Biogeosciences, 12, 2893–2905, 2015
www.biogeosciences.net/12/2893/2015/
doi:10.5194/bg-12-2893-2015
© Author(s) 2015. CC Attribution 3.0 License.
Modeling forest lines and forest distribution patterns with
remote-sensing data in a mountainous region of
semiarid central Asia
M. Klinge1, J. Böhner2, and S. Erasmi1
1Institute of Geography, University of Göttingen, Goldschmidtstr. 5, 37077 Göttingen, Germany
2Institute of Geography, University of Hamburg, Bundesstraße 55, 20146 Hamburg, Germany
Correspondence to: M. Klinge (mklinge1@gwdg.de)
Received: 17 July 2014 – Published in Biogeosciences Discuss.: 13 October 2014
Revised: 19 April 2015 – Accepted: 21 April 2015 – Published: 20 May 2015
Abstract. Satellite images and digital elevation models pro-
vide an excellent database to analyze forest distribution pat-
terns and forest limits in the mountain regions of semiarid
central Asia on the regional scale. For the investigation area
in the northern Tien Shan, a strong relationship between for-
est distribution and climate conditions could be found. Addi-
tionally areas of potential human impact on forested areas are
identified at lower elevations near the edge of the mountains
based on an analysis of the differences in climatic precondi-
tions and the present occurrence of forest stands.
The distribution of spruce (Picea schrenkiana) forests is
hydrologically limited by a minimum annual precipitation of
250 mm and thermally by a minimum monthly mean tem-
perature of 5 ◦C during the growing season. While the ac-
tual lower forest limit increases from 1600 m a.s.l. (above
sea level) in the northwest to 2600 m a.s.l. in the south-
east, the upper forest limit rises in the same direction from
1800 m a.s.l. to 2900 m a.s.l.. In accordance with the main
wind directions, the steepest gradient of both forest lines and
the greatest local vertical extent of the forest belt of 500 to
600 m to a maximum of 900 m occur at the northern and
western mountain fronts.
The forests in the investigation area are strongly restricted
to north-facing slopes, which is a common feature in semi-
arid central Asia. Based on the presumption that variations in
local climate conditions are a function of topography, the po-
tential forest extent was analyzed with regard to the param-
eters slope, aspect, solar radiation input and elevation. All
four parameters showed a strong relationship to forest distri-
bution, yielding a total potential forest area that is 3.5 times
larger than the present forest remains of 502 km2.
1 Introduction
The latitudinal and elevational variation in distinct plant
associations and geomorphologic landscape units has been
used for a long time to deduce regional environmental and
climatical conditions in geosciences (e.g., von Humboldt,
1845–1862; Troll, 1973a, b; Hövermann, 1985). Image clas-
sification and GIS modeling of remote-sensing data are stan-
dard methods to map landscape elements and their distribu-
tion in remote areas, which are poorly accessible due to lo-
gistic or political difficulties. Satellite analysis based on auto-
mated image processing offers a quick and useful alternative
to field mapping or manual digitalization from aerial images
(Mayer and Bussemer, 2001). While satellite images such as
Landsat data provide excellent information to delineate the
spatial forest distribution (Hansen et al., 2013), SRTM (Shut-
tle Radar Topography Mission) data can be used to examine
relief-dependent distribution patterns with a digital terrain
model (DTM). The combination of these two data sets en-
ables high-resolution mapping on a regional to local scale.
The geoecologic and climatic environmental settings con-
trol the natural distribution of forest stands (Holtmeier, 2000;
Körner, 2012; Miehe et al., 2003). In addition, the actual sit-
uation can strongly be influenced by human activities such
as logging, fire clearing and animal grazing, which decreases
the potential natural forest area (PFA). This often makes it
difficult to differentiate between natural factors and human
impact on the distribution of timbered areas. In general, hu-
man activity has reduced the forest area since prehistorical
times so that the actual forest area (AFA) pattern mostly
represents the minimum of the potential environmental dis-
Published by Copernicus Publications on behalf of the European Geosciences Union.
2894 M. Klinge et al.: Modeling forest lines and forest distribution patterns
tribution range. However, due to the possibility of anthro-
pogenic forest management and afforestation during the last
centuries, forests may occur at sites less favorable for natural
tree growth.
Due to the highly continental, cold and semiarid climate
in central Asia, tree growth is mostly determined by topog-
raphy parameters. Forest stands beyond sites with more fa-
vorable conditions regarding groundwater are predominantly
limited to north-facing slopes in the mountains with an up-
per and lower forest limit (Dulamsuren et al., 2014; Hilbig,
1995; Klinge et al., 2003; Treter, 1996, 2000).
Different definitions have been used for tree and forest
lines (Körner, 2012; Körner and Paulsen, 2004). The tree-
line ecotone covers three main boundary lines at the upper
limit of forest distribution. The highest is the tree species
line, where tree seedlings occur but no adult trees. The tree-
line is the maximum elevation where patches of forest can
exist at topographically favorable places. In our investigation
we refer to the forest line, which is defined as the limit of
closed forest at the upper (timberline) and lower boundary of
forest distribution.
For the region of the northern Tien Shan in China, Dai
et al. (2013) state an upper forest line beginning with
2900 m a.s.l. in the west, which decreases eastward down to
2500 m a.s.l. and then rises again to 2900 m a.s.l. in the east.
In the northwestern Tien Shan, Fickert (1998) reports an up-
per forest line of 2900 and 2850 m a.s.l. and a lower forest
line of 2400 and 2500 m a.s.l., respectively for the Sailijski-
Alatau and Kungeij-Alatau. In the Altai Mountains, Klinge et
al. (2003) found upper forest lines increasing eastward from
1800 to 2600 m a.s.l. and, concurrently, lower forest lines in-
creasing from 1000 to 2200 m a.s.l., while the vertical exten-
sion of the forest belt varies between 400 and 1200 m.
