pixel image compositing [11], was conducted within a tiling framework covering the extent of both prototyping locations (see Figure 1 for NW and GC LANDFIRE tiles). A total of 7637 and 27,921 Landsat scenes were processed for the GC and NW, respectively ( Figure 2).…”

Section: Data Sourcesmentioning

confidence: 99%

“…The first was previously developed by LANDFIRE to be used for mapping disturbance (i.e., LF 2010, LF 2012, and LF 2014) and is based upon imagery from two periods of time within the same calendar year, including early season (Julian days 135-227) and late season (Julian days 228-306) composites. Landsat imagery within the Julian date ranges is obtained from the Landsat archive (https://earthexplorer.usgs.gov/, accessed 11 June 2019) for each year of interest, placed in a virtual stack of all images, and processed to the specifications established in the best-pixel algorithm [11]. As an example, to identify disturbance in 2015, early and late season best-pixel composites are produced by year for 2014, 2015, and 2016.…”

Section: Data Sourcesmentioning

confidence: 99%

“…For this paper, we are mainly concerned with the development of EVT, EVC, and EVH products and not focused on how derivatives of those products are created.Subsequent iterations of LANDFIRE, including LF 2008, LF 2010, LF 2012, and LF 2014, consist of updated yearly EVT, EVC, and EVH products adjusted according to disturbance type for each year between 1999 and 2014 [4,5,9]. Landsat was again the base dataset used in the mapping process; however, because Landsat TM and ETM+ data became freely available after 2008 [10], the cost of Landsat was no longer a constraint, and in 2010 a Landsat compositing method was developed to better leverage all the available Landsat data [11]. Additionally, Landsat Operational Land Imager (OLI) data became available in 2013 [12], which increased data availability and subsequently enabled the development of new data intensive approaches for the image compositing process.For each iteration of LANDFIRE, disturbance was classified from imagery using change detection algorithms (e.g., forest clear-cut and regeneration); from maps provided by partners (e.g., forest clear-cut and fire perimeters); and by analyst visual interpretation [5].…”

mentioning

confidence: 99%

“…Subsequent iterations of LANDFIRE, including LF 2008, LF 2010, LF 2012, and LF 2014, consist of updated yearly EVT, EVC, and EVH products adjusted according to disturbance type for each year between 1999 and 2014 [4,5,9]. Landsat was again the base dataset used in the mapping process; however, because Landsat TM and ETM+ data became freely available after 2008 [10], the cost of Landsat was no longer a constraint, and in 2010 a Landsat compositing method was developed to better leverage all the available Landsat data [11]. Additionally, Landsat Operational Land Imager (OLI) data became available in 2013 [12], which increased data availability and subsequently enabled the development of new data intensive approaches for the image compositing process.…”

mentioning

confidence: 99%

See 2 more Smart Citations

LANDFIRE Remap Prototype Mapping Effort: Developing a New Framework for Mapping Vegetation Classification, Change, and Structure

