Recent forest cover changes (2002 2015) in the Southern Carpathians: A [603790]

Recent forest cover changes (2002 –2015) in the Southern Carpathians: A
case study of the Iezer Mountains, Romania
Bogdan Mihaia, Ionu țSăvulescua, Marina Rujoiu-Marea,b, Constantin Nistora,⁎
aUniversity of Bucharest, Faculty of Geography, 1, Nicolae B ălcescu Blvd., 010041 Bucharest, Romania
bSimion Mehedin ți Doctoral School, University of Bucharest, Faculty of Geography, 1, Nicolae B ălcescu Blvd., 010041 Bucharest, Romania
HIGHLIGHTS
•The forest cover changes in Iezer Moun-
tains between 2002 and 2015 wereanalysed.
•A change detection approach based on
Landsat ETM+/OLI images was used.
•Clear cuttings and windthrows are the
main forms of deforestation.
•The leading force of deforestation is the
ownership recovery process after 2005.GRAPHICAL ABSTRACT
abstract article info
Article history:
Received 6 March 2017Received in revised form 28 April 2017Accepted 28 April 2017Available online 29 May 2017
Editor: Elena PaolettiThe paper explores the dynamics of the forest cover change i n the Iezer Mountains, part of Southern Carpathians, in
the context of the forest ownership recovery and deforestation processes, combined with the effects of biotic andabiotic disturbances. The aim of the study is to map and eva luate the typology and the spatial extension of changes
in the montane forest cover between 700 and 2462 m a.s.l., sampling all the representative Carpathian ecosystems,
from the European beech zone up to the spruce- firz o n ea n dt h es u b a l p i n e – a l p i n ep a s t u r e s .T h em e t h o d o l o g yu s e s
a change detection analysis of satellite imagery with Landsat ETM+/OLI and Sentinel-2 MSI data. The work flow
started with a complete calibration of multispectral data from 2002, before the massive forest restitution to private
owners, after the Law 247/2005 empowerment, and 2015, the intensi fication of deforestation process. For the data
classi fication, a Maximum Likelihood supervised classi fication algorithm was utilized. The forest change map was
developed after combining the classi fications in a unitary formula using image difference. The principal outcome of
the research identi fies the type of forest cover change using a quantitative formula. This information can be inte-
grated in the future decision-making strategies for forest stand management and sustainable development.
© 2017 Elsevier B.V. All rights reserved.Keywords:
DeforestationChange detectionLandsat ETM+/OLI
Sentinel-2 MSI
Forest management
1. Introduction
Over the last three decades, social and economic changes in Central
and Eastern Europe have left their imprint on the relationship between
the environment and society. After 1990, the changes in forest cover,
particularly featuring the high-value mountain broadleaf and conifer
forests of the Carpathian region, have been one of the key indicators of
the environmental change. This process occurred in the context of theScience of the Total Environment 599 –600 (2017) 2166 –2174
⁎Corresponding author.
E-mail addresses: [anonimizat] (B. Mihai), [anonimizat]
(I. S ăvulescu), [anonimizat] (M. Rujoiu-Mare), [anonimizat]
(C. Nistor).
http://dx.doi.org/10.1016/j.scitotenv.2017.04.226
0048-9697/© 2017 Elsevier B.V. All rights reserved.
Contents lists available at ScienceDirect
Science of the Total Environment
journal homepage: www.elsevier.com/locate/scitotenv

gradual land restitution or land ownership recovery ( Cvitanovic et al.,
2016 ) and the re-orientation of national economies from centralized in-
dustrial production of the socialist period to private enterprise. The
transition period was motivated by immediate financial gain and
characterised by limited short- and long-term investments in infra-
structure and human resources and unsustainable natural resource ex-
ploitation practices.
Various researchers ( Song et al., 2001; Woodcock et al., 2001 )h a v e
demonstrated and explained the phenomena from the point of view
of its spatio-temporal dimension and socio-economic driving factors
(Munteanu et al., 2014 ), with the help of Landsat remote sensing imag-
ery time series. To date, much of the research has focused on the effects
of forest restitution ( Griffiths et al., 2012; Knorn et al., 2012; Stefanski et
al., 2014 ) as well as on the interaction of environmental, socio-econom-
ical and historical factors on forest canopy dynamics ( Griffiths et al.,
2014; Shandra et al., 2013 ) and subsequent forest management issues
(Munteanu et al., 2016 ). A convergent approach has followed the inte-
gration of these environmental changes into a broader context for a bet-
ter understanding of the continental and global effects of deforestation
(Kim et al., 2014; Levers et al., 2014; Plutzar et al., 2016 ).