Tree growth in high mountains is generally restricted by
temperature conditions (Haase et al., 1964; Holtmeier, 2000;
Jobbagy and Jackson, 2000; Körner, 2012). The upper for-
est line is a thermally determined distribution boundary that
is generally defined by the mean July temperature (Walter
and Breckle, 1994) or the warmest-month isotherm of 10 ◦C.
According to Körner (2012) and Körner and Paulsen (2004),
this parameter is not appropriate for all parts of the world.
This can be seen in Fickert (1998), who shows that the upper
forest line in the northern Tien Shan coincides well with the
10 ◦C July isotherm, while further south in the northwest-
ern Himalaya and northwestern Karakorum, it is connected
to the July isotherms of 16 and 12 ◦C, respectively. For the
eastern side of the northern Tien Shan, Dai et al. (2013) re-
port a mean temperature of the warmest month of 10.5 ◦C
at the mean position of the alpine forest line. Moreover, the
mean annual air temperature (MAAT) is weakly correlated
to the forest line because it includes temperatures from the
nongrowing season, which play a minor role in tree growth
(Jobbagy and Jackson, 2000; Körner, 2012).
A suitable way of describing the temperature environment
at the upper forest line is by using a minimum threshold
value for the mean air temperature during the growing sea-
son, which is defined as the period of monthly mean tem-
peratures above 5 ◦C (Dai et al., 2013; Körner, 2012; Körner
and Paulsen, 2004). Based on the strong correlation between
soil and air temperatures, Körner and Paulsen (2004) state a
global range of 5.5 to 7.5 ◦C for minimum mean air tempera-
tures during the growing season. Paulsen and Körner (2014)
developed a climate-based model for treeline prediction by
defining the growing season as days with a mean temperature
above 0.9 ◦C and a mean temperature of more than 6.4 ◦C
during that time. For the upper forest line between 2750 and
2920 m a.s.l. in the Tien Shan in Kyrgyzstan Körner (2012)
found a mean temperature of 6.5 ◦C during the 155 days of
the growing season (late April until late September).
The forest expansion into dry regions is controlled by pre-
cipitation and soil water supply (Dulamsuren et al., 2010,
2014; Kastner, 2000; Klinge et al., 2003). Between the more
humid mountain regions and the arid basins of central Asia, a
lower limit of forest distribution occurs, which is termed the
lower forest line. According to Walter and Breckle (1994)
this forest distribution boundary coincides with an annual
precipitation of at least 300 mm, while Holdridge (1947) pro-
poses 250 mm and Miehe et al. (2003) found Juniperus trees
in southern Tibet growing in regions with an annual precipi-
tation of between 200 and 250 mm. Dulamsuren et al. (2010)
state an annual precipitation between 230 and 400 mm at
lower elevations for larch forests in northern and central
Mongolia. In western Mongolia Dulamsuren et al. (2014)
found coniferous forests existing at an annual precipitation
of around 120 mm, which are explained by soil water bene-
fits due to the occurrence of permafrost ice in the soil.
Everywhere in mountainous areas of the semiarid inner-
Asian forest steppe coniferous forests are restricted to north-
facing slopes. While the north-facing slopes are dominated
by larch trees (Larix sibirica) in Mongolia, spruce trees
(Picea schrenkiana) occur in the Tien Shan (Dai et al., 2013;
Fickert, 1998; Liu et al., 2013; Wang 2005, 2006). Thus,
the restriction of conifers to north-facing slopes in the inner-
Asian forest steppe is not bound to certain tree species but
rather to the environmental settings.
The semiarid climate conditions generate an overall de-
ficiency of moisture which considerably influences the ele-
vational forest distribution and may even control the upper
forest limit (Liang et al., 2012; Liu et al., 2013; Miehe et al.,
2008). A specific relief position is combined with particular
climate conditions such as temperature, precipitation, evapo-
ration and insulation, which are similar at comparable sites in
the surroundings. For this reason the relief parameters eleva-
tion, aspect, slope angle and solar radiation input can be used
to define topoclimatic conditions in mountain regions (Miehe
et al., 2003). However, to identify potential forest sites based
on those definitions, the geologic and soil properties have to
be comparable.
The impact of human activity on vegetation and especially
on the forest since prehistorical times is a permanent question
Biogeosciences, 12, 2893–2905, 2015 www.biogeosciences.net/12/2893/2015/
M. Klinge et al.: Modeling forest lines and forest distribution patterns 2895
Figure 1. Map overview showing the investigation area detailed
(trapezoid) in central Asia.
that needs to be investigated to clarify the environmental sig-
nificance of any actual forest line (Miehe and Miehe, 2000).
Dulamsuren et al. (2014) found a considerable anthropo-
zoogenic influence on the actual lower forest line in the Mon-
golian Altai. For northern Mongolia, Schlütz et al. (2008)
showed that the present vegetation pattern in the mountain
taiga where steppes occur on south-facing slopes is caused
by climate conditions and relief and does not originate from
human activities.
Human impact on natural forests in Kazakhstan goes back
to prehistoric times, with nomadism and animal grazing as
a way of life adapted to the natural environmental condi-
tions of the steppe (Karger, 1965; Giese, 1981, 1983). During
summertime the alpine meadows and mountain steppes in the
upper mountains were regularly used as pastures for the live-
stock. Even during Soviet times in Kazakhstan the nomadic
movements were generally adopted by the sovkhoz system.
Even today the alpine pastures are still in use. Extensive an-
imal grazing prevents the rejuvenation of trees, and nomads
may expand the grassland by setting fire to it.
Spatial models, which are able to predict the climatically
induced forest distribution and especially the upper forest
line on a global scale by exclusively using spatial climate
data, already exist (Paulsen and Körner, 2014). However,
a clear method to empirically distinguish the actual forest
distribution and its elevational limits for small areas cov-
ering a single mountain system and to simultaneously in-
vestigate the potential human impact is lacking. In this in-
vestigation we introduce a procedure to solve this problem
based on medium-resolution remote-sensing data. In addi-
tion, spatially explicit climate data and tree-growth-limiting
climate parameters serve to differentiate potential human im-
pact from natural conditions in the forest distribution.
