We introduce and test LandTrendr (Landsat-based detection of Trends in Disturbance and Recovery), a new approach to extract spectral trajectories of land surface change from yearly Landsat ...time-series stacks (LTS). The method brings together two themes in time-series analysis of LTS: capture of short-duration events and smoothing of long-term trends. Our strategy is founded on the recognition that change is not simply a contrast between conditions at two points in time, but rather a continual process operating at both fast and slow rates on landscapes. This concept requires both new algorithms to extract change and new interpretation tools to validate those algorithms. The challenge is to resolve salient features of the time series while eliminating noise introduced by ephemeral changes in illumination, phenology, atmospheric condition, and geometric registration. In the LandTrendr approach, we use relative radiometric normalization and simple cloud screening rules to create on-the-fly mosaics of multiple images per year, and extract temporal trajectories of spectral data on a pixel-by-pixel basis. We then apply temporal segmentation strategies with both regression-based and point-to-point fitting of spectral indices as a function of time, allowing capture of both slowly-evolving processes, such as regrowth, and abrupt events, such as forest harvest. Because any temporal trajectory pattern is allowable, we use control parameters and threshold-based filtering to reduce the role of false positive detections. No suitable reference data are available to assess the role of these control parameters or to test overall algorithm performance. Therefore, we also developed a companion interpretation approach founded on the same conceptual framework of capturing both long and short-duration processes, and developed a software tool to apply this concept to expert interpretation and segmentation of spectral trajectories (TimeSync, described in a companion paper by Cohen et al., 2010). These data were used as a truth set against which to evaluate the behavior of the LandTrendr algorithms applied to three spectral indices. We applied the LandTrendr algorithms to several hundred points across western Oregon and Washington (U.S.A.). Because of the diversity of potential outputs from the LTS data, we evaluated algorithm performance against summary metrics for disturbance, recovery, and stability, both for capture of events and longer-duration processes. Despite the apparent complexity of parameters, our results suggest a simple grouping of parameters along a single axis that balances the detection of abrupt events with capture of long-duration trends. Overall algorithm performance was good, capturing a wide range of disturbance and recovery phenomena, even when evaluated against a truth set that contained new targets (recovery and stability) with much subtler thresholds of change than available from prior validation datasets. Temporal segmentation of the archive appears to be a feasible and robust means of increasing information extraction from the Landsat archive.
Availability of free, high quality Landsat data portends a new era in remote sensing change detection. Using dense (~
annual) Landsat time series (LTS), we can now characterize vegetation change over ...large areas at an annual time step and at the spatial grain of anthropogenic disturbance. Additionally, we expect more accurate detection of subtle disturbances and improved characterization in terms of both timing and intensity. For Landsat change detection in this new era of dense LTS, new detection algorithms are required, and new approaches are needed to calibrate those algorithms and to examine the veracity of their output. This paper addresses that need by presenting a new tool called TimeSync for syncing algorithm and human interpretations of LTS. The tool consists of four components: (1) a chip window within which an area of user-defined size around an area of interest (i.e., plot) is displayed as a time series of image chips which are viewed simultaneously, (2) a trajectory window within which the plot spectral properties are displayed as a trajectory of Landsat band reflectance or index through time in any band or index desired, (3) a Google Earth window where a recent high-resolution image of the plot and its neighborhood can be viewed for context, and (4) an Access database where observations about the LTS for the plot of interest are entered. In this paper, we describe how to use TimeSync to collect data over forested plots in Oregon and Washington, USA, examine the data collected with it, and then compare those data with the output from a new LTS algorithm, LandTrendr, described in a companion paper (Kennedy et al., 2010). For any given plot, both TimeSync and LandTrendr partitioned its spectral trajectory into linear sequential segments. Depending on the direction of spectral change associated with any given segment in a trajectory, the segment was assigned a label of disturbance, recovery, or stable. Each segment was associated with a start and end vertex which describe its duration. We explore a variety of ways to summarize the trajectory data and compare those summaries derived from both TimeSync and LandTrendr. One comparison, involving start vertex date and segment label, provides a direct linkage to existing change detection validation approaches that rely on contingency (error) matrices and kappa statistics. All other comparisons are unique to this study, and provide a rich set of means by which to examine algorithm veracity. One of the strengths of TimeSync is its flexibility with respect to sample design, particularly the ability to sample an area of interest with statistical validity through space and time. This is in comparison to the use of existing reference data (e.g., field or airphoto data), which, at best, exist for only parts of the area of interest, for only specific time periods, or are restricted thematically. The extant data, even though biased in their representation, can be used to ascertain the veracity of TimeSync interpretation of change. We demonstrate that process here, learning that what we cannot see with TimeSync are those changes that are not expressed in the forest canopy (e.g., pre-commercial harvest or understory burning) and that these extant reference datasets have numerous omissions that render them less than desirable for representing truth.