The effects of forest cover transformation at the regional or local scale
were also the subject of numerous of research papers from throughout
the Carpathian ecological region. Most of these studies mapped the spa-
tial pattern of changes and brie fly analysed the factors driving forest dis-
turbance and the relationship with environmental change issues ( Kozak,
2010; Kozak et al., 2007; Kuemmerle and Muller et al., 2009; Mihai et al.,
2007; Mina et al., 2015 ). With the exception of these contributions, the
most recent changes in forest stand distribution on a detailed scale has
been less studied and documented with maps.
The purpose of this paper is to quantify the forest cover changes in
the Iezer Mountains in the context of the intensi fication of deforestation
and timber harvesting, as a consequence of massive restitution of forest
patches to their former owners – special Law 247/2005 ( Parliament,
2005 ) – and the privatization of forest m anagement bodies. It is supposed
that the newly legislation introduced a non-linear evolution of forest
zones and forest stand dynamics with clear-cuts and road building
resulting in intensi fication of soil erosion and sedimentation on the ex-
posed sites. This is a consequence of the average density of the forest
roads (9.4 m/ha for the study area), which is higher than the national
value (6.7 m/ha). This current research continues a previous work inthe same study area ( Mihai et al., 2007 ), using a new Landsat data series.
2. Materials and methods2.1. Study area
The Iezer Mountains are a representative mountain region in South-
ern Carpathians of Romania ( Fig. 1 ). The study area was selected accord-
ing as a representative site of the entire Southern branch of the
Carpathian Mountain massif. The altitude ranges between 700 m and
2469 m a.s.l. with a glacial relief featuring cirques and valleys. The relief
is actually modelled by periglacial processes with debris accumulation at
the higher elevations followed by landslides and gully erosion at the
lower altitudes. This area has been intensely exploited by sheep grazing
and timber harvesting since the medieval period.
The selection of this study area is strongly related to the variety of
mountain environments and to the high degree of mountain accessibility
(Savulescu and Mihai, 2011 ) along the main river valleys, characterised
by a dense network of forest roads from the foot of the mountains and
up to some higher elevatio ns near the timberline (1900 –2000 m), with
an average density of forestry roads between 3 and 7 m/ha. These road
networks are the main access points for timber harvesting and transport
activities. This geographic context is typical for all mountain regions in
the Southern Carpathians –a dense network of roads connecting to
densely clustered settlements along valleys and depressions. The devel-
opment of road building in the mountains has played an increasing rolein exploitation and management of n atural resources d uring the last two
decades of unbridled economic activity.
2.2. Geospatial data and data sources
Following the previous time series of remote sensing imagery, the
current analysis integrates Landsat imagery from the latest sensors gen-
eration (ETM+ for the year 2002, OLI for 2015). These images were inte-
grated with ancillary data, used for calibration of the spectral signatures
and the production of vegetation grid coverages ( Table 1 ).
Since the beginning of 2000s, a variety of authors have evaluated
Landsat time series imagery from the USGS archive for larger regions at
30 m resolution ( Woodcock et al., 2001 ). This data with a little more
than two weeks revisiting time and a higher spectral resolution (seven
bands to nine bands at latest versions) is freely available and easy to be
processed with elevation data like SRTM30 or EU-DEM in the context
of data calibration stage.
The analysis used two cloud-free Landsat coverages (scene, mosaic
depending on the WRS-2 scene footprint arrangement) from USGS ar-
chives that was selected for an anniversary date more close to the end
of the peak of the vegetation season from August to September, for sam-
ple data production.
Thefirst Landsat data coverage for August 30, 2002 corresponds to
the period before the massive forest land ownership recovery when
the initial pattern of forest zones and stands was the main feature with
primarily secondary forest stands, 50 –70 years old ( Săvulescu, 2014 ).