2 Study area
The investigation area detailed, Uzynkara Ridge, also known
as the Ketmen Mountain range, the Ketmen Mountain range,
is located in the northernmost part of the Tien Shan in cen-
tral Asia on the border between Kazakhstan and China (79–
81◦ E/42◦45′–43◦45′ N) (Fig. 1). This closed mountain sys-
tem was chosen for investigation because it provides excel-
lent topographic preconditions to clearly indicate the lower
and upper boundaries of forest distribution between the mid-
dle and central part of Asia. It fills a knowledge gap about
forest lines between regions of the Tien Shan in the south
and east, and the Altai mountains in the north (Fickert, 1998;
Dai et al., 2013; Klinge et al., 2003). The main cities in
the region are Shonzy and Kegen. The complete mountain
range is part of the catchment area (CA) of the Ili River
in the north. While the northern mountain side is directly
drained to the Ili River, the Kegen River in the southern
intermountain basin first flows westward and then, turning
into the Sharyn River, flows in a northerly direction, and
the Tekes River in the southernmost part runs eastward into
Chinese territory. The mountain system is structured by two
main ridges – a northern front range (NFR) and a southern
mountain range (SMR) – which converge in the east and en-
close an intermountain basin in the west. The highest peak
is the Nebesnaja, reaching 3652 m a.s.l. A high mountain
plateau at ∼ 3400 m a.s.l. drops southward, while the north-
facing slopes are cut steeply by Pleistocene cirques. Today,
no glaciation but permafrost occurs in the uppermost areas.
The edge of the mountains is tectonically clearly distin-
guished from the alluvial fans and fanglomerates at approx-
imately 1500 m a.s.l. in the north and at 2000 m a.s.l. in the
southern intermountain basin, following west to east trend-
ing fault lines. The mountains mainly consist of metamor-
phic and volcanic Carboniferous and Devonian rocks, includ-
ing several Palaeozoic granite bodies. Permian, Silurian and
Jurassic rocks are also distributed locally.
The MAAT in Almaty (848 m a.s.l.) is 8.7 ◦C and in
Karakol (1744 m a.s.l.), which is situated south of the
investigation area, 6.3 ◦C (Fickert, 1998). According to
Medeu (2010) the mean air temperature is between −8 and
−10 ◦C in January and between 20 and 24 ◦C in July. In
wintertime the Siberian anticyclone produces weather con-
ditions with cold air masses in the basins and warmer air
temperatures above the inversion layer between 1000 and
1550 m a.s.l. (Giese, 1973).
The majority of precipitation in Kazakhstan comes with
air masses from the west and southwest. In the mountains of
the northern Tien Shan, mainly convective rainfall occurs in
spring and autumn. Additionally, cold air masses from north-
ern directions bring precipitation to the northern Tien Shan
in summertime (Böhner, 2006; Lydolph, 1977). According
to Giese (1973) the annual precipitation in the basins of the
foreland lies between 100 and 300 mm; in the lower moun-
tains and in the intermountain basin, it is between 300 and
www.biogeosciences.net/12/2893/2015/ Biogeosciences, 12, 2893–2905, 2015
2896 M. Klinge et al.: Modeling forest lines and forest distribution patterns
Figure 2. Workflow of DEM (digital elevation model) and satellite image processing to determine the spatial forest distribution patterns in
semiarid mountain systems of central Asia on a high-resolution scale.
400 mm. In the mountains it increases to more than 800 mm.
The precipitation maxima occur in May and June, with a mi-
nor, secondary maximum in September.
The foreland, basins and treeless mountain areas are
covered by steppe vegetation with forb and bunch grass
(Medeu, 2010). In the drier regions to the north, it changes
to grassland, sagebrush desert, saltwort and sedge vegeta-
tion. The forest belt mainly consists of spruce trees (Picea
schrenkiana). In the westernmost part, aspen trees (Populus
tremula) and in the northeastern part birch trees (Betula pen-
dula) also occur. On the southern slopes shrub areas also ex-
ist.
The soils are distributed according to the climate condi-
tions and the vegetation zones (Medeu, 2010). In the fore-
land, desert soils occur. In the lower front ranges and in the
intermountain basin, mountain steppe soils of castanozem
and chernozem type are distributed. In the forest belt dark
chernozems, which are locally bleached and podzolized, oc-
Biogeosciences, 12, 2893–2905, 2015 www.biogeosciences.net/12/2893/2015/
M. Klinge et al.: Modeling forest lines and forest distribution patterns 2897
Figure 3. Spatial distribution of actual forested area (AFA) and potential forest area (PFA) in the mountainous region of the northernmost
Tien Shan shown above a true-color composite of a Landsat 7 satellite image from the 13 September 2000.
Table 1. Confusion matrix showing the accuracy report of the supervised maximum likelihood classification (area in ha).
Reference Classification data Sum Producer’s Omission
data Forest No forest reference accuracy error
Forest 2696.7 333.6 3030.3 0.890 0.110
No forest 14.7 76 426.2 76 440.9 0.9998
Sum classification 2711.4 76 759.8 Total Sum of Overall
User’s accuracy 0.995 0.995 test data accuracy
Commission error 0.0054 79 471.2 0.996
cur in forests, and pheaozem soils exist at meadow steppe
sites. At high elevations, alpine and subalpine soils occur in
mountain meadows and meadow steppes.
Arable land in eastern Kazakhstan is located at the foot
of the mountain ranges on the alluvial fans in the basins and
on the foothills at a lower elevation. In this transition zone
between the pediments and mountain ranges, the soils are
improved by a certain amount of Pleistocene loess (Giese,
1983; Karger, 1965; Machalett et al., 2006). After leaving
the mountains, the water from the rivers is used for irriga-
tion cultivation on the pediments. In the foothills agricul-
ture is supported by sufficient rainfall; this is the so-called
bogar cultivation (Giese, 1983). According to these require-
ments the settlements are located along the main valleys at
the mountain boundary. Around the settlements, there is sig-
nificant woodcutting for construction and fuel.
