Preserving a Comprehensive Vegetation Knowledge Base – An Evaluation of Four Historical Soviet Vegetation Maps of the Western Pamirs (Tajikistan)

We edited, redrew, and evaluated four unpublished historical vegetation maps of the Western Pamirs (Tajikistan) by the Soviet geobotanist Okmir E. Agakhanjanz. These maps cover an area of 5,188 km2 and date from 1958 to 1960. The purpose of this article is to make the historic vegetation data available to the scientific community and thus preserve a hitherto non available and up to now neglected or forgotten data source with great potential for studies on vegetation and ecosystem response to global change. The original hand-drawn maps were scanned, georeferenced, and digitized and the corresponding land cover class was assigned to each polygon. The partly differing legends were harmonized and plant names updated. Furthermore, a digital elevation model and generalized additive models were used to calculate response curves of the land cover classes and to explore vegetation-topography relationships quantitatively. In total, 2,216 polygons belonging to 13 major land cover classes were included that are characterized by 252 different plant species. As such, the presented maps provide excellent comparison data for studies on vegetation and ecosystem change in an area that is deemed to be an important water tower in Central Asia.


Introduction
The Western Pamirs of Tajikistan constitute an area of high biodiversity with 1,500-2,000 vascular plant species, including 160 endemics, that perform important ecosystem functions and services for the region and the adjacent lowlands [1][2][3]. However, there is strong evidence that land cover and vegetation in the Pamirs are changing with negative impacts on ecosystem properties [4][5][6][7]. Particularly after the dissolution of the Soviet Union pressure on natural resources and human induced land cover change strongly increased [8]. This is primarily associated with a demand for fuel and agricultural products leading to deforestation and overgrazing. For example, the area of Juniper woods strongly decreased during the last decades [2] and the number of cattle increased since 1990 [9]. Overstocking led to a reduced pasture potential, including the expansion of unpalatable and harmful plants [4]. In contrast to pastures, which cover vast areas of the slopes, arable land and associated villages are limited to narrow river terraces and alluvial fans. In this area, riparian Tugai forests play the dominant role in river discharge regulation and embankment stability, and hence in the protection of soils and infrastructure [2]. These forests are strongly degraded because of the energy crisis after the Soviet breakdown that forced the local population to use the Tugai for fuelwood, leading to an increased vulnerability of arable land [10]. This situation was intensified by increased severity of weather conditions, such as torrential rains [5,11], and a rapid decrease in mass balance and extent of local glaciers, resulting in increased summer run-off of the rivers [11][12][13][14][15][16], trends which are considered to be linked to climate change. In summary, this led to increased erosive forces and decreased erosion control at the same time. A consistent warming within the next decades [17] will further accelerate this development. The temperature increase might also destabilize harvests and therefore intensify food scarcity [6,18], for example due to increased infestations of insects on fruit trees [5]. Furthermore, it might affect high-altitude species [19,20], which encompass many endemics and medicinal plants [7,21]. These plants become threatened by the upward shift of more competitive species from below, which might cause their decline or even extinction because they are 'trapped' on the summit and thus lack an escape route [20,22,23]. The discussed findings indicate that the Western Pamirs are a highly dynamic region where anthropogenic and climatic impacts affected vegetation patterns and ecosystem properties in the last decades. Hence, this area provides an ideal field laboratory for detailed studies on the impact of vegetation and ecosystem change on ecosystem functions and services and on the livelihoods of the people. However, such studies require baseline comparison data from the past. Here, we edited, redrew, and evaluated four unpublished historical vegetation maps and the corresponding field notes of the Soviet geobotanist Okmir E. Agakhanjanz (see section 2 and Fig 1) that cover 5,188 km 2 of the Western Pamirs' districts Jazgulom, Rushan, Shugnan and Roshtkala and date from 1958 to 1960. A few other maps are available but not yet evaluated. The purpose of this article is to make the historic vegetation data available to the scientific community and thus preserve a hitherto non available and up to now neglected or forgotten data source with great potential for studies on vegetation and ecosystem response to global change [24,25]. For some vegetation units, where it is feasible, we give an estimation of recent developments based on own observations. Okmir Agakhanjanz and the History of the Vegetation Maps "I am a geobotanist. I investigate the plant cover of the Central Asian Mountains and prepare vegetations maps. Attentively I study the plants. I am interested to know how they form communities among themselves and how they thrive in their mountainous environment. At which altitudes do they grow? What kind of slopes and soils do they colonize? And why specifically those?" ( [26], p. 7-8).
Prof. Dr. Okmir E. Agakhanjanz ( Ã 5 January 1927 in St. Petersburg; † 28 October 2002 in Minsk; Fig 1) started his geobotanical career on the Taimyr-Peninsula in 1946. Since 1949 he lived in Dushanbe (Tajik SSR) and was member of the Department of Ecology and Experimental Geobotany at the Academy of Sciences of the Tajik SSR. His main duty was to do geobotanical mapping in various parts of the Soviet Union during many self-organized expeditions. The difficulties and the special circumstances of many of these expeditions during Soviet times are described in AGACHANJANZ [27]. The main goals of the geobotanical mapping expeditions were to establish sound data on grazing potential and biomass production of the natural vegetation in Darwaz, the Fergana Valley, in Southern Tajikistan and predominantly in the Pamirs. From the latter area several geobotanical maps were produced. Vegetation types were characterized

