Elsevier Editorial System(tm) for Remote Sensing of Environment Manuscript Draft Manuscript Number: Title: Object-based mapping of the circumpolar taiga-tundra ecotone with MODIS tree cover Article Type: Full length article Keywords: taiga-tundra ecotone, MODIS, VCF, CAVM, CAPI Corresponding Author: Mr. Ross F. Nelson, Corresponding Author's Institution: First Author: Kenneth J Ranson, PhD Order of Authors: Kenneth J Ranson, PhD; Paul M Montesano, MS; Ross F. Nelson Research Highlights Research Highlights: Object-based mapping of the circumpolar taiga-tundra ecotone with MODIS tree cover - MODIS Vegtation Continuous Fields data used to map northern transition. - Circumpolar northern boreal transition from taiga to tundra mapped as an ecotone. - Compare MODIS transition zone to 2 tree lines and Canadian Landsat % tree cover map: - Tree line 1 – CAVM: corresponds to 1- 23% MODIS % tree cover. - Tree line 2 – CAPI: corresponds to 1-12% MODIS % tree cover. - MODIS-Landsat differences of ~25% (worst case) demonstrated. *Manuscript 1 Object-based mapping of the circumpolar taiga-tundra ecotone 2 with MODIS tree cover 3 4 K.J. Ransona, P.M. Montesanob, and R. Nelsona,c 5 6 a. Code 614.4, Biospheric Sciences Branch, NASA/Goddard Space Flight Center, Greenbelt, Maryland 7 20771, USA 8 b. Sigma Space Corp., Lanham, MD,20706 USA 9 c. corresponding author: email: [email protected], telephone: 1-301-614-6632 10 11 Abstract 12 The circumpolar taiga-tundra ecotone was delineated using an image 13 segmentation based mapping approach with multi-annual MODIS Vegetation Continuous 14 Fields (VCF) tree cover data. Circumpolar tree canopy cover (TCC) throughout the 15 ecotone was derived by averaging MODIS VCF data from 2000 - 2005 and adjusting the 16 averaged values using linear equations relating MODIS TCC to Quickbird-derived tree 17 cover estimates. The adjustment helped mitigate VCF’s overestimation of tree cover in 18 lightly forested regions. An image segmentation grouped pixels representing similar tree 19 cover into polygonal features (objects) that form the map of the transition zone. Each 20 feature represents an area much larger than the 500m MODIS pixel to characterize the 21 patterns of sparse forest patches on a regional scale. Comparisons of the adjusted average 22 tree cover data were made with (1) two existing tree line definitions aggregated for each 23 1° longitudinal interval in North America and Eurasia and (2) Landsat-derived Canadian 24 proportion of forest cover for Canada. The adjusted TCC from MODIS VCF shows, on 25 average, <12% TCC for all but one regional zone at the intersection with independently 26 delineated tree lines. Adjusted values track closely with Canadian proportion of forest 27 cover data in areas of low tree cover. Those polygons near the boreal/tundra interface 28 with either (1) mean adjusted TCC values between 5-20% , or (2) mean adjusted TCC 29 values <5% but with a standard deviation > 5% were used to identify the ecotone. 1 30 31 Introduction 32 Earth’s longest vegetation transition zone, the circumpolar taiga-tundra ecotone 33 (TTE), stretches for more than 13,400 km around Arctic North America, Scandinavia, 34 and Eurasia (Callaghan et al. 2002a). This ecotone is subjected to climate change 35 (Serreze et al. 2000), and sensitive to climate change (Callaghan et al. 2002b). The 36 shifting of local subarctic tree lines throughout the forest-tundra biome is linked to 37 ecological processes at different spatiotemporal scales and will reflect future global 38 climatic changes (Payette et al. 2001). Monitoring the dynamics of this forest-tundra 39 boundary is critical for our understanding of the causes and consequences of the changes. 