Monitoring landscape change in multi-use west-central Alberta, Canada using the disturbance-inventory framework
Highlights
► Annual change was tracked in the foothills of Alberta. ► Focus on the spatio-temporal distribution of industrial disturbance features. ► Industrial disturbances include clearcuts, oil and gas wells, coal mines, and roads. ► Cumulative disturbance density and fragmentation increased significantly in '98-`05. ► Rapid change and fragmentation call for ongoing monitoring of this landscape.
Introduction
Human activities have become the source of much contemporary landscape change, in part by altering the amount, spatial pattern, and character of global vegetation communities (Foley et al., 2005, Houghton, 1994, Lambin et al., 2001). These human-induced modifications have been identified as a major cause of biodiversity decline and species endangerment (Balmford et al., 2003, Hansen et al., 2001), stimulating a growing emphasis on monitoring programs designed to reveal the consequences of anthropogenic development on natural systems. The public lands that comprise much of Alberta's Rocky Mountain foothills are no exception to this global trend, and are an example of a fast-changing forested landscape that supports intensive use by a variety of resource-extraction industries, including forestry, coal mining, and petroleum development (Linke et al., 2008, Linke et al., 2005, Schneider et al., 2003). At the same time, this area is host to native wildlife and iconic species, including the threatened woodland caribou (Rangifer tarandus caribou) (ASRD/ACA, 2010a) and the grizzly bear (Ursus arctos), recently also designated as threatened in this Canadian province (ASRD/ACA, 2010b). These issues greatly contribute to the monitoring obligations borne by managers and regulators in this landscape (AGBRP, 2008) and evoke inquiries regarding the distribution, extent and proximity of industry-related disturbances and their associated changes in landscape structure over time.
Remote sensing has long been considered an essential tool for monitoring landscape change (Kerr and Ostrovsky, 2003, Skole et al., 1997, Turner et al., 2003) but the challenges associated with moving from change detection to landscape monitoring are complex. In essence, landscape monitoring involves the comparison of landscape conditions across two or more dates in time, and may involve the use of landscape pattern analysis (LPA) (Li and Wu, 2007, O'Neill et al., 1988) to quantify transitions in structural composition (i.e., area of cover-types; e.g., Fry et al., 2011) and/or configuration (i.e., edge density, patch connectivity; e.g., Southworth et al., 2002). While various remote-sensing techniques exist for detecting and analyzing change (Blaschke, 2005, Coppin et al., 2004, Desclée et al., 2006, Lu et al., 2004), conventional approaches commonly rely on independently classified land-cover maps (i.e., post-classification analysis), often with little attention paid to issues of classification accuracy (Hess, 1994, Newton et al., 2009) and the propagation of errors (Mas, 2005, Singh, 1989). Spurious changes are differences between maps that are not caused by real changes on the ground, but rather by classification errors arising from differences in atmosphere, illumination, vegetation phenology, soil moisture, satellite-sensor configuration, image-to-ground registration, map-to-map alignment, and classification performance between two or more dates (Carmel et al., 2001, Mas, 2005, Yuan and Elvidge, 1998). While propagation of classification errors were not viewed as a serious hindrance to LPA in early work (e.g., Wickham et al., 1997), more recent studies have demonstrated their large and mainly unpredictable impact on landscape pattern indices (Brown et al., 2000, Langford et al., 2006, Shao et al., 2001), calling into question the reliability of nearly every LPA study ever published (Gergel, 2006, Langford et al., 2006).
While post-classification change analysis may work well under conditions where changes are reported in an aggregated, aspatial manner (Ahlqvist, 2008, Vogelman et al., 2001), there is a growing need to identify the pattern, nature, and magnitude of change more explicitly (Xian et al., 2009), and to incorporate the temporal variability of landscape pattern dynamics into ecological studies (Cushman & McGarigal, 2006). As such, we require the development of processing strategies capable of producing consistent, multi-temporal series of land-cover products, thereby enabling reliable and repeatable landscape monitoring (e.g., Gillanders et al., 2008, Shao and Wu, 2008).
An alternative to post-classification analysis is map updating, wherein an existing map product (i.e., reference map T0) is updated to a second point in time (Tn) through its reclassification only within the regions of identified change between the two dates (Change Tn − T0). This strategy precludes the occurrence of any spurious change outside the areas being updated, thereby increasing the thematic and spatial consistency of map products across the entire monitoring horizon (Fry et al., 2011, McDermid et al., 2008). Despite these advantages, map updating is not free of challenges. For example, slight spatial mismatches between the boundaries of change regions and existing features in the reference map occur regularly, arising from the fact that it is practically impossible to delineate dynamic objects in a spatially consistent manner across two or more time periods (McDermid et al., 2008). These mismatches introduce small, spurious artifacts, such as slivers and gaps, in the updated or backdated maps (McDermid et al., 2008), and are similar to those generated from polygon-overlay operations in Geographic Information Systems (GIS) analysis (Chrisman, 1989, Goodchild, 1978). Despite their small size, these slivers and gaps can seriously distort the rate and direction of change trajectories for landscape pattern indices, thereby compromising their ability to monitor trends over time (Linke et al., 2009b).
While accurate, precise, and consistent map-updating of land-cover polygons is undoubtedly best-achieved through human image-interpretation and manual editing (e.g., Feranec et al., 2007, Loveland et al., 2002, Sohl et al., 2004), this is an exceptionally labor-intensive process, and not feasible for monitoring projects extending over large areas and/or frequent time intervals. Automated processing strategies for generating multi-temporal map series that reduce labor costs while maintaining high standards of accuracy and consistency are still highly sought after, and remain “the Holy Grail of change detection” (Loveland et al., 2002 p. 1098).