Study Area
The Western Pamirs are located in the east of Tajikistan, in the Gorno Badakhshan Autonomous Oblast (GBAO). The four maps discussed in this article cover 5,188 km 2 of the districts Jazgulom, Rushan, Shugnan and Roshtkala, approximately between 37°N/71°21'E and 38°2 2'N/72°E (see Fig 2). Elevations range from less than 1,600 m asl in deeply incised valleys up to 6,231 m asl (Peak Vudor). The climate is strongly continental and mainly characterized by the influence of the Westerlies bringing precipitation in winter whereas the summer is dry. Monsoonal influences are assumed to be blocked by the mountain ranges of Hindu Kush and Karakoram. Nevertheless, in summer minor rainfall occurs, which might be related to monsoonal dynamics [28,29]. WALTER and BRECKLE [30] determined the annual mean precipitation within a range of 90 to 217 mm per year. However, the amount of precipitation shows great local differences that are mainly linked to elevation and aspect. It can reach more than 500 mm per year near the snow line at 4,000 m asl, or it can be below 100 mm per year in shielded valleys [11,30,31]. The annual mean temperature ranges between 0.2 and 1.6°C [4,32,33].

Material and Methods
The original hand-drawn maps were scanned, georeferenced, and digitized. Then, the corresponding land cover class was assigned to each polygon. In order to test for spatial accuracy and to eliminate allocation errors we carried out GPS-based field spot checks for 60 polygons.
Then, the polygons were used to extract pixel based values of the variables elevation, slope, north-exposedness, and east-exposedness with a spatial resolution of 90 m, derived from the Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) [34]. Aspect, as a circular variable, was transformed to north-and east-exposedness (i.e. cosine and sinus of aspect, see [35]). For discussion and comparison of the values (particularly with elevation values given in the original map legends and in AGACHANJANC [31]) minimum, maximum, arithmetric mean, and median were calculated.
Furthermore, we applied generalized additive models (GAMs, [36]) to calculate response curves of the land cover classes and to explore vegetation-topography relationships quantitatively. GAMs are an extension of generalized linear models (GLMs, [37]) that allow for more complex response shapes than a linear one and hence for ecologically more meaningful environmental gradients. We evaluated the results of the GAMs based on the D 2 value (100 Ã (null deviance-deviance)/null deviance), which represents the percentage of deviance explained and is analogous to the R 2 as produced by simple linear regression [38]. The GAMs were fitted using the function gam from the mgcv R-package [39] with logit as the link function, binomial error distribution, and smoothed spline fits with two degress of freedom. Plant species were named according to the original map legends. Subsequently, the names were checked for validity with the Vascular plants of Russia and adjacent states [40], the Afghan checklist of vascular plants [41], and by expert W. B. Dickoré. Outdated names were supplemented by the new accepted name mentioned in square brackets behind the original name. Still, few names are under dispute, we then use both names, not indicating which the synonymy is.