40 High-latitude ecosystems, i.e. boreal forests and tundra, play an important role in the 41 climate system (Bonan et al. 1995) and have been warming in recent decades (Chapin et 42 al. 2005). Modeled effects of increased deciduous tree cover in high-latitudes show a 43 positive feedback with temperature in the vegetated regions of the arctic (Swann et al. 44 2010). Improved understanding of the role of the forest-tundra boundary requires a 45 concerted research effort to be conducted over a long enough time period to detect and 46 quantify ecosystem feedbacks (Chapin et al. 2000). The objective of this study was to 47 develop a framework for long term monitoring of the circumpolar tundra-taiga ecotone 48 using satellite data. We accomplish this by preparing a validated map of MODIS percent 49 forest cover for the entire ecotone with repeatable methods. This goal is to provide a 50 benchmark for future studies of the tundra-taiga ecotone. We accomplish this by 51 preparing a validated map of MODIS percent forest cover for the entire ecotone using 2 52 procedures easily replicated by others. The results provide a benchmark for future 53 studies of the tundra-taiga ecotone. 54 55 Background 56 The TTE is dynamic because it is very sensitive to human activity and climate 57 change. For example, during the last 6000 years in northern Eurasia a general cooling 58 trend of about 2-4oC was associated with a 400 – 500km southward retreat of larch and 59 birch forest stands (Callaghan et al. 2002b). In the past three decades global average 60 surface air temperatures have risen by approximately 0.6oC (Hansen et al. 2006) while 61 temperatures in parts of the Northern Hemisphere have warmed by as much as 2oC 62 (Hansen et al. 1999). If it is assumed that growth and reproduction are controlled by 63 temperature, a rapid advance of the tree line would be predicted (Grace et al. 2002). The 64 northward movement of the TTE may result if climatic warming persists over centuries 65 or millennia (Skre et al. 2002). Some studies predict that up to about one half of the 66 tundra could be colonized by trees by 2100 (Callaghan et al. 2002b; Harding et al. 2001). 67 There is general agreement that temperature is important in determining the 68 northern extent of the boreal forest. Widespread degradation of permafrost has been 69 shown in numerous studies (Pavlov 1994; Osterkamp and Romanovsky 1996; Osterkamp 70 and Jorgenson 2006; Jorgenson et al. 2006). Recent observations suggest that, in an 71 indication of northern climate warming, shrubs and forests are expanding into the tundra 72 (e.g., Kharuk et al. 1998, 2002; Esper and Schweingruber 2004; Tape et al. 2006; Blok et 73 al. 2010) and experimental evidence shows that increased shrub cover in the tundra is a 74 response to warming (Walker et al. 2006). In a case study conducted by Rees et al. 3 75 (2002) in a portion of the West Siberian plain (66.5oN, 70.75oE) from 1968 to 1997, 76 seven of their 20 test sites showed colonization by advancing forests, one showed an 77 inconclusive shift in the tree line and one test site, which was surrounded by water and 78 wetlands, showed evidence of forest retreat. The remaining 11 sites showed no 79 appreciable forest advance but four of these sites had developed denser forest cover. In 80 the Polar Ural mountains crown closure of stands increased 4-5 times and the tree line 81 boundary shifted 100-300 m into the tundra (Kharuk et al. 1998). Devi et al. (2008) 82 estimate the past century’s forest expansion in this region has led to an increase in 83 biomass of 40-75 Mg/ha. These observations are consistent with those in northern 84 Canada (Payette and Gagnon 1985) while transects through the forest-tundra in eastern 85 Canada have shown no significant correlations in the short term (5-20 years) with 86 climatic variables, but do reveal a rise in tree line (Gamache and Payette 2005). In the 87 TTE in western, central and eastern Canada white spruce density increased without any 88 significant displacements of the Arctic tree line over the past 100-150 years (Payette et al. 89 2001). Recently established trees near the forest limit at coastal sites east of Hudson Bay 90 are associated with warmer conditions over the past 100 years (Laliberté and Payette 91 2008). However, from the 1970s – 1990s, a Landsat-based study of the TTE in two areas 92 of northern Canada has also shown that the extent of the boreal forest in these areas 93 remain stable (Masek 2001). It is important to note that the position of northern portion of 94 the boreal forest is likely influenced by a range of conditions that vary by region 95 (Callaghan et al. 2002a; Crawford and Jeffree 2007; Crawford 2005). 