Ninth International Conference on Permafrost ... - IARC Research

Potential Inclusion of Vegetation Indices in Mountain Permafrost ModelingMarian Kremer, Antoni G. Lewkowicz, Michael Sawada, Philip P. BonnaventureDepartment of Geography, University of Ottawa, Ottawa, CanadaMark EdnieGeological Survey of Canada, Ottawa, CanadaIntroductionA widely-used method to model the distribution ofmountain permafrost employs basal temperature of snow(BTS) measurements as an indicator of the probability ofpermafrost presence. Multiple regression is used to developa spatial field of BTS values, and these values are used topredict the distribution of permafrost either through the BTS“rules-of-thumb” or through ground-truthing relationships(e.g., Lewkowicz & Ednie 2004). The best independentvariables for modeling BTS values are generally elevationand potential incoming solar radiation (PISR) (e.g., Gruber& Hoelzle 2001, Lewkowicz & Ednie 2004, Ødegård et al.1999). Multiple regression of BTS against these variablesgenerally results in r 2 values of 0.3–0.4, indicating that thereare other important factors affecting BTS values and hencepermafrost (Gruber & Hoelzle 2001, Lewkowicz & Ednie2004), one of which is vegetation.Vegetation affects the surface offset (Smith & Riseborough2002) by influencing turbulent energy fluxes, by shading theground surface in summer and by altering snow distributionin winter, especially in mountain catchments wheresignificant redistribution of snow may occur (e.g., Pomeroyet al. 2006). A small number of attempts have been madeto include vegetation in permafrost spatial models usingvegetation indices and land cover classifications createdfrom remotely sensed satellite images. However, there is nogenerally accepted method to represent vegetation for thispurpose.Despite its known theoretical significance, vegetation hasproven to be of little importance in the few European mountainpermafrost studies that have included it. For example, theNormalized Difference Vegetation Index (NDVI) which wasused by Ødegård et al. (1999) in southern Norway did notsubstantially improve statistical explanation because it washighly correlated with elevation, one of the other independentvariables. In Switzerland, Gruber and Hoelzle (2001) usedthe Soil Adjusted Vegetation Index (SAVI) to correct for thehigh reflectance of soil in the imagery, but obtained similarresults. Without it the r 2 value was 0.386, and with SAVI ther 2 increased by only 0.012 (Gruber & Hoelzle 2001).Attempts to incorporate vegetation into permafrostmodeling using land cover classifications have been moresuccessful. In the Yukon-Tanana Uplands of Alaska, LandsatThematic Mapper (TM) imagery was used to generate landcover classifications that included types of canopy cover(closed or open) and types of vegetation (coniferous forest,deciduous forest, mixed forest, shrub) (Morrissey & Strong1986). Using logistic discriminant functions with thisclassification and data from the thermal band of Landsat TMimagery, which provides similar information to PISR, threeclasses of permafrost distribution (frozen, discontinuouslyfrozen, and unfrozen) were predicted with reasonablesuccess. In the Mayo region, Yukon, Leverington, andDuguay (1996) also developed a vegetation classificationfrom Landsat TM imagery and used it to test models forpredicting active layer depths and the presence or absenceof permafrost. The land cover classification was found to beone of the most useful factors for predicting the presenceof permafrost. Etzelmüller et al. (2006) combined both landcover classification and NDVI in their multicriteria analysisof mountain permafrost distribution in Mongolia. Theyfound that NDVI was useful only after an initial land coverdivision into forested and nonforested areas.ObjectivesThe goal of this project is to examine which, if any, ofthe ways discussed above to incorporate vegetation maybe suitable for permafrost modeling in the mountains ofnorthwest Canada. Despite their lack of effectiveness inEurope, it is possible that vegetation indices may yet proveuseful, given differing patterns of permafrost distributionin relation to vegetation zones. Alternatively, land coverclassification or a hybrid approach may be the most effective.By testing models at sites in several climatological zones, wehope to determine if there is a single method that is effectiveor if adjustments to the methodology must be made to takelocal vegetation-climate relations into account.Study AreasThe eight field areas that are being examined for this studyrepresent all the major climatological regions of the southernYukon and include Wolf Creek near Whitehorse, Johnson’sCrossing, Sa Dena Hes mine north of Watson Lake, Faro,Keno, the Top-of-the-World Highway near Dawson, theRuby Range, and Haines Summit in extreme NW BritishColumbia. These sites span almost 5° of latitude (59°36′ to64°05′N), and all fall into zones of discontinuous permafrost(Heginbottom et al. 1995). Elevations in the areas generallyvary from about 700 m to 2000 m a.s.l., with the Dawsonarea extending down to about 320 m. Permafrost is presentat higher elevations in all of the areas as well as below treeline in most of them as a result of cold air drainage andhydrological variability.149