Mapping resource use over a Russian landscape: an integrated look at harvesting of a non-timber forest product in central Kamchatka

Small-scale resource use became an important adaptive mechanism in remote logging communities in Russia at the onset of the post-Soviet period in 1991. We focused on harvesting of a non-timber forest product, lingonberry (Vaccinium vitis-idaea), in the forests of the Kamchatka Peninsula (Russian Far East). We employed an integrated geographical approach to make quantifiable connections between harvesting and the landscape, and to interpret these relationships in their broader contexts. Landsat TM images were used for a new classification; the resulting land-cover map was the basis for linking non-spatial data on harvesters’ gathering behaviors to spatial data within delineated lingonberry gathering sites. Several significant relationships emerged: (1) mature forests negatively affected harvesters’ initial choice to gather in a site, while young forests had a positive effect; (2) land-cover type was critical in determining how and why gathering occurred: post-disturbance young and maturing forests were significantly associated with higher gathering intensity and with the choice to market harvests; and (3) distance from gathering sites to villages and main roads also mattered: longer distances were significantly correlated to more time spent gathering and to increased marketing of harvests. We further considered our findings in light of the larger ecological and social dynamics at play in central Kamchatka. This unique study is an important starting point for conservation- and sustainable development-based work, and for additional research into the drivers of human–landscape interactions in the Russian Far East.