The second data coverage (September 2, 2015) shows the near pres-
ent-day con figuration of forest zones. This OLI data features the intensive
logging period, when private forest land administration replaced the
state management.
For topographic corrections, a hillshade grid generated from EU-DEM
elevation model is integrated toget her with Landsat images in an equa-
tion, in order to produce calibrated and nearly pure spectral signatures
of forest vegetation.
Sentinel-2 MSI data with a better spatial, spectral and temporal reso-
lution (up to 10 m, 13 spectral channels, 10 days revisiting time) are se-
lected for validating the preview Landsat change detection grid and fordetailed mapping.
2.3. Data processing and analysis
The analysis of forest cover change in the Iezer Mountains followed
two sections as shown in the flow chart ( Fig. 2 ). The first is remote sens-
ing data processing for change data production, while the second part is
the GIS data integration and analysis for mapping and statistic ap-
proaches. The process starts with the production of training samples
which is based on the forest stand vector data from INCDS Bucharest
(National Institute for Research and Development in Forestry) ( INCDS,
2002; INCDS, 2015 ), expert knowledge from the comparative visual
analysis of imagery as well as field data collected in different periods of
the vegetation season, from September 2003 until May 2016 (it includes
GPS points and field photos, referenced with orthophotos 0.5 m from
2012 by ANCPI Bucharest).
Satellite imagery (L1T level of processing) needs a special calibration
approach in order to reduce the in fluence of sensor parameters, atmo-
sphere and topography/illumination on the current data from the both
sensors ( Schowengerdt, 2007 ). The metadata file connected with the
spectral bands as found in the archives from USGS help the application
of gain and offset parameters from Landsat ETM+ (2002) and Landsat
OLI (2015) sensors at the dates of images for Top of the Atmosphere
(TOA) radiance calculation. Atmospheric correction by FLAASH tool
(Fast Line-of-sight Atmospheric Analysis of Hypercubes) returned the
spectral re flectance of the ground surface on Bottom of the Atmosphere
(BOA). The topographic C-correction method ( Chuvieco, 2016 )r e t u r n e d
the surface re flectance grid for all the spectral channels of both imagery2167 B. Mihai et al. / Science of the Total Environment 599 –600 (2017) 2166 –2174

sets, normalized on a 0 –1 scale in order to extract the forest cover re flec-
tance, free from shadows at the same scale.
The co-registered calibrated images are analysed within the training
stage of the thematic classi fication of land cover/forest vegetation types
(seven classes). This is done with the help of the expert knowledge(field updated) combined with the forest stand vector database by
INCDS Bucharest, resized on the study area. The training areas de fined
are plotted in a separability chart for all available spectral bands to ex-
plain the difference between classes within spectral channel of imagery
(Fig. 3 ).
Fig. 1. Location map of the Iezer Mountains, Southern Carpathians, Romania. Lands at OLI image 564 band combination at 30 m spatial resolution from August 201 5. Data source: USGS Earth
Explorer Landsat archive.
Table 1Geospatial data used for change detection analysis of forest cover in the Iezer Mountains (2002 –2015).
No. Datasets Spatial resolution/scale Year Data source
1 Landsat 7 ETM+ SLC-on LE71830282002242SGS00
LE71830292002242MTI0030 m 2002 US Geological Survey, Earth explorer
2 Landsat 8 OLI LC81840282015245LGN00 30 m 2015 US Geological Survey, Earth explorer
3 EU-DEM Digital elevation dataset 3035 25 m 2013 EEA (European Environment Agency)4 Sentinel-2 A MSI S2A_OPER_PRD_MSIL1C_
PDMC_20150808T09254410 m 2015 Scihub Copernicus/European Space Agency
5 Forest Stand Vector Database,
converted to rasterRomanian Forest 1:20000,
Raster 10 m2002,
2015INCDS (National Institute for Research and
Development in Forestry)2168 B. Mihai et al. / Science of the Total Environment 599 –600 (2017) 2166 –2174

Fig. 2. Flow chart of the change detection analysis (2002 –2015) of forest cover changes in the Iezer Mountains, Romania.