3 Methods
A schematic workflow of the GIS analysis procedure with
input data, intermediate data and output data is presented in
Fig. 2. The analysis is divided into two main processes: the
first is working on the relief parameters to estimate the PFA
and the second conducts the delineation of the upper and
lower forest lines. The forest lines are defined as the distribu-
tion boundaries of closed forest stands with areas larger than
0.5 ha, disregarding single trees, which may represent special
environmental places or remnants of former forests. Trees
near the rivers in the valley bottoms were excluded from the
examination because these are sites which have more favor-
able conditions regarding groundwater and which are mostly
occupied by deciduous trees.
The determination of the AFA in the investigation area was
achieved based on a supervised maximum likelihood classi-
fication from multispectral satellite images (visible light and
infrared channels) of Landsat 7/ETM+ from the 13 Septem-
www.biogeosciences.net/12/2893/2015/ Biogeosciences, 12, 2893–2905, 2015
2898 M. Klinge et al.: Modeling forest lines and forest distribution patterns
Figure 4. Frequency distribution of relief parameters in relation to
actual forest area (AFA, green; in the diagrams of aspect, slope gra-
dient and solar radiation input, the light green areas represent the
standard deviation of 95 range excluded from PFA delineation) and
total study area (TSA, brown).
ber 2000 (Fig. 3). Aerial photos provided as imagery and
Bing basemaps by ESRI were used to detect forest area refer-
ence sites for the training and validation of the classifier. Two
classes were built and manually digitized for the classifica-
tion and validation process. One class represents the forest
areas and the other class includes different kinds of no-forest
landscape. Depending on the ground resolution of 30× 30 m,
one pixel covers several individual trees so that small clear-
ings and aisles may have been disregarded. The confusion
matrix in Table 1 shows a producer’s accuracy of 89 % for
forest areas, where ∼ 6 % of the forest area were used for
validation. This possible underestimation of the forest area
of up to 11 % mainly occurs at the edges of closed forest,
where the classification depends on the quantity of trees in
one Landsat pixel.
The relief parameters elevation, aspect, slope gradient and
total solar radiation input were derived from a DTM based
on SRTM data (Rabus et al., 2003), which was converted
to UTM zone 44 N with a spatial resolution of 90× 90 m.
The polygons of the delineated forest stands were intersected
with the relief parameters in order to investigate the relief-
dependent spatial distribution of forest sites in the study area.
In addition, the statistics of all relief parameters were com-
puted for the total study area (TSA) to indicate the poten-
tial impact of topography on the spatial distribution of for-
est stands (Fig. 4). The PFA was then identified based on
the assumption of confidence ranges for all four relief pa-
rameters, which were found responsible for forest distribu-
tion. This range was defined by the standard deviation (95 %
confidence interval) from the single-frequency distribution of
the relief parameters aspect, slope gradient and sum of solar
radiation input during the mean growing season (March to
November). While these parameters are not systematically
influenced by human impact, the vertical distribution may
have been changed by forest clearing at the lower and up-
per boundary. Therefore 99 % of the frequency distribution
of the elevation parameter was chosen in this case.
Baseline climate data sets for central Asia, comprising
monthly radiation, temperature and precipitation data at
a horizontal resolution of 0.5 arc seconds (approximately
1200 m in longitudinal and 850 m in latitudinal direction),
are provided by Böhner (2006). The regular-grid climate
layers were estimated using an empirical modeling ap-
proach, which basically integrates statistical downscaling of
coarse-resolution atmospheric fields (NCAR/NCEP-CDAS
reanalyses series – National Center for Atmospheric Re-
search/National Center for Environmental Prediction – cli-
mate data assimilation system; Kalnay et al., 1996) and GIS-
based surface parameterization techniques, to sufficiently ac-
count for the topographic heterogeneity of the target area.
A comprehensive description of data bases and modeling
techniques is given in Böhner (2006) and Böhner and An-
tonic (2009). The suitability and precision of the model-
ing approach is discussed in Gerlitz et al. (2013, 2014) and
Soria-Auza et al. (2010).
The frequency distribution of selected climate parameters
related to the AFA and the TSA shown in Fig. 5 was cal-
culated in the same way as described above. In contrast to
the high resolution of the SRTM data, the climate data has
a resolution about 10 times lower, which leads to a gener-
alization and coarser-scale of relief positions where climatic
differences between slope aspects inside the valleys are av-
eraged. The climate data related to the forest stands are ana-
lyzed by the climatic limitation values for forest development
to detect potential human impact on the forest distribution
patterns when obvious discrepancies occur.
Biogeosciences, 12, 2893–2905, 2015 www.biogeosciences.net/12/2893/2015/
M. Klinge et al.: Modeling forest lines and forest distribution patterns 2899
Figure 5. Frequency distribution of climate parameters for actual forest area (AFA: columns and left axes, in km2) and the total study area
(TSA: graph and right axes, in km2).
Table 2. Statistical values of the relief parameters related to the forest distribution.
Parameter Unit Maximum Total 95 % of the
distribution distribution distribution
value range range
Elevation m a.s.l. 2500 1575–2900 1925–2775
Aspect degree horizontal 315 (NW) 0–360 260–70
Slope gradient degree vertical 28 0–62 12–38
Solar radiation input kWh m−2 1075 450–1550 800–1325
To outline the actual forest lines, it is initially necessary
to segment the relief into small CAs, which represent small
side valleys or slope niches divided by convex ridges. This is
done by computing the surficial hydrology regime from the
DTM. The size of a CA is given by the threshold value for the
stream definition function, which assigns the minimum num-
ber of cells that must discharge into a specific cell to start a
depth contour. In this study a value of 200 was found practi-
cal for the lower forest line and a value of 100 was suitable
for the upper forest line. The single CAs generally consist
of sections on the left and on the right side of a valley. Hav-
ing different aspects in one segment is expedient to receive a
general forest line value for one valley section.
