Scott Mitchell

Use of vector polygons for the accuracy assessment of pixel-based land cover maps

Identifying appropriate validation sources for large-area land cover products is a challenge, with logistical constraints frequently necessitating the use of preexisting data sources. Several issues exist when comparing polygon (vector-based) datasets to raster imagery: geolocational mismatches, differences in features or classes mapped, disparity between the scale of polygon delineation and the spatial resolution of the image, and temporal discrepancies. To evaluate the potential impact of using vector coverages to assess the accuracy of pixel-based land cover maps, five evaluation protocols are applied to test sites located in British Columbia and Newfoundland and Labrador, Canada. One protocol directly compared the land cover of the sample unit to the land cover of the forest inventory polygon within which the sample unit fell, two protocols used different regions around the sample unit to define the land cover class, and two protocols were based on homogeneity criteria that restricted the selection of sample units. For the protocols tested, the overall accuracy values ranged from 34% to 58%. Given the broad range of accuracies achieved, the results suggest that caution is needed when making spatially explicit comparisons between raster and vector datasets. When possible, the use of purpose-collected validation data is recommended for the accuracy assessment of maps derived from remotely sensed data; if preexisting vector-based data are the only option for the validation, approaches accounting for the heterogeneity of classes within a given polygon are recommended.

Impacts of Spatial Partitioning in Hydroecological Models: Predicting Grassland Productivity with RHESSys

Environmental models constructed with a spatial domain require choices about the representation of space. Decisions in the adaptation of a spatial data model can have significant consequences on the ability to predict environmental function as a result of changes to levels of aggregation of input parameters and scaling issues in the processes being modelled. In some cases, it is possible to construct a systematic framework to evaluate the uncertainty in predictions using different spatial models; in other cases, the realm of possibilities plus the complexity of the environmental model in question may inhibit numeric uncertainty estimates. We demonstrate a range of potential spatial data models to parameterize a landscape-level hydroecological model (RHESSys). The effects of data model choice are illustrated, both in terms of input parameter distributions and resulting ecophysiological predictions. Predicted productivity varied widely, as a function of both the number of modelling units, and of arbitrary decisions such as the origin of a raster grid. It is therefore important to use as much information about the modelled environment as possible. Combinations of adaptive methods to evaluate distributions of input data, plus knowledge of dominant controls of ecosystem processes, can help evaluate potential representations. In this case, variance-based delineation of vegetation patches is shown to improve the ability to intelligently choose a patch distribution that minimizes the number of patches, while maintaining a degree of aggregation that does not overly bias the predictions.

Developing a geocollaboratory for Indigenous tourism research

Collaboratories have been defined as virtual places where collaborative research can be undertaken. As part of the Aboriginal Tourism Network (ABORINET), a geocollaboratory was developed to support Indigenous tourism research. Indigenous communities are culturally distinct and remotely located and this presents geographic and sociocultural constraints when conducting research on issues affecting these communities. ABORINETs development focused on the specific goal of enabling collaboration between researchers and Indigenous peoples on issues related to Indigenous tourism planning and management, and the general issue of enabling the sharing of differing knowledge and management approaches among research and Indigenous communities. The purpose was to develop a multi-scale and multi-method data collection and analysis protocol for better understanding Indigenous tourism in a way that supports multi-site and longitudinal comparisons, for connecting Indigenous communities across the world, and for sharing the results in ways that are meaningful to stakeholders within and beyond Indigenous communities. This paper outlines the development of the geocollaboratory and describes the lessons learned with specific attention afforded the geographical nature of the collaboratory. Recommendations for mitigating challenges are proposed and future research opportunities are identified.

