A word on restoration ecology: Researchers, practitioners, and policymakers alike have a hard time agreeing on how to define the nascent field of restoration ecology. Part of the reason for this is that the field, by nature, is multidisciplinary and informed by a wide range of study areas. The articles reviewed here are similarly diverse and cover a broad spectrum of topics relevant to restoration. 

Ralph J. Garono, McFall, Kerry, Burke, Jennifer, and Sutherland, Bruce. 2003. “Critical habitat on the lower Columbia River." Geospatial Solutions 13(11): pp36-41.

This trade journal article detailed the process by which a collaboration of universities, non-profits, and consultants created a GIS dataset used to evaluate habitat change and make management recommendations in the Lower Columbia River Estuary. The authors underscore the challenges of mapping in such a temporally and spatially dynamic system, subject to small-scale changes related to storm surges, light reflectivity, etc. Their solution is a hierarchical data collection approach, which relies on coarse-scale (30m resolution) satellite imagery of the entire reach combined with fine-scale (1.5m resolution) aerial imagery targeting only focal areas. Both datasets were geo-referenced to allow their combination despite their different scales. I was impressed with their ground-truthing process, by which volunteers took cover estimates, digital photos, and multispectral data with a radiometer within each flight path. This information was then used to ‘train’ processing software to match pixels collected by satellite and air to the information collected on the ground. The processing formula then was used to classify similar areas without needing to ground-truth them. The applications of the resulting dataset is manifold: Digitized topographical maps ca. 1800 were overlaid to assess habitat change; restoration and conservation sites were prioritized vis a vis connectivity and invasive plant cover; and habitat classes were matched to known salmonid responses to quickly evaluate habitat quality.  

Fistikogul, Okan and Nilgun Harmancioglu. “Integration of GIS with USLE in assessment of soil erosion.” Water Resources Management 16: pp. 447–467.

This article underscores the efficacy of GIS use in natural resource management as a tool used to predict potential scenarios and to evaluate alternative land management plans, but cautions that the accuracy and utility of sophisticated tools like GIS are often hampered by data limitations. The authors demonstrate the use of GIS in land management by integrating a simple soil erosion model, the Universal Soil Loss Equation, with GIS to predict erosion and consequent runoff nutrient loads in a small watershed in Turkey. GIS is particularly suitable to this task as erosion processes have spatial character; the impact of alternative management scenarios can be rapidly evaluated and displayed; and celll size can be controlled to reflect the level of data resolution. The authors first digitized soil, topographic, rainfall, land use, and hydrological maps into raster format, then multiplied cell values of each layer in a raster overlay to calculate erosion potential for each cell in the watershed. The authors point out that sophistication in the data model may in fact become a liability if data equal in complexity to the model are unavailable. They also emphasize that a substantial environmental database must be established to support such calculations, and lament that a primary challenge to doing so is a lack of a data clearinghouse making locally collected information available to a wider range of users.

Wilson, Jenny and Kim Lowe. 2003. “Planning for the restoration of native biodiversity within the Goulburn Broken Catchment, Victoria, using spatial modeling.” Ecological Management and Restoration 4(3): pp. 212-219.

These workers take advantage of the explicitly spatial nature of conservation biology principles to model habitat planning in GIS.  They adapted conservation axioms including remnant size and proximity, extent and diversity of vegetation within that remnant, and connectivity of remnants within the landscape to develop a set of rules to spatially analyze land use change.  This was done by overlaying a 400m grid on tree cover, hydrology, road, vegetation class, and bioregion data layers, and then performing a series of buffers on various objects to expand their spatial extent until various conservation criteria were met. For instance, vegetation remnants should be at least 40ha large and connected to their nearest neighbor via buffered riparian corridors. 

Several issues were raised by these researchers about the inability of the data layers used to satisfactorily answer some questions.  For instance, the vegetation layer was based on satellite imagery that failed to capture vegetation other than trees, or tree layers with tree density less than 10%; this obviously leaves many areas of potential biological importance out of analysis.  Additional data layers, such as land-use surrounding remnants, would also add essential planning information.  Spatial and temporal scale also posed difficulties.  At the scale of their data sources (1: 100,000 – 1:500,000), some of the important features might have been missed in the original mapping, and therefore excluded from analysis. (However, I would argue objects overlooked due to low resolution mapping are not likely to be biologically relevant in the conservation planning context.)  Perhaps more importantly, static analysis at such low resolution inevitably fails to capture not only those ecological processes that operate at high resolution, but also those that operate at multiple spatial and temporal scales.