In an effort to contribute towards this goal, we have developed an innovative approach to multi-temporal mapping and landscape monitoring: the disturbance-inventory (D-I) framework (Linke and McDermid, 2011, Linke, McDermid, Laskin, et al., 2009). The D-I framework enables the generation of a spatially consistent time series of land-cover maps in a semi-automated, repeatable manner; without the need for manual alterations of the boundaries of change regions. It is designed to account for land-cover conversions specifically related to disturbance events and uses a combination of raster- and vector-operations in a GIS environment to: (1) store, classify, and manipulate dynamic objects (i.e., objects that appear, disappear, and/or change thematically over the monitoring horizon); and (2) seamlessly integrate these objects into an existing thematic map. In identifying the need for this research (McDermid et al., 2008), demonstrating the issues to be overcome (Linke et al., 2009b), and articulating the solution (Linke and McDermid, 2011, Linke, McDermid, Laskin, et al., 2009), we have developed a foundation for spatially consistent monitoring. Our next goal is to demonstrate the application of the D-I framework in operational monitoring programs, and establish the value of our approach to projects that aim to understand the impacts of human-induced disturbance on our natural landscapes.
The objective of this paper is to present and discuss the results of a multi-temporal monitoring program designed to track changes in the multi-use foothills of west-central Alberta, Canada. Specifically, we describe the spatio-temporal distribution of disturbance features brought about by industrial development, and track the associated annual changes in land-cover pattern for a large, 8800-km2 area between the years 1998 and 2005. In order to make this paper self-contained, we first provide a background summary of the conceptual foundations of the D-I framework. Then, we describe the methods used for the change monitoring in this application, followed by the delivery of the monitoring results. The paper concludes with a discussion of key findings and implications for future monitoring studies.
Section snippets
Basic components, workflow, and output products
In order to generate a spatially consistent time series of land-cover using the D-I framework, two basic components are needed: (1) a reference map; and (2) a D-I GIS-vector layer (Fig. 1). The reference map consists of a mosaic of non-overlapping map objects, wherein each object is a contiguous area sharing the same land-cover attribute. This map represents land-cover conditions at time T0, and serves as the basis for any projections backward (T-n) (i.e., backdating) and forward (T+ n) (i.e.,
Study area
The 8721-km2 study area is located in the west-central core of the Alberta foothills of western Canada (Fig. 3), just east of Jasper National Park. The area is situated south of Hinton and is occupied primarily by closed-canopied, pure- and mixed-coniferous (Picea glauca, Picea engelmanii, Pinus contorta, Abies lasiocarpa) and deciduous (Populus spp.) forests (Beckingham et al., 1996, Strong, 1992). The region encompasses an elevation gain from about 1000 m on the east, to about 2400 m along its
Results
Over the seven-year monitoring horizon, the total area of change detected in the study area was 385 km2, (i.e., 4.4% of the study area), corresponding to a mean annual rate of change of 0.63 ha/km2/yr (i.e., 0.63%/yr) (Fig. 4). Annual rates of change remained relatively close to the mean, with a maximum fluctuation of 22% (i.e., maximum rate of 0.77 ha/km2/yr) occurring between the years 2002 and 2003 (Fig. 4).
Disturbances and change between 1998 and 2005
The D-I framework to landscape monitoring generated a complete collection of spatially and temporally discrete disturbance features for the foothills study area, which enabled the estimation of landscape change between 1998 and 2005. Over this time frame, the region as a whole experienced a mean annual rate of change of 0.63%, leading to an increase in mean cumulative area covered by disturbances from 6.3 to 10.7 ha/km2. As a means of comparison, the following mean annual rates of change were
Acknowledgements
We thank Dr. Marie-Josée Fortin for her support and critical feedback to the first version of this manuscript. We are also grateful to Dr. Guillermo Castilla for constructive review comments throughout the research period of this manuscript. This work has been funded in part through the Natural Science and Engineering Research Council of Canada (NSERC) grant to Gregory J. McDermid. Julia Linke was directly supported by an Alberta Ingenuity Award for the data preparation phase of this research,
References (73)
Extending post-classification change detection using semantic similarity metrics to overcome class heterogeneity: A study of 1992 and 2001 U.S. National Land Cover Database changes
Remote Sensing of Environment
(2008)- et al.
Measuring the changing state of nature
Trends in Ecology and Evolution
(2003) - et al.
Estimating error in an analysis of forest fragmentation change using North American Landscape Characterization (NALC) data
Remote Sensing of Environment
(2000) - et al.
Parsimony in landscape metrics: Strength, universality, and consistency
Ecological Indicators
(2008) - et al.
Forest change detection by statistical object-based method
Remote Sensing of Environment
(2006) - et al.
Corine land-cover change detection in Europe (case studies of the Netherlands and Slovakia)
Land Use Policy
(2007) A new method for gridding elevation and stream line data with automatic removal of pits
Journal of Hydrology
(1989)- et al.
From space to species: Ecological applications for remote sensing
Trends in Ecology & Evolution
(2003) - et al.
The causes of land-use and land-cover change: Moving beyond the myths
Global Environmental Change
(2001) - et al.
Fragmentation of a forested Rocky Mountain landscape, 1950–1993
Biological Conservation
(1996)