Fig. 3. The ROI (Region of Interest) separability index for the training sets extracted from Landsat OLI image.2169 B. Mihai et al. / Science of the Total Environment 599 –600 (2017) 2166 –2174

The diagram in Fig. 3 confirms the sharpest differences between forest
zones and the alpine/subalpine pastures and the barren grounds at the
highest elevations, above 1800 –2000 m a. s. l. These occur in the near in-
frared and short wave infrared spectra. In the same intervals, there is an
important difference between deciduous forest (mainly beech stands),
dwarf pines ( Pinus mugo in subalpine zone above the timberline) as
well as mixed and coniferous forest (mainly spruce- fir stands). For this
reason, the most suitable colour composites for land cover analysis are
beyond the edge of the visible spectrum, with a peak in the mid- or
short-wave infrared, where forest stands have different re flectance (de-
c i d u o u sv sc o n i f e r o u s ) .
The thematic classi fications were performed with the help of the
Maximum Likelihood algorithm ( Chuvieco, 2016; Schowengerdt,
2007 ), which is based on the Gaussian probability calculation around
each training set of pixels. This classi fication algorithm is selected be-
cause of the relatively limited spectral information for one reference
time, which need to be strictly compared with the training samples.
Two reference raster coverages for land cover types (seven classes) are
obtained from the analysis.
The results are validated with the help of ground truth polygons de-
rived from field data collection and INCDS (2002, 2015) forest stand da-
tabase. Both datasets are evaluated with the help of the confusion
matrices which returned accuracies and the Kappa coef ficients.
Change detection analysis is based on the method used in the previ-
ous work in the study area ( Mihai et al., 2007 ). The image classi fication
difference method ( Lillesand et al., 2015 ) can easily discriminate the
transformation of land cover classes, and allow the mapping of them
into a GIS application. The forest cover change grid is reclassi fied and
simpli fied with a median filter for preserving the useful clusters with rel-
evant information at the resolution of the original imagery (30 m). There
are 12 change classes to be mapped focusing on forest vegetation zones.
Validation of this new layer is also done in a qualitative formula, by
superposing the clusters and the corresponding vectors on Sentinel-2
MSI image. In this context, the vectors of the forest stands (2002) are
overlaid, only for the attribute check between land cover classes and
the transformation classes. The quantitative validation is performedwith the corresponding data subset from the Global Forest Loss (GFL)
data layer ( Hansen et al., 2013 ). In this context, a cross tabulation be-
tween the current deforestation dataset (2002 –2015) and the GFC
dataset (2000 –2014) is done for the entire study area to evaluate the
change detection analysis accuracy.
Mapping of forest cover change continues with the statistical analysis
of changes within the matrix of class transformation and the diagram ofchanges. This allows the separation between the direction of class
change and the area of changes.
3. Results
The forest cover change analysis in the Iezer Mountains continues the
previously established approach (1986 –2002), with the new interval of
about 15 years (2002 –2015) to map the effects of the ownership change
and private forest management within a representative mountain area of
the Southern Carpathians.
The analysis produces the basic land cover data, and mainly the forest
cover classes affected by transformation in the local/regional context, the
current approach being adapted to the quality of the satellite imagery
available and fi
eld data.
The results of Landsat calibration process ( Fig. 4 ) show visible differ-
ences in the brightness values of the same forest zones in the original
data and the corrected data, where atmospheric effect of water vapours,
together with illumination condition are mostly removed.
For both time reference data, a set of two land cover classi fications are
obtained ( Fig. 4 ). Seven classes are described and adapted to the local sit-
uation as well as to the objectives of the analysis, in continuity with the
previous one. These follow the mapping of the main vegetation zones:
deciduous forests (beech), coniferous forests (spruce- fir), mixed forest
stand, dwarf pine, pasture, barren land and reservoirs. Some issues
needs special attention: the subalpine dwarf pine stands classi fication
with nearly similar signatures with the neighbouring spruce- firs t a n d s ;
the classi fication of barren ground as different from the rare vegetation
and even from settlements and riverbeds. The first problem is solved
by improving the training area for dwarf pine with the help of ground
Fig. 4. Land cover classi fication of Landsat imagery, Iezer Mountains. A. ETM+ data from August 30, 2002. B. OLI data from September 2, 2015.2170 B. Mihai et al. / Science of the Total Environment 599 –600 (2017) 2166 –2174

data and INCDS database. The second problem is only partly solved, al-
though calibration theoretically returns nearly pure spectral data. Barren
ground is correctly classi fied above timberline and along floodplains, but
it is no longer differentiated from the compact and dense built-up areas.
