Applied Geography 35 (2012) 345e352

Contents lists available at SciVerse ScienceDirect

Applied Geography
journal homepage: www.elsevier.com/locate/apgeog

Deforestation in Heilongjiang Province of China, 1896e2000: Severity,

spatiotemporal patterns and causes
Jay Gao a, b, Yansui Liu a, *
a
b

Institute of Geographical Sciences and Natural Resources Research, CAS, Beijing 100101, China
School of Environment, University of Auckland, Private Bag 92019, Auckland, New Zealand

a b s t r a c t
Keywords:
Deforestation
Change detection
Spatiotemporal pattern
Cause analysis
Heilongjiang

This study attempts to ascertain the spatiotemporal patterns of de(re)forestation and its causes in Heilongjiang Province, China during the last century. In 1896, there were 308,020 km2 of forest covering 68%
of the Province. Forest area was reduced to 247,256 km2 by 1949 at an annual rate of 1146 km2. By 1958
primary forest was reduced to 169,533 km2 while secondary forest remained at 68,801 km2. Thus,
9421 km2 of forest were logged at a rate of 1046 km2 per annum. From 1958 to 1980, forest as a whole
was reduced by 22,326 km2 at an annual rate of 1014 km2. The amount of deforestation was reduced to
9211 km2 for dense forest, but sparse forest gained 831 km2 during 1980e2000. The net decrease of
8379 km2 represents an annual loss of 419 km2. Spatially, deforested areas used to be extensive and
expansive, but have become fragmented with thousands of patches that have a shrinking mean size.
These deforested sites were located in low-lying at terrains with a close proximity to rivers and roads.
Such land was replaced primarily by farmland and secondarily by grassland. Therefore, the causes of
deforestation are identied as demand for timber, population-driven land reclamation up to 1980, and
urbanisation in the last two decades.
2012 Elsevier Ltd. All rights reserved.

Introduction
Forests are one of the most signicant components of the Earths
biosphere. They play a critical role in regulating the Earths surface
temperature and precipitation, preserving soil nutrients, minimizing ooding, and xing carbon, in addition to serving as
a valuable habitat for wildlife. In spite of these functions, however,
forest resources around the world are being depleted for various
reasons. Forest depletion or deforestation is a process of removal of
a forest or stand of trees that results ultimately in non-woodland
uses or tree covers at a reduced density. According to FAO (2005),
about half of mature forests in the tropics have been logged.
Deforestation has been occurring at an annul rate of 5.8 million ha
between 1990 and 1997 (Achard et al., 2002) while 2.3 million ha is
degraded annually (Mayaux et al., 2005). This conversion from
forest to non-forest covers have contributed to environmental
degradation, increased ooding, exacerbated soil erosion, and
reduced biodiversity. Deforestation must be timely monitored so
that decision-makers can have the most recent knowledge on the
status of forest land in order to achieve sustainable forestry, and
minimize carbon footprint on the environment.

* Corresponding author: Tel.: 86 10 64889037; fax: 86 10 64857065.

E-mail addresses: jg.gao@auckland.ac.nz (J. Gao), liuys@igsnrr.ac.cn (Y. Liu).
0143-6228/$ e see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.apgeog.2012.08.001

As a panoramic, systematic, and objective method of detection

and assessment, remote sensing enables deforestation to be efciently monitored. With an ability to faithfully preserve the Earths
surface (Achard et al., 2002), aerial photographs and satellite images
have found applications in detecting forest cover change and forest
fragmentation (Harper, Steininger, Tucker, Juhn, & Hawkins, 2007).
Satellite imagery with a broad coverage, a moderate resolution, and
multispectral bands are particularly useful for monitoring deforestation (Baker & Williamson, 2006) owing to its capability of
supplying transparent and reliable information on forest cover and
condition (Fuller, 2006). Space-borne imagery enables detection and
differentiation of forest and non-forest covers (Townshend, Justice,
Li, Gurney, & McManus, 1991). Because of its synoptic views,
space-borne imagery plays a vital role in identifying and estimating
deforested areas (Buchanan et al., 2008). Coarse resolution data such
as Advanced Very High Resolution Radiometer (AVHRR) imagery at
a resolution of 1 km are well suited to identication of deforestation
(Tucker, Holben, & Goff, 1984), assessment of deforestation rate
(Gastellu-Etchegorry, Estreguil, Mougin, & Laumonier, 1993), and
documentation of land-clearing activities over an extensive range
(Stone, Brown, & Woodwell, 1991). The ne temporal resolution of
AVHRR images makes them especially valuable in real-time identication and locating of res, but hampers their utility in estimating
burned areas (Setzer, Pereira, & Pereira, 1994). Such a task is more
competently achieved from images of a ner spatial resolution