Wilson and Lowe recognize the immense utility of GIS to aid in a more strategic application of limited resources to restoration planning.  The map produced is of use in answering several land management questions recognized by these authors: What is the distribution pattern of the buffered (biologically relevant) areas? Are some remnant sizes or vegetation types underrepresented in certain areas? How much land is needed to meet existing conservation targets?  Given potential land use scenarios, what are the predicted restoration outcomes? Having worked in restoration, I applaud their much-needed efforts to increase the efficacy of conservation planning. However, I would add another indispensable use to maps such as theirs: selecting restoration sites in the first place.

Von Holle, Betsy, and Glenn Motzkin. 2007. “Historical land use and environmental determinants of nonnative plant distribution in coastal southern New England.” Biological Conservation 136: pp. 33-43.

Holle and Motzkin set out to shed further light on what factors govern the susceptibility of habitats to biological invasion.  They map nonnative species distributions in upland sandplain vegetation communities to address three hypothesis put out by other workers: Nonnative species are more frequent and abundant in areas that have younger stands, open canopy, and low native species richness; nonnative species distributions are a result of spatial variation in recent and historical anthropogenic disturbance; and, especially in localities with dry, nutrient poor soils, nonnative species distribution and abundance are largely dictated by edaphic (soil) conditions.  Nonnative species distributions were mapped via vegetation and soil samples in 776, 20m x 20m plots.  These researchers rely heavily on a spatially-referenced GIS database to analyze species distributions by overlaying major roads, geology, political boundaries, historic land cover, and historical land use on their sampling data.  The layers containing historical information had to first be digitized from other records; often the resolution of these older maps determined the grain at which analysis could take place.  Analyses of plant distributions relative to distance from roads, in particular, would have been unwieldy without GIS (although this was not ultimately an important factor, it is often used as a proxy for likely human disturbance).  Holle and Motzkin found nonnative species to be widespread and common but not abundant. Of all the variables tested, those that best predicted higher nonnative species occurrence were glaciolacustrine and till deposits (as opposed to drier, coarser soils like moraine or sand); open-canopy, historically disturbed or cultivated areas; and soils with greater calcium and phosphorous.  All of these factors are likely interrelated, as human activities like agriculture tend to be more common on more mesic, rich glaciolacustrine soils, and agricultural practices such as liming and fertilization can raise some soil nutrients. Interestingly, native species richness seemed to have little effect on nonnative cover. Instead, the type of native species present, especially woody, broad-canopy species, have more mitigating influence.

Platt, Rutherford, Thomas Veblen, and Rosemary L. Sherriffz. 2006. “Are wildfire mitigation and restoration of historic forest structure compatible? A spatial modeling assessment.” Annals of the Association of American Geographers 96(3): pp. 455–470.

Much money and effort are currently being directed towards forest fuels reduction, based on the premise that fire suppression has resulted in unnaturally dense forests that will burn with unnaturally high intensity.  Fuels reductions projects often have duel goals of reducing the danger of catastrophic fires, and restoring forests to historic stand structure.  Platt et al. use an overlay of three data layers to identify areas near Boulder, Colorado, that should be targeted for fuels reduction.  Their first data layer is a prediction of potential fire behavior developed from 30m DEMs, canopy cover estimated with Landsat imagery, and fuels data hand-drawn on aerial photos. These fire behavior models were used to assign raster cells to low, high, and extreme potential intensities. Cells were assigned to reconstructed historical fire frequency classes in the second layer. The third layer was a land ownership raster.  By executing a select procedure that chose cells of high potential fire behavior where fire frequency was lower than it had been historically, they identified forests most in need of treatment.  Their study showed that, surprisingly, much less area fell into this high-priority category than had been assumed, and that most target area was owned by those agencies not receiving the bulk of fuels reduction funds.

These researchers dealt with the common challenges of representing fire, an extremely spatially and temporally variable phenomenon, with static, coarse scale data.  (The failure by biologists to incorporate such heterogeneity has, in fact, resulted in mis-reconstruction of ‘historical’ burning patterns for many areas.)  Nonetheless, expense and time necessitate that these researchers move forward with the model after making several simplifying assumptions.  They dealt with temporal and spatial variability by assigning static average values in some cases, interpolating between known values, and being careful to apply map-based inferences to broad geographical areas instead of local sites.  Platt et al. also took an additional step to address the challenge of spatial and temporal variability that I have not often seen in other studies: They validated their maps by selected ground-truthing, and refined their model by studying how inferences differed when individual variables were manipulated.  They stress that actual management plans should be developed using finer resolution data than were available here.

Dömötörfy, Zsolt, David Reeder, and Piroska Pomogyi. 2003. “Changes in the macro-vegetation of the Kis-Balaton Wetlands over the last two centuries: a GIS perspective.” Hydrobiologia 506-509(1-3): pp. 671–679.