Introduction
Research into human dimensions of environmental change seeks to understand the coupled relationships of natural resource use with ecological and socio-economic changes Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. 1 Author to whom any correspondence should be addressed. at the local, regional, and global scales (Lambin andGeist 2006, Liu et al 2007). In the Russian Federation (Russia), the changes associated with the dissolution of the Soviet Union in 1991 have greatly reshaped resource use (Backman 1999, Lerman andShagaida 2007). The demise of the state-command economy particularly affected Russia's large forestry and agricultural sectors. In 1991 the forest harvest dropped to approximately one-quarter of its late 1980 levels (Bergen et al 2008) and remains behind this mark today. Large-scale reduction of collective agriculture also occurred (Ioffe et al 2004(Ioffe et al , 2012. For instance, cultivated land in European Russia shrunk by 31% between 1990 and 2000 (Prishchepov et al 2012).
Studies have demonstrated the utility of spaceborne satellite imagery in directly observing how such dramatic socio-economic transformation has reconfigured landscape patterns in Russia , Bergen et al 2008, 2013, de Beurs et al 2009, Krankina et al 2004, Peterson et al 2009, Prishchepov et al 2012, Wright et al 2012. These studies employed imagery from moderate or coarse spatial resolution sensors to quantify land-cover changes, and confirmed decreases in industrial-scale logging and widespread abandonment of large collective agricultural lands. Not as directly quantifiable through remote sensing alone, however, are the patterns of small-scale adaptive resource use on these landscapes of major socio-economic and land-cover change.
An emerging set of detailed ethnographic and sociological studies have suggested the importance of small-scale resource use in post-Soviet communities in Russia's boreal zone. These studies show that household garden and livestock production became a primary means of sustenance for many people (Metzo 2001, O'Brien et al 2005, Pallot and Moran 2000, Pallot and Nefedova 2007, Seeth et al 1998, Southworth 2006. People also increasingly turned to their surrounding landscapes from which they acquired a variety of resources, including non-timber forest products (NTFPs; e.g. wild berry species, a variety of mushrooms, and medicinal plants) (Crate 2003, 2008, Pickup and White 2003, Yamin-Pasternak 2008. Households also marketed these resources to generate critical income (Lerman et al 2008, Pallot andNefedova 2003).
Because small-scale resource use often occurs over landscapes that have been modified by anthropogenic activities, the development of more explicitly spatial approaches to analysis has been a methodological objective of recent studies applied in other areas of the globe. Innovative methods have been employed in household-level case studies to link combinations of biophysical and socio-economic data in geographic and temporal representations (Evans and Moran 2002, Lambin 2003, Liu et al 1999, Rindfuss et al 2003a, Walsh et al 2003. In these endeavors, GIS and spatial analysis have been powerful tools in connecting land cover (often derived from remote sensing) to land use through the linking of resource-use data to land parcels and the land-cover types within these parcels (Evans and Moran 2002, Evans et al 2005, Rindfuss et al 2003b. We combined detailed ethnographic data on small-scale resource use with remotely sensed data and an explicitly spatial methodology in our study situated in central Kamchatka in the Russian Far East (figure 1). Here the shift from state-driven, large-scale resource exploitation to household-based, small-scale resource consumption in its logging and farming communities occurred abruptly following the dissolution of the Soviet Union. We looked specifically at harvesting of the NTFP species lingonberry (Vaccinium vitis-idaea) given its wide distribution throughout disturbed forest landscapes in the study area and its high socio-cultural and economic importance (Hitztaler 2010, Figure 1. Shown is the location of the Kamchatka Peninsula study area in Northern Eurasia (inset) and the study site within central Kamchatka; also shown are ecoregions adapted from WWF (Olson et al 2001) and the Kamchatka River. Kabanov 1963). Our guiding research questions were: (1) which landscape factors were associated with harvesters' choice of lingonberry gathering sites? and (2) which landscape factors affected four gathering behaviors: gathering intensity, time allocated to gathering, number of household members gathering, and the choice to market harvests?
Our initial expectations were formulated within two theoretical contexts. Land-cover/land-use change (LCLUC) science considers how humans have altered landscape patterns, as well as the consequences of this change for subsequent human activities . Optimal foraging theory (OFT) explains how humans or animals optimize their gathering (or foraging) choices under conditions of food and time scarcity (Aswani and Lauer 2006, Smith 1983, Winterhalder and Smith 2000. Thus, we expected that the factors of land cover, distance, and accessibility would influence harvesters' decisions to gather in particular sites, and also influence their gathering behaviors. More specifically, we predicted that harvesters would choose to gather in sites with more recently disturbed land cover, and that these sites would be further from villages, yet still highly accessible due to Soviet-era industrial logging and road patterns that began in proximity to villages and over time extended to increasingly remote sites (Eikeland andRiabova 2002, Hitztaler 2003). We expected to find positive relationships between favorable land cover and increases in all gathering behavior metrics. We also expected to see positive relationships between increasing distance (from a gathering site to the nearest village and main road) and greater intensity, time allocation, and marketing; at the same time, an increase in distance should be associated with a decrease in the number of gatherers. Finally, we expected positive relationships between greater accessibility via the extensive forest road network and increases in all four gathering behavior metrics.
The approach that we employed allowed us to test these predictions and make quantifiable connections between resource use and the landscape-level patterns of the study area through the integration of spatial, ecological, and ethnographic data (see Nyerges and Green 2000). Our investigation was conducted through four objectives: (1) Map land-cover type through Landsat image classification coupled with ancillary spatial data; (2) Map lingonberry gathering sites using our existing forest ecological plot data and ethnographic data; (3) Link ethnographic data on harvesters' gathering decisions and behaviors to specific gathering sites in the landscape; (4) Develop several types of statistical models to identify coupled relationships between complex landscapes and harvesting of lingonberry.