346

J. Gao, Y. Liu / Applied Geography 35 (2012) 345e352

(e.g., multispectral scanner MSS imagery at 79 m resolution) that are

good at determining net forest removal, regrowth, and forest edges
(Huang et al., 2007; Westman, Strong, & Wilcox, 1989). New satellite
images with improved spatial resolution have become available
since then.
Multi-temporal remote sensing data that have been in existence
for decades are especially good at detecting changes in land use in
general and deforestation in particular. Goldewijk and Ramankutty
(2004) reviewed the detection of land cover change over the last
three centuries. The areas of historical studies range from regional
to country, such as the Nestos delta, Greece (Mallinis,
Emmanoloudis, Giannakopoulos, Maris, & Koutsias, 2011) and
Belgian Ardennes (Petit & Lambin, 2002). Since earth resources
satellites did not come into existence until the early 1970s, the
change detection span is restricted to tens of years (Abd El-Kawy,
Rd, Ismail, & Suliman, 2011; Gao & Liu, 2010; Hadeel, Jabbar, &
Chen, 2010; Masoud & Koike, 2006; Onur, Maktav, Sari, & Kemal
Snmez, 2009). This span can be extended further to the pre1970s era if coarser resolution data such as AVHRR are used
(Laneve & Castronuovo, 2005). Longer period of change detection
has to rely on non-remote sensing data (Mallinis et al., 2011). So far
the post-classication change detection method is the most
popular (Chen & Wang, 2010; Faid & Abdulaziz, 2012; Gao & Liu,
2010). Images are also processed to derive vegetation index that
is used for subsequent detection of vegetation-related change
(Laneve & Castronuovo, 2005).
Remotely sensed data are indispensable in studying the rapidly
changing forest environment and in determining the trend of
deforestation (Peralta & Mather, 2000), such as detecting spatial
changes in forest cover, and identifying critical areas of forest cover
changes (Lele & Joshi, 2009). Once the deforested areas are identied, their spatial patterns can be quantied (Steininger et al.,
2001; Wang et al., 2003), their relationship with the size of forest
cover analysed (Zhang, Devers, Desch, Justice, & Townshend, 2005),
and the reasons behind deforestation explored. After deforestation
is analysed from multi-temporal satellite images, and the trend of
deforestation rates estimated, it is possible to assess the impact of
spatial, cultural, and economic factors on deforestation (e.g., the
distance from roads, rivers, research facilities, oil facilities, markets
and towns, and land ownership) (Greenberg, Kefauver, Stimson,
Yeaton, & Ustin, 2005).
The causal relationship between deforestation and its contributors is most effectively determined via overlay of forestry maps
with other physiographic layers in a geographic information system
(GIS). This spatial analysis can reveal the signicance and magnitude of the relationship between forest cover and environmental
variables (Apan & Peterson, 1998). A proposed vegetation classication logic for remote-sensing data (Running, Loveland, & Pierce,
1994) and recent computer advances with GIS make estimation
of deforestation repeatable and quantiable (Iverson et al., 1994).
However, no comprehensive studies have been carried out to
systematically examine the change between forest and other land
covers over a long (e.g., >100 years) period and the causes of
deforestation over different periods. Such a long-time frame is
essential in order to assess the validity of the forest transition
theory that states over time, forest cover declines, but at some
point a transition occurs, such that the decline halts and reverses,
and forest cover thereafter expands (Perz, 2007). This study aims
at modifying this theory using the remote sensing-GIS approach in
accurate quantication of deforestation in Heilongjiang Province of
Northeast China. Specically, this research aims (1) to ascertain the
level and extent of deforestation in this subarctic region of China
over the last century; (2) to detect structural changes of primary
(e.g., non-plantation) and secondary (e.g., planted and regenerated)
forest resources and their spatiotemporal patterns; (3) to establish

the association between the observed pattern and physiographic

features; and (4) to explore the causes of deforestation.
Study area
Heilongjiang is located in Northeast China between 121130 e
135 050 E and 43 220 e53 24N. This Province has a territory of
452,000 km2 and a population of 38.17 million (2010 census data). It
has a continental monsoon climate with an annual temperature
of 4 to 4  C. Winters are long and frigid (temperature averages 31
to 15  C), and summers are short and cool with an average
temperature of 18e23  C. Annual rainfall averages 500e600 mm,
concentrated mostly in the summer months. The terrain is dominated by a few mountain ranges, two most important ones being the
Greater Khingan and Lesser Khingan (Fig. 1). This Province has
relatively plentiful water resources with several major rivers, of
which the Amor (black dragon) in the north-east bordering Russian,
and the Songhua River that ows across it are the most important
ones. Wetlands are distributed in low lying plains of 50e250 m
above sea level. The Province has the countrys largest natural
forests that constitute the principal component of the forest
resources. They are also one of the most distinctive temperate forests
in the world. Forestry covers 19.19 million ha of the Provinces total
land area (41.9%), with a total reserve of 1.5 billion m3 of timber.
The Province has a long history of deforestation. Large-scale
logging started in the early 1920s with the construction of the
mid-eastern railway. During 1929e1944 forest area in Northeast
China decreased by 18%, with the timber reserve decreased by
14.3%. Deforestation has caused grave environmental degradation,
such as soil erosion and ooding.
Methodology
Remotely sensed data and land use map
Two historical forest cover maps produced by the forest
resources investigation and management unit of the Headquarters
of Forestry and Industry Bureau in Heilongjiang Province were
collected. They showed the distribution of forest (including Korean
pine) in 1896 and 1949 at a scale of 1:2.5 million. These maps were
produced by the Forestry Resource Investigation and Management
Ofce of (then) Heilongjiang Forest Chief Bureau to the highest
accuracy possible (Wang et al., 2003). A land use map of Northeast
China was collected from the Institute of Geographical Science and
Natural Resources, Chinese Academy of Science. It was produced at
a scale of 1:3 million from visual interpretation of aerial photographs taken in 1958. Displayed in this map were ten categories of
general land covers, such as primary and secondary forest. In
addition, Landsat TM/Enhanced TM Plus (ETM) images recorded
in 1980 and 2000 were collected from the ground receiving station
in Beijing. About 58 images were acquired to cover the entire
Province in each time. These images were recorded during the
JuneeSeptember period when forest has a maximum spectral
disparity from other covers.
GIS data
These data include a digital elevation model (DEM), roads, rails
and stream data. The DEM was downloaded from the NASAs EOS
data archive (http://asterweb.jpl.nasa.gov/gdem.asp). These
archived global DEM data are publicly available free of charge. They
cover the Earths land surface between 83 N and 83 S. Global DEM
is produced at 30 m from stereoscopic ASTER VNIR bands. These
data are available in 1 by 1 tiles in the GeoTIFF format. Roads, rails
and stream data were extracted from the Digital Charts of the

