Elsevier

Remote Sensing of Environment

Volume 159, 15 March 2015, Pages 28-43
Remote Sensing of Environment

Eastern Europe's forest cover dynamics from 1985 to 2012 quantified from the full Landsat archive

https://doi.org/10.1016/j.rse.2014.11.027Get rights and content

Highlights

  • Annual quantification of forest dynamics for 27 years performed with high accuracy

  • Results show spatial and temporal variation of forest cover change.

  • Wall-to-wall forest dynamics maps available at http://glad.geog.umd.edu/europe/

  • Overall, forest cover increased by 4.7%, but decreased in some regions.

  • Forest dynamics mirrored the collapse of the USSR and the recent economic crisis.

Abstract

In the former “Eastern Bloc” countries, there have been dramatic changes in forest disturbance and forest recovery rates since the collapse of the Soviet Union, due to the transition to open-market economies, and the recent economic crisis. Unfortunately though, Eastern European countries collected their forest statistics inconsistently, and their boundaries have changed, making it difficult to analyze forest dynamics over time. Our goal here was to consistently quantify forest cover change across Eastern Europe since the 1980s based on the Landsat image archive. We developed an algorithm to simultaneously process data from different Landsat platforms and sensors (TM and ETM +) to map annual forest cover loss and decadal forest cover gain. We processed 59,539 Landsat images for 527 footprints across Eastern Europe and European Russia. Our results were highly accurate, with gross forest loss producer's and user's accuracy of > 88% and > 89%, respectively, and gross forest gain producer's and user's accuracy of > 75% and > 91%, based on a sample of probability-based validation points. We found substantial changes in the forest cover of Eastern Europe. Net forest cover increased from 1985 to 2012 by 4.7% across the region, but decreased in Estonia and Latvia. Average annual gross forest cover loss was 0.41% of total forest cover area, with a statistically significant increase from 1985 to 2012. Timber harvesting was the main cause of forest loss, accompanied by some insect defoliation and forest conversion, while only 7.4% of the total forest cover loss was due to large-scale wildfires and windstorms. Overall, the countries of Eastern Europe experienced constant levels or declines in forest loss after the collapse of socialism in the late 1980s, but a pronounced increase in loss in the early 2000s. By the late 2000s, however, the global economic crisis coincided with reduced timber harvesting in most countries, except Poland, Czech Republic, Slovakia, and the Baltic states. Most forest disturbance did not result in a permanent forest loss during our study period. Indeed, forest generally recovered fast and only 12% of the areas of forest loss prior to 1995 had not yet recovered by 2012. Our results allow national and sub-national level analysis and are available on-line (http://glad.geog.umd.edu/europe/) to serve as a baseline for further analyses of forest dynamics and its drivers.

Introduction

European forests co-evolved with humans since the beginning of the Holocene, and their current distribution, structure, and dynamics represent a long history of clearing, alteration, and management (Fuchs et al., 2013, Johann, 2004, Kalyakin et al., 2004, Kaplan et al., 2009, Kaplan et al., 2012). Shaped by human activities, forests were a main sector of the economy providing food (e.g., hunting, livestock grazing, and plant products), timber products (e.g., lumber for construction and naval fleets, and pulp for paper), fuel (e.g., firewood, and charcoal), and other important resources (e.g., potash, and tar). The importance of forest resources, which can be quickly exhausted by unrestricted use, provided the impetus for forest mapping, inventory, and management. Forest mapping techniques were developed concomitantly with land tenure systems, and the first forest maps were already produced in the 14th century (Morse, 2007). In North and Central Europe, exhaustion of timber resources for naval ship building, lumber, and charcoal used for iron production, were the main factors why forest inventories were established in the 19th century (Eliasson, 2002, Tomppo et al., 2010). Forest inventories and management expanded into Eastern Europe and European Russia in the 19th and 20th centuries. In the 20th century, national forest inventory and monitoring incorporated various instrumental measurement methods, statistical sampling, and, later, remote sensing technology. As a result, the forests of Europe are among the most well-monitored ecosystems of the world.

Despite the wealth of forest inventory data, this information is unfortunately not readily available, nor well suited for region-wide analyses. One problem is that forest definitions and inventory methods vary among countries and have changed over time, making cross-national and multi-temporal comparisons complicated or even impossible (Seebach, Strobl, San Miguel-Ayanz, Gallego, & Bastrup-Birk, 2011). The lack of accessibility to national forest data poses another complication because many countries in Eastern Europe treat forest maps and precise forest statistics as either commercially sensitive or even a matter of national security, and thus prohibit its distribution beyond governmental agencies. Even where forest inventory information is in principle available, it is often hard to obtain from national (or sometimes regional) agencies where it is stored in a variety of formats.