After combining the catchment polygons with the forest
polygons, it is possible to determine the maximum and min-
imum elevation values for forests inside a single CA. The
calculated values are spatially allocated as points to the po-
sition of those pixels for which forest line values have been
determined. To eliminate the preconditions on the lower for-
est line given by the elevation limits of the relief, only those
www.biogeosciences.net/12/2893/2015/ Biogeosciences, 12, 2893–2905, 2015
2900 M. Klinge et al.: Modeling forest lines and forest distribution patterns
Figure 6. The lower forest line and the catchment areas providing lower forest line values in the investigation area.
Figure 7. The upper forest line and the catchment areas providing upper forest line values in the investigation area.
minimum values of forest stands were chosen which are more
than 50 m higher than the total minimum value of the CA.
The distance between the highest forest stands and the crest
line above has a special influence on the upper forest line;
this is called the “summit syndrome” by Körner (2012). Near
the summits the local climate conditions strongly suppress
tree growth by stronger wind, reduced temperature and snow
drift. To receive a reasonable value for the climatic upper for-
est line and to eliminate preconditions by relief height, only
those maximum forest values were chosen which lie more
than 100 m below the total maximum elevation of the catch-
ment. Finally, the forest lines were calculated from the re-
maining points by a Natural Neighbor interpolation method.
Biogeosciences, 12, 2893–2905, 2015 www.biogeosciences.net/12/2893/2015/
M. Klinge et al.: Modeling forest lines and forest distribution patterns 2901
Figure 8. Distribution of AFA in relation to the hydrological climatic environment.
Figure 9. Distribution of AFA and PFA and the July isotherms.
4 Results
4.1 Relief parameterization
The total AFA in the investigation area is 502 km2 (Fig. 3).
Frequency distributions of relief parameters for the AFA are
shown in Fig. 4. The slope gradient and solar radiation of
the forest stands show a normal distribution. The values with
the maximum distribution are 28◦ for slope gradient and
1075 kWh m−2 for the sum of solar radiation input (Table 2).
Less than 5 % of the forests occur on southern slopes (SE–
S–SW). The curve of the parameter aspect has a steeper left
slope and a maximum value in the northwestern direction
(315◦), which is strongly related to the diurnal air temper-
ature trend caused by insolation and heating processes on
different slope directions. This underlines the fact that the
strong relation of forest distribution to slope aspect is caused
by natural environmental conditions. The logistic problems
of access in the relief influence how forest is transformed by
www.biogeosciences.net/12/2893/2015/ Biogeosciences, 12, 2893–2905, 2015
2902 M. Klinge et al.: Modeling forest lines and forest distribution patterns
Figure 10. Distribution of AFA and the 5 ◦C monthly isotherms during the growing season
Table 3. Comparison between of the modeled area values of single relief parameter classes and of the combination of all four relief parame-
ters. (AFA: actual forest area; PFA: potential forest area.)
Relief parameter: Elevation Solar radiation input Slope gradient Slope aspect All four parameters
Site classification km2 % FAAP % TMA km2 % FAAP % TMA km2 % FAAP % TMA km2 % FAAP % TMA km2 % FAAP % TMA
(3) PFA without AFA 5358.7 91.4 65.9 4791.2 90.5 59.0 4279.9 89.5 52.7 3850.5 88.5 47.4 1323.6 72.5 16.3
(1) PFA with AFA 502.0 8.6 6.2 480.5 9.1 5.9 483.8 10.1 6.0 474.2 10.9 5.8 446.7 24.5 5.5
(4) No PFA with AFA 0.3 0.004 0.003 21.5 0.4 0.3 18.0 0.4 0.2 27.9 0.6 0.3 55.4 3.0 0.7
(2) No PFA without AFA 2265.2 27.9 2832.9 34.9 3344.4 41.2 3773.6 46.4 6300.7 77.5
Sum of all classifications which 2767.2 34.1 3313.4 40.8 3828.3 47.1 4247.8 52.3 6747.4 83.0
represent the actual situation
% FAAP: percentage of the total actual and potential forest area. % TMA: percentage of total mountain area with 8126 km2.
nomads and woodcutters, which reduces the pure signal of
elevation in the data. Climate controls environmental condi-
tions in a coarser regional scale and creates sharper eleva-
tional boundaries. The curve of the parameter elevation has
a shallow left and a steep right slope, which indicates hu-
man impact on forest distribution at the lower boundary. The
lower forest lines start at 1575 m a.s.l. and the upper forest
lines exceed 2900 m a.s.l. so that the maximal vertical dis-
tance of the forest limits for the entire investigation area is
1325 m (Table 2).
The TSA represents the complete area of the elevation belt
between the forest lines from 1500 and 2900 m a.s.l.. Except
for the slope gradient diagram, the flat slope positions < 5◦
were additionally excluded from the TSA. The resulting TSA
is approximately 4975 km2. In regard to the independent fre-
quency distribution curves of the relief parameters in Fig. 4
between forest stands and the TSA, no statistical influence
of the main topographic pattern on the forest distribution is
detectable.
From a statistical point of view, all four relief parameters
control the forest distribution. Therefore, it is necessary to
check the modeling accuracy of the PFA received from one
single relief parameter against the combination of all four re-
lief parameters (Table 3). Comparing the modeled PFA and
the AFA, four different classes can be built: (1) PFA with
AFA and (2) no PFA without AFA represent the mapped sit-
uation; (3) PFA without AFA and (4) no PFA with AFA rep-
resent the differences between modeling and mapping. To re-
ceive a statistical background for the evaluation of the mod-
eling quality of the delineated PFA, it is once related to the
sum (FAAP) of AFA and the PFA and twice referred to the
total mountain area (TMA) of 8126 km2, while the boundary
between the mountains and the pediments of the foreland is
generally defined by the changeover line of the slope gradient
at 2.5◦. From all four single relief parameters the modeling
based on the slope aspect coincides best with the actual situa-
tion, but, in any case, the combination of all four parameters
obviously enhances the prediction accuracy (Table 3). The
PFA calculated from all four relief parameters is 1825 km2
and therefore 3.5 times larger than the AFA. Figure 3 shows
the spatial differences between the AFA and PFA. In relation
to the AFA the PFA generally extends to the lower and upper
elevations.