J. Gao, Y. Liu / Applied Geography 35 (2012) 345e352

347

Fig. 1. Location of the study area, and its major physiographic features.

World at http://www.maproom.psu.edu/dcw/in the ERSI interchange format. The data used in this study had a scale of 1:1 million
(version 1993).
Data processing
All the collected forest and general land use maps were scanned
at 300 dpi in Adobe Photoshop using a atbed scanner. The scanned
forest maps of 1896 and 1949 were then classied into 20 clusters
using the iterative K-means clustering method in ERDAS Imagine.
These clusters were subsequently grouped into two categories of
forest and non-forest. After median ltering to remove noise, the
derived forest cover maps in the raster format were geo-referenced
to the UTM coordinate system (zone 51 north) in ArcGIS using the
current digital map to around 1 pixel size, vectorized and saved as
shape les. They were then intersected with the provincial
boundary to retain only those areas inside the study area.
The 1958 land use map contained multiple covers that were
extracted in Adobe Illustrator. A separate layer was created for
a unique land cover in the original map. Afterwards, each layer
was exported to ArcGIS as a raster image separately. It was

vectorized using the spatial analyst conversion tool. Similar steps

were taken for both primary and secondary forest to ensure that
the smallest land use type maintained its presence on the relatively larger layers. Finally, a new shape le was created, into
which primary and secondary forest shape les were added. A
new eld was added to that new shape le for calculating the area
of each polygon.
The false colour composites of bands 4, 3 and 2 of the TM/
ETM satellite data were visually interpreted to produce land
cover maps. Prior to interpretation, the images were geometrically
rectied based on distinct ground points to an accuracy of less than
two pixels (about 50 m) in PCI Geomatics (version 7.0). Land cover
information was delineated through on-screen digitization
according to a pre-determined classication scheme of six categories of land covers (farmland, woodland, grassland, built-up
areas, water, and fallow land). Farmland included both dry land
and rice paddy. Woodland is dened as areas covered by natural
forest, shrubs, or sparsely populated saplings. Grassland is dened
as any land used for grazing. The accuracy of these mapped covers
was assessed through a combination of eld visits and comparison
with existing statistics, and was found to be over 90%.

348

J. Gao, Y. Liu / Applied Geography 35 (2012) 345e352

Detection of deforestation
All the forest/land cover maps were overlaid with each other
using the union logic in ArcGIS after they had been projected to
the same UTM coordinate system (i.e., zone 51 north). The newly
derived output maps retained the identity of all polygons in both
the input and output layers, which enables the detection of
conversion between forest and other land covers. These resultant
land cover change maps were then queried to generate the quantity
of change from forest to non-forest covers, or deforestation. The
queried results were then exported to Excel for quantitative analysis, and visualized in ArcGIS to illustrate the spatial pattern of
deforestation.
Spatial analysis
The deforestation maps were overlaid with physiographic
features in order to establish the relationship of deforestation with
physical accessibility and natural settings (e.g., slopes and elevation). This was achieved by generating a buffer of 2 km around
roads, railways and streams after they had been extracted from the
Digital World Chart. After being dissolved, this buffered le was
then intersected with the deforestation maps in ArcGIS. Those areas
inside the buffered zone were analysed statistically to identify the
spatial patterns of deforestation. Later the deforestation maps were
also overlaid with the DEM and slope derived from it individually to
examine the topographic properties of deforested sites.
Results
Trend of deforestation
1896e1949
Forested areas stood at 308,020 km2 in 1896, covering 68% of the
Province. Some of the forest patches are extensive spatially, suggesting a holistic landscape with a low degree of fragmentation
(Fig. 2A). Relatively small patches of forest are distributed in the
lower west, juxtaposed with non-woodland area. Another unique
feature of non-forest land is its distribution in valleys along stream
channels in the linear form. Forest area shrank to 247,756 km2 in
1949 (Fig. 2B), or a reduction of 62,697 km2 at 1183 km2 per annum.
The most noticeable change to the spatial distribution of forest land
is characterized by the disappearance of small patches of forest in