Remote sensing (RS) data can provide an alternative data source to quantify forest cover and change independent of official governmental data sources. Information derived from satellite imagery, however, is not equivalent to inventory data collected by forest managers. Optical remote sensing data is suitable for mapping land-cover (tree canopy cover, dominant tree species composition) while national forest inventory data focuses on land-use (e.g., forest land). This means that while tree canopy cover change can be readily observed with remote sensing data, it is not directly comparable to harvested timber volumes reported by the national forest statistics. As a result, remote sensing data are rarely used as a primary source for national forest inventories, and statistical reports due to differences between land-use and land-cover forest definitions (Tomppo et al., 2010). The recent expansion in remote sensing-based forest monitoring products, however, highlights that these data could be valuable for many applications. First, remote sensing-based products can cover vast areas consistently, avoiding discontinuities due to administrative and national boundaries (Hansen et al., 2013, Kuemmerle et al., 2006, Pekkarinen et al., 2009, Potapov et al., 2011). Second, long-term records of satellite observations now available in image archives allow forest change quantification over several decades (Baumann et al., 2012, Griffiths et al., 2013, Margono et al., 2012, Potapov et al., 2012).

Spatial and temporal consistency is an inherent property of remote sensing-based forest cover and change products, alleviating the need for harmonization procedures commonly applied to regional and national forestry inventory data (Seebach et al., 2011, Tomppo et al., 2010). Simple biophysical criteria such as forest cover (defined using certain tree canopy cover thresholds without attribution to specific land cover categories and land use) make remote sensing-based products more suitable to assess carbon change than national forest inventories that are based on land use definitions (DeFries et al., 2002, Harris et al., 2012, Tyukavina et al., 2013). At the same time, remote sensing-based forest cover change analysis requires less effort and time than ground surveys, and can be performed in areas of limited ground access. This is why remote sensing-based products are widely used for multi-national forest assessments and change estimations, and their results serve as a baseline for carbon modeling and socio-economic analyses as well as for studies of landscape dynamics and biodiversity patterns (Burgess et al., 2012, Griffiths et al., 2012, Hansen et al., 2013, Harris et al., 2012, Kuemmerle et al., 2007, Tyukavina et al., 2013, Wendland et al., 2011).

While there have been prior assessments of forests in Europe with remote sensing (e.g., Gallaun et al., 2010, Pekkarinen et al., 2009, Schuck et al., 2003), none of them analyzed the full Landsat record for all of Eastern Europe. The lack of a comprehensive analysis of forest dynamics in Eastern Europe is unfortunate, because the region has witnessed numerous changes in forest cover since the collapse of socialism. Several remote sensing-based forest cover change projects have documented some of these changes (Baumann et al., 2012, European Environment Agency, 2007, Griffiths et al., 2013, Kuemmerle et al., 2009, Pekkarinen et al., 2009, Potapov et al., 2011). However, prior projects have several limitations precluding their use for analyses of forests dynamics across Eastern Europe: (i) none of these products cover the entire region; (ii) the methodologies used in different studies are not compatible; (iii) validation results are inconsistent and hard to compare; and (iv) with few exceptions (Potapov et al., 2011), products are not readily available.

Our research goal here was to fill these gaps and to produce a forest cover change product for all of Eastern Europe for nearly three decades using a consistent set of remote sensing data, methodology, and definitions. Our first objective was to develop a methodology that would allow multi-sensor data integration and seamless forest cover and change mapping. The methodology that we developed was then implemented to map forest cover change in Eastern Europe from 1985 to 2012. Our second objective was to provide consistent and rigorous validation of the reported forest cover change. Lastly, our third objective was the unrestricted sharing of the resulting product for further analyses (http://glad.geog.umd.edu/europe/). While we provide here an overview of the results and discuss potential forest change factors, the in-depth analysis of social and economic drivers of the observed forest changes was outside the scope of this project.

Section snippets

Study area

Our study area included the Eastern European countries that formed the “Eastern Bloc” until the end of the 1980s, except the former German Democratic Republic (aka East Germany, now part of Germany), and Albania (which disassociated from the Eastern Bloc in 1961). The study area included several former USSR republics (Estonia, Latvia, Lithuania, Belarus, and Ukraine) and the European part of Russia (Fig. 3A). The 2012 national and administrative boundaries of the countries were obtained from

Results

The forest cover for year 1985 (sum of dynamics types B, D, E, F, see Fig. 2 for explanation) was 216 million ha (Table 3). By 2012, forest area increased by 10 million ha (4.7%) and reached 226 million ha (sum of types B, C, E). In total, there were 24 million ha of forest loss (including areas that experienced forest gain after loss), which was substantially lower than the 34 million ha of forest gain. Forest loss of 1985 forests (sum of types D, E, F) represented 11% of the 1985 forest area, or 0.41%