96 Recent work indicates increased Arctic vegetation growth with climate warming. 97 Experiments have shown that a 1-3°C temperature warming can increase the height and 4 98 cover of deciduous shrubs in the tundra after two growing seasons (Walker et al. 2006). 99 Analysis of shrub growth rings in northwestern Russian Arctic tundra has revealed field 100 data that closely follows the patterns found in the greening signal of shrubs in the satellite 101 record since the 1980s (Forbes et al. 2010). Results from the analysis of Landsat time 102 series in Canada’s forest-tundra interface suggests shrub replacement of bare ground and 103 lichen cover (Olthof et al. 2008). Kharuk et al. (2005) studied the expansion of evergreen 104 conifers into larch-dominated forest, noting the increase in tree-stand density and the 105 spread of larch in the ecotone. However, the patterns of vegetation growth may not be 106 consistent across the taiga-tundra transition (Bunn and Goetz 2006; Verbyla 2008) or 107 may be interpreted differently depending on the scale of the analysis (Alcaraz-Segura et 108 al. 2010). Furthermore, vegetation change may in many cases depend on site-specific 109 characteristics, particularly where there is a direct effect from human activities (Crawford 110 et al. 2003; Virtanen et al. 2002; Vlassova 2002; Toutoubalina and Rees 1999; Rees et al. 111 2002; Hagner and Rigina 1998). 112 The transition from taiga to tundra is characterized by a change in tree cover 113 density. The TTE is not a distinct edge but a transition area where patches of tundra and 114 forest are mixed. It is not easily defined and can be difficult to identify and quantify. 115 Timoney et al. (1992) described the TTE in Canada in terms of the ratio forest to tundra 116 identified on aerial photographs and Timoney (1995) highlights major differences in the 117 ecotone from west to east. Across the circumpolar ecotone the abundance of trees has 118 been used to identify vegetation patterns and monitor these patterns for shifts indicative 119 of vegetation’s response to human activities and climate changes. Rees et al. (2002) 120 discuss identifying and monitoring the TTE in Russia using Landsat and Synthetic 5 121 Aperture Radar (SAR) data and the potential and limitations associated with satellite 122 observation. Esper and Schweingruber (2004) suggest that there is a circumpolar trend of 123 changes in treeline occurring where there are notable temperature changes and consistent 124 monitoring methods are needed (Frey and Smith 2007). Spectral un-mixing methods 125 (Sun et al. 2004) and multi-angular and –temporal data (Heiskanen and Kivinen 2008) 126 have been used to characterize tree cover in the TTE. Stow et al. (2004) discuss multi- 127 temporal remote sensing of land cover change applications in the Arctic, presenting case 128 studies with a range of spatial resolutions and extents with durations spanning months to 129 decades. These studies have shown that the climatically induced changes in vegetation 130 cover and composition in the Arctic need to be monitored at different spatial scales. 131 Ranson et al. (2004) examined the capabilities of remote sensing data for identifying the 132 tundra-taiga transition zone in Ary-mas, Siberia by mapping the tree abundances. Their 133 results showed that Landsat-7 ETM+ image and C-band SAR (RADARSAT) were 134 adequate for mapping tree density in this area. They also demonstrated that lower 135 resolution data such as MODIS can also be used to characterize the transition zone when 136 adequate training data are available. 137 Image segmentation has been used with MODIS data (Tilton et al. 2006) and to 138 delineate forest features and analyze forested areas at multiple scales (Pekkarinen 2002; 139 van Aardt et al. 2006; Achard et al. 2009). Segmentations of high resolution digital aerial 140 imagery were used to delineate vegetation (seagrass) percent cover (Lathrop et al. 2006). 