the west and the drastic shrinkage of forest in the Songnen plain
(Fig. 3A). The number of forest patches was reduced while the
remaining ones were all relatively large in size. Consequently,
deforested areas had a relatively large and consistent patch size
throughout the Province with the most expansive patches located
in the lower west and east of the Province.
Although some areas were identied as reforested in Fig. 3A,
such changes are likely not genuine as they have a linear shape
along rivers or the border. These spurious changes resulted from
the misalignment of the two sets of maps used in the change
detection, a common problem of overestimation in land cover
change detection caused by positional error in multi-temporal
images that is related to the number of classes (Verbyla & Boles,
2000). Given the enormous quantity of forest present (nearly
a quarter of million square kilometres), such inaccuracy exerted an
inconsequential impact on the accuracy of change as skinny polygons enclosed little area.
1949e1958
In total, there were 142,184 km2 of primary forest (Table 1) while
secondary forest stood at 68,801 km2 in 1958. During 1949e1958
deforested area mounted to a net loss of 60,265 km2. Consequently, forested area was reduced to 247,755 km2 at an annual rate
of 1183 km2. The pace of deforestation in this period drastically
quickened than in the previous period. Primary forest lost
78,222 km2 (8638 km2 per annum), 54,230 km2 of which were
turned to secondary forest in southern Heilongjiang while less than
half of this amount (20,548 km2) was turned into dry land (Table 2).
A similar quantity (21,335 km2) was changed to marshes.
Spatially, most of the deforested areas are distributed in
southern Heilongjiang in large patches (Fig. 3B). These lowland
areas have a higher concentration along the Songhua River in the
east and the southwest that can be inundated by water easily.
Deforestation was particularly severe in the Songnen plain, with
former forest replaced by farmland. Another heavily logged area is
the southern Greater Khingan range where primary forest had been
succeeded by secondary forest. Deforested areas in central Heilongjiang have a relatively small patch size, especially in the west.
1958e1980
During 1958e1980 the general trend of deforestation is characterized by an increase of 16,659 km2 in primary forest and
a decrease of 38,985 km2 in secondary forest (Table 1). Overall,

Fig. 2. Distribution of forested areas in 1896 (A) and 1949 (B).

J. Gao, Y. Liu / Applied Geography 35 (2012) 345e352

349

Fig. 3. Change in the distribution of deforested areas in four periods. A: 1896e1949; B: 1949e1958; C: 1958e1980; D: 1980e2000.

forested area decreased by a net value of 22,326 km2. In contrast to

this decrease, dry land gained 28,657 km2 while paddy elds
increased 11,006 km2. Most of the gain in dry land is from dense
forest at 12,192 km2 and from sparse forest at 10,245 km2 (Table 2).
The next highest destination cover of forests is grassland with
a combined area of 18,256 km2. In addition, 45,950 km2 of primary
forest became sparse forest, against only 18,502 km2 of dense forest
Table 1
General land covers with an emphasis on forestry and their change in two periods
(unit: km2).a
Land cover

1958

1980

Change
1958e1980b

2000

Change
1980e2000

Farmland
Primary/dense
forest
Secondary/sparse
forest
Grassland/barren
Built-up areas
Water
Swamp
Sandy
Sum

102,304
169,534

141,968
186,193

39,663
16,659

160,088
176,982

18,120
9211

68,801

29,816

38,985

30,648

831

converted from secondary forest. Deforested areas can be found

throughout the Province, especially towards the eastern part
(Fig. 3C). While the number of deforested patches is numerous,
their size has decreased considerably. Such a spatial pattern
suggests that logging was more localized. This could be attributed
to the difcult terrain over which remaining trees were standing.
1980e2000
The trend of deforestation continued during this period. The
dense forest area decreased by 9211 km2 while sparse forest gained
831 km2 (Table 1), resulting in a net loss of 8380 km2. All the

Table 2
Land covers that have replaced forest at the end of a study period (unit: km2).
Land cover

26,652
385
1576
80,066
1777
451,095

37,871
8650
15,537
28,480
4046
452,561

11,218
8265
13,961
51,586
2269
1466

31,966
8784
14,905
25,203
3976
452,551

5905
134
631
3277
71
10

a
Results for 1896 and 1949 were not shown as they are binary (forest versus
non-forest).
b
means increase;  means decrease.

Primary forest
Secondary forest
Farmland
Grassland
Build-up Area
Water
Sandy land
Marsh
Total

1949 / 1958

1958 / 1980

1980 / 2000

Dense

Dense

Sparse
45,950

54,230
21,515
6019
60
406
2004
21,335
105,569

18,502
13,499
13,102
865
2292
480
5589
54,627

11,201
5154
617
703
32
2121
81,574

Sparse
3330

5560
7769
1078
16
4
7
2
14,436

1670
14
1994
15
74
1
7096

350

J. Gao, Y. Liu / Applied Geography 35 (2012) 345e352

deforested areas have a small size and are widely distributed in

a mountainous terrain with a small physical size (Fig. 3D). The
majority of the cleared land was used as farmland. Dry land gained
10,665 km2 while paddy elds increased by 7455 km2. These results
are consistent with what Zhang, Liu, and Zhang (2003) have found.
They reported that land cover change in the Greater and Lesser
Khingans was characterized by the conversion from forest/grassland to farmland. During 1980e1995 a total of 3078 km2 of farmland was gained from woodland. This amount more than doubled
to 6223 km2 during 1980e2000, even though not all newly gained
farmland originated from forest (e.g., from grassland). A similar
trend of conversion was also observed in East Heilongjiang.