We developed a new algorithm for COntinuous monitoring of Land Disturbance (COLD) using Landsat time series. COLD can detect many kinds of land disturbance continuously as new images are collected ...and provide historical land disturbance maps retrospectively. To better detect land disturbance, we tested different kinds of input data and explored many time series analysis techniques. We have several major observations as follows. First, time series of surface reflectance provides much better detection results than time series of Top-Of-Atmosphere (TOA) reflectance, and with some adjustments to the temporal density, time series from Landsat Analysis Ready Data (ARD) is better than it is from the same Landsat scene. Second, the combined use of spectral bands is always better than using a single spectral band or index, and if all the essential spectral bands have been employed, the inclusion of other indices does not further improve the algorithm performance. Third, the remaining outliers in the time series can be removed based on their deviation from model predicted values based on probability-based thresholds derived from normal or chi-squared distributions. Fourth, model initialization is pivotal for monitoring land disturbance, and a good initialization stability test can influence algorithm performance substantially. Fifth, time series model estimation with eight coefficients model, updated for every single observation, based on all available clear observations achieves the best result. Sixth, a change probability of 0.99 (chi-squared distribution) with six consecutive anomaly observations and a mean included angle < 45° to confirm a change provide the best results, and the combined use of temporally-adjusted Root Mean Square Error (RMSE) and minimum RMSE is recommended. Finally, spectral changes (or “breaks”) contributed from vegetation regrowth should be excluded from land disturbance maps. The COLD algorithm was developed and calibrated based on all these lessons learned above. The accuracy assessment shows that COLD results were accurate for detecting land disturbance, with an omission error of 27% and a commission error of 28%.
•We developed a new algorithm for continuous monitoring of land disturbance.•All land disturbance types are detected in the disturbance map.•Both omission and commission errors are <30%.•No training data is needed for detecting land disturbance.•Disturbance maps with high spatial and high temporal resolutions are derived.
Improved monitoring of forest biomass and biomass change is needed to quantify natural and anthropogenic effects on the terrestrial carbon cycle. Landsat's temporal and spatial coverage, moderate ...spatial resolution, and long history of earth observations provide a unique opportunity for characterizing vegetation changes across large areas and long time scales. However, like with other multi-spectral passive optical sensors, Landsat's relationship of single-date reflectance with forest biomass diminishes under high leaf area and complex canopy conditions. Because the condition of a forest stand at any point in time is largely determined by its disturbance and recovery history, we conceived a method that enhances Landsat's spectral relationships with biomass by including information on vegetation trends prior to the date for which estimates are desired. With recently developed algorithms that characterize trends in disturbance (e.g. year of onset, duration, and magnitude) and post-disturbance regrowth, it should now be possible to realize improved Landsat-based mapping of current biomass across large regions. Moreover, given that we now have 40years of Landsat data, it should also be possible to use this approach to map historic biomass densities.
In this study, we developed regression tree models to predict current forest aboveground biomass (AGB) for a mixed-conifer region in eastern Oregon (USA) using Landsat-based disturbance and recovery (DR) metrics. We employed the trajectory-fitting algorithm LandTrendr to characterize DR trends from yearly Landsat time series between 1972 and 2010. The most important DR predictors of AGB were associated with magnitude of disturbance, post-disturbance condition and post-disturbance recovery, whereas time since disturbance and pre-disturbance trends showed only weak correlations with AGB. Including DR metrics substantially improved predictions of AGB (RMSE=30.3Mgha−1, 27%) compared to models based on only single-date reflectance (RMSE=39.6Mgha−1, 35%). To determine the number of years required to adequately capture the effect of DR on AGB, we explored the relationship between time-series length and model prediction accuracy. Prediction accuracy increased exponentially with increasing number of years across the entire observation period, suggesting that in this forest region the longer the historic record of disturbance and recovery metrics the more accurate the mapping of AGB. However, time series lengths of between 10 and 20years were adequate to significantly improve model predictions, and lengths of as little as 5years still had a meaningful impact. To test the concept of historic biomass prediction, we applied our model to Landsat time series from 1972–1993 and estimated AGB biomass change between 1993 and 2007. Our estimates compared well with historic inventory data, demonstrating that long-term Landsat observations of DR processes can aid in monitoring AGB and AGB change.