Description and Discussion of the Mapped Land Cover Classes
In this section, we present and discuss the digitized vegetation maps, the associated descriptions, and information on the altitudinal distribution, slope and aspect. Agakhanjanz mapped altogether 13 major land cover classes (Fig 3) for an area of 5,188 km 2 that consists of 2,216 polygons (i.e. spatially coherent patches or biotopes). These classes were further divided in various subunits. In total, the descriptions of these subunits list 252 different plant species. The polygons were used to extract the information on the four topographic variables under  Table 1.
We found a relatively strong and significant relationship between elevation and 12 out of 13 land cover classes (D 2 between 0.3 and 34.6%, p<0.0001, see Table 1). Slope was important for the distribution of cultivated land (D 2 = 9.0%). Furthermore, Juniper vegetation, cushion plant vegetation, and floodplain meadows showed D 2 values of nearly 6%. Aspect, regardless of wether north-or east-exposedness, showed only minor relation to the distribution of the land cover classes. Only nival areas respond to north-exposedness (D 2 = 4.5%). Furthermore, tall forb communities and Rosaceae scrub show relatively high values, however this result is influenced by spatial autocorrelation due to the low number of polygons representing these land cover classes.

Mountain Tugai
Tugai is the local name of Central Asian alluvial scrub and forest (Fig 5). Various woody species can dominate and therefore form different 'complexes'. Most important are species of willow (Salix), birch (Betula), poplar (Populus), and sea buckthorn (Hippophaë). Tugai vegetation occurs, rather locally, in all investigated areas and covers 67.6 km 2 (1.3%) of the mapped area. According to the map legends, this formation occurs up to 3,700 m asl. The statistical analysis shows elevations between 1,787 and 4,505 m asl, with an average of 3,036 m asl (median   Determined by smoothed spline fits between the predicted probability of occurrence of land cover class presence (yaxis) and elevation (x-axis; the last two graphs refer to slope and north-exposedness) estimated by a GAM using binomial distribution (logit link) for binary data. Shaded areas indicate the 95% confidence bands.
3,059). The response curve indicates a high probability of occurence from the lowest elevations in the study area up to 3,500 m asl and then sharply drops. Tugai forests degraded in the aspect of area and structure especially after the Soviet breakdown, when they were used for firewood. Meanwhile, however, the situation improved due to programs of communal forest management [10].
Willow Tugai. In the Jazgulom area ( Fig 6) willow species form floodplain and gallery forests. Salix turanica frequently grows up to 15 m tall and covers 70 to 80%. Other Salix species form a second lower layer. A third layer consists of scrubs (Ribes janczewskii, Lonicera species), tall grasses  Other Tugai. Tugai dominated by sea buckthorn (Hippophaë rhamnoides) associated with Tamarix ramosissima and open grass vegetation. On alluvial sands and screes in the Rushan area (Fig 7). Currant vegetation with Ribes janzcewskii associated with Polygonum coriarium [Aconogonon coriarium] and Rosa fedtschenkoana. Near springs and river inlets in the Roshtkala area (Fig 8).

