Continuity of Landsat observations: Short term considerations

As of writing in mid-2010, both Landsat-5 and -7 continue to function, with sufficient fuel to enable data collection until the launch of the Landsat Data Continuity Mission (LDCM) scheduled for December of 2012. Failure of one or both of Landsat-5 or -7 may result in a lack of Landsat data for a period of time until the 2012 launch. Although the potential risk of a component failure increases the longer the sensor's design life is exceeded, the possible gap in Landsat data acquisition is reduced with each passing day and the risk of Landsat imagery being unavailable diminishes for all except a handful of applications that are particularly data demanding. Advances in Landsat data compositing and fusion are providing opportunities to address issues associated with Landsat-7 SLC-off imagery and to mitigate a potential acquisition gap through the integration of imagery from different sensors. The latter will likely also provide short-term, regional solutions to application-specific needs for the continuity of Landsat-like observations. Our goal in this communication is not to minimize the community's concerns regarding a gap in Landsat observations, but rather to clarify how the current situation has evolved and provide an up-to-date understanding of the circumstances, implications, and mitigation options related to a potential gap in the Landsat data record.     

Updating the 2001 National Land Cover Database land cover classification to 2006 by using Landsat imagery change detection methods

The recent release of the U.S. Geological Survey (USGS) National Land Cover Database (NLCD) 2001, which represents the nation's land cover status based on a nominal date of 2001, is widely used as a baseline for national land cover conditions. To enable the updating of this land cover information in a consistent and continuous manner, a prototype method was developed to update land cover by an individual Landsat path and row. This method updates NLCD 2001 to a nominal date of 2006 by using both Landsat imagery and data from NLCD 2001 as the baseline. Pairs of Landsat scenes in the same season in 2001 and 2006 were acquired according to satellite paths and rows and normalized to allow calculation of change vectors between the two dates. Conservative thresholds based on Anderson Level I land cover classes were used to segregate the change vectors and determine areas of change and no-change. Once change areas had been identified, land cover classifications at the full NLCD resolution for 2006 areas of change were completed by sampling from NLCD 2001 in unchanged areas. Methods were developed and tested across five Landsat path/row study sites that contain several metropolitan areas including Seattle, Washington; San Diego, California; Sioux Falls, South Dakota; Jackson, Mississippi; and Manchester, New Hampshire. Results from the five study areas show that the vast majority of land cover change was captured and updated with overall land cover classification accuracies of 78.32%, 87.5%, 88.57%, 78.36%, and 83.33% for these areas. The method optimizes mapping efficiency and has the potential to provide users a flexible method to generate updated land cover at national and regional scales by using NLCD 2001 as the baseline.     

A method for integrating MODIS and Landsat data for systematic monitoring of forest cover and change in the Congo Basin

In this paper we demonstrate a new approach that uses regional/continental MODIS (MODerate Resolution Imaging Spectroradiometer) derived forest cover products to calibrate Landsat data for exhaustive high spatial resolution mapping of forest cover and clearing in the Congo River Basin. The approach employs multi-temporal Landsat acquisitions to account for cloud cover, a primary limiting factor in humid tropical forest mapping. A Basin-wide MODIS 250 m Vegetation Continuous Field (VCF) percent tree cover product is used as a regionally consistent reference data set to train Landsat imagery. The approach is automated and greatly shortens mapping time. Results for approximately one third of the Congo Basin are shown. Derived high spatial resolution forest change estimates indicate that less than 1% of the forests were cleared from 1990 to 2000. However, forest clearing is spatially pervasive and fragmented in the landscapes studied to date, with implications for sustaining the region's biodiversity. The forest cover and change data are being used by the Central African Regional Program for the Environment (CARPE) program to study deforestation and biodiversity loss in the Congo Basin forest zone. Data from this study are available at http://carpe.umd.edu.     