141 The purpose of this technique is to group pixels into meaningful image objects that 142 represent landscape features. While groups of adjacent remote sensing pixels within a 143 particular land cover category may be similar, they typically exhibit internal 6 144 heterogeneity. The segmentation process aims to minimize internal segment (image 145 object) heterogeneity by accounting for both local image texture and the size of groups of 146 pixels (Baatz and Schäpe 2000). Image segmentation procedures are useful for 147 characterizing boundaries between ecological regions, where patches of vegetation of 148 varying densities and gaps can signal a transition between land cover types. McMahon et 149 al. (2004) suggest that research on ecological regions should consider the hierarchical 150 relationship of landscape elements. The mean and variance associated with individual 151 image objects, coupled with the hierarchical relationship of objects and sub-objects can 152 be used to quantify changes in vegetation patches and gaps across a geographic gradient. 153 Furthermore, the segmentation process can facilitate remote sensing data fusion, by 154 incorporating information from multiple datasets on a per-segment basis. 155 156 Methods 157 MODIS data acquisition and processing 158 Figure 1 is a composite Terra MODIS image of the northern hemisphere above 159 50o latitude. This image represents the geographic scope and diversity in the vegetated 160 boreal and arctic zones. Collection 4 (C4) MODIS VCF (MOD44b) data (USGS 161 LPDAAC 2010) was acquired as 10° x 10° tiles for 6 years (2000 – 2005) for the 21 tiles 162 shown in Figure 2. Tiles for continental North America spanned 50°N – 70°N and tiles 163 for Eurasia spanned 60°N - 70°N. No C4 MODIS VCF data existed above 70°N, which 164 accounts for the data gap in northern Siberia. The data tiles were mosaicked by year for 165 both North America and Eurasia using ENVI 4.5 software (ENVI 2010). The 6 years of 166 data were then combined to derive an average VCF dataset, where each pixel represented 7 167 the mean percent tree canopy cover value of the corresponding pixel across all of the 6 168 VCF yearly datasets. Finally, the Collection 5 MODIS land cover (MCD12Q1, IGBP 169 global vegetation classification scheme) was used to change those VCF pixels whose 170 corresponding land cover type was water to tree canopy cover values of zero. Two multi- 171 annual datasets, one for continental North America and one for Eurasia resulted from this 172 processing sequence, each maintaining the original sinusoidal projection of the 463.3m 173 (nominally 500m) pixel grid. 174 VCF Adjustment 175 Each continental multi-annual VCF mosaic was adjusted based on inversions of 176 the results of linear regressions, shown in Table 1, relating Quickbird-validated percent 177 tree cover to MODIS VCF percent tree canopy cover (Montesano et al. 2009). The 178 adjustments were applied for seven broad longitudinal zones representing very general 179 circumpolar regions on a pixel-by-pixel basis (Figure 3). A value of zero was assigned to 180 those pixels for which the adjustment equations resulted in negative values, while a value 181 of 100 was assigned to those pixels whose adjusted values exceeded 100 percent tree 182 canopy cover. 183 The tree canopy cover values for pixels of the adjusted multi-annual VCF mosaics 184 (VCF ) were examined within the taiga-tundra ecotone using two existing definitions of adj 185 tree line. Vector polylines, approximating the northern limit of trees, from both the 186 Circumpolar Arctic Vegetation Map (CAVM) (Walker et al. 2005) and the Circum-Arctic 187 Map of Permafrost and Ground-Ice Conditions (CAPI) (Heginbottom et al. 1993) were 188 decomposed into points at intervals of 463.3m (the size of a VCF pixel). These points 189 were used to sample corresponding VCF pixels within one-degree longitudinal adj 8