Detecting natural succession on abandoned agricultural land in
1. Detecting natural succession on
abandoned agricultural land in the
war-affected northeast Bosnia-
Herzegovina using Landsat TM
imagery
Msc thesis under the supervision of: prof. Jacek Kozak
2. The aim of this study is to identify the magnitude of landscape
change that occurred due to several impacts of the war on
agricultural land in Bosnia-Herzegovina. Main question of this
research is to find places where significant natural succession
process occurred.
• Subregion of Bosnia and
Herzegovina was chosen because,
this place was the focus of intense
fighting, ethnic cleansing and
therefore underwent a substantial
depopulation. Moreover, it
contains minefields. The land use
of this area was predominantly
agricultural where natural
succession is the most expected.
• The study is also an attempt to use
remote sensing data in research
when fieldwork is too dangerous
because of civil conflicts.
3. The study was carried out using two Landsat TM images
from June 1991 and July 2011
• Especially advantageous and interesting is possibility to conduct research
in places where work in field is dangerous or difficult.
• Knowledge about natural succession process is important in environmental
management and for this reason this topic is considered as relevant.
4. This study gives insight for expedience of using various change
detection techniques for evaluation environmental changes.
• For this purpose qualitative and quantitative description environmental changes are
necessary. Post-classification comparison was used as qualitative method and
NDVI differencing as quantitative.
• Additionally changes were evaluated in dependency on elevation and distance from
Srebrenica.
• Shuttle Radar Topography Mission (SRTM) elevation data were used to analyze
changes in land cover in dependency on elevation
5. • For qualitative changes
description supervised
classification was
conducted.
• In case of this study
landscape heterogeneity
poses a problem which
results in high spectral
variation within the same
land-cover class. Supervised
classification was chosen as
a good method to reduce
this problem. Maximum
Likelihood Classification
(MLC) was used as a
classification algorithm.
6. • Visually and statistical classification accuracy
assessment based on a sample of points was
performed. Overall accuracy and overall kappa
coefficient were calculated.
• High resolution images available in Google Earth were
used for accuracy assessment purposes.
7. • To assess land cover changes in a quantitative
way, NDVI differencing was used. This method
was chosen because it emphasizes differences in
the spectral response of different classes. NDVI
differences were then analyzed separately for land
cover classes identified at 1991 image.
• The image of 1991 was classified with a
supervised approach into three classes
‘Settlements and agriculture’, ‘Water’ and ‘Forest’.
To calculate NDVI characteristics for various
areas, zonal mean function available in Erdas
Imagine was used. Zones were the three
delineated land cover classes.
8. • Additionally calculations
were carried out separately
for the Bosnian and
Serbian parts of the study
area. SRTM data were
used to recognize
variations of land cover
changes expressed with
NDVI differencing in
dependency on elevation.
• Changes of NDVI were
described in dependency
on distance from
Srebrenica. To calculate
NDVI characteristics for
various areas, zonal mean
function was applied.
Zones were classes of
distance and classes of
elevation.
9. Results
• The overall land use classification accuracy for image
from 1991 is 92%. Result of classification is significantly
better than random (at the 95 percent confidence level).
• Because the aim of this work was to identify natural
succession and to assess the impact of the war (through
depopulation or landmines) on agricultural land, the most
important was selecting places with significant changes in
vegetation. For this qualitative change detection purpose,
results of supervised classification was performed for both
Landsat scenes.
11. To support the research hypothesis on land abandonment and
succession on agricultural land, NDVI for the pre-and post-war
imagery was calculated. This provides a directly comparable
measure of vegetation changes.
The analysis was focused only on agricultural areas, with an
assumption that forests and water bodies had stable NDVI over
time and were not affected by the war. The higher NDVI
values reflect the higher amount of vegetation in abandoned
agricultural land.
12. NDVI values from both sets of scenes were differenced and
combined into a single map of NDVI differences for the
“Settlements and agriculture”
13. • Both positive changes (increase of NDVI) and negative
changes (decrease of NDVI) were observed in ‘settlements
and agriculture’ class,.
• The most significant positive NDVI changes were observed in
Srebrenica region and also in cities Banovici and Zivinice.
Quantitative analysis confirmed thesis that vegetation cover
increased in abandoned agricultural land of study area.
• Moreover, NDVI differences increased also in the ‘Forest’
class in area around Srebrenica. This pattern in the
‘Settlements and agriculture’ class further confirms that there
was a significant land abandonment around Srebrenica, with
more intense natural succession than elsewhere in the study
area.
• No significant variation of NDVI differences were observed in
dependency on elevation, in the ‘Settlements and agriculture’
class. Mean NDVI differences in this class for all study area
very similar to values obtained for particular elevation ranges.
Similar patterns were observed in the ‘Forest’ class.
14. Key research findings and conclusions
• Comparison of independently classified images confirms
thesis that amount of vegetation increased over the analyzed
20 years period. The greatest amount of vegetation growing on
abandoned agricultural land was found in the Srebrenica
region.
• Ethnic cleansing, re-settlement actions and danger caused by
minefields were likely a major factor in process of ecological
succession.
• Bosnian War have had significant influence on land use
pattern. In case of this study, the war has not put land use
systems toward intensification trajectories but allowed
landscapes to ‘rewild’, and gave opportunities for
conservation.
15. • It is difficult to forecast how long-lasting land use changes
will be. The vegetation cover probably would increase in
region of Srebrenica but the future of minefields remains
open. Evidence from other areas suggests that farmland
abandonment may persist for a long time.
• Both post-classification comparison and vegetation index
differencing (NDVI differencing) were found as useful in
analyzing natural succession. Qualitative analysis in post-classification
comparison may be used as a complementary
methods to quantitative approach such as vegetation index
differencing (NDVI differencing).