Integrating profiling LIDAR with Landsat data for regional boreal forest canopy attribute estimation and change characterization

Forest dynamics are characterized by both continuous (i.e., growth) and discontinuous (i.e., disturbance) changes. Change detection techniques that use optical remotely sensed data to capture disturbance related changes are established and commonly applied; however, approaches for the capture of continuous forest changes are less mature. Optical remotely sensed imagery is well suited for capturing horizontally distributed conditions, structures, and changes, while Light Detection And Ranging (LIDAR) data are more appropriate for capturing vertically distributed elements of forest structure and change. The integration of optical remotely sensed imagery and LIDAR data provides improved opportunities to fully characterize forest canopy attributes and dynamics.The study described in this paper captures forest conditions along a corridor approximately 600 km long through the boreal forest of Canada. Two coincident LIDAR transects, representing 1997 and 2002 forest conditions respectively, are compared using image segments generated from Landsat ETM+ imagery. The image segments are used to provide a spatial framework within which the attributes and temporal dynamics of the forest canopy are estimated and compared. Segmented and classified Landsat imagery provides a context for the comparison of sufficiently spatially related LIDAR profiles and for the provision of categories to aid in the application of empirical models requiring knowledge of land cover.Global and local approaches were employed for characterizing changes in forest attributes over time. The global approach, emphasized the overall trend in forest change along the length of the entire transect, and indicated that key canopy attributes were stable, and transect characteristics, including forest canopy height, did not change significantly over the five-year period of this study (two sample t-test, p = 0.08). The local approach analyzed segment-based changes in canopy attributes, providing spatially explicit indications of forest growth and depletion. The local approach identified that 84% of the Landsat segments intercepted by both LIDAR transects either have no change, or have a small average increase in canopy height (0.7 m), while the other 16% of segments have an average decrease in canopy height of 1.6 m. As expected, the difference in the magnitude of the changes was markedly greater for depletions than it was for growth, but was less spatially extensive. Growth tends to occur incrementally over broad areas; whereas, depletions are dramatic and spatially constrained. The approach presented holds potential for investigating the impacts of climate change across a latitudinal gradient of boreal forest.     

Validation of a large area land cover product using purpose-acquired airborne video

Large area land cover products generated from remotely sensed data are difficult to validate in a timely and cost effective manner. As a result, pre-existing data are often used for validation. Temporal, spatial, and attribute differences between the land cover product and pre-existing validation data can result in inconclusive depictions of map accuracy. This approach may therefore misrepresent the true accuracy of the land cover product, as well as the accuracy of the validation data, which is not assumed to be without error. Hence, purpose-acquired validation data is preferred; however, logistical constraints often preclude its use — especially for large area land cover products. Airborne digital video provides a cost-effective tool for collecting purpose-acquired validation data over large areas. An operational trial was conducted, involving the collection of airborne video for the validation of a 31,000 km² sub-sample of the Canadian large area Earth Observation for Sustainable Development of Forests (EOSD) land cover map (Vancouver Island, British Columbia, Canada). In this trial, one form of agreement between the EOSD product and the airborne video data was defined as a match between the mode land cover class of a 3 by 3 pixel neighbourhood surrounding the sample pixel and the primary or secondary choice of land cover for the interpreted video. This scenario produced the highest level of overall accuracy at 77% for level 4 of classification hierarchy (13 classes). The coniferous treed class, which represented 71% of Vancouver Island, had an estimated user's accuracy of 86%. Purpose acquired video was found to be a useful and cost-effective data source for validation of the EOSD land cover product. The impact of using multiple interpreters was also tested and documented. Improvements to the sampling and response designs that emerged from this trial will benefit a full-scale accuracy assessment of the EOSD product and also provides insights for other regional and global land cover mapping programs.     