Geography and ecology
The study area (figures 1 and 2) is situated within an 8071 km 2 area within the Central Depression of Kamchatka, formed during the glacial and interglacial periods of the late Pleistocene (Braitseva et al 2005). The Kamchatka River bisects the study area and proluvial fans formed from the transport of volcanic material are well distributed along the broad valley. Elevation ranges from approximately 20-500 m asl in the river floodplain to 1566 m asl for a study area peak (figure 2, lower right) lying to the southwest of the Tolbachik volcano. Flanked by mountain ranges to the west and east, climate is continental in contrast to the maritime regions of the peninsula. Winters are long with high precipitation levels and frozen topsoil up to 90 cm; summers are short and dry with a vegetative period of only up to 80 days (Hitztaler 2003, Newell 2004). The Central Depression provides the primary productive habitat for coniferous forests in Kamchatka. Referred to in the vernacular as 'Conifer Island' (Newell 2004), these forests constitute the easternmost stretch of Russia's taiga forests. Larch (Larix dahurica) (Krestov 2003) has historically dominated the low-diversity forest cover, while spruce (Picea ajanensis) is found in sub-montane settings. Today, however, early-successional broadleaved species, primarily birch (Betula platyphylla) accompanied by aspen (Populus tremula), are widespread owing to major anthropogenic disturbance (i.e. logging and fire). The few non-anthropogenic fires in this region are usually sparked by volcanic activity. Our species of interest, lingonberry, is reported to be most prevalent and productive in disturbed forests where logging and fires have created gaps in the forest canopy that enable its growth due to increased light availability (Kabanov 1963, Yudina et al 1986.

Human history
Recognizing the great potential in exploiting Kamchatka's rich natural resources, the Soviet state initiated an ambitious development program for the peninsula, which included the commencement of large-scale logging and timber processing, in the early 1900s. A steady influx of people from mainland Russia followed, attracted to this remote region by high salaries (Heleniak 2001, Mikheeva 2002. In central Kamchatka people settled in isolated villages, including Kozyrevsk, Lazo, and Atlasovo (figure 2), which were the bases of large forest enterprises (lespromkhozy). In addition to employment, these enterprises built and supported nearly all infrastructure critical to village life. People also came to work in the state and collective farms (sovkhozy and kolkhozy) that supported central Kamchatka's burgeoning population. The villages of Dolinovka, and later Lazo (figure 2), became the sites of these farms.
The escalating ecological challenge in Kamchatka as available timber was increasingly exploited was compounded by the pervasive socio-economic crisis that unfolded in Russia after the dissolution of the Soviet Union. The ensuing loss of livelihood and opportunity was devastating: villages changed almost overnight as homes, buildings, and construction projects were quickly abandoned (Hitztaler 2003(Hitztaler , 2004(Hitztaler , 2010. As self-sufficiency became the survival mechanism for those who remained, a distinct shift in resource-use patterns occurred. In addition to household garden plots, the gathering and preservation of NTFPs became a critical livelihood activity. This dependence on NTFPs, both for sustenance and income, resulted in a sizable increase in volume harvested. To illustrate, one household documented in the ethnographic data (Hitztaler 2010) reported gathering 10 ls of mushrooms and lingonberry apiece each year during the Soviet period. In the post-Soviet period these amounts jumped to 70 l of mushrooms and 300 l of lingonberry annually.

Landsat land-cover classification
We obtained two mostly cloud-free Landsat TM (Thematic Mapper) Level 1T (terrain corrected) satellite images from 23 July 2007 for path 99, rows 21 and 22. Image pre-processing included cloud removal, atmospheric correction using the COST method (Chavez 1996), and subsetting the images to the study area. We performed unsupervised classification using the ISODATA algorithm in ERDAS IMAGINE 9.2 (Leica Geosystems 2009). Ancillary data were important in developing a land-cover classification scheme (table 1) and assigning clusters to land-cover classes. These data included our own large set of geolocated field photos, ecological literature, and Russian 1:100 000 topographic maps (Roskartografia 2001). Reference data had been previously created from the forest plot data (N = 43, Hitztaler 2010) and 364 additional testing pixels were collected from visual interpretation of unclassified Landsat imagery and topographic maps. Based on these, we calculated the overall, producer's, and user's accuracies.
We created vector data in ArcMap 9.3.1 (ESRI 2009) based on the topographic maps and the Landsat imagery (figure 2). Land cover that was difficult to distinguish using digital image classification included villages and agricultural fields. These land covers were manually digitized, rasterized, and then merged with the results of the digital classification. Roads (all unpaved) were digitized and classified into primary, intermediate, and forest classes: primary roads connect villages in central Kamchatka to the cities of Petropavlovsk-Kamchatsky (capital) and Elizovo in southeast Kamchatka; intermediate roads are arterial roads within forested areas; and forest roads are least traveled and maintained.