Effect of Different Grazing Intensities on the Spatial-Temporal Variability in Above-Ground Live Plant Biomass in North American Mixed Grasslands

Abstract. Areas with relatively high spatial heterogeneity generally have more biodiversity than those that are spatially homogeneous areas due to increased potential as habitat. Management practices such as controlled grazing also affect the biodiversity in grasslands, and we hypothesize that this is due in part to the impacts of variation in grazing on plant heterogeneity and its spatial patterns. Understanding these mechanisms is important for designing an effective grazing system from a livestock management point of view. We used satellite-based, above-ground, live plant biomass (ALB) estimates at a pasture scale, in an experimental area located across the border of the East Block of Grasslands National Park (GNP) and an adjacent community pasture, to assess the effects of 5 intensities of grazing on the spatiotemporal pattern of ALB in mixed grasslands. Overall, heterogeneity increased with grazing intensity, whereas the spatial range decreased, except at the highest intensity, which had no impact on heterogeneity. Résumé. Les zones à relativement forte hétérogénéité spatiale ont généralement une plus grande biodiversité que les zones spatialement homogènes en raison de l’habitat potentiel accru. Les pratiques de gestion telles que le pâturage contrôlé influencent également la biodiversité dans les prairies. Nous émettons l’hypothèse que cela est dû en partie à l’impact des variations du pâturage sur l’hétérogénéité de la végétation et ses structures spatiales. La compréhension de ces mécanismes est importante pour la conception d’un système de pâturage efficace d’un point de vue de la gestion du bétail. Nous avons utilisé des estimations satellitaires de la biomasse végétale vivante aérienne à l’échelle du pâturage dans une zone expérimentale située à la frontière du bloc Est du parc national des Prairies «GNP» et un pâturage communautaire adjacent pour évaluer les effets de 5 intensités de pâturage sur la structure spatio-temporelle de la biomasse végétale vivante aérienne dans les prairies mixtes. Dans l’ensemble, l’hétérogénéité a augmenté avec l’intensité du pâturage tandis que la portée spatiale a diminué, sauf à la plus haute intensité qui n’a eu aucune incidence sur l’hétérogénéité.

Monte Carlo-Based Analysis of the Effect of Positional and Thematic Uncertainties on Biodiversity Models

Monte Carlo methods are a common approach to quantifying uncertainty propagation. We used Monte Carlo simulation to quantify the effects of positional and thematic uncertainties in a set of landscape maps on model averaged regression coefficients that were based on metrics derived from these maps. Results indicate that the uncertainty estimates from model averaging outweigh the effects of positional and thematic uncertainties in the landscape maps. Shifts between reference and simulated coefficients indicate a need for further research into simulation approaches that account for spatial autocorrelation.

The importance of accurate visibility parameterization during atmospheric correction: impact on boreal forest classification

Observation of the Earths surface from spaceborne platforms is complicated by the various layers of the Earths atmosphere that reflect, scatter, and attenuate electromagnetic radiation passing through them, thus influencing (upward or downward) the signal strength recorded at the sensor relative to the true quantity of radiance reflected from the observed surfaces. The magnitude and spatial distribution of atmospheric effects is non-stationary and will vary due to numerous factors. While the effect of these factors cannot be eliminated completely, the understanding of radiative transfer physics, atmospheric states, and electromagnetic wave propagation permits much of these effects to be appropriately modelled and minimized. Such corrections for atmospheric effects permit the extraction of more accurate physical properties of surface materials and states from imagery than if atmospheric effects were ignored. Modelling of atmospheric effects with radiative transfer models, however, requires appropriate parameterization. We explore the sensitivity of the important visibility parameter of the popular Atmospheric and Topographic Correction (ATCOR) model for atmospheric correction over boreal forest land cover. Further, we provide a methodology for estimating reasonable values for the visibility parameter in the event that this information is not readily available. Our sensitivity analyses, performed on Landsat 7 Enhanced Thematic Mapper Plus (ETM+) imagery from northern Québec and Ontario, rely on both incremental adjustments to the visibility parameter to assess the degree of atmospheric effect removal and the cascading effect on land-cover classification. We build confidence around our measures using a spatial bootstrapping analysis within each of the two images we analyse. Our analysis demonstrates that exceeding a magnitude of error of approximately 2 km in estimating a visibility parameter values can decrease classification accuracy by nearly 10%. Our assessments of the spatial structure of the mitigated atmospheric component within our scenes, testing for complete spatial randomness, clustering of like values, or evenness in value distributions are inconclusive, but hint towards more clustered results with greater magnitudes of parameterization error.