Mapping northern land cover fractions using Landsat ETM+

The goal of fractional mapping is to obtain land cover fraction estimates within each pixel over a region. Using field, Ikonos and Landsat data at three sites in northern Canada, we evaluate a physical unmixing method against two modeling approaches to map five land cover fractions that include bare, grass, deciduous shrub, conifer, and water along an 1100 km north-south transect crossing the tree-line of northern Canada. Error analyses are presented to assess factors that affect fractional mapping results, including modeling method (linear least squares inversion (LLSI) vs. linear regression vs. regression trees), number of Landsat spectral bands (3 vs. 5), local and distant fraction estimation using locally and globally calibrated models, and spatial resolution (30 m vs. 90 m). The ultimate purpose of this study is to determine if reliable land cover fractions can be obtained for biophysical modeling over northern Canada from a three band, resampled 90 m Landsat ETM+ mosaic north of the tree-line. Of the three modeling methods tested, linear regression and regression trees with five spectral bands produced the best local fraction estimates, while LLSI produced comparable results when unmixing was sufficiently determined. However, distant fraction estimation using both locally and globally calibrated models was most accurate using the three spectral bands available in the Landsat mosaic of northern Canada at 30 m resolution, and only slightly worse at 90 m resolution. While local calibrations produced more accurate fractions than global calibrations, application of local calibration models requires stratification of areas where local endmembers and models are representative. In the absence of such information, globally calibrated linear regression and regression trees to estimate separate fractions is an acceptable alternative, producing similar root mean square error, and an average absolute bias of less than 2%.     

North American forest disturbance mapped from a decadal Landsat record

Forest disturbance and recovery are critical ecosystem processes, but the spatial pattern of disturbance has never been mapped across North America. The LEDAPS (Landsat Ecosystem Disturbance Adaptive Processing System) project has assembled a wall-to-wall record of stand-clearing disturbance (clearcut harvest, fire) for the United States and Canada for the period 1990-2000 using the Landsat satellite archive. Landsat TM and ETM+ data were first converted to surface reflectance using the MODIS/6S atmospheric correction approach. Disturbance and early recovery were mapped using the temporal change in a Tasseled-Cap “Disturbance Index” calculated from the early (~ 1990) and later (~ 2000) images. Validation of the continental mapping has been carried out using a sample of biennial Landsat time series from 23 locations across the United States. Although a significant amount of disturbance (30-60%) cannot be mapped due to the long interval between image acquisition dates, the biennial analyses allow a first-order correction of the decadal mapping. Our results indicate disturbance rates of up to 2-3% per year are common across the US and Canada due primarily to harvest and forest fire. Rates are highest in the southeastern US, the Pacific Northwest, Maine, and Quebec. The mean disturbance rate for the conterminous United States (the “lower 48” states and District of Columbia) is calculated as 0.9 +/− 0.2% per year, corresponding to a turnover period of 110 years.     

Cross-sensor change detection over a forested landscape: Options to enable continuity of medium spatial resolution measures

Landsat data have been widely used for change detection studies of forest ecosystems. Technical issues related to the longevity and quality of the Landsat-5 and -7 instruments prompted this investigation into how data from other sensors may be integrated with the existing Landsat image archive. Change maps indicating the location and extent of stand replacing disturbances occurring between 1999 and 2004 were developed using a rank-order change detection approach. The near-infrared (NIR) band from an image representing initial stand conditions (T1: Landsat-7 ETM+), and the NIR band of images acquired on subsequent dates and with different sensors (T2: ASTER, SPOT-4, and Landsat-5 TM) were selected, essentially acting as three different T2 images. Pair-wise comparisons between the T1 image and each of the T2 images required the pixel values to be sorted, ranked, and differenced; a threshold was then applied to the difference values to identify the stand replacing disturbances. The rank-order change detection approach precluded the need for an additional image normalization process. When compared to a manually interpreted map of change events, the output from the ASTER, SPOT-4, and Landsat-5 TM data were all equally effective in identifying all of the stand replacing disturbances that occurred between 1999 and the year of T2 image acquisition, and errors of commission were minimal. Important logistical limitations to cross-sensor change do exist however and include the lack of spatially or temporally extensive image archives for sensors other than Landsat, incompatible image footprints, and data cost and policy. This rank-order change detection approach is suitable for applications involving multi-temporal datasets where problems may exist due to image normalization, cross-sensor radiometric calibration, or unavailability of a desired sensor type.     