The confusion matrices con firm the reliability of data for further
change detection analysis. In our case, only the overall accuracy (higher
than 90%) together with Kappa coef ficient (more than 0.9), made possi-
ble the integration of both classi fications into the change detection anal-
ysis. The change detection map ( Fig. 5 ) as a result of the difference
between classi fication grids, allows the production of a new dataset in
vector format, which illustrates the transformation of the forest cover
during the reference period (gain and loss in stand surface).
For an easier understanding of the nature of changes in their spatial
and temporal patterns, a separate layer derived from a Sentinel-2 MSI
L1C image is used. This shows at 10 m resolution the con figuration of for-
est zones and also cleared areas. For mapping purposes, this new data
layer is resampled in SNAP 5software at 10 m resolution, because of
the smaller differences in spatial resolution consistency.The interpretation of the change detection results together with the
significance of class transformation is done on detailed maps for selected
areas in the Iezer Mountains (presented in the Supplementary material),
which illustrate different situations:
a) recent clear-cuts are the effect of logging activities within the owner-
ship recovered forested lands;
b) a continuity of the older clearings with new deforested stands, in-
cluding the areas affected by windthrows;
The windthrow areas form July 2005 are mapped on the basis of the
orthophotos ( ANCPI, 2006 ). The other surfaces mapped as forest stand
loss are the same with the clear-cuts.
4. Discussion
Mapping of forest cover change as a main objective of this study was
possible after the accuracy assessment of the basic datasets. The classi-
fication confusion matrices are generated for the both reference times
Fig. 5. Change detection map of the forest zones in the Iezer Mountains, 2002 –2015.2171 B. Mihai et al. / Science of the Total Environment 599 –600 (2017) 2166 –2174

in order to prepare the change clusters. Table 2 corresponds to the Max-
imum Likelihood classi fication of the Landsat OLI data (2015), that
returns high values of accuracy (92.66%). It is interesting to notice that
mixed forest (large extension) and dwarf pines have an accuracy
equal or more than 97%. Pastures are poorly classi fied (88%) as an effect
of the dif ficulties in the separation between grass and other isolated
vegetation patches like trees, dwarf pine, juniper and other stands.
The land cover change matrix and the statistics of the class transfor-
mation ( Fig. 6 ) show the type of changes in its spatial dimension (hect-
ares of class transformation). This chart is drawn on the basis of change
detection data processing, where the classes previously de fined returned
12 types of forest class transformation, same as we mapped. It shows a
strong transformation of forest cover in the study area, beyond the gen-
eral trend of vegetation zones shifting in altitude, also con firmed in the
previously published papers ( Mihai et al., 2007; S ăvulescu, 2014 ).
Forest cover change model (forest loss) in Iezer Mountains returns
more detailed results than the existing Global Forest Loss data 2000 –
2014 ( Hansen et al., 2013 ). The linear regression coef ficient of 0.55
(Fig. 7 ) is the result of the degree of generalization of the GFC (Global
Forest Change) dataset in comparison with the currently produced
dataset. GFC features the forest stands with a density of canopies higher
than 30% and a height of trees of minimum 5 m. Our approach is based on
spectral data derived from 30 m resolution imagery and training areas
validated by field mapping and a GPS survey for ground control points
collection.
The most spectacular transformation is the loss of more than 6100 ha
of forest stands in the study area, where conifers were converted to pas-
tures and barren land. This process also features the deciduous forest at
lower altitudes (700 –1200 m), where villagers of the Roma populationpractice illegal timber harvesting around the village which is, for the
most part, the only one economic alternative. After 2007, intensive log-
ging was much larger than the natural potential capacity.