Rosaceae scrub
Rosaceae scrubs cover only very small areas of the study area (3.3 km 2 ; <0.1%) and consist of almond, rose, and cotoneaster vegetation. According to the map legends and AGACHANJANC [31] they occur from 2,500 up to 2,900 m asl, topographic analysis shows occurrences between 2,087 and 3,535 m asl with a mean at 2,716 m asl (median 2698). The response curve shows a very high probability of occurrence up to 4,000 m asl and, thereafter, a steep decline towards zero probability.
Almond vegetation. Scrubs, dominated by shrubs or small trees of Amygdalus bucharica associated with tall herbs (Eremurus stenophyllus [E. ambigens], Incarvillea olgae, Silene Cotoneaster vegetation. Composed of Cotoneaster uniflorus and C. multiflorus associated with Lonicera microphylla. Occurs on alluvial fans with abundant amounts of snow. Juniper vegetation. Juniper vegetation occurs in 132.5 km 2 (2.6%) of the study area in elevations between 1,929 and 4,607 m asl (mean 3,254, median 3,265). The probability of occurrence is very high between 2,500 and 4,000 m asl and sharply drops for elevations below and above that values. According to the map legends Juniper vegetation is abundant up to 3,800 (the latter only in the Roshtkala area). A characteristic feature of this vegetation type is a distinct shrub layer with Rosa kokanica, Rosa maracandica, Rosa korshinskiana, Lonicera korolkowii, and near springs Betula pamirica. Furthermore, associated with dwarf shrubs and herbs (e.g. Artemisia persica, Artemisia lehmanniana, Cousinia pannosa, Cousinia rubiginosa, and on screes Acantholimon korolkovii, Acantholimon parviflorum) and grasses (Stipa caucasica, Stipa bella [S. drobovii], Stipa kirghisorum, Poa relaxa). Small patches of this association are widespread in higher altitudes (up to 3800 m asl). On conglomerate slopes and on rocks in the Roshtkala area (Fig 8)

Mountain deserts
Mountain deserts cover 350 km 2 (6.8%) of the mapped area and show occurrences between 1,553 and 4,367 m asl (mean 2,894, median 2,917). The response curve indicates a very high probability of occurrence up to 3,500 m asl and, thereafter, a sharp drop of the curve that reaches zero probability at 4,500 m asl. AGACHANJANC [31] states a range from 2,000 to 3,500 m asl, the map legends give values between 2,500 and 3,400 m asl. Mountain deserts can be divided into four main types.

Cushion plant vegetation
Cushion plant vegetation occurs in 342 km 2 (6.6%) of the mapped area in elevations between 2,303 and 4,605 m asl (mean 3,580, median 3,582). In AGACHANJANC [31] and the map legends 3,000 m asl are stated as the minimum and 4,000 m asl (partly 4,700) as the maximum elevation. The response curve displays a probability peak of 0.8 at 3,800 m asl.
Meadow-like type of Acantholimon korolkovii vegetation with open layers of Prangos pabularia. High elevations with abundant snow cover of the Rushan area (Fig 7).

Mountain steppes
Mountain steppes cover 206 km 2 (4.0%) of the mapped area an occur between 1,992 and 4,691 m asl (mean 3,664, median 3,701), with a probability peak of 0.9 around 3,600 m asl. AGACHAN-JANC [31] states an altitudinal distribution between 3,100 and 4,000 m asl, in the map legends a Four Historical Soviet Vegetation Maps of the Western Pamirs range from 3,200 to 4,400 m asl is given. Four different steppe types can be differentiated: Grass steppes, Herbaceous steppes, Prickly Herbaceous steppes, and Wormwood steppes. However, according to the authors' perception, prickly Cousinia herbs spread intensively since the completion of Agakhanjanz's work therefore a differentiation between the two types of herbaceous steppes is not reasonable anymore.

Tall forb communities
With 3 km 2 (<0.1%), tall forb communities cover only very small parts of the study area. Their altitudinal distribution on the mapped area ranges between 2,178 and 3,930 (mean 2,754, median 2,631), the map legends list 3,900 m asl as the maximum elevation, and AGACHANJANC [31] states a range between 3,100 and 3,300 m asl. The response curves shows a maximum probability of occurrence up to 3,000 m asl, followed by a decrease that reaches zero probability at 4,500 m asl.

Mountain meadows
Mountain meadows occur in 12.8 km 2 (0.3%) of the mapped area in elevations from 2,079 to 4,220 m asl (mean 3,276; median 3,335), with a flat probability peak between 3,000 and 3,600 m asl. In the map legend an altitudinal distribution between 1,700 and 4,100 m asl and in AGA-CHANJANC [31] between 4,000 and 4,500 m asl is given. The rather big variability of this vegetation type and the discrepancy between the different data sources should be verified by future field work.