Accuracy assessment using sub-pixel fractional error matrices of global land cover products derived from satellite data

Information on land cover distribution at regional and global scales has become fundamental for studying global changes affecting ecological and climatic systems. The remote sensing community has responded to this increased interest by improving data quality and methodologies for extracting land cover information. However, in addition to the advantages provided by satellite products, certain limitations exist that need to be objectively quantified and clearly communicated to users so that they can make informed decisions on whether and how land cover products should be used. Accuracy assessment is the procedure used to quantify product quality. Some aspects of accuracy assessment for evaluating four global land cover maps over Canada are discussed in this paper. Attempts are made to quantify limiting factors resulting from the coarse spatial resolution of data used for generating land cover information at regional and global levels. Sub-pixel fractional error matrices are introduced as a more appropriate way for assessing the accuracy of mixed pixels. For classification with coarse spatial resolution data, limitations of the classification method produce a maximum achievable accuracy defined as the average percent fraction of dominant land cover of all pixels in the mapped area. Relationships among spatial resolution, landscape heterogeneity and thematic resolution were studied and reported. Other factors that can affect accuracy, such as misregistration and legend conversion, are also discussed.     

Exploiting synergies of global land cover products for carbon cycle modeling

Within the past decade, several global land cover data sets derived from satellite observations have become available to the scientific community. They offer valuable information on the current state of the Earth's land surface. However, considerable disagreements among them and classification legends not primarily suited for specific applications such as carbon cycle model parameterizations pose significant challenges and uncertainties in the use of such data sets.This paper addresses the user community of global land cover products. We first review and compare several global land cover products, i.e. the Global Land Cover Characterization Database (GLCC), Global Land Cover 2000 (GLC2000), and the MODIS land cover product, and highlight individual strengths and weaknesses of mapping approaches. Our overall objective is to present a straightforward method that merges existing products into a desired classification legend. This process follows the idea of convergence of evidence and generates a ‘best-estimate’ data set using fuzzy agreement. We apply our method to develop a new joint 1-km global land cover product (SYNMAP) with improved characteristics for land cover parameterization of the carbon cycle models that reduces land cover uncertainties in carbon budget calculations.The overall advantage of the SYNMAP legend is that all classes are properly defined in terms of plant functional type mixtures, which can be remotely sensed and include the definitions of leaf type and longevity for each class with a tree component. SYNMAP is currently used for parameterization in a European model intercomparison initiative of three global vegetation models: BIOME-BGC, LPJ, and ORCHIDEE.Corroboration of SYNMAP against GLCC, GLC2000 and MODIS land cover products reveals improved agreement of SYNMAP with all other land cover products and therefore indicates the successful exploration of synergies between the different products. However, given that we cannot provide extensive validation using reference data we are unable to prove that SYNMAP is actually more accurate. SYNMAP is available on request from Martin Jung.     

The availability of cloud-free Landsat ETM+ data over the conterminous United States and globally

The U.S. Landsat satellite series provide the longest dedicated land remote sensing data record with a balance between requirements for localized high spatial resolution studies and global monitoring. As with any other optical wavelength satellite sensor, cloud contamination greatly compromises image usability for land surface studies. Additionally, selective scene acquisition due to payload, ground station and mission cost constraints further reduces Landsat image availability. Since the 1999 launch of the Landsat Enhanced Thematic Mapper Plus (ETM+) a Long-term Acquisition Plan (LTAP) has been used to anticipate user requests with the goal of annually refreshing a global daytime archive of cloud-free ETM+ data. This research evaluates the availability of cloud-free Landsat ETM+ data over the conterminous U.S. and globally using 3 years of ETM+ cloud fraction metadata archived by the U.S. Landsat project. Landsat application requirements including obtaining at least one cloud-free observation in a year, a season, and two different seasons, or at least a pair of cloud-free observations occurring no more than 16, 32, 48, 64, and 80 days apart within a year and season are considered. Probabilistic analyses indicate that over the conterminous U.S., land applications requiring at least one cloud-free observation in a year, a season, two different seasons, or at least two cloud-free observations occurring within any period of the year, are on average largely unaffected by cloud cover, except for certain Winter applications and cloudy scenes near the U.S.-Canada border and the Great Lakes. Cloud becomes a constraint when at least two cloud-free observations are required from the same season over the conterminous U.S., especially when the separation between observations is restricted to short time intervals. Global applications requiring at least one cloud-free observation in a season, in two different seasons, and applications requiring at least two cloud-free observations in a year, are all severely affected by cloud and data availability constraints; and globally it is generally not practical to consider land applications that require at least two cloud-free observations in any season. Globally, only land applications requiring at least one cloud-free observation per year are largely unaffected by cloud cover and the reduced global ETM+ data availability. These results are specific only to the U.S. Landsat ETM+ archive; they suggest the need for an increased global Landsat acquisition rate for the current and future Landsat missions and/or the development of new approaches to mitigating cloud contamination in the U.S. global Landsat ETM+ archive.     