Restoration of the Kis-Balaton wetlands surrounding Lake Balaton, Hungary, began 150 years after they were canalized and drained. Historical records and maps of the area kept for a variety of purposes were interpreted to model the reestablishment of biofiltration and related water quality processes through time. Domotorfy et al. set out to demonstrate the use of GIS in mapping and quantifying historical vegetation change, and to construct a Digital Elevation Model of the area to predict vegetation composition over time from water depth calculated from the DEM.  Their main data sources were digitized military survey maps, digitized aerial photographs, TINs produced from interpolated contour maps, and GPS vegetation surveys.  Because these many sources were not in a consistent coordinate system, they had to rubbersheet basemaps using historically fixed-object Ground Control Points.  Vegetation maps were incorporated via heads-up digitizing and extensive interpretation of metadata necessary to resolve the works of different cartographers using divergent and often contradicting symbologies.  Finally, these authors used the DEMs they created to model water depths across the grid area based on known draining and reflooding activities around the Kis-Balaton wetlands.  These data sources were used together to try to understand the process of wetland survival and regeneration.

Of particular interest to me is Domotorfy et al.’s description of the problems they faced in reconstructing historical biotic patterns from sources not intended for such a purpose.  For instance, the objective of the military in their mapping of the area was to record nature of the land as it was suitable to militia, not necessarily emphasizing accuracy in classing land in terms of biotic communities. Additionally, such historical snapshots can depict one point in time that is not representative of an average trend for that period. These researchers do an admirable job of using a range of secondary records to resolve such conflicts. 

Robb, James T. 2002. “Assessing wetland compensatory mitigation sites to aid in establishing mitigation ratios.” Wetlands 22(2): pp. 435-440.

Permitting agencies often require creation, restoration, enhancement, or preservation of wetlands to compensate for those lost through development.  Because many of these compensation projects either fail to be executed, or fail to produce viable replacement wetlands, state and federal agencies usually require more wetland area to be installed than will be lost; the ratio of compensation area to lost area is the ‘mitigation ratio.’ In order to help amass data on which to establish mitigation ratios, Robb inventoried mitigation wetlands across Indiana, categorized each as constructed, incomplete, or no attempt at construction, and then measured area within a subsample of constructed wetlands to assess establishment success.  Each subsampled site wetland line, and each vegetation cover type with the wetland area, was mapped with a highly accurate Trimble GPS.  Wetland boundaries were interpolated from a point taken every 5 seconds, and the area of each wetland vegetation class polygon calculated.  This survey and subsequent GIS analysis allowed the author to report that wetland area had been established equivalent to that lost, but far less than that required.  He also described patterns in which vegetation classes tended to fail or succeed; based on these rates, he then recommends empirically-based mitigation ratios needed to overcome failure rates specific to each vegetation class. Even though this study provides much-needed data on which to base sound land development policy, the author acknowledges that his analysis did not take into account the quality of created wetlands, and he forwards several suggestions intended to help assure mitigation requirements are not only followed, but also produce viable replacement wetland habitats.        
 

Appleton, Katy, and Andrew Lovett. 2002. “GIS-based visualization of rural landscapes: Defining ‘sufficient’ realism for environmental decision-making.” Landscape and Urban Planning, 65: pp. 117-113.

Decision-making often relies heavily on the visual portrayal of environmental data to increase transparency, participation, and broad understanding.  As current technology is capable of presenting data with a greater level of realism and detail than is actually supported by the resolution at which information was gathered, there is a danger that decision-makers give artificially realistic maps more weight than those that more representatively display available information.  Appleton and Lovett set out to define not only what level of detail in a map is sufficient for respondents to ‘relate’ to it, but also which map elements were most important to represent in detail.  They used a database based on a small rural area in the UK with a fairly fine scale (1:2500) and a 10m DEM, that was further supported by ecological surveys, georeferenced aerial imagery at 25cm resolution, and interviews with the inhabitants of the area.  On three standard maps, the detail of main elements (those under consideration in a decision-making process), and auxiliary elements (those merely providing context), were varied from plain color, to computer-generated texture, to photographed texture.  62 respondents then rated each image on a 1-9 scale based on how easily the map image shown allowed them to imagine the real landscape under consideration.  Respondents also rated the realism of the images relative to a photo. 

These researchers found that realism ratings differed between the three different maps used rather than the level of detail varied within a map, and that the relative importance of a map element was not related to its gross area.  Realism was most easily conveyed by increasing the detail of vegetation.  However, they did not request respondents to rate map realism or react to individually altered map elements in the context of an explicit land planning decision.  I find this work further hard to relate to how map realism affects the environmental decision-making process because neither the small area depicted (2km x 4km), nor the eye-level view, are necessarily related to those maps and images actually used in land management decision-making. 

