slider slider slider slider slider slider slider slider slider slider slider slider slider slider slider
forecom

Results

Task 3 - Reconstruction of forest cover based on historical maps

The Polish partner (Jagiellonian University, JU) was responsible for the reconstruction of the forest cover changes in the whole area of the Polish Carpathians (nearly 20 000 km2), based on three historical maps:

  • Second Austro-Hungarian military survey (~ 1850, scale of 1:28 800)
  • Polish military topographic maps (1930s, 1:100 000)
  • Polish topographic maps (1970s, 1: 25 000)

Task3 - Polish study area

Task3 - source data

 

Second Military Survey

The maps of the Second Military Survey (1:28 000) were based on the cadastral mapping (1:2 880), giving very detailed information about the forest cover. In the territory of the Polish Carpathians, three different sets of the map, varied by the date of publishing were used: 1822-1824, 1839-1840, 1861-1862. The maps were rectified before vectorization.

In most cases, the extraction of the forest, due to high map quality, was not problematic; however, difficulties in the reconstruction have been defined on the oldest edition of the maps, on the overlapping, adjacent sheets and on the borders between regions vectorised by different people.

Task3 - Polish study area

Three main kind of bias were identified:

  • overlapping forest patches,
  • gaps in the polygons not existing in the reality,
  • omitted, small parts of the bigger patch, continuing on adjacent polygon.

Additional uncertainty layer with comments confirmed the abovementioned problems.

 

Polish military maps

The Polish military maps (1:100 000) were prepared for the military purposes, so the quality is very high and representation of the forest cover is very detailed. Although the maps were available in the georeferenced form, we have decided to improve the results. Using additional control points improved the geometric correctness significantly.

There were no important variations in the quality of the map sheets for manual vectorization. The results of vectorising Polish military maps are consistent over the whole study area.

Polish military maps

Additionally the procedure of the automatic extraction was tested on the military maps as well (for results, see below).

 

Polish topographic maps

The maps created in the 1970s were the official Polish topographic maps, produced by the National Office of Geodesy and Cartography in the scale 1:25 000. Maps were available as a high quality georeferenced geotiff files.

Polish topographic maps

The Polish topographical map of 1970s was processed fully using the algorithms developed and tested for automatic extraction of forest information.

 

The Swiss partner (WSL) was responsible for the reconstruction of the forest cover changes in the Swiss Alps, based on four historical maps:

  • Dufour Map - circa 1850s
  • Siegfried map 1st edition - 1880s
  • Siegfried map last edition - 1940s
  • Landeskarte der Schweiz - 1970s

Vectorization of forest cover information was performed for the eastern part of the Swiss Alps (Cantons: Grisons, Glarus, Uri, Nidwalden and Obwalden).

Task 3 - Swiss study area

 

 

Dufour Map

The original images for the Dufour Map are dated from 1845‐1864 and were created in 1:25,000 scale for the Swiss plateau and 1:50,000 for the mountain area.

All the sheets were manually vectorized according to defined rules.

Dufour MapDufour Map

 

Siegfried map 1st and last edition

The Siegfried map is the second federal map series (1870‐1949) with map sheets at 1:25'000 scale for the Jura and the Central Plateau and 1:50'000 in the Alps.

Although the vectorization is relatively simple from a technical perspective, there were some semantic challenges to solve. They were mostly related to unclear forest representation / tree symbology on the maps.

Dufour mapSigfried map

 

Landeskarte der Schweiz

From the 1940s the modern 'Landeskarte der Schweiz' gradually replaced the Siegfried map.

Task 3 - Swiss study area

The Landeskarte der Schweiz of 1970s was processed fully using the algorithms developed and tested for automatic extraction of forest information.

 

Automatic forest extraction

Automatic forest extraction was based on the methods of morphological image processing (Iwanowski, Kozak 2012). To distinguish forest patches the following steps were employed:

  • linear filtering (local averaging of the color variations,
  • color segmentation (RGB‐>HSV color space transformation),
  • orphological filtering to shape the regions (set of subsequent operations of filtering the patches outlines.

The method of automatic forest extraction was tested on 20 map samples of Polish military maps (1930s) and on 20 map samples of Polish topographic maps (1970s).