Land cover mapping of large areas using chain classification of neighboring Landsat satellite images

Satellite imagery is the major data source for regional to global land cover maps. However, land cover mapping of large areas with medium-resolution imagery is costly and often constrained by the lack of good training and validation data. Our goal was to overcome these limitations, and to test chain classifications, i.e., the classification of Landsat images based on the information in the overlapping areas of neighboring scenes. The basic idea was to classify one Landsat scene first where good ground truth data is available, and then to classify the neighboring Landsat scene using the land cover classification of the first scene in the overlap area as training data. We tested chain classification for a forest/non-forest classification in the Carpathian Mountains on one horizontal chain of six Landsat scenes, and two vertical chains of two Landsat scenes each. We collected extensive training data from Quickbird imagery for classifying radiometrically uncorrected data with Support Vector Machines (SVMs). The SVMs classified 8 scenes with overall accuracies between 92.1% and 98.9% (average of 96.3%). Accuracy loss when automatically classifying neighboring scenes with chain classification was 1.9% on average. Even a chain of six images resulted only in an accuracy loss of 5.1% for the last image compared to a reference classification from independent training data for the last image. Chain classification thus performed well, but we note that chain classification can only be applied when land cover classes are well represented in the overlap area of neighboring Landsat scenes. As long as this constraint is met though, chain classification is a powerful approach for large area land cover classifications, especially in areas of varying training data availability.     

Fully fuzzy supervised classification of land cover from remotely sensed imagery with an artificial neural network

Mixed pixels are a major problem in mapping land cover from remotely sensed imagery. Unfortunately, such imagery may be dominated by mixed pixels, and the conventional hard image classification techniques used in mapping applications are unable to appropriately represent the land cover of mixed pixels. Fuzzy classification techniques can, however, accommodate the partial and multiple class membership of mixed pixels, and be used to derive an appropriate land cover representation. This is, however, only a partial solution to the mixed pixel problem in supervised image classification. It must be reognised that the land cover on the ground is fuzzy, at the scale of the pixel, and so it is inappropriate to use procedures designed for hard data in the training and testing stages of the classification. Here an approach for land cover classification in which fuzziness is accommodated in all three stages of a supervised classification is presented. Attention focuses on the classification of airborne thematic mapper data with an artificial neural network. Mixed pixels could be accommodated in training the artificial neural network, since the desired output for each training pixel can be specified. A fuzzy land cover representation was derived by outputting the activation level of the network's output units. The activation level of each output unit was significantly correlated with the proportion of the area represented by a pixel which was covered with the class associated with the unit (r>0.88, significant at the 99% level of confidence). Finally, the distance between the fuzzy land cover classification derived from the artificial neural network and the fuzzy ground data was used to illustrate the accuracy of the land cover representation derived. The dangers of hardening the classification output and ground data sets to enable a conventional assessment of classification accuracy are also illustrated; the hardened data sets were over three times more distant from each other than the fuzzy data sets.     

Detecting trends in forest disturbance and recovery using yearly Landsat time series: 1. LandTrendr — Temporal segmentation algorithms

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.     