Table 3 shows the administrative change in forest cover management
as an effect of the empowerment of Law 247/2005 in the study area. The
Central and Eastern mountain slopes, belonging to Câmpulung and Ruc ăr
forest inspectorates –Fig. 5 , features an important ownership transfer to
the private sector. This is correlated with the most intensive forest loss
areas. The forest ratio was calculated from the current data (0.20), and
confirms the deforestation process in comparison with the reference
value (1.0) which means no change in forest stand surface. For the entire
Romanian Carpathians region this index is 0.6 and for the Southern
Carpathians it is 0.72, as it is con firmed from the GFC datasets spatial
analysis.
State control over forested lands was partly lost after 2005. This
meant that private regional and local forest administration bodies were
free to apply a new forest stand management strategy, based on timber
harvesting from stands and less on forest stand recovery.
This forest cover change trend features the former socialist countries
and mainly some Carpathian countries. After 1990, forest cover dynam-
ics followed a new direction in Central and Eastern Europe, as a conse-
quence of changes in the political system and ownership structures.
Forest cover disturbances showed different patterns ( Cvitanovic et al.,
2016; Levers et al., 2014 ), from an intensi fication of deforestation in
Croatia ( Cvitanovic et al., 2016 ), Romania ( Knorn et al., 2012;
Kuemmerle and Muller et al., 2009; Plutzar et al., 2016 ), Czech Republic
(Griffiths et al., 2014 ) and Ukraine ( Kuemmerle and Chaskovskyy et al.,
2009; Stefanski et al., 2014 ) to reforestation in Hungary ( Griffiths et al.,
2014 ).Table 2
Classi fication confusion matrix for land cover based on Landsat OLI image, from September 02, 2015.
Classes Ground reference data
BL P CF DP DF MF W Total User's Accuracy %
Classi fied Image Barren land BL 94.12 12.21 0.45 0.00 0.00 0.00 0.00 7.79 74.42
Pastures P 5.88 87.79 0.00 1.39 1.30 0.00 0.00 14.49 94.38
Coniferous forest CF 0.00 0.00 90.50 0.00 0.00 1.08 0.00 18.30 99.01
Dwarf pine DP 0.00 0.00 8.14 97.22 0.00 0.00 0.00 7.97 79.55
Deciduous forest DF 0.00 0.00 0.00 0.00 90.91 1.62 0.00 25.63 98.94
Mixed forest MF 0.00 0.00 0.90 1.39 7.79 97.30 0.00 18.75 86.96
Water W 0.00 0.00 0.00 0.00 0.00 0.00 100.00 7.07 100.00
Producer's accuracy % 94.12 87.79 90.50 97.22 90.91 97.30 100.00
Overall accuracy =92.66% Kappa coef ficient =0.9108
The bold values on the diagonal represent the percentage of correlation between the classi fication result and ground truth data. The users's accuracy and producer's accuracy values are
related to commission and omission errors from classi fication.
Fig. 6. Change detection diagram of land cover classes in the Iezer Mountains, between 2002 and 2015.2172 B. Mihai et al. / Science of the Total Environment 599 –600 (2017) 2166 –2174

As a consequence of this situation, the Romanian Government Act
Number 51/2016 ( Government, 2016 ) introduced protection measures
for a strict control of logging, timber harvesting and transportation.
Since 2014, the Romanian Ministry of Environment launched a mobile
application called “Inspectorul P ădurii ”(The Forest's Inspector). This al-
lows any user to write and send messages to environmental authorities
about timber transports identi fied on highways and roads. Between
2014 and 2016, there were received 30,000 records (25% were con-
firmed as illegal transports).
5. Conclusion
The change detection analysis at 30 m spatial resolution satis fies the
requirements for mapping and evaluat ion of forest cover change at local
scales and for obtaining the spatial dimension of the phenomenon aswell as the typology of change. The recently produced satellite imagery
(Sentinel-2 MSI) is a more useful as regards their spatial and spectral res-
olution, but they have a limited in temporal resolution and were used for
result validation.
The analysis con firms the continuation of the natural trend of vertical
shifting of the mountain vegetation zones. There is a trend of deciduous
forest shifting upon the coniferous one, while barren grounds above the
timberline are colonized by dwarf pines and sometimes by pastures,
when grazing is not intensive, mainly on the lee sides. This is partly the
effect of the climatic evolution proven by weather data. The results can
be extended to many of the regions in the Southern Carpathians.