Floodplain meadows
Floodplain meadows are widespread in all mapped areas (94.0 km 2 , 1.8%) in elevations between 1,729 and 4,616 m asl (mean 3,695, median 3,667). In the map legends values between 2,700 and 4,000 m asl (sometimes 4,800 m asl) are given. The response curve shows a distinct probability peak of 0.85 at 3,700 m asl. These meadows are limited to riparian habitats under influence of groundwater or melting snow. Several associations can be differentiated, whereof the most important are dominated by sedges (Carex) and bog sedges (Kobresia) and show, according to field observations by the authors, degraded conditions due to a strong grazing impact.

Nival areas
Nival areas consist of glaciers and firn fields without higher plants and cover 394.3 km 2 (7.6%) of the mapped area. According to the response curves the probability for the occurrence is between 3,069 and 6,029 m asl (mean 4,679, median 4,673), while according to the map legends they are found above 4,600-4,800 m asl and above 3,900 m asl in the moister north (Jazgulom area). AGACHANJANC [31] gives 4,500 m asl as the threshold for the occurrence of nival areas. This value is also displayed by the response curve that increases above 3,000 m asl and reaches a probability of 1.0 at around 4,500 m asl. Furthermore, for this land cover class a relatively high relevance of north-exposedness could be verified by the data analysis. The response curve shows lowest probability values for values below zero (i.e. south-exposedness) and a linear increase of probability from 0 to 1.0 (i.e. fully north-exposed). Compared to the 1960s the extension of nival areas decreased due to strong warming in this region [17].

Cultivated land
This land cover class predominantly consists of settlements, agricultural land, clover meadows, gardens (see Fig 17), and in the Jazgulom area (Fig 6) also of walnut plantations with Juglans regia. Cultivated land covers 133.5 km 2 (2.6%) of the mapped area in elevations between 1,596 and 3,930 m asl (mean 2,560, median 2,570). The response curve indicates a maximum probability of occurrence up to 3,800 m asl, followed by a sharp drop towards zero probability. According to the map legends cultivated land reaches up to 3,400 m asl (rarely 3,700 m asl). For this land cover class also slope was identified as an important environmental variable. The response curve shows a nearly linear trend of a maximum probability at 0°via a 0.5 probability at just under 30°towards zero probability at 70°.

Conclusion
The presented maps depict a detailed description of the distribution and the status of vegetation in the study area in 1958-1960. However, they also have some weaknesses that need to be outlined. Ground checks revealed spatial inaccuracies of the polygons. This applies particularly to highly elevated polygons, e.g. of cryophytic and subnival vegetation, and to very narrow polygons depicting Mountain Tugai, floodplain meadows and cultivated land. For example, edges of narrow Mountain Tugai polygons reached into the valley slope and hence into neighbouring vegetation classes like e.g. Artemisia deserts. Another problem is the unclear botanical taxonomy. Many species names are outdated or need to be verified and existing determinations are often doubtful or highly debated by taxonomists. In addition, the number of species listed in the map descriptions is far from being complete. The estimated number of species for the entire Western Pamirian flora is between 1,500 [2] and 2,000 [3]. Obviously, in the map descriptions only prominent and/or dominant species are listed. This is mainly due to the fact that mapping at that time had the main goal to provide data on productivity of the various vegetation types. Therefore, further taxonomic efforts are necessary, including the collection of herbarium specimen. Nevertheless, we are confident that the maps presented in this article provide a sound basis for the study of environmental changes that e.g. occurred widespread after the breakdown of the Soviet Union, and recently because of increasing temperatures and heavy rains that can lead to extreme events, such as floods, debris flows, or glacial lake outbursts [5,42,43]. For example, on the 7 th August 2002 a glacial lake outburst destroyed the village of Dasht in the Shakhdara valley, killed 24 people and displaced the Shakhadar river bed by about 1 km. A similar event occurred in the Red valley (a tributary valley of the Bartang valley in the Rushan area) where heavy rains caused a debris flow in summer 2011 that destroyed the cultivated land and Tugai forests almost completely causing the abandonment of three small villages. Finally, since vascular plants react with some delay on changed environmental conditions, they are considered to display long term trends and thus are useful indicators for an ecological assessment of the impact of climate change [23,25]. Particularly, high mountain areas are ideal sites for comparative studies of cold habitats [25] and the maps presented here can serve as a comparison baseline.