New version of Pan-European Forest Map available
Earth Observation data are regarded as a cost-efficient means for locating different types of vegetation cover at the ground level. Two different earth-observation products (Kempeneers et al. 2011 and Schuck et al. 2002 / Päivinen et al. 2001) have been combined with statistical forest inventory data to produce a new pan-European forest map at 1km resolution that corresponds to the official statistics at national and/or regional level. The map shows the percentage share of forest from land area. Besides geographical Europe, it also covers Turkey. The map is now available for download from EFI website.
The utilized forest/non-forest map of the EC Joint Research Centre (Kempeneers et al. 2011) is based on IRS and SPOT data and covers the EU27 countries + Norway, Switzerland, Turkey and the Balkan area. It was aggregated from 25m resolution to 1km by summing up the forest area for each 1x1km pixel. NOAA-AVHRR forest share estimates at 1km resolution (Schuck et al. 2002 / Päivinen et al. 2001) have been used to extend the map up to the Ural mountains, namely covering Belarus, Ukraine, Moldova, European part of the Russian Federation. The combined forest cover map shows the %share of forest in each 1x1km-pixel, but differs from official statistics which was tackled in the next step.
Statistical data on forest area and its distribution for different forest classes are traditionally available through national forest inventory statistics and other national and international forest statistical sources. Various sources of data were used to compile the map according to the availability of data: 1) Recent national forest inventory (NFI) statistics on forest area were used at the sub-national level for 19 European countries, including the European part of the Russian Federation. 2) In addition, country-level statistics on forest area published by Forest Europe 2011 were applied for all countries covered by the map.
The satellite-based forest cover data was first calibrated to sum up to the forest area statistics within a given administrative region. The calibration iteratively adjusts the pixel values in a region to sum up with the statistics while keeping the forest share per pixel below the maximum possible limit of 100%. This is achieved by first determining the ratio between the forest area in the map and in the statistics for each region. All pixel values in the respective regions are then multiplied by this ratio. Finally all pixels exceeding a forest share of 100% after multiplication, are set back to a value of 100. This process is repeated until the difference between forest area in map and statistics falls below a specified threshold (0.05%).
A European timberline mask compiled by Schuck et al. (2002) was implemented to exclude areas considered above the timberline from the calibration process. Such areas were automatically assigned a 0 value. In a first calibration run, the 19 countries with available NFI statistics (incl. Russian Federation) were adjusted to the values at the level of sub-national administrative regions. In a second run, all European countries (excl. Russian Federation) were adjusted to the internationally harmonized statistics by Forest Europe 2011 at national level, to allow for comparability between the countries. This second calibration slightly changed the results of the previous regional-level calibration if there were differences between NFI statistics and Forest Europe 2011 data. Still, the overall distribution of forest area between the different regions in a country (as stated in NFI statistics) remained the same also after the country-level calibration.
The calibration achieves an overall fit of the map with the statistics at regional and country level, i.e. when summing up all forest area in a country the result corresponds to the respective statistics. However, at the local level the map might differ from the real situation due to uncertainties in the applied remote sensing products and the changes introduced by the calibration procedure. Uncertainties are higher for Belarus, Ukraine, Moldova and Russia, since the input map used for these areas is of much lower resolution (1000m) than the one used for the rest of Europe (25m). For more background information on the quality of the input maps please refer to Kempeneer et al. (2011) and Schuck et al. (2002).
More information and opportunity to download the map is available here.