Picotte

Dockter²,

Long³

et al. 2019

Fire

View full text Add to dashboard Cite

LANDFIRE (LF) National (2001) was the original product suite of the LANDFIRE program, which included Existing Vegetation Cover (EVC), Height (EVH), and Type (EVT). Subsequent refinements after feedback from data users resulted in updated products, referred to as LF 2001, that now served as LANDFIRE's baseline datasets and are the basis for all subsequent LANDFIRE updates. These updates account for disturbances and vegetation transition changes that may not represent current vegetation conditions. Therefore, in 2016 LANDFIRE initiated the Remap prototype to determine how to undertake a national-scale remap of the LANDFIRE primary vegetation datasets. EVC, EVH, and EVT were produced (circa 2015) via modeling for ecologically variable prototyping areas in the Pacific Northwest (NW) and Grand Canyon (GC). An error analysis within the GC suggested an overall accuracy of 52% (N = 800) for EVT, and a goodness of fit of 51% (N = 38) for percent cover (continuous EVC) and 53% (N = 38) for height (continuous EVH). The prototyping effort included a new 81-class map using the National Vegetation Classification (NVC) within the NW. This paper presents a narrative of the innovative methodologies in image processing and mapping used to create the new LANDFIRE vegetation products. Fire 2019, 2, 35 2 of 26 LANDFIRE (LF) 2001 was a joint 5-year project between the U.S. Department of Agriculture (USDA) Forest Service and U.S. Department of the Interior (DOI) to provide vegetation and fuel datasets for the conterminous United States circa 2001 [6]. A foundation for this suite of data products was the LANDFIRE Reference Database (LFRDB), which was developed to hold ground-referenced plot data information about vegetation types and structure metrics (i.e., height and cover). Data sources included Forest Inventory and Analysis (FIA), U.S. Geological Survey (USGS) National Gap Analysis Program (GAP), the Nature Conservancy, and other federal, state, and local datasets [6].Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+) are the foundational geospatial datasets used by LANDFIRE. Descriptions of the Landsat bands and indices used by LANDFIRE are shown in Table 1a,b, respectively. TM and ETM+ images were not available cost-free during the production timeframe of LF 2001; however, LANDFIRE was a member of the Multi-Resolution Land Characteristics (MRLC) Consortium [8], which gave the program access to any previously purchased Landsat scenes. Landsat scenes not in the MRLC archive were purchased to meet the goal of three cloud-free images per Landsat World Wide Referencing System 2 (WRS2) Path/Row for the conterminous United States, Alaska, and Hawaii [6]. Additional remotely sensed data sources including a digital elevation model (DEM), slope, elevation, and biophysical gradients (e.g., temperature and precipitation) were also acquired or produced as needed.Classification and regression tree (CART) models were developed to determine vegetation types, while regression tree models were used to classify vegetation structur...

show abstract

Beyond deforestation: Differences in long-term regrowth dynamics across land use regimes in southern Amazonia

Müller

Rufin

Griffiths

et al. 2016

Remote Sensing of Environment

View full text Add to dashboard Cite

The loss of tropical forests threatens a broad range of ecosystem services. Particularly, tropical deforestation is a major driver of biodiversity decline and substantial carbon emissions. Regrowth of secondary vegetation may help to restore habitat for many threatened species and improve ecosystem services that deteriorated due to deforestation. However, spatial-temporal patterns of regrowing secondary vegetation in the tropics remain weakly understood. We therefore analyze regrowth dynamics across two different land use systems in southern Amazonia: the extensive pastoralism in Pará and the capital-intensive agriculture in Mato Grosso. Both systems are connected by the BR-163 highway, which represents a major axis of deforestation and agricultural development in the Brazilian Amazon since decades. We used a 29 year time series of Landsat images to extract regrowth extent, duration, lag time between deforestation and regrowth, and frequency of regrowth cycles. Our results reached an overall accuracy of 89% and showed regrowth on up to 50% of the deforested area in Pará and a maximum of 25% in Mato Grosso. In both states, annual regrowth rates drastically dropped after 1996, which coincided with socioeconomic transformations and drought events. The majority of regrowth was concentrated within 60m distance to forest edges and cleared again after an average of 5 years. Our approach bears great potential for mapping post deforestation regrowth dynamics within and beyond the Brazilian Amazon based on long-term and freely accessible remote sensing data collections, such as the Landsat and the forthcoming Sentinel-2 archives.

show abstract

A Landsat Data Tiling and Compositing Approach Optimized for Change Detection in the Conterminous United States

Cited by 15 publications

References 33 publications

Long-Term Land Use/Land Cover Change Assessment of the Kilombero Catchment in Tanzania Using Random Forest Classification and Robust Change Vector Analysis

Long-Term Land Use/Land Cover Change Assessment of the Kilombero Catchment in Tanzania Using Random Forest Classification and Robust Change Vector Analysis

LANDFIRE Remap Prototype Mapping Effort: Developing a New Framework for Mapping Vegetation Classification, Change, and Structure

Beyond deforestation: Differences in long-term regrowth dynamics across land use regimes in southern Amazonia

Contact Info

Product

Resources

About