Mapping lingonberry gathering sites
Using the forest plot and ethnographic data collected in 2004, 2006(Hitztaler 2010, we delineated initial lingonberry gathering sites on the land-cover type map. The GPS coordinates associated with these data were used to represent the center of gathering sites from which radial buffers with a 1-km radius were grown. This radius was chosen to create an area that could encompass patchy lingonberry harvests; it was also a reasonable amount to travel within a site, on foot or by vehicle. We then dissolved the boundaries of each radial buffer to create 32 polygons representing lingonberry gathering areas, and assigned each with a GIS site ID. We created a household-level gathering behavior dataset based on the ethnographic survey data. We then grouped responses together based on gathering site ID and calculated the following variables for each of the 32 sites: sum of households that gathered; sum of individual gatherers and per cent of the total number of gatherers (in all sites); mean gathering intensity (liters gathered/hour); mean number of hours gathered; and per cent of households that sold their harvests. We added these aggregate data as variables to the GIS attribute table for the 32 gathering sites. These steps enabled the linking of harvesting households to the landscape, or the joining of non-spatial and spatial data.
We added other essential data on each gathering site to the GIS attribute tables, including its total area and the proportion of each land-cover type within the site. We also entered least-cost distances from the gathering sites to the villages and to the primary road. Least-cost distances were based on actual routes that gatherers could take via roads; they were calculated using the cost-distance and cost-path functions in ArcMap. Finally, to estimate general accessibility levels within the sites, we computed and entered the road density in each site.

Statistical analysis
We first analyzed correlations between mapped land-cover types in gathering sites and (1) distance from villages or primary roads, and (2) density, using bivariate correlation models in SPSS 17.0.3 (2009). (Note that for these and subsequent models only the land-cover types 1-9 were tested, see table 1.) Next, to establish whether the factors of land-cover type, distance, road density, and year (2003 or 2006) affected harvesters' decisions to gather in a certain site, we ran binary logistic models in SPSS. In these models we combined the 2003 and 2006 data sets to obtain a sample size of N = 64 sites and entered year as the within-subject effect and gathering site ID number as the subject effect. The dependent binary variable was gathering occurrence (yes or no); the covariates (i.e. continuous variables) were the following: proportions of land-cover types within sites, least-cost distance in km from site to nearest village, least-cost distance from site to primary road artery, and road density in sites. Lastly, to investigate the effects of land-cover type, distance, and road density on four gathering metrics of intensity, total hours spent gathering, total number of gatherers, and marketing of harvests we used linear mixed models. Similar to the binary logistic models, we combined the same data sets and ran these models in SPSS where year was treated as a repeated effect. The gathering metrics were

Land cover and anthropogenic features of the study area
The overall classification accuracy of the land-cover map was 88%; producer's/user's accuracies were between 71%-100% (table 1 and appendix). The map (figure 3) and derived land-cover proportions (table 1) indicate a complex landscape with forest types patterned by elevation, and by disturbance, regeneration, or maturation stage. Villages are small and located near the Kamchatka River or one of its tributaries. Gathering sites are generally located in low to mid-elevation upland terrain and distributed over the north-south range of the study site. Predominant land-cover types in gathering sites are: fragmented mixed-larch, shrub, maturing broadleaved, fragmented mixed-birch, and young broadleaved. A primary  road bisects the study site from north to south and there are concentrations of numerous forest roads (figure 2).