Quality of selected sheets of the scanned Polish military maps (1930s) were not good enough to be used for automatic extraction of forest cover as a basic reconstruction method. The average number of misclassified pixels on the Polish maps from 1930s was about 14%.

The results of the automatic extraction procedure on the Polish topographic maps (1970s) obtained during the work had very high accuracy. During tests, around 98% of pixels have been correctly recognized.

First results of the automatic extraction procedure on the Swiss topographic maps (1970s) were promising, with >95% of pixels correctly recognized. Map processing will be completed in September 2013.

Automatic vectorization

Task 4 - Reconstruction of forest cover changes based on aerial and satellite imagery

We selected six communes in the Swiss Alps (Zweisimmen, Lenk, Giswil, Bürglen, Valendas and Lostallo) and six communes in the Polish Carpathians (Budzów, Milówka, Niedźwiedź, Rzepiennik Strzyżewski, Szczawnica, Zawoja) which were our test areas in a local scale. Each of these communes represented a gradient of environmental (elevation, slope) and socio-economic (type and main function of commune) conditions.

 

In Polish study areas three imagery datasets have been used:

  • old aerial photographs from 1977 (black-white, orthorectified before vectorization),
  • orthophotomaps from 1996-2003 (black and white or color RGB, at scale of 1:5 000 or 1:10 000, ground pixel size ranging from 0.25 m to 0.75 m),
  • orthophotomaps from 2009 (RGB, at scale 1:5 000, ground pixel size of 0.25 m).

orthophoto1977 orthophoto1997 orthophoto1997

Auxiliary data were administrative borders of the communes and digital terrain model.

 

Manual vectorization

The Polish partner (Jagiellonian University, JU) defined forest as trees older than 10 years including all single trees, patches of trees and large continuous forest area. Moreover, forest succession was digitized for periods between 70s - 90s and 90s – 10s and included trees younger than 10 years, bushes and early state of vegetation.

On the base of the above definition forest cover was manually vectorized for Polish communes of: Szczawnica, Milówka, Zawoja and Budzów. Manual vectorization of forest succession was performed for Szczawnica and Budzów for periods of 2009-1990 and 1990-1970. The final forest cover and forest succession maps were generated after rasterization and clean-up.

Budzow vectorization

The Swiss partner (WSL) digitized single trees and patches and continuous forest. All the six communes were vectorized manually. The final forest mask was created using Alpha-Shape Algorithm, that groups individual trees and tree-patches into forest, based on distance and other criteria.

Forest mask

 

Automatic forest and succession mask generating

For the remaining Polish communes the authomatic procedure was introduced. It utilized object based image analysis (OBIA) and multiresolution segmentation.

Forest mask Forest mask Forest mask

We distinguished the following classes: forest, middle vegetation (succession), low vegetation (succession), built-up areas and others. The most problematic was classification of shadows.

 

Database

All the manually vectorized and semi-automatically derived forest and succession masks were collected in geodatabase.

Task 5 - Estimates of the current abandonment rate and forest succession

Workflow

We developed a conceptual framework for mapping forest succession patterns using vegetation structure information derived from Airborne Laser Scanning (ALS) data supported by national topographic vector data (BDOT10k). Secondary forest succession on abandoned land (SFSAL)was defined as any area with vegetation typical of secondary forest succession, particularly tall herb communities, shrubs, bushes, and single young trees, that occupies land with agricultural use.

The SFSAL mapping conceptThe SFSAL mapping concept

 

This work was performed in the Szczawnica commune in the Polish Carpathians. The methodology of ALS data processing described in detail in the paper of Kolecka et al. (2015).

Workflow of the SFSAL mappingWorkflow of the SFSAL mapping

 

Sampling strategy

To map secondary forest succession throughout the entire Polish Carpathians, we used the spatial sampling strategy described in detail in Kolecka et al. (2016). We focused on agricultural land, and therefore, we excluded some areas beyond our interest.

Workflow of the sampling strategy developmentWorkflow of the sampling strategy development

 

As a result, we obtained a sample of 230 tiles (of 2 km by 2 km), one within each of the Carpathian communes.