The anthropogenic pressure given by economic reasons limits the bi-
ological productivity increase and the general natural trend is not suf fi-
cient for the forest and vegetation recovery.The intensi fication in deforestation after 2007 overtakes the forest
natural recovery capacity. The ecological stability is characterised by fra-
gility and the environmental components are seriously affected (soil,
surface and groundwater, fauna, etc.). Even if the forest can be rehabil-
itated, the new, arti ficially-created ecosystems will have a higher degree
of exposure to biotic and abiotic hazards.
Acknowledgements
A special thank you to Yurij Bihun, Director, Shelterwood Systems,
Jericho, Vermont and Science Committee member of Science for the
Carpathians (S4C), for assistance in editing the material.
Appendix A. Supplementary data
Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.scitotenv.2017.04.226 .
References
ANCPI, 2006. Orthophotos. INSPIRE INIS VIEWER, Bucharest.
Chuvieco, E., 2016. Fundamentals of Satellite Remote Sensing: An Environmental Ap-
proach. CRC Press Taylor & Francis, Boca Raton – London – New York.
Cvitanovic, M., Blackburn, G.A., Jepsen, M.R., 2016. Characteristics and drivers of forest
cover change in the post-socialist era in Croatia: evidence from a mixed-methods ap-
proach. Reg. Environ. Chang. 16, 1751 –1763.
Government R, 2016. Government act no. 51/2016 51/2016. Monitorul O ficial al
României.
Griffiths, P., Kuemmerle, T., Kennedy, R.E., Abrudan, I.V., Knorn, J., Hostert, P., 2012. Using
annual time-series of Landsat images to assess the effects of forest restitution in
post-socialist Romania. Remote Sens. Environ. 118, 199 –214.
Griffiths, P., Kuemmerle, T., Baumann, M., Radeloff, V.C., Abrudan, I.V., Lieskovsky, J., et al.,
2014. Forest disturbances, forest recovery, and changes in forest types across the
Carpathian ecoregion from 1985 to 2010 based on Landsat image composites. Remote
Sens. Environ. 151, 72 –88.
Hansen, M.C., Potapov, P.V., Moore, R., Hancher, M., Turubanova, S.A., Tyukavina, A., et al.,
2013. High-resolution global maps of 21st-century forest cover change. Science 342,
850 –853.
INCDS, 2015. In: INCDS (Ed.), Romanian Forest Stands.
INCDS, 2002. In: INCDS (Ed.), Romanian Forest Stands.
Kim, D.H., Sexton, J.O., Noojipady, P., Huang, C.Q., Anand, A., Channan, S., et al., 2014. Global,
Landsat-based forest-cover change from 1990 to 2000. Remote Sens. Environ. 155,
178 –193.
Knorn, J., Kuemmerle, T., Radeloff, V.C., Szabo, A., Mindrescu, M., Keeton, W.S., et al., 2012.
Forest restitution and protected area effectiveness in post-socialist Romania. Biol.
Conserv. 146, 204 –212.
Kozak, J., 2010. Forest cover changes and their drivers in the Polish Carpathian mountains
since 1800. Reforesting Landscapes. 10 (253-+).
Fig. 7. Cross-validation between GFL (2000 –2014) and forest cover loss (2002 –2015) derived from Landsat images in the Iezer Mountains.
Table 3
Forest stands surface recovery in the Iezer Mountains after the empowerment of Law 247/
2005.
Forest
inspectorateArea (ha) in 1996 (from
Forest Management Map)Area (ha) in 2017
(www.rosilva.ro )Forest stand
recovered surface
(Law 247/2005)
Area (ha) %
Câmpulung 29,567.80 17,473.00 12,130.80 41.03
Rucăr 27,078.60 0.00 27,078.60 100.00
Domne ști 30,033.40 22,393.00 7640.40 25.44
Aninoasa 22,553.00 16,023.00 6530.00 28.952173 B. Mihai et al. / Science of the Total Environment 599 –600 (2017) 2166 –2174

Kozak, J., Estreguil, C., Troll, M., 2007. Forest cover changes in the northern Carpathians in
the 20th century: a slow transition. J. Land Use Sci. 2, 127 –146.