Statistical results
The results of the bivariate correlation models showed significant associations of distance and road density with land-cover type (table 2). As distance from gathering sites to the nearest village increased (1) larch forests in gathering sites significantly decreased; and (2) herbaceous-dominated land cover in sites significantly increased (table 2, matrix 1). Further, with each 1-km increase from the primary road to gathering sites, fragmented mixed-birch forests in sites decreased (table 2, matrix 2). These models also indicated a significant positive correlation between increasing densities of primary and forest roads in gathering sites and increases in maturing broadleaved and fragmented mixed-birch forests in these sites (table 2, matrices 3 and 4). The strongest correlation was between forest road density and maturing broadleaved forests. Greater forest road density was also significantly related to decreasing proportions of bare land. Binary logistic models corroborated a significant effect of year, land-cover type, and distance on harvesters' decisions to gather. Of the significant results, the parameter estimate of β (beta coefficient) showed that gathering was more likely to occur in 2003 than it was in 2006 (table 3, model 1). Second, the land-cover types of spruce forest, young broadleaved, and bare land in sites significantly influenced the probability of gathering (table 3, models 2-4). As spruce forests increased within a gathering site, the probability of gathering in this site decreased. In contrast, higher proportions of young broadleaved and bare land in sites corresponded to a higher probability of gathering in these sites. Third, increasing distance from a gathering site to the nearest village and primary road significantly corresponded to decreased likelihood of gathering in that site (table 3, models 5 and 6).
Linear mixed models confirmed a significant negative relationship between larch forests and the metrics of intensity Table 3. Binary logistic models (type III). The dependent variable is the response of gathering in a particular site: 1 = yes, or 0 = no. The covariates (or continuous independent variables) are: land-cover type (types 1-9, table 1) and distance (km) from gathering site to nearest village or primary road. Each covariate was modeled separately, and year was controlled for in each model.   and hours spent gathering (table 4, models 1 and 4). In contrast, rising intensity was significantly associated with higher proportions of maturing broadleaved forests and young broadleaved (table 4, models 2 and 3). Further, increases in bare land and maturing broadleaved forests were significantly related to more gatherers present at a site, and to a greater occurrence of marketing among gatherers, respectively (table 4, models 5 and 6). Models showed no significant effect of year (not shown in table 4) on gathering intensity and marketing. Year did, however, have a significant effect on total hours spent gathering in a season (p = 0.023).
In 2003 harvesters spent less time gathering than in 2006. A second set of linear mixed models established that longer distances from a gathering site to the nearest village and primary road were significantly associated with more hours spent gathering (table 5, models 2 and 6). There was also a significant relationship between increasing distance from a gathering site to the primary road and an increase in the percentage of gatherers who sold their harvests (table 5, model 8).

Discussion
In this section we elaborate upon the several significant relationships that emerged in our analyses to gain a deeper perspective on NTFP harvesting dynamics in central Kamchatka. We do so by contextualizing them vis-à-vis theory as well as the broader spatial, ecological, social, and temporal backgrounds in which they are embedded.

Land-cover/land-use characteristics of gathering locations
We first sought to understand the location of gathering sites in relation to anthropogenic features (e.g. villages and roads) and LCLUC. Higher proportions of disturbed land-cover types, i.e. fragmented mixed-larch, fragmented mixed-birch, maturing broadleaved, and shrub, all of which provide ample light penetration conducive to lingonberry fruiting, clearly distinguished gathering sites (table 1). Given the region's logging history, we predicted a positive correlation between longer distances (from gathering sites to the nearest village and primary road) and these disturbed land-cover types. While our statistical results did not show significant associations of distance with specific young and maturing forest types, they nonetheless outlined an expected general pattern: as distance from sites to villages grew, mature forests decreased, while landscapes in the earliest regeneration stages increased. Recent logging in the few remaining primary forests on the northwestern rim of the Central Depression has continued this pattern (Hitztaler 2010).
We also considered the accessibility of favorable gathering sites by looking at road density within them. Given that the forest road network steadily expanded in step with heightened logging activity in the late Soviet period we expected high road density to be significantly associated with post-disturbance young or maturing forests. Our results firmly corroborated the link between such forests and higher road densities. In sum, these analyses presented a clearer picture of spatial and ecological patterns: favorable sites characterized by higher proportions of post-disturbance young or maturing forests occurred further from villages, yet were nonetheless accessible due to the high density of roads within them.