Scott, Daniel, Jay Malcolm, and Christopher Lemieux. 2002. “Climate change and modeled biome representation in Canada’s national park system: Implications for system planning and park mandates.” Global Ecology & Biogeography, 11: pp. 475–484.

Because global warming is expected to be more pronounced at higher latitudes, changes in ecosystem composition are projected to be particularly striking in northerly countries like Canada.  Using two Global Vegetation Models (GVMs) and five General Circulation Models (GCMs), Scott et al. examine the potential biome changes in Canada’s national parks system.  As the directive of Canada’s national parks governing body (Parks Canada) is first and foremost to ensure ecological integrity and representation ‘for all time’, they also look at potential changes in park planning made necessary by global warming.  Both GVMs, which simulate potential distribution of plant types based on plant physiology, average seasonal climate, and hydrological conditions, were used in conjunction with each of the five GCMs, which were all based on a scenario of doubled CO2.  These authors used a raster representation of Canada’s 39 national parks (cell resolution = 0.5 degrees lat-long) to overlay park boundaries on potential vegetation maps predicted by the climate change and vegetation models.  Though not explicit, these authors seemed to satisfy any mixed pixel problems with a greater area approach.  All Global Vegetation Model and General Circulation Model combinations showed significant change in biome respresentation, including the appearance of at least one new biome in most parks, and a biome change in at least half of each park’s grid cells.  Specifically, tundra and taiga biome area decreased, while temperate forest and mixed forest generally increased.  Models inconsistently predicted boreal forest and savanna/woodland changes.  In this study, GIS was instrumental in facilitating not just whether change would occur, but also a magnitude of that change.  This greater level of information is essential if land managers are to develop tractable conservation plans in the face of global warming.  To the end of more effective conservation, Scott et al. stress that protecting ecosystem representation should be attempted by facilitating ecosystem resiliency, for instance by planning for connectivity, protecting outlying populations, and focusing on northern biomes,  Like other researchers, these authors caution that models presented here are equilibrium models, and can therefore only be a general guideline as they fail to incorporate many complexities such as changed migration rates, disturbance regimes, and competitive relationships.  

Vanreusel, Wouter, and Hans Van Dyck. 2007. “When functional habitat does not match vegetation types: A resource-based approach to map butterfly habitat.” Biological Conservation 135: pp. 202-211.

These authors assert that though much of modern ecology has been focused on mapping species population structure as a function of habitat, habitat has been most often represented simply by vegetation or land cover types.  As these variables are often insufficient to delineate areas that are actually usable to an organism, Vanreusel and Van Dyck seek to better define habitat based on the presence of key resources and conditions, and whether the spatial relationships between resources are on a scale appropriate to the organism’s behavior.  In the present case, they define habitat based on resources essential to larvae and adults of the green hairstreak butterfly (Belgium), combined with travel distances of the butterfly from mark-recapture studies. They then compare habitat defined in this way to habitat defined only by vegetation.  Over two seasons, workers GPS’ed the location of butterfly sighting, and mapped their mark-recapture movement on aerial photos.  Also recorded were use of essential resources (e.g. host plant, microclimate, nectar plants, mate-locating sites, etc.).  The entire area was mapped, areas of homogenous vegetation delineated as polygons on aerial photos, and the photos digitized.  A 5m x 5m grid was overlaid on this vector data to further classify vegetation polygons in terms of additional essential resources.  The hill-shade function, based on vegetation height, was used to determine insolation and microclimatic suitability.  After using the reclassify and map calculator functions, the resulting selection identified the varying quality of adult and larval habitat zones.  These zone boundaries were buffered, and if overlapping, were considered contiguous in the sense that an adult butterfly could travel between zones in a single day, modified by the vegetation-dependent ease of moving between grid cells (“Cost distance” function in Spatial Analyst).  Finally, the butterfly sightings layer, with sighting points buffered by 25m to account for error, was overlaid on both the habitat resource quality map, and on the simple vegetation-only map.  When actual butterfly sightings were compared between these maps, Vanreusel and Van Dyck found that the vegetation map accounted for only 22% of the area known to be used habitat.  These authors conclude that not only does the resource-based map contain more of the actual sightings than the vegetation-only map, but that the resource map also better identifies habitat shape and configuration, all of which are essential for good conservation and restoration planning.  This study is an excellent example of how GIS provides a platform for analyses that not only recognize the complex and multivariate nature of ecological models, but also provides a more intuitive and interpretable way to communicate findings.

Photo note: The photo above is the uplifted (Western) side of San Clemente Island, one of the Channel Islands. Having worked in restoration on the Channel Islands, I was excited to see them again as a focal point in lab datasets!