Biogeosciences, 12, 2893–2905, 2015 www.biogeosciences.net/12/2893/2015/
M. Klinge et al.: Modeling forest lines and forest distribution patterns 2903
4.2 Forest line patterns
Figure 6 shows the lower forest line in the investigation area
starting at 1600 m a.s.l. in the northwest and increasing to
2600 m a.s.l. in the southeast. Values for the lower forest
line are mostly derived from the lower CAs but there are
also many CAs at the higher elevations of the NFR, where
the forest stands do not reach the valley bottom. This phe-
nomenon may be caused by the local relief of tree-free flat
valley bottoms, which would be a climate rather than a to-
pographic signal. However, regarding the lower forest line
in the second mountain range southeast of the intermountain
basin and behind the NFR, it remains at a higher elevation
around 2400 m a.s.l.. Here the high lower forest line position
is obviously caused by the drier conditions of the rain shadow
position, which may also be true for the upper valleys in the
NFR.
The upper forest line distribution and the area above the
forest line are shown in Fig. 7. In the NFR the upper forest
line at the edge of the mountains starts at 1800 m a.s.l. in the
west and increases to 2200 m a.s.l. in the east, maintaining
a vertical distance of 200 m to the lower forest line. From
the edge of the mountains in the north to the crest line, the
upper forest line rises to 2800 m a.s.l., and, crossing the in-
termountain basin, it lies at an elevation between 2400 and
2800 m a.s.l. in the SMR. The local vertical distance of the
forest belt reaches its maximum value of more than 900 m on
the northern side of the NFR. On the southern side and in the
SMR, the forest belt is very narrow, with vertical distances
between 50 and 400 m.
4.3 Climate environmental conditions
The environmental conditions were analyzed in terms of
frequency distribution of climate parameters for the AFA
(Fig. 5) and were mapped together with the AFA (Figs. 8–
10). The diagrams in Fig. 5 show the differences between
AFA and TSA for all climate parameters except for the
MAAT, which was already excluded as a significant forest
limitation parameter.
The lowest value class of forest stands for annual precip-
itation is 250 mm, while the highest potential evapotranspi-
ration is up to 1100 mm a−1. One third of the AFA lies in
areas with a negative potential water balance (pWB, i.e., the
difference between annual precipitation and potential evap-
otranspiration; cf. Fig. 5) but with a precipitation amount of
between 300 and 700 mm a−1 (Fig. 8). These areas are sit-
uated at the westernmost edges and on the southern slopes
of the mountain ranges. While the westernmost sites are ex-
posed to the westerlies, which transport most of the humidity,
the southern slopes lie in the rain shadow but at a higher el-
evation, and therefore the lower forest line is around 600 m
higher than on the northern side of the NFR.
In the eastern part of the northern side of the NFR, the
AFA belt is very small and, concurrently, the lower forest
line increases to 2000 m a.s.l., 400 m higher than in the west-
ern part of the NFR. Here, the lower forest line occurs with
a precipitation of 700 mm and at a positive pWB of 150
to 300 mm a−1, while the PFA extends more into the lower
slope positions; this corresponds to the mean values of the re-
gions described above. This is an indication for a nonnatural
distribution and points to greater human influence on forests
in this region.
The forest distribution related to mean air temperature in
July ranges between 7 and 17 ◦C, with a maximum of around
11 to 12 ◦C (Fig. 5). Comparing the AFA and PFA with the
July isotherms (Fig. 9) shows that the upper AFA is mainly
bordered by the 10 ◦C July isotherm but also extends to the
8 ◦C July isotherm in many places. The upper PFA is gen-
erally aligned to the 8 ◦C July isotherm. Figure 10 shows
the distribution of the monthly mean air temperature 5 ◦C
isotherm during the growing season and the AFA. Except at
the westernmost part, the 5 ◦C isotherm is above the AFA
between June and September, and the upper AFA bound-
ary coincides well with the 5 ◦C isotherm of September. The
PFA at the upper forest boundary extends up to the position
of 5 ◦C isotherm in June; the growing season obviously be-
comes very short at these highly elevated places. As shown
in Figs. 9 and 10, the PFA at the upper limit is overestimated
and the upper AFA boundary generally has a natural limita-
tion.
5 Discussion and conclusions
It was shown that the AFA and the forest lines coincide well
with the local climate conditions. At the lower limit, forests
are restricted to a minimum annual precipitation of 250 mm.
The upper forest line coincides with the 10 ◦C July isotherm
in most places and to the minimum monthly mean tempera-
ture of 5 ◦C for the period between June and September. In
the more humid parts of the investigation area at the western
and northern slopes of the NFR both forest lines have a steep
gradient and the forest belt has its greatest vertical extension
between 500 and 600 m and locally up to 900 m. This agrees
well with the findings of increasing vertical forest extension
concurrent with increasing humidity and vice versa by Fick-
ert (1998), Dai et al. (2013) and Klinge et al. (2003) in the
surrounding regions. Besides temperature, rainfall influences
the upper forest line because clouds reduce the air tempera-
ture by shadowing and by the reflection of solar insulation.
This explains the steep gradient of the upper forest line at the
windward side of the mountain ridges.
The comparison of the AFA with climate data reveals a
strong relation between the distribution patterns at the up-
per boundary, but divergences occur at the lower boundary.
This indicates human impact on the forests at the edge of the
mountains, modifying the lower forest line, while the upper
forest line represents the natural condition. Accordingly the
PFA derived from relief parameters at lower elevations in-
www.biogeosciences.net/12/2893/2015/ Biogeosciences, 12, 2893–2905, 2015
2904 M. Klinge et al.: Modeling forest lines and forest distribution patterns
dicates additional area for more potential natural forest. The
PFA at the upper boundary is overestimated by highest forest
stands occurring at few places with favorable climatic con-
ditions because we used the total vertical distance of forest
distribution as a relief parameter instead of the standard vari-
ation, presuming that extensive logging may also occur in the
alpine meadow pastures. GIS analysis combined with multi-
spectral satellite images and DTM is well suited to determine
forest lines and potential forest areas for semiarid regions on
a local to regional scale. For forest line delineation it is neces-
sary to eliminate elevation values which are restricted by the
relief conditions and do not represent climatic limitations.