Relationship with environmental variables

With terrain
The entire Province has a relatively low elevation that averages
only 317 m above sea level with a small standard deviation of 209 m
(Table 3). The topography is at with a mean gradient of only 2.74 .
During 1896e1949, deforestation took place at a low elevation
where the land is mostly at with a gradient (0.84 ) much lower
than the mean slope gradient of the entire Province. During 1958e
1980 deforestation shifted to a higher ground with the mean
elevation of deforested areas rising to 299.91 m above mean sea
level (amsl), and further to 337.81 m during 1980e2000, higher
than the average elevation of the entire Province. This migration to
a higher ground suggests that trees at lower elevations have long
been logged, so there are not many left for further logging. This
trend can also be appreciated from the increasing steepness in
topography. The mean slope gradient rose to 1.33 during 1958e
1980, and to 1.93 during 1980e2000. In spite of the increase, the
slope is still gentler than the mean gradient of the entire Province.
Thus, deforestation took place mostly in gentle terrain of a low
elevation.
With proximity to channels
In the earliest period over half of the deforested areas were
located within 2 km from a river channel (Table 3). Thus, rivers
were rather important to deforestation as roads were not well
developed back then. Logs were transported mainly through rivers.
Rivers became even more important during the 1949e1958 period
as testied by the huge increase in the proportion of <2 km
deforested areas (91.3%), but this proportion diminished to the low
50s during 1958e1980. The portion was the lowest during 1980e
2000 at only 39.2%. This trend of decreasing proportions reects
the fact that fewer and fewer trees in close proximity to river
channels were left for logging as time went by. In contrast, those
trees more distant from rivers are more economically and conveniently transported overland.
With accessibility
A similar pattern is observable with distance to roads and railways. In the earliest period 44.7% of the deforested areas were

Table 3
Topographic properties of deforested sites in relation to those of the Province.
Topographic variable

Whole province
1896e1949
1949e1958
1958e1980
1980e2000

Elevation (m)

Gradient (degrees)

Mean

St dev

Mean

St dev

317.09
194.57
262.85
299.91
337.81

209.9
124.1
151.58
164.02
164.90

2.74
0.84
2.21
1.33
1.93

4.46
2.20
3.82
2.57
3.09

located within 2 km from existing roads. This percentage then

steadily decreased over the years to 36.7% during 1980e2000
(Table 3), a percentage similar to that with streams. This decrease
must be examined in the context of road construction. Given that
more roads have been built since 1896, it is easy to conjecture that
trees further away from roads were logged and transported overland. The relatively high proportion of deforested areas along roads
and channels makes these two parameters reliable predictors of
deforestation in the future.
The above ndings about the relationship between deforestation and other variables are in agreement with that of Greenberg
et al. (2005) who concluded that the strongest predictor of deforestation is proximity to roads. The highest rate of forest clearing
(1.5% gross) at a Peruvian frontier took place along roads (Alvarez &
Naughton-Treves, 2003). The model developed by Gunter, Hodges,
Swalm, and Regens (2000) in estimating the probability of urban
development in a pine or mixed forest parcel identied the distance
to the nearest state or federal highway, distance to the access
points, and distance to the nearest interstate interchange as the
signicant inuencers of a parcel being deforested.
Causes of deforestation
Elsewhere the replacement of undisturbed forest with cropland,
grassland, and secondary scrub demonstrates that the proximate
cause of deforestation is chiey agricultural expansion as a result of
the synergistic convergence of climatic, technological, and socioeconomic factors (Zak, Cabido, Cceres, & Daz, 2008). Other
determinants of deforestation in the Amazon are identied as land
characteristics (such as soil quality and vegetation density) and
factors affecting transport costs (such as distance to major markets
and roads) (Pfaff, 1999). While not all of these factors are equally
applicable to the study area, some factors (e.g., the pressure from
agriculture) appear to be universal, together with the demand for
timber and land resources.
Demand for timber
Deforestation has been historically acute in Northeast China
where 600 million m3 of timbers were produced during 1949e
1979, causing a drastic decline in forest resources (Zhang, 2008).
During 1948e1950 alone, approximately 6 million m3 of timber
was output from Northeast China. In 1950, Heilongjiang produced
5.21 million m3 of timber. By 1977, half of the national timber was
generated in Heilongjiang. After nearly 60 years of logging from
1896, the total timber reserve in this Province decreased from 4
billion m3 to about 1.4 billion m3 (Zhang, 2008). The fact that
primary forest was logged at a faster pace than secondary forest
indicates that the demand for timber was a critical cause of
deforestation in the earlier periods. In fact, the demand for timber
was the sole determinant of deforestation prior to 1949, and
a major cause until 1980.
Agricultural pressure
Around the middle of last century one of the most dominant
land covers that replaced forest was dry land. During 1949e1958,
20,548 km2 of forest were converted to dry land, accounting for
nearly 20% of the total deforested areas (Table 4). Although the
absolute amount of dry land created out of dense or sparse forest
individually was reduced to over 10,000 km2 during 1958e1980,
their combined area of 22,437 km2 is much higher than that
created during the previous period. This magnitude is also the
second highest next to the change from primary forest to sparse
forest. In particular, the proportion of dry land rose sharply to 22.3%