1 – national borders, 2 – boundaries of administrative units, 3 – physiographical boundary of the Carpathians,
4 – sampled tiles, 5 – excluded tiles, 6 – tiles available for sampling, 7 – tiles not covered with ALS data

 

SFSAL mapping

All 230 tiles were analysed according to the developed strategy, returning two indices: (1) the overall SFSAL rate – the ratio of the identified secondary forest succession area to the total available agricultural land area per tile and (2) the high vegetation subset of SFSAL (hvSFSAL) – the ratio of the identified secondary forest succession area with high vegetation to the total available agricultural land area per tile.

The average SFSAL rate was 13.9%, ranging from 1.7% to 38.4%. SFSAL was observed rather on grasslands (78%) and much less on arable land (22%). Medium-high vegetation constituted majority (95.7%), and only 4.3% of the secondary forest succession included high vegetation. Higher SFSAL rates were observed in the western part of the study area and in the communes surrounding larger cities.

 

SFSAL drivers

We investigate the determinants of the current SFSAL on the base of variables that described the following attributes: (1) tile location, (2) environmental conditions and land cover in a tile, and (3) environmental and socioeconomic characteristics of the commune in which the given tile is located. Agricultural land located on steep slopes and at lower altitudes in the fragmented forest-agricultural landscape is most likely to be abandoned. Also proximity of current provincial capital cities and administrative centres of the communes were important factors of abandonment. Therefore, we can indicate that two types of factors influence SFSAL: topography and employment outside of agriculture.

The detailed description of the results will be available soon.

Task 6 - Estimation of climate change and land use contribution to past forest cover change

Estimates of the land use drivers and climate change contributions to forest cover change in the Swiss Alps and Polish Carpathians was the major research task resulting in creating additional climate datasets for the Carpathians and the Alps (Wypych et al. submitted; see also above) and socio-economic data for the Carpathians. Identified warming in both regions since mid-19th century that will also continue in future, however, was found not to impact past forest cover changes in both regions. Forest cover gain and loss modelling for the entire period covered by forest masks elaborated in the project (1850-2010) was carried out using ensemble approach known from other domains, and for the first time operationally applied in land cover change analysis (Bolliger et al. in revision). In the Swiss Alps a major determinant of forest cover gains and losses was distance from forest area in the previous period (Bolliger et al. in revision), similarly as in the Carpathians for two periods (1860s-1930s; 1970s-2010s). Forest gains in the Carpathians between 1930s-1970s were driven mostly by demographic changes and post-war depopulation in the eastern part of the region (Ostapowicz et al. 2015). In general, forest cover changes in the Carpathians followed a theoretical model of forest transition based on attributing forest cover change trends to increasing interactions among regions and relative differences of their agricultural suitability (Kozak and Szwagrzyk 2016).

Forest gain drivers for the Polish Carpathians, 1930s-1970s and 1970s-2010s. Note lower importance of ‘distance to forest’ variable in the first period (from Ostapowicz et al. 2015).

 

Task 7 - Refine existing models for predicting land use change for Swiss Alps and develop improved forest cover change estimates

Task 8 - Adjustment of SA model of land use change scenarios to Polish Carpathians; future forest cover change estimates

As the Swiss partner had a substantial expertise in land use and land cover predictions for Switzerland, models existing for Switzerland were refined, and the applied to simulate forest cover changes in both regions based on important variables identified in modelling past forest changes (both regions) and patterns of contemporary secondary forest succession (Polish Carpathians). Dyna-CLUE land use allocation algorithm was applied for three scenarios (T, L, S, see above). Forest cover predictions showed forest increase in both regions depending on the scenario (Swiss Alps: 2.4-36%, Polish Carpathians: 5-22%) (Price et al. in revision). For the Polish Carpathians, these predictions are consistent with estimates of the present-day secondary forest succession described above.

 

Task 9 - Quantify implications of forest cover change on carbon pools and biodiversity

Models elaborated in the project documented a continuous increase of total aboveground carbon stocks over time in both study areas and pointed at forest locations (hotspots) that will play a key role in the terrestrial carbon sequestration in next decades (Pazur et al. 2016, also in preparation).

Above-ground biomass (ABG) for a sample location and trends over time (1850-2060) for the entire studied section of the Swiss Alps (L, S, T are different scenarios of land use and forest cover change till 2060).

Project supported by a grant from Switzerland through the Swiss Contribution to the enlarged European Union.
Value of co-financing: 681 420, 22 CHF.

Swiss Contribution
Bundess Logo