The DTM-derived relief parameters slope aspect, gradient
and solar radiation serve well as indicators of the climatic
environment in the investigation area and help to transfer en-
vironmental settings to other places in the broader study area.
Human impact is recognized by the evaluation of the parame-
ter elevation. Therefore, a forest line evaluation with respect
to the general climatic conditions has to be performed be-
fore the parameter elevation is incorporated into the spatial
delineation process of the PFA. In conclusion, the proposed
workflow is a helpful method for the evaluation of the poten-
tial forest distribution and the delineation of human impact.
It can be used to indicate local climate variability, for land-
scape analysis and for effective reforestation planning.
Acknowledgements. The authors would like to thank the US
Geological Survey for making the satellite data freely available
for scientific research. We acknowledge support by the Open
Access Publication Funds of Göttingen University. We also thank
F. Lehmkuhl and two anonymous referees for their great support in
improving the manuscript of this publication.
This open-access publication was funded
by the University of Göttingen.
Edited by: A. Ito
References
Böhner, J.: General climatic controls and topoclimatic variations of
Central and High Mountain Asia, Boreas, 35, 279–295, 2006.
Böhner, J. and Antonic, O: Land-Surface Parameters Specific to
Topo-Climatology, in: Geomorphometry: Concepts, Software,
Applications, edited by: Hengl, T. and Reuter, H. I., Dev. Soil
Science, 33, 195–226, 2009.
Dai, L., Li, Y., Luo, G., Xu, W., Lu, L., Li, C., and Feng, Y.: The
spatial variation of alpine timberlines and their biogeographical
characteristics in the northern Tianshan Mountains of China, En-
viron. Earth Sci., 68, 129–137, 2013.
Dulamsuren, C., Hauck, M., Khishigjargal, M., Leuschner, H. H.,
and Leuschner, C.: Diverging climate trends in Mongolian taiga
forests influence growth and regeneration of Larix sibirica, Oe-
cologia, 63, 1091–1102, 2010.
Dulamsuren, C., Khishigjargal, M., Leuschner, C., and Hauck, M.:
Response of tree-ring width to climate warming and selective
logging in larch forests of the Mongolian Altai, J. Plant Ecol.,
7, 24–38, 2014.
Fickert, T.: Vergleichende Beobachtungen zu Solifluktions- und
Frostmustererscheinungen im Westteil Hochasiens, Erlanger Ge-
ogr. Arb., 60, 150 pp., 1998.
Gerlitz, L., Bechtel, B., Zaksek, K., Kawohl, T., and Böhner, J.:
SAGA GIS based processing of spatial high resolution tempera-
ture data, Proceedings of the 27th EnviroInfo-Conference, 2013,
693–702, 2013.
Gerlitz, L., Conrad, O., Thomas, A., and Böhner, J.: Assessment of
Warming Patterns for the Tibetan Plateau and its adjacent Low-
lands based on an elevation and bias corrected ERA-Interim Data
Set, Climate Res., 58, 235–246, 2014.
Giese, E.: Wetterwirksamkeit atmosphärischer Zustände und
Prozesse in Sowjet-Mittelasien, Westfälische Geogr. Stud., 37,
395–410, 1973.
Giese, E.: Seßhaftwerden von Nomaden, Erfahrungen über die
Dynamik traditioneller sozialer Einrichtungen (am Beispiel des
kasachischen Volkes), in: Die Nomaden in Geschichte und
Gegenwart, Beiträge zu einem internationalen Nomadismus-
Symposium am 11 und 12 Dezember 1975 im Museum für Völk-
erkunde Leipzig, Berlin, 175–197, 1981.
Giese, E.: Nomaden in Kasachstan – Ihre Seßhaftwerdung und
Einordnung in das Kolchos- und Sowchossystem, Geogr. Rund-
schau, 11, 575–588, 1983.
Haase, G., Richer, H., and Barthel, H.: Zum Problem landschaft-
sökologischer Gliederung, dargestellt am Beispiel des Changai-
Gebirges in der Mongolischen Volksrepublik, Wiss. Veröff.
Deutschen Inst. Länderk., 21/22, 489–516, 1964.
Hansen, M. C., Potapov, P. V., Moore, R., Hancher, M., Turubanova,
S. A., Tyukavina, A., Thau, D., Stehman, S. V., Goetz, S. J.,
Loveland, T. R., Kommareddy, A., Egorov, A., Chini, L., Justice,
C. O., and Townshend, J. R. G.: High-Resolution Global Maps
of 21st-Century Forest Cover Change, Science, 342, 850–853,
2013.
Hilbig, W.: The Vegetation of Mongolia, Amsterdam, 258 pp., 1995.
Holdridge, L. R.: Determination of world plant formations from
simple climatic data, Science, 105, 367–368, 1947.
Holtmeier, F.-K.: Die Höhengrenze der Gebirgswälder, Arb. Inst.
Landschaftsökologie, 8, 337 pp., 2000.
Hövermann, J.: Das System der klimatischen Geomorphologie auf
landschaftskundlicher Grundlage, Z. Geomorphol. N. F., Supp.,
56, 143–153, 1985.
Jobbagy, E. G. and Jackson, R. B.: Global controls of forest line
elevation in the northern and southern hemispheres, Global Ecol.
and Biogeogr., 9, 253–268, 2000.
Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D.,
Gandin, L., Iredell, M., Saha, S., White, G., Woollen, J., Zhu, Y.,
Leetmaa, A., Reynolds, R., Chelliah, M., Ebisuzaki, W., Higgins,
W., Janowiak, J., Mo, K. C., Ropelewski, C., Wang, J., Jenne,
R., and Joseph, D.: The NCEP/NCAR 40-year reanalysis project,
Bull. Amer. Meteor. Soc., 77, 437–470, 1996.