Harvesters' decisions of where to gather on the landscape
The results on factors affecting harvesters' initial decision to gather in a site further explained the patterns we identified. While distance (from sites to both nearest village and primary road) was a limiting factor, land-cover type proved more influential on these decisions, reinforcing other studies that identified vegetation type as a key factor in natural resource use (Cocks et al 2008). Although our statistical tests emphasized the effect of individual land-cover types on the choice to gather, the resulting relationships should be interpreted within a landscape context. Consider, for instance, the strong positive influence of young broadleaved forests on harvesters' choice to gather in a site. Lingonberry typically does not fruit in such dense regrowth. Abundant harvests, however, may be found within other land-cover types (e.g. fragmented mixed forest and shrub) among which young broadleaved forests are interspersed in a complex mosaic, as is often the case in a patchy post-fire or logging landscape. (The gathering site of Lokhmataya, figure 3, is an excellent example of such a landscape.) These spatial relationships that we corroborated quantitatively were intuitively understood by most gatherers in central Kamchatka. Referred to as local ecological knowledge (LEK) in the literature (Huntington 2000), gatherers have cultivated a working knowledge of ecological processes over a disturbed and expansive landscape through years spent in the forest, often as both loggers and gatherers. Our ethnographic work drew out this LEK, revealing gatherers' often clear grasp of lingonberry ecology and the connection between abundant lingonberry harvests and more recently disturbed forest landscapes (Hitztaler 2010). Gatherers were also keenly aware of the dynamic spatial redistribution of NTFPs, noting the disappearance of nearby harvests as logged forests surrounding villages enter into later maturation stages.

How gathering occurs on the landscape
We demonstrated that land-cover type, distance, and road density influenced where gathering activity takes place. Next we specified how these factors shaped gathering behaviors. Harvesters gathered more intensely on landscapes where young or maturing forests were present, and were more likely to sell their harvests collected from these forests. They also spent more time gathering in sites further from villages and primary roads. In this case, in accordance with OFT, gatherers appeared to optimize time allocation as well as patch (i.e. site) selection upon the basis of mean harvest rate for that site (see Smith 1983). Applying this patch-choice optimality concept, we would also expect to find a significant positive association between increasing distances and gathering intensity. Our results, however, did not support this correlation, pushing us to think further about how shifting socio-economic circumstances have conditioned post-Soviet gathering behaviors. For instance, the urgency of NTFP harvesting may have subsided as economic hardship has more recently lessened in central Kamchatka. Moreover, better times have meant a rise in car ownership, making it easier for more harvesters, including less focused ones, to travel further distances to sites.

Conclusions
Our study linked the coupled relationships of NTFP harvesting activity to the striking ecological and social changes that have defined central Kamchatkan landscapes in the late Soviet and more recent post-Soviet periods. Linking non-spatial data on lingonberry gathering practices to the spatial data on gathering sites enabled us to do quantitative analyses of (1) the relationship between specific land-cover types and anthropogenic landscape features; and (2) the effects of land-cover type, distance, and accessibility on gathering decisions and behaviors. Occurrence of young or maturing forests proved to be the most important factor influencing gathering choices and behaviors. Distance was also a factor; although its effect was not as clear, especially as the post-Soviet phenomenon of individual car ownership has increasingly facilitated travel to more remote gathering sites.
In sum our examination revealed an intricate picture of resource use in central Kamchatka that is a product of both past and present land-and resource-use practices and is changing in response to dynamic ecological and social Table A.1. Contingency matrix produced from the accuracy assessment of the Landsat land-cover type classification. Producer's quantifies how well the set of reference pixels of a particular class were classified and user's quantifies the probability that a pixel classified into a given class actually represents that class on the ground (Lillesand et al 2007). changes. We have to date conducted the most comprehensive and integrative analyses of small-scale, non-timber forest product gathering in a remote forest area in Russia. These analyses have strong potential implications for both future research and policy-making on natural resource use and conservation in the region. Our results also elicit further inquiries into the major processes that have driven the observed interactions between inhabitants and their landscape during a period of heightened dependence on local natural resources.