J. Gao, Y. Liu / Applied Geography 35 (2012) 345e352

351

Urbanisation

Table 4
Proximity of deforested sites to roads and streams.
Period

Deforested
area (km2)a

Annual
rate (km2)

Distance to roads
and railways

Distance
to streams

<2 km

<2 km

1896e1949
1949e1958
1958e1980
1980e2000

60,265a
51,341
56,657
9483

1137
5704
2575
474

26,938
18,839
21,802
3074

44.7
39.7
38.5
36.7

30,340
46,874
29,582
3721

50.3
91.3
52.2
39.2

a
These areas are larger than the difference values presented in Table 1 because
they included newly forested areas.

from dense forest and 12.6% from sparse forest, the highest among
all the new uses of deforested land. Additionally, a large portion of
the forest has been turned into grazing land, especially dense forest
at 13,102 km2. Farmland/grassland replacing dense forest reached
a combined proportion of 48.7%, or nearly half of the cleared dense
forest. The pace of conversion from forest to farmland was subdued
during 1980e2000. However, farmland creation was still the major
drive of deforestation as it was still the most dominant land cover
to replace forest. By comparison, grassland did not gain much from
forest. Therefore, the need to create more farmland is the major
cause of deforestation from 1949 to 1980.
This detected trend of decrease in forest and increase in cropland is highly consistent with the national trend identied by Ge,
Dai, He, Pan, and Wang (2008). They reported that cropland in
China kept increasing at the expense of forest land during 1700e
1949. This trend of primary forest replaced by other land covers
was also observed in Argentina with the main cause of deforestation identied as agricultural expansion (e.g., soybean cultivation in
particular) (Zak et al., 2008). Within less than half a century most of
the forests in the hill country were cleared for planting such crops
as corns, wheat and soy beans. Shifting cultivation was responsible
for deforestation in the drier parts of the hill country in Sri Lanka
hill country (Wickramagamage, 1998).
The detected conversion from forest to farmland was a direct
response to the ever increasing rural population caused by a high
fertility rate in Heilongjiang Province. It rose steadily from less than
10 million in 1958 to over 20 million in 1978, an increase of more
than 100% (Fig. 4). The newly gained population required more food
and hence cultivation of more land. Farmland was created via
deforestation. However, rural population started to decline after
reaching the peak in 1978, relieving the necessity of farmland
creation. This explains the subdued conversion from forest to
farmland during 1980e2000.

During 1980e2000 the largest change to sparse forest among all

of its changes is the conversion to built-up areas at 1994 km2
(Table 2), the second highest destination cover of sparse forest,
higher than dry land (1621 km2). Thus, urbanisation encroached
upon sparse forest in the last two decades. The necessity of clearing
forest to construct settlements can be traced to the growth in urban
population of the Province that rose sharply from 12.3 million in
1980 to over 20 million in the late 1990s. The inux of a huge rural
population to cities and towns led to the expansion of human
settlements and required construction of associated infrastructure
in land that was created through deforestation. Unlike the demand
for timber that affected mostly primary forest, urbanisation affects
only secondary forest as a result of tightened control over the
clearing of forest.
Conclusions
The deforestation trend in Heilongjiang Province is characterized by a drastic reduction in forest from 308,020 km2 to
247,256 km2 at an annual rate of 1146 km2 during 1896e1949.
Deforestation continued at an unabated pace during 1949e1958
when primary forest was reduced to 169,533 km2 while
secondary forest remained at 68,801 km2. If both types of forest are
lumped together, deforestation sums up to 9421 km2 at an annual
rate of 1046 km2. From 1958 to 1980, forest lost 22,326 km2 at an
annual rate of 1014 km2. The pace of deforestation was reduced to
9211 km2 for dense forest and actually gained 831 km2 for sparse
forest during 1980e2000. The net decrease of 8379 km2 represents
an annual rate of only 419 km2. Apparently, the trend of deforestation has not been reversed by natural regeneration as the deforested land was used for non-forest purposes.
The deforested areas have a at topography. At the early stage of
deforestation these sites were low-lying with a close proximity to
river channels and roads. As the deforestation process persisted, the
proportion of deforested sites within the 2 km buffer zone
decreased. A similar pattern is found with distance to roads and rail
roads. Initially, deforestation took place in a spatially expansive
manner, resulting in the smallest number of deforested patches. In
subsequent periods the number of deforested patches became
increasingly larger while the maximum size of deforestation
decreased correspondingly. The net effect is a highly fragmented
landscape. Accompanying these changes was the shift of deforested
areas from a low elevation to higher grounds and the deforested
terrain became ever steeper.
The disappearance of forest in the early stage was caused by
the demand for timber. This cause of deforestation prevailed
until 1980. From the middle of last century, dry land and grassland had replaced dense and sparse forest, suggesting that the
pressure to produce enough food to feed the ever growing
population was the main cause of deforestation. During 1980e
2000 sparse forest was turned into built-up areas, suggesting
urbanisation was the major cause of deforestation. Because of
such diverse uses of the deforested land, it is impossible to
imagine that the forest transition theory applies here. This theory
needs to be modied to take into account the nal use of the
deforested land.
Acknowledgements

Fig. 4. Change in rural and urban population between 1949 and 2000 in Heilongjiang
Province (source: National Bureau of Statistics Department of Comprehensive
Statistics, 2005).