Data availability

The USGS Landsat program is the oldest provider of operational medium-resolution satellite data. The data record of Landsat TM/ETM +/OLI instruments spans over last 30 years, and the free data access and redistribution policy made it the best data source for the analysis of long time-series of land cover change. The main problem of the existing archive, however, is an inconsistent data acquisition record. A dramatic decline in acquisitions occurred after the commercialization of the Landsat 5

Conclusion

Our analysis proved the feasibility of a Landsat-based, long-term (27 years) forest cover change assessment for a large region. Despite data limitations, especially the incomplete Landsat archive for the 1990s, our proposed approach for the mapping of gross forest cover loss and gain events was successful and enabled the estimation of net forest cover change. Our validation, which was performed using probability-based sampling and specifically focused on classification errors within edge pixels,

Acknowledgements

The project was supported by NASA Land-Cover/Land-Use Change Program research grants NNX13AC66G and NNX12AG74G. We greatly appreciate help in fieldwork from our colleagues E. Boren, M. Doktorova, M. Dubinin, A. Manisha, and A. Purekhovsky; and valuable comments by Dr. L. Laestadius and two anonymous reviewers.

References (63)

  • J.O. Kaplan et al.

    The prehistoric and preindustrial deforestation of Europe

    Quaternary Science Reviews

    (2009)
  • T. Kuemmerle et al.

    Forest cover change and illegal logging in the Ukrainian Carpathians in the transition period from 1988 to 2007

    Remote Sensing of Environment

    (2009)
  • T. Kuemmerle et al.

    Cross-border comparison of land cover and landscape pattern in Eastern Europe using a hybrid classification technique

    Remote Sensing of Environment

    (2006)
  • P. Olofsson et al.

    Good practices for estimating area and assessing accuracy of land change

    Remote Sensing of Environment

    (2014)
  • A. Pekkarinen et al.

    Pan-European forest/non-forest mapping with Landsat ETM + and CORINE Land Cover 2000 data

    ISPRS Journal of Photogrammetry and Remote Sensing

    (2009)
  • D. Pflugmacher et al.

    Using Landsat-derived disturbance history (1972–2010) to predict current forest structure

    Remote Sensing of Environment

    (2012)
  • D. Pflugmacher et al.

    Using Landsat-derived disturbance and recovery history and lidar to map forest biomass dynamics

    Remote Sensing of Environment

    (2014)
  • P. Potapov et al.

    Regional-scale boreal forest cover and change mapping using Landsat data composites for European Russia

    Remote Sensing of Environment

    (2011)
  • P.V. Potapov et al.

    Quantifying forest cover loss in Democratic Republic of the Congo, 2000–2010, with Landsat ETM + data

    Remote Sensing of Environment

    (2012)
  • A.V. Prishchepov et al.

    The effect of Landsat ETM/ETM + image acquisition dates on the detection of agricultural land abandonment in Eastern Europe

    Remote Sensing of Environment

    (2012)
  • A. Schuck et al.

    Compilation of a European forest map from Portugal to the Ural mountains based on earth observation data and forest statistics

    Forest Policy and Economics

    (2003)
  • S.V. Stehman

    Estimating area from an accuracy assessment error matrix

    Remote Sensing of Environment

    (2013)
  • S.V. Stehman et al.

    Design and analysis of thematic map accuracy assessment: Fundamental principles

    Remote Sensing of Environment

    (1998)
  • E.F. Vermote et al.

    Atmospheric correction of MODIS data in the visible to middle infrared: first results

    Remote Sensing of Environment

    (2002)
  • K.J. Wendland et al.

    Regional- and district-level drivers of timber harvesting in European Russia after the collapse of the Soviet Union

    Global Environmental Change

    (2011)
  • L. Breiman

    Bagging predictors

    Machine Learning

    (1996)
  • L. Breiman et al.

    Classification and regression trees

    (1984)
  • V. Brukas et al.

    Policy drivers behind forest utilisation in Lithuania in 1986–2007

    Baltic Forestry

    (2009)
  • R. Burgess et al.

    The political economy of deforestation in the tropics

    Quarterly Journal of Economics

    (2012)
  • M. Carroll et al.

    Vegetative cover conversion and vegetation continuous fields

  • J.J. Danielson et al.

    Global multi-resolution terrain elevation data 2010 (GMTED2010): U.S. Geological Survey

  • Cited by (244)

    • Evaluation of Landsat image compositing algorithms

      2023, Remote Sensing of Environment
    View all citing articles on Scopus
    View full text