An automated approach for segmenting and classifying a large sample of multi-date Landsat imagery for pan-tropical forest monitoring

The TREES-3 project of the Joint Research Centre aims at assessing tropical forest cover changes for the periods 1990-2000 and 2000-2010 using a sample-based approach. This paper refers to the 1990-2000 assessment. Extracts of Landsat satellite imagery (20 km × 20 km) are analyzed for these reference dates for more than 4000 sample sites distributed systematically across the tropical belt. For the processing and analysis of such a large amount of satellite imagery a robust methodological approach for forest mapping and change detection has been developed. This approach comprises two automated steps of multi-date image segmentation and object-based land cover classification (based on a supervised spectral library), followed by an intense phase of visual control and expert refinement. Image segmentation is done at two spatial scales, introducing the concept of a minimum mapping unit via the automated selection of a site-specific scale parameter. The automated segmentation of land cover polygons and the pre-classification of land cover types mainly aim at avoiding manual delineation and at reducing the efforts of visual interpretation of land cover to a reasonable level, making the analysis of 4000 sample sites feasible. The importance of visual control and correction can be perceived when comparing to the initial automatic classification result: about 20% of the polygon labels were changed through expert knowledge by visual interpretation. The component of visual refinement of the mapping results had also a notable impact on forest area and change estimates: for a set of sample sites in Southeast Asia (~ 90% of all sites of SE-Asia) the rate of change in tree cover (deforestation) was assessed at 0.9% and 1.6% before and after visual control, respectively.     

Integration of radar and Landsat-derived foliage projected cover for woody regrowth mapping, Queensland, Australia

In Queensland, Australia, forest areas are discriminated from non-forest by applying a threshold (∼ 12%) to Landsat-derived Foliage Projected Cover (FPC) layers (equating to ∼ 20% canopy cover), which are produced routinely for the State. However, separation of woody regrowth following agricultural clearing cannot be undertaken with confidence, and is therefore not mapped routinely by State Agencies. Using fully polarimetric C-, L- and P-band NASA AIRSAR and Landsat FPC data for forests and agricultural land near Injune, central Queensland, we corroborate that woody regrowth dominated by Brigalow (Acacia harpophylla) cannot be discriminated using either FPC or indeed C-band data alone, because the rapid attainment of a canopy cover leads to similarities in both reflectance and backscatter with remnant forest. We also show that regrowth cannot be discriminated from non-forest areas using either L-band or P-band data alone. However, mapping can be achieved by thresholding and intersecting these layers, as regrowth is unique in supporting both a high FPC (> ∼ 12%) and C-band SAR backscatter (> ~ − 18 dB at HV polarisation) and low L-band and P-band SAR backscatter (e.g. < =∼ 14 dB at L-band HH polarisation). To provide a theoretical explanation, a wave scattering model based on that of Durden et al. [Durden, S.L., Van Zyl, J.J. & Zebker, H.A. (1989). Modelling and observation of radar polarization signature of forested areas. IEEE Trans. Geoscience and Remote Sensing, 27, 290-301.] was used to demonstrate that volume scattering from leaves and small branches in the upper canopy leads to increases in C-band backscattering (particularly HV polarisations) from regrowth, which increases proportionally with FPC. By contrast, low L-band and P-band backscatter occurs because of the lack of double bounce interactions at co-polarisations (particularly HH) and volume scattering at HV polarisation from the stems and branches, respectively, when their dimensions are smaller than the wavelength. Regrowth maps generated by applying simple thresholds to both FPC and AIRSAR L-band data showed a very close correspondence with those mapped using same-date 2.5 m Hymap data and an average 73.7% overlap with those mapped through time-series comparison of Landsat-derived land cover classifications. Regrowth mapped using Landsat-derived FPC from 1995 and JER-1 SAR data from 1994-1995 also corresponded with areas identified within the time-series classification and true colour stereo photographs for the same period. The integration of Landsat FPC and L-band SAR data is therefore expected to facilitate regrowth mapping across Queensland and other regions of Australia, particularly as Japan's Advanced Land Observing System (ALOS) Phase Arrayed L-band SAR (PALSAR), to be launched in 2006, will observe at both L-band HH and HV polarisations.     