This research was supported by a research grant from the

Faculty of Science at the University of Auckland. Additional funding
was received from the National Natural Science Foundation of
China (grant numbers 41130748 and 40871257), and the

352

J. Gao, Y. Liu / Applied Geography 35 (2012) 345e352

Knowledge Innovation Program of the Chinese Academy of

Sciences (grant number KZCX2-EW-304). We are grateful to two
anonymous reviewers who made valuable and constructive
comments on an earlier version of this manuscript.

References
Abd El-Kawy, O. R., Rd, J. K., Ismail, H. A., & Suliman, A. S. (2011). Land use and land
cover change detection in the western Nile delta of Egypt using remote sensing
data. Applied Geography, 31, 483e494.
Achard, F., Eva, H. D., Stibig, H. J., Mayaux, P., Gallego, J., Richards, T., et al. (2002).
Determination of deforestation rates of the worlds humid tropical forests.
Science, 297, 999e1002.
Alvarez, N. L., & Naughton-Treves, L. (2003). Linking national agrarian policy to
deforestation in the Peruvian Amazon: a case study of Tambopata 1986e1997.
Ambio, 32, 269e274.
Apan, A. A., & Peterson, J. A. (1998). Probing tropical deforestation: the use of GIS
and statistical analysis of georeferenced data. Applied Geography, 18, 137e152.
Baker, J. C., & Williamson, R. A. (2006). Satellite imagery activism: sharpening the
focus on tropical deforestation. Singapore Journal of Tropical Geography, 27, 4e14.
Buchanan, G. M., Butchart, S. H. M., Dutson, G., Pilgrim, J. D., Steininger, M. K.,
Bishop, K. D., et al. (2008). Using remote sensing to inform conservation status
assessment: estimates of recent deforestation rates on New Britain and the
impacts upon endemic birds. Biological Conservation, 141, 56e66.
Chen, Z., & Wang, J. (2010). Land use and land cover change detection using satellite
remote sensing techniques in the mountainous Three Gorges Area, China.
International Journal of Remote Sensing, 31, 1519e1542.
Faid, A. M., & Abdulaziz, A. M. (2012). Monitoring land-use change-associated land
development using multitemporal Landsat data and geoinformatics in Kom
Ombo area, South Egypt. International Journal of Remote Sensing, 33, 7024e7046.
FAO (Food and Agriculture Organization). (2005). Global forest resources assessment
2005: Progress towards sustainable forest management. Rome.
Fuller, D. O. (2006). Tropical forest monitoring and remote sensing: a new era of transparency in forest governance? Singapore Journal of Tropical Geography, 27, 15e29.
Gao, J., & Liu, Y. (2010). Determination of land degradation causes in Tongyu County,
Northeast China via land cover change detection. International Journal of Applied
Earth Observation and Geoinformation, 12, 9e16.
Gastellu-Etchegorry, J. P., Estreguil, C., Mougin, E., & Laumonier, Y. (1993). A GIS
based methodology for small scale monitoring of tropical forests e a case study
in Sumatra. International Journal of Remote Sensing, 14, 2349e2368.
Ge, Q., Dai, J., He, F., Pan, Y., & Wang, M. (2008). Land use changes and their relations
with carbon cycles over the past 300 a in China. Science in China Series D: Earth
Sciences, 51, 871e884.
Goldewijk, K. K., & Ramankutty, N. (2004). Land cover change over the last three
centuries due to human activities: the availability of new global data sets.
Geojournal, 61, 335e344.
Greenberg, J. A., Kefauver, S. C., Stimson, H. C., Yeaton, C. J., & Ustin, S. L. (2005).
Survival analysis of a neotropical rainforest using multitemporal satellite
imagery. Remote Sensing of Environment, 96, 202e211.
Gunter, J. T., Hodges, D. G., Swalm, C. M., & Regens, J. L. (2000). Predicting the
urbanization of pine and mixed forests in Saint Tammany Parish Louisiana.
Photogrammetric Engineering and Remote Sensing, 66, 1469e1476.
Hadeel, A. S., Jabbar, M. T., & Chen, X. (2010). Environmental change monitoring in
the arid and semi-arid regions: a case study Al-Basrah Province, Iraq. Environmental Monitoring and Assessment, 167, 371e385.
Harper, G., Steininger, M., Tucker, C., Juhn, D., & Hawkins, F. (2007). Fifty years of
deforestation and forest fragmentation in Madagascar. Environmental Conservation, 34, 325e333.
Huang, C., Kim, S., Altstatt, A., Townshend, J. R. G., Davis, P., Song, K., et al. (2007).
Rapid loss of Paraguays Atlantic forest and the status of protected areas e
a Landsat assessment. Remote Sensing of Environment, 106, 460e466.
Iverson, L. R. S., Brown, A., Prasad, H., Mitasova, A. J., Gillespie, R., & Lugo, A. E.
(1994). Use of GIS for estimating potential and actual forest biomass for
continental South and Southeast Asia. In V. H. Dale (Ed.), Effects of land-use
change on atmospheric CO2 concentrations: South and Southeast Asia as a case
study (pp. 67e116). New York: Springer-Verlag.
Laneve, G., & Castronuovo, M. M. (2005). Comparison between vegetation change
analysis in Kenya based on AVHRR and SeaWiFS images. International Journal of
Remote Sensing, 26, 2549e2559.