Karger, A.: Historisch-geographische Wandlungen der Wei-
dewirtschaft in den Trockengebieten der Sowjetunion am
Beispiel Kasachstans, in: Weide-Wirtschaft in Trockengebieten,
edited by: Knapp, R., Gießen, 37–49, 1965.
Biogeosciences, 12, 2893–2905, 2015 www.biogeosciences.net/12/2893/2015/
M. Klinge et al.: Modeling forest lines and forest distribution patterns 2905
Kastner, M.: Patterns of forest distribution in Western Mongolia,
Berliner Geowissensch. Abh., 205, 67–71, 2000.
Klinge, M., Böhner, J., and Lehmkuhl, F.: Climate patterns, snow-
and timberlines in the Altai Mountains, Central Asia, Erdkunde,
57, 296–308, 2003.
Körner, C.: Alpine Treelines, Functional Ecology of the Global
High Elevation Tree Limits, Springer, Basel, 220 pp., 2012.
Körner, C. and Paulsen, J.: A world-wide study of high altitude tree-
line temperatures, J. Biogeogr., 31, 713–732, 2004.
Liang, E., Lu, X., Ren, P., Li, X., Zhu, L., and Eckstein, D.: Annual
increments of juniper dwarf shrubs above the tree line on the cen-
tral Tibetan Plateau: a useful climatic proxy, Ann. Bot.-London,
109, 721–728, 2012.
Liu, H., Williams, A. P., Allen, C. D., Guo, D., Wu, X.,
Anenkhonov, O. A., Liang, E., Sandanov, D. V., Yin, Y., Qi, Z.,
and Badmaeva, N. K.: Rapid warming accelerates tree growth
decline in semi-arid forests of Inner Asia, Glob. Change Biol.,
19, 2500–2510, 2013.
Lydolph, P. E.: Climates of the Soviet Union, World Survey of Cli-
matology, 7, Amsterdam, 443 pp., 1977.
Machalett, B., Frechen, M., Hambach, U., Oches, E. A., Zöller,
L., and Markovic, S. B.: The loess sequence from Remisowka
(northern boundary of the Tien Shan Mountains, Kazakhstan) -
Part I: Luminescence dating, Quaternary Int., 152/153, 192–201,
2006.
Mayer, T. and Bussemer, S.: Die Waldsteppe Südsibiriens. Ökosys-
temanalyse mit Hilfe der Radarfernerkundung, Mitt. Geogr.
Gesell. München, 85, 161–180, 2001.
Medeu, A. R.: The national atlas of the Republic of Kazakhstan /
Institute of Geography, Part 1: Natural conditions and resources,
Almaty, 2010, 150 pp., 2010.
Miehe, G. and Miehe, S.: Comparative high mountain research on
the treeline ecotone under human impact, Erdkunde, 54, 34–50,
2000.
Miehe, G., Miehe, S., Koch, K., and Will, M.: sacred forests in Ti-
bet – using geographical information systems for forest rehabili-
tation, Mt. Res. Dev., 23, 324–328, 2003.
Miehe, G., Miehe, S., Will, M., Opgenoorth, L., Duo, L., Dorgeh,
T., and Liu, J.: An inventory of forest relicts in the pastures of
Southern Tibet, editd by: Xizang, A. R., China, Plant Ecol., 194,
157–177, 2008.
Paulsen, J. and Körner, C.: A climate-based model to predict poten-
tial treeline position around the globe, Alpine Botany, 124, 1–12,
2014.
Rabus, B., Eineder, M., Roth, A., and Bamler, R: The shuttle radar
topography mission – a new class of digital elevation models
acquired by spaceborne radar, ISPRS J. Photogramm., 57, 241–
262, 2003.
Schlütz, F., Dulamsuren, C., Wieckowska, M., Mühlenberg, M.,
and Hauck, M.: Late Holocene vegetation history suggests natu-
ral origin of steppes in the northern Mongolian mountain taiga,
Palaeogeogr. Palaeocl., 261, 203–217, 2008.
Soria-Auza, R. W., Kessler, M., Bach, K., Barajas-Barbosa,
P. M., Lehnert, M., Herzog, S. K., and Böhner, J.: Im-
pact of the quality of climate models for modelling species
occurrences in countries with poor climatic documentation:
a case study from Bolivia, Ecol. Model., 221, 1221–1229,
doi:10.1016/j.ecolmodel.2010.01.004, 2010.
Treter, U.: Gebirgs-Waldsteppe in der Mongolei, Geogr. Rund-
schau, 48, 655–661, 1996.
Treter, U.: Recent extension and regeneration of the larch forest in
the mountain forest steppe of north-west Mongolia, Marburger
Geogr. Schriften, 135, 156–170, 2000.
Troll, C.: The upper timberlines in different climate zones, Arctic
Alpine Res., 5, A3–A18, 1973a.
Troll, C.: High mountain belts between the polar caps and the equa-
tor: Their definition and lower limit, Arctic Alpine Res., 5, A19–
A27, 1973b.
von Humboldt, A.: Kosmos: Entwurf einer physischen
Weltbeschreibung, Stuttgart, 5 volumes, 1847–1862.
Walter, H. and Breckle, S. W.: Spezielle Ökologie der Gemäßigten
und Arktischen Zonen Euro-Nordasiens, Ökologie der Erde, 3,
Jena, 726 pp., 1994.
Wang, T., Ren, H., and Ma, K.: Climatic signals in tree ring of Picea
schrenkiana along an altitudinal gradient in the central Tianshan
Mountains, northwestern China, Trees, 19, 735–741, 2005.
Wang, T., Zhang, Q. B., and Ma, K.: Treeline dynamics in relation
to climatic variability in the central Tianshan Mountains, north-
western China, Global Ecol. Biogeogr., 15, 406–415, 2006.
www.biogeosciences.net/12/2893/2015/ Biogeosciences, 12, 2893–2905, 2015