Changes in the extent of surface mining and reclamation in the Central Appalachians detected using a 1976-2006 Landsat time series

Surface mining and reclamation is the dominant driver of land cover land use change (LCLUC) in the Central Appalachian Mountain region of the Eastern U.S. Accurate quantification of the extent of mining activities is important for assessing how this LCLUC affects ecosystem services such as aesthetics, biodiversity, and mitigation of flooding. We used Landsat imagery from 1976, 1987, 1999 and 2006 to map the extent of surface mines and mine reclamation for eight large watersheds in the Central Appalachian region of West Virginia, Maryland and Pennsylvania. We employed standard image processing techniques in conjunction with a temporal decision tree and GIS maps of mine permits and wetlands to map active and reclaimed mines and track changes through time. For the entire study area, active surface mine extent was highest in 1976, prior to implementation of the Surface Mine Control and Reclamation Act in 1977, with 1.76% of the study area in active mines, declining to 0.44% in 2006. The most extensively mined watershed, Georges Creek in Maryland, was 5.45% active mines in 1976, declining to 1.83% in 2006. For the entire study area, the area of reclaimed mines increased from 1.35% to 4.99% from 1976 to 2006, and from 4.71% to 15.42% in Georges Creek. Land cover conversion to mines and then reclaimed mines after 1976 was almost exclusively from forest. Accuracy levels for mined and reclaimed cover was above 85% for all time periods, and was generally above 80% for mapping active and reclaimed mines separately, especially for the later time periods in which good accuracy assessment data were available. Among other implications, the mapped patterns of LCLUC are likely to significantly affect watershed hydrology, as mined and reclaimed areas have lower infiltration capacity and thus more rapid runoff than unmined forest watersheds, leading to greater potential for extreme flooding during heavy rainfall events.     

Detecting trends in forest disturbance and recovery using yearly Landsat time series: 2. TimeSync — Tools for calibration and validation

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.     

Change detection with heterogeneous data using ecoregional stratification, statistical summaries and a land allocation algorithm

A ca. 1980 national-scale land-cover classification based on aerial photo interpretation was combined with 2000 AVHRR satellite imagery to derive land cover and land-cover change information for forest, urban, and agriculture categories over a seven-state region in the U.S. To derive useful land-cover change data using a heterogeneous dataset and to validate our results, we a) stratified the classification using predefined ecoregions, b) developed statistical relationships by ecoregion between land-cover proportions derived from the 1980 national-level classification and aggregate statistical data that were available in time series for all regions in the U.S., c) classified multi-temporal AVHRR data using a process that constrained the results to the estimated proportions of land covers in ecoregions within a multi-objective land allocation (MOLA) procedure, d) interpreted land cover from a sample of aerial photographs from 2000, following the protocols used to produce the 1980 classification for use in accuracy assessment of land cover and land-cover change data, and e) compared land cover and land-cover change results for the MOLA method with an unsupervised classification alone. Overall accuracies for the 2000 MOLA and unsupervised land-cover classifications were 85% and 82%, respectively. On average, the 1980-2000 land-cover change RMSEs were one order of magnitude lower using the MOLA method compared with those based on the unsupervised data.     

Using artificial intelligence and data fusion for environmental monitoring: A review and future perspectives

Analyzing satellite images and remote sensing (RS) data using artificial intelligence (AI) tools and data fusion strategies has recently opened new perspectives for environmental monitoring and assessment. This is mainly due to the advancement of machine learning (ML) and data mining approaches, which facilitate extracting meaningful information at a large scale from geo-referenced and heterogeneous sources. This paper presents the first review of AI-based methodologies and data fusion strategies used for environmental monitoring, to the best of the authors’ knowledge. The first part of the article discusses the main challenges of geographical image analysis. Thereafter, a well-designed taxonomy is introduced to overview the existing frameworks, which have been focused on: (i) detecting different environmental impacts, e.g. land cover land use (LULC) change, gully erosion susceptibility (GES), waterlogging susceptibility (WLS), and land salinity and infertility (LSI); (ii) analyzing AI models deployed for extracting the pertinent features from RS images in addition to data fusion techniques used for combining images and/or features from heterogeneous sources; (iii) describing existing publicly-shared and open-access datasets; (iv) highlighting most frequent evaluation metrics; and (v) describing the most significant applications of ML and data fusion for RS image analysis. This is followed by an overview of existing works and discussions highlighting some of the challenges, limitations and shortcomings. To provide the reader with insight into real-world applications, two case studies illustrate the use of AI for classifying LULC changes and monitoring the environmental impacts due to dams’ construction, where classification accuracies of 98.57% and 97.05% have been reached, respectively. Lastly, recommendations and future directions are drawn.