Instead of directly linking Landsat data with the limited amount of available field-based AGB data, in this study we used the field data to map AGB with airborne lidar and then sampled the lidar data for model training and error assessment. By using lidar data to build and test our prediction model, this study illustrates that lidar data have great value for scaling between field measurements and Landsat data.
•We characterized forest disturbance and recovery using annual Landsat time series.•We mapped forest biomass and biomass change using disturbance history metrics.•Airborne lidar data was used for model training and testing.•Disturbance history metrics were important predictors of forest biomass.•Increasing time-series length improved model predictions.
New and previously unimaginable Landsat applications have been fostered by a policy change in 2008 that made analysis-ready Landsat data free and open access. Since 1972, Landsat has been collecting ...images of the Earth, with the early years of the program constrained by onboard satellite and ground systems, as well as limitations across the range of required computing, networking, and storage capabilities. Rather than robust on-satellite storage for transmission via high bandwidth downlink to a centralized storage and distribution facility as with Landsat-8, a network of receiving stations, one operated by the U.S. government, the other operated by a community of International Cooperators (ICs), were utilized. ICs paid a fee for the right to receive and distribute Landsat data and over time, more Landsat data was held outside the archive of the United State Geological Survey (USGS) than was held inside, much of it unique. Recognizing the critical value of these data, the USGS began a Landsat Global Archive Consolidation (LGAC) initiative in 2010 to bring these data into a single, universally accessible, centralized global archive, housed at the Earth Resources Observation and Science (EROS) Center in Sioux Falls, South Dakota. The primary LGAC goals are to inventory the data held by ICs, acquire the data, and ingest and apply standard ground station processing to generate an L1T analysis-ready product. As of January 1, 2015 there were 5,532,454 images in the USGS archive. LGAC has contributed approximately 3.2 million of those images, more than doubling the original USGS archive holdings. Moreover, an additional 2.3 million images have been identified to date through the LGAC initiative and are in the process of being added to the archive. The impact of LGAC is significant and, in terms of images in the collection, analogous to that of having had two additional Landsat-5 missions. As a result of LGAC, there are regions of the globe that now have markedly improved Landsat data coverage, resulting in an enhanced capacity for mapping, monitoring change, and capturing historic conditions. Although future missions can be planned and implemented, the past cannot be revisited, underscoring the value and enhanced significance of historical Landsat data and the LGAC initiative. The aim of this paper is to report the current status of the global USGS Landsat archive, document the existing and anticipated contributions of LGAC to the archive, and characterize the current acquisitions of Landsat-7 and Landsat-8. Landsat-8 is adding data to the archive at an unprecedented rate as nearly all terrestrial images are now collected. We also offer key lessons learned so far from the LGAC initiative, plus insights regarding other critical elements of the Landsat program looking forward, such as acquisition, continuity, temporal revisit, and the importance of continuing to operationalize the Landsat program.
•USGS Landsat archive contained 5.5 million images as of January 1, 2015.•To date 3.2 million images were added by the Landsat Global Archive Consolidation (LGAC).•LGAC will consolidate an additional of ~2.3 million images into the UGSG archive.•As of January 1, 2015, LGAC had contributed 57% of the images in the USGS archive.•Ground systems are an important element of operational land imaging activities.
Monitoring and classifying forest disturbance using Landsat time series has improved greatly over the past decade, with many new algorithms taking advantage of the high-quality, cost free data in the ...archive. Much of the innovation has been focused on use of sophisticated workflows that consist of a logical sequence of processes and rules, multiple statistical functions, and parameter sets that must be calibrated to accurately classify disturbance. For many algorithms, calibration has been local to areas of interest and the algorithm's classification performance has been good under those circumstances. When applied elsewhere, however, algorithm performance has suffered. An alternative strategy for calibration may be to use the locally tested parameter values in conjunction with a statistical approach (e.g., Random Forests; RF) to align algorithm classification with a reference disturbance dataset, a process we call secondary classification. We tested that strategy here using RF with LandTrendr, an algorithm that runs on one spectral band or index. Disturbance detection using secondary classification was spectral band- or index-dependent, with each spectral dimension providing some unique detections and different error rates. Using secondary classification, we tested whether an integrated multispectral LandTrendr ensemble, with various combinations of the six basic Landsat reflectance bands and seven common spectral indices, improves algorithm performance. Results indicated a substantial reduction in errors relative to secondary classification based on single bands/indices, revealing the importance of a multispectral approach to forest disturbance detection. To explain the importance of specific bands and spectral indices in the multispectral ensemble, we developed a disturbance signal-to-noise metric that clearly highlighted the value of shortwave-infrared reflectance, especially when paired with near-infrared reflectance.