Kuemmerle, T., Chaskovskyy, O., Knorn, J., Radeloff, V.C., Kruhlov, I., Keeton, W.S., et al.,
2009a. Forest cover change and illegal logging in the Ukrainian Carpathians in the
transition period from 1988 to 2007. Remote Sens. Environ. 113, 1194 –1207.
Kuemmerle, T., Muller, D., Grif fiths, P., Rusu, M., 2009b. Land use change in Southern Roma-
nia after the collapse of socialism. Reg. Environ. Chang. 9, 1 –12.
Levers, C., Verkerk, P.J., Muller, D., Verburg, P.H., Butsic, V., Leitao, P.J., et al., 2014. Drivers of
forest harvesting intensity patterns in Europe. For. Ecol. Manag. 315, 160 –172.
L i l l e s a n d ,T M a ,K i e f e r ,R W a ,C h i p m a n ,J W a ,2 0 1 5 . Remote Sensing and Image Interpretation.
Mihai, B., Savulescu, I., Sandric, I., 2007. Change detection analysis (1986 –2002) of vegeta-
tion cover in Romania: a study of alpine, subalpine, and forest landscapes in the Lezer
mountains, Southern Carpathians. Mt. Res. Dev. 27, 250 –258.
Mina, M., Bugmann, H., Klopcic, M., Cailleret, M., 2015. Accurate modeling of harvesting is
key for projecting future forest dynamics: a case study in the Slovenian mountains.
Reg. Environ. Chang. 1 –16.
Munteanu, C., Kuemmerle, T., Boltiziar, M., Butsic, V., Gimmi, U., Halada, L., et al., 2014.
Forest and agricultural land change in the Carpathian region-a meta-analysis oflong-term patterns and drivers of change. Land Use Policy 38, 685 –697.
Munteanu, C., Nita, M.D., Abrudan, L.V., Radeloff, V.C., 2016. Historical forest management
in Romania is imposing strong legacies on contemporary forests and their manage-
ment. For. Ecol. Manag. 361, 179 –193.
Parliament R, 2005. Legea nr. 247 din 2005 privind reforma în domeniile propriet ățiiși
justi ției, precum și unele m ăsuri adiacente. Monitorul O ficial.Plutzar, C., Kroisleitner, C., Haberl, H., Fetzel, T., Bulgheroni, C., Beringer, T., et al., 2016.
Changes in the spatial patterns of human appropriation of net primary production
(HANPP) in Europe 1990 –2006. Reg. Environ. Chang. 16, 1225 –1238.
Săvulescu, I., 2014. Vegeta ția forestier ăaM u n ților Iezer. Editura Etnologic ă, Bucharest.
Savulescu, I., Mihai, B., 2011. Mapping forest landscape change in Iezer Mountains, Roma-
nian Carpathians. AGIS approach based on cartographic heritage, forestry data and re-
mote sensing imagery. J. Maps 7, 429 –446.
Schowengerdt, R.A., 2007. Remote Sensing: Models and Methods for Image Processing.
Academic Press, London.
Shandra, O., Weisberg, P., Martazinova, V., 2013. Influences of climate and land use history
on forest and timberline dynamics in the Carpathian Mountains during the twentieth
century. In: Kozak, J., Ostapowicz, K., Bytnerowicz, A., Wy żga, B. (Eds.), The
Carpathians: Integrating Nature and Society Towards Sustainability. Springer, Berlin
Heidelberg, Berlin, Heidelberg, pp. 209 –223.
Song, C., Woodcock, C.E., Seto, K.C., Lenney, M.P., Macomber, S.A., 2001. Classi fication and
change detection using Landsat TM data: when and how to correct atmospheric ef-
fects? Remote Sens. Environ. 75, 230 –244.
Stefanski, J., Chaskovskyy, O., Waske, B., 2014. Mapping and monitoring of land use changes
in post-soviet western Ukraine using re mote sensing data. Appl. Geogr. 55, 155 –164.
Woodcock, C.E., Macomber, S.A., Pax-Lenney, M., Cohen, W.B., 2001. Monitoring large areas
for forest change using Landsat: generaliza tion across space, time and Landsat sensors.
R e m o t eS e n s .E n v i r o n .7 8 ,1 9 4 –203.2174 B. Mihai et al. / Science of the Total Environment 599 –600 (2017) 2166 –2174