Lele, N., & Joshi, P. K. (2009). Analyzing deforestation rates spatial forest cover
changes and identifying critical areas of forest cover changes in North-East
India during 1972e1999. Environmental Monitoring and Assessment, 156,
159e170.
Mallinis, G., Emmanoloudis, D., Giannakopoulos, V., Maris, F., & Koutsias, N. (2011).
Mapping and interpreting historical land cover/land use changes in a Natura
2000 site using earth observational data: the case of Nestos delta, Greece.
Applied Geography, 31, 312e320.
Masoud, A. A., & Koike, K. (2006). Arid land salinization detected by remotelysensed landcover changes: a case study in the Siwa region, NW Egypt. Journal
of Arid Environments, 66, 151e167.
Mayaux, P., Holmgren, P., Achard, F., Eva, H., Stibig, H. J., & Branthomme, A. (2005).
Tropical forest cover change in the 1990s and options for future monitoring.
Philosophical Transactions of the Royal Society of London Series B Biological
Sciences, 360, 373e384.
National Bureau of Statistics Department of Comprehensive Statistics. (2005). China
compendium of statistics 1949e2004. Beijing: China Statistics Press.
Onur, I., Maktav, D., Sari, M., & Kemal Snmez, N. (2009). Change detection of land
cover and land use using remote sensing and GIS: a case study in Kemer,
Turkey. International Journal of Remote Sensing, 30, 1749e1757.
Peralta, P., & Mather, P. (2000). An analysis of deforestation patterns in the
extractive reserves of acre Amazonia from satellite imagery: a landscape
ecological approach. International Journal of Remote Sensing, 21, 2555e2570.
Perz, S. G. (2007). Grand theory and context e specicity in the study of forest
dynamics: forest transition theory and other directions. Professional Geographer,
59, 105e114.
Petit, C. C., & Lambin, E. F. (2002). Impact of data integration technique on historical
land-use/land-cover change: comparing historical maps with remote sensing
data in the Belgian Ardennes. Landscape Ecology, 17, 117e132.
Pfaff, A. S. P. (1999). What drives deforestation in the Brazilian Amazon? Evidence
from satellite and socioeconomic data. Journal of Environmental Economics and
Management, 37, 26e43.
Running, S. W., Loveland, T. R., & Pierce, L. L. (1994). A vegetation classication logic
based on remote sensing for use in global biogeochemical models. Ambio, 23,
77e81.
Setzer, A. W., Pereira, M. C., & Pereira, A. C., Jr. (1994). Satellite studies of biomass
burning in Amazonia e some practical aspects. Remote Sensing Reviews, 10,
91e103.
Steininger, M. K., Tucker, C. J., Ersts, P., Killeen, T. J., Villegas, Z., & Hecht, S. B. (2001).
Clearance and fragmentation of tropical deciduous forest in the Tierras Bajas
Santa Cruz, Bolivia. Conservation Biology, 15, 856e866.
Stone, T. A., Brown, I. F., & Woodwell, G. M. (1991). Estimation by remote sensing of
deforestation in central Rondonia Brazil. Forest Ecology and Management, 38,
291e304.
Townshend, J., Justice, C., Li, W., Gurney, C., & McManus, J. (1991). Global land cover
classication by remote sensing: present capabilities and future possibilities.
Remote Sensing and the Environment, 35, 243e255.
Tucker, C. J., Holben, B. N., & Goff, T. E. (1984). Intensive forest clearing in Rondonia
Brazil as detected by satellite remote sensing. Remote Sensing of Environment, 15,
255e261.
Verbyla, D. L., & Boles, S. H. (2000). Bias in land cover change estimates due to
misregistration. International Journal of Remote Sensing, 21, 3553e3560.
Wang, X.-C., Sun, L., Zhou, X.-F., Wang, T.-M., Li, S.-J., & Guo, Q.-X. (2003). Dynamic
of forest landscape in Heilongjiang Province for one century. Journal of Forestry
Research, 14, 39e45.
Westman, W. E., Strong, L. L., & Wilcox, B. A. (1989). Tropical deforestation and
species endangerment: the role of remote sensing. Landscape Ecology, 3, 97e109.
Wickramagamage, P. (1998). Large-scale deforestation for plantation agriculture in
the hill country of Sri Lanka and its impacts. Hydrological Processes, 12, 2015e
2028.
Zak, M. R., Cabido, M., Cceres, D., & Daz, S. (2008). What drives accelerated land
cover change in central Argentina? Synergistic consequences of climatic
socioeconomic and technological factors. Environmental Management, 42,
181e189.
Zhang, S. F. (2008). Northeast timber: the pillar of the new China. Chinese National
Geography, 576, 240e247, (in Chinese).
Zhang, Q., Devers, D., Desch, A., Justice, C. O., & Townshend, J. (2005). Mapping
tropical deforestation in central Africa. Environmental Monitoring and Assessment, 101, 69e83.
Zhang, G. P., Liu, J., & Zhang, Z. (2003). Spatial-temporal changes of cropland in
China for the past 10 years based on remote sensing. Acta Geographica Sinica, 58,
323e332.