•A Landsat disturbance signal-to-noise ratio (DSNR) metric was developed.•DSNR shown to be strongly correlated to forest disturbance detection error•Secondary classification used to improve forest disturbance detection accuracy.•SWIR bands and SWIR-based indices most important•NIR band important when used in combination with SWIR
Tropical forests hold most of Earth’s biodiversity. Their continued loss through deforestation and agriculture is the main threat to species globally, more than disease, invasive species, and climate ...change. However, not all tropical forests have the same ability to sustain biodiversity. Those that have been disturbed by humans, including forests previously cleared and regrown (secondary growth), have lower levels of species richness compared with undisturbed (primary) forests. The difference is even greater considering extinctions that will later emanate from the disturbance (extinction debt). Here,we find that Haiti has less than 1%of its original primary forest and is therefore among the most deforested countries. Primary forest has declined over three decades inside national parks, and 42 of the 50 highest and largest mountains have lost all primary forest. Our surveys of vertebrate diversity (especially amphibians and reptiles) on mountaintops indicates that endemic species have been lost along with the loss of forest. At the current rate, Haiti will lose essentially all of its primary forest during the next two decades and is already undergoing a mass extinction of its biodiversity because of deforestation. These findings point to the need, in general, for better reporting of forest cover data of relevance to biodiversity, instead of “total forest” as defined by the United Nation’s Food and Agricultural Organization. Expanded detection and monitoring of primary forest globally will improve the efficiency of conservation measures, inside and outside of protected areas.
•We characterized annual rates of forest disturbance between 1985 and 2012.•Landsat time series were visually interpreted, with support of ancillary data.•A probability design was used to scale ...estimates and provide uncertainties.•Harvest was the most important disturbance agent class prior to the mid-90s.•Forest decline is now more important than harvest as the dominant agent class.
Evidence of shifting dominance among major forest disturbance agent classes regionally to globally has been emerging in the literature. For example, climate-related stress and secondary stressors on forests (e.g., insect and disease, fire) have dramatically increased since the turn of the century globally, while harvest rates in the western US and elsewhere have declined. For shifts to be quantified, accurate historical forest disturbance estimates are required as a baseline for examining current trends. We report annual disturbance rates (with uncertainties) in the aggregate and by major change causal agent class for the conterminous US and five geographic subregions between 1985 and 2012. Results are based on human interpretations of Landsat time series from a probability sample of 7200 plots (30m) distributed throughout the study area. Forest disturbance information was recorded with a Landsat time series visualization and data collection tool that incorporates ancillary high-resolution data. National rates of disturbance varied between 1.5% and 4.5% of forest area per year, with trends being strongly affected by shifting dominance among specific disturbance agent influences at the regional scale. Throughout the time series, national harvest disturbance rates varied between one and two percent, and were largely a function of harvest in the more heavily forested regions of the US (Mountain West, Northeast, and Southeast). During the first part of the time series, national disturbance rates largely reflected trends in harvest disturbance. Beginning in the mid-90s, forest decline-related disturbances associated with diminishing forest health (e.g., physiological stress leading to tree canopy cover loss, increases in tree mortality above background levels), especially in the Mountain West and Lowland West regions of the US, increased dramatically. Consequently, national disturbance rates greatly increased by 2000, and remained high for much of the decade. Decline-related disturbance rates reached as high as 8% per year in the western regions during the early-2000s. Although low compared to harvest and decline, fire disturbance rates also increased in the early- to mid-2000s. We segmented annual decline-related disturbance rates to distinguish between newly impacted areas and areas undergoing gradual but consistent decline over multiple years. We also translated Landsat reflectance change into tree canopy cover change information for greater relevance to ecosystem modelers and forest managers, who can derive better understanding of forest-climate interactions and better adapt management strategies to changing climate regimes. Similar studies could be carried out for other countries where there are sufficient Landsat data and historic temporal snapshots of high-resolution imagery.
Insects are important forest disturbance agents, and mapping their effects on tree mortality and surface fuels represents a critical research challenge. Although various remote sensing approaches ...have been developed to monitor insect impacts, most studies have focused on single insect agents or single locations and have not related observed changes to ground-based measurements. This study presents a remote sensing framework to (1) characterize spectral trajectories associated with insect activity of varying duration and severity and (2) relate those trajectories to ground-based measurements of tree mortality and surface fuels in the Cascade Range, Oregon, USA. We leverage a Landsat time series change detection algorithm (LandTrendr), annual forest health aerial detection surveys (ADS), and field measurements to investigate two study landscapes broadly applicable to conifer forests and dominant insect agents of western North America. We distributed 38 plots across multiple forest types (ranging from mesic mixed-conifer to xeric lodgepole pine) and insect agents (defoliator western spruce budworm and bark beetle mountain pine beetle). Insect effects were evident in the Landsat time series as combinations of both short- and long-duration changes in the Normalized Burn Ratio spectral index. Western spruce budworm trajectories appeared to show a consistent temporal evolution of long-duration spectral decline (loss of vegetation) followed by recovery, whereas mountain pine beetle plots exhibited both short- and long-duration spectral declines and variable recovery rates. Although temporally variable, insect-affected stands generally conformed to four spectral trajectories: short-duration decline then recovery, short- then long-duration decline, long-duration decline, long-duration decline then recovery. When comparing remote sensing data with field measurements of insect impacts, we found that spectral changes were related to cover-based estimates (tree basal area mortality
R
2
adj
=
0.40,
F
1,34
=
24.76,
P
<
0.0001 and down coarse woody detritus
R
2
adj
=
0.29,
F
1,32
=
14.72,
P
=
0.0006). In contrast, ADS changes were related to count-based estimates (e.g., ADS mortality from mountain pine beetle positively correlated with ground-based counts
R
2
adj
=
0.37,
F
1,22
=
14.71,
P
=
0.0009). Fine woody detritus and forest floor depth were not well correlated with Landsat- or aerial survey-based change metrics. By characterizing several distinct temporal manifestations of insect activity in conifer forests, this study demonstrates the utility of insect mapping methods that capture a wide range of spectral trajectories. This study also confirms the key role that satellite imagery can play in understanding the interactions among insects, fuels, and wildfire.
► This study integrates Landsat, aircraft, and field estimates of forest disturbance. ► Landsat spectral time series capture variable timing and severity of insect impacts. ► Bark beetles and defoliators cause both short- and long-duration spectral changes. ► Insect-caused spectral changes are related to tree mortality and some surface fuels. ► This approach can help quantify the interactions among insects, fuels, and wildfire.
Automated cloud and cloud shadow identification algorithms designed for Landsat Thematic Mapper (TM) and Thematic Mapper Plus (ETM+) satellite images have greatly expanded the use of these Earth ...observation data by providing a means of including only clear-view pixels in image analysis and efficient cloud-free compositing. In an effort to extend these capabilities to Landsat Multispectal Scanner (MSS) imagery, we introduce MSS clear-view-mask (MSScvm), an automated cloud and shadow identification algorithm for MSS imagery. The algorithm is specific to the unique spectral characteristics of MSS data, relying on a simple, rule-based approach. Clouds are identified based on green band brightness and the normalized difference between the green and red bands, while cloud shadows are identified by near infrared band darkness and cloud projection. A digital elevation model is incorporated to correct for topography-induced illumination variation and aid in identifying water. Based on an accuracy assessment of 1981 points stratified by land cover and algorithm mask class for 12 images throughout the United States, MSScvm achieved an overall accuracy of 84.0%. Omission of thin clouds and bright cloud shadows constituted much of the error. Perennial ice and snow, misidentified as cloud, also contributed disproportionally to algorithm error. Comparison against a corresponding assessment of the Fmask algorithm, applied to coincident TM imagery, showed similar error patterns and a general reduction in accuracy commensurate with differences in the radiometric and spectral richness of the two sensors. MSScvm provides a suitable automated method for creating cloud and cloud shadow masks for MSS imagery required for time series analyses in temperate ecosystems.
•Automated cloud/shadow identification algorithm for Landsat MSS data.•Enhances integration of MSS data into time series analysis with TM/ETM+/OLI.•Overall algorithm accuracy of 84.0%, with better performance for thick clouds.•Compared relatively well to coincident assessment of Fmask applied to TM data.•Performs well for most conditions except semi-transparent clouds and snow.