Ústav pro hospodářskou úpravu lesů Brandýs nad Labem

The photogrammetric survey within the Czech NFI

The combination of field survey and remote sensing (RS) data has become a standard methodological approach of many NFIs. Photogrammetric interpretation of digital aerial images (AI) in a densified sampling grid reduces field survey costs and increases the accuracy of estimates of target NFI parameters. Results of photo-interpretation are further used as training data for semi-automated generation of digital maps (based on aerial orthophoto maps or satellite imagery).

Why photogrammetry

In the second cycle of the National Forest Inventory (NFI2), photogrammetric interpretation (further used as photo-interpretation) of sample points was carried out in a 0.5 × 0.5 km square grid and independently also in a 2 × 2 km grid (two sample points within a square). As a result, 354 869 sample points were interpreted in the whole Czech Republic, many of which were evaluated repeatedly for the sake of checks (interpretation of the same points by more operators).

The main benefits of photo-interpretation can be summarised as follows:

  • Reduction in fieldwork by roughly 60 % as a result of excluding sample points clearly outside the target land categories (e.g. Forest, OWL) from the field assessment.
  • Refining information provided by the NFI – combining a relatively small amount of correct but expensive field assessment information with a large amount of less accurate, but cheaper photogrammetric information leads to a more accurate estimation of many target NFI parameters. This is important for the provision of NFI results at the level of smaller sub-regions and for periods shorter than the whole inventory cycle (currently five years).
  • A cost-effective estimation of landscape characteristics, i.e. without the need for a dedicated field survey. If landscape characteristics were required without the possibility of using photo-interpretation, the NFI field survey would have to be extended beyond the target land categories. For this purpose, the photo-interpretation of a 500 m long transect with random orientation has been used within the Czech NFI.
  • Training data for image classification – a large amount of photo-interpreted sample points evenly covering the whole territory of the Czech Republic is used for automated production of wall-to-wall digital map layers of the whole country (e.g. forest land category maps, growth stages and tree species composition of forests, woody vegetation outside forests, etc.).

Surveys at sample points are carried out on the basis of digital images, which enable a fully digital processing line with all its advantages. These digital images are captured by a camera simultaneously in four channels, three of which are sensitive in the visible part of the electromagnetic spectrum (i.e. in true RGB colours) and one in the near-infrared band. The presence of the near-infrared channel significantly improves the interpretation quality regarding the vegetation assessment (e.g. distinction of coniferous and broadleaved species, dead trees, grasslands from cropland, etc.).

A comprehensive overview of the use of photogrammetry in the Czech NFI can be found in [3].

Technology and personnel

The photogrammetric interpretation takes place simultaneously on six workstations equipped with a specific application (designed by ÚHÚL, a GUI implemented by Topol Software, s. r. o.). This application provides a GUI (Graphical User Interface) that guides operators through the steps of the photo-interpretation procedure. ÚHÚL specialists designed and implemented a specific functionality within the PostgreSQL database, which ensures the performance of the whole photo-interpretation process including quality assurance. The logic implemented in the PostgreSQL database controls and guides the work plan of a particular NFI cycle including preparation of annual field survey plans (based on the NFI design and results of the photo-interpretation) and regular field data uploads and its quality control. In short, the functionality implemented in PostgreSQL interlinks all NFI data collection and quality control activities carried out within the Czech NFI.

Interpretation of sample points

The first step is to interpret a square 4 × 4 matrix of points evenly distributed in the interpretation 51 × 51 m square which is centred at the position of a sample point (determined by the NFI sampling grid) [3, p. 36]. The altitude above sea level is measured and the type of surface (a conifer, deciduous tree, shrub, terrain or another surface) is assessed on each of the sixteen points. Within the NFI2, approximately 5.7 million grid points were evaluated in this way throughout the whole Czech Republic.

In the next step, a particular sample point is assigned to a land category according to the FAO (Food and Agriculture Organization of the United Nations) FRA (Global Forest Resource Assessment) guidelines. There are two sets of criteria used to classify sample points into a particular land category:

  • Physiognomic criteria – occurrence (or absence) and properties of trees, i.e. stand height, canopy cover, a total area, and width.
  • Current and actual land use – the assessment consists in determining whether agricultural or urban use prevails at the position of the particular sample point at the time of the NFI survey. Should none of the given land uses prevail, the land can be placed in the forest land or OWL category, depending on the outcomes of the assessment of physiognomic criteria.

Definitions of the FAO FRA land categories used by the Czech NFI were adjusted to European conditions by the European National Forest Inventory Network (ENFIN). Five categories are distinguished: Forest, OWL (Other Wooded Land), OLwTC (Other Land with Tree Cover), Other Land (OL) and Water Bodies (WB). Each sample point is assigned to one of these categories considering the above two types of criteria, irrespective of the status of land according to the cadastre, forest management plans or any other pre-existing registers, maps, etc.

The photogrammetric classification of sample points to the FAO FRA classes follows the same decision tree as it is used during the Czech NFI field survey. The only differences are that operators can assign a sample point to the category of ‘unclassifiable’ (due to image imperfections, clouds, etc.) and they do not have to strictly decide whether urban or agricultural land use prevails or not (see the schema nodes no. 11). Land use is not always apparent on the imagery so the result can be a mixture of the two respective classes (depending on the particular part of the decision tree which has been reached).

Sample points in the forest land category are further analysed [3, p. 42]. All recorded parameters are related to the specific stand segment in which the sample point is located. The following forest attributes are assessed:

  • type of forest land – a productive forest area (including clearcuts), unstocked forest land (mainly earth roads and roadside landings easily convertible to a productive forest area), other land (mainly impervious, reinforced roads and landings)
  • damage to a stand – by wind or snow, fire
  • growth stage – an unstocked forest area, young plantation, thicket (up to 2.5 m), small pole stage (up to 8 m), pole-stage stand (up to 20 m), large-diameter stand (above 20 m), mixed stages (a selection forest)
  • species composition – mostly conifers, mostly broadleaves, a mixed stand, stand up to 8 m in height (species composition cannot be reliably distinguished in the case of the youngest stands)
  • canopy density – normal, broken, dense
  • type of species mixture – an unmixed stand, individual mixing, different species occurring in rows or lines, a species mixture exhibiting clumps or groups

Apart from the classification according to the FAO FRA land categories, the sample points are as well classified following the LULUCF (Land Use, Land-Use Change and Forestry) guidelines [4, chap. 2.2]. The following land categories are distinguished: Forest (the same definition as in the case of FAO FRA), Cropland, Grassland, Wetland, Settlements and Other land.

Interpretation of landscape transects

In addition to the sample points, photogrammetric transects – 500 m long lines with random orientation – are also interpreted. Each photogrammetric transect has got the same orientation as a field survey transect, which is, however, a bit shorter (300 m). The centre of each photogrammetric transect coincides with the sample point, belonging to the 1 × 1 km sub-grid. In terms of a sampling methodology, the photogrammetric transects are based on:

  • Line intersect sampling (LIS) – this method is used to estimate total lengths and densities of line objects; total lengths, an area and canopy cover of strip objects (tree belts); a total area, areal representation and an average size of areal objects [1], [2, pp. 279–325], [3, p. 73].
  • A strip area of a fixed shape and size – geometrically it is a strip 100 m wide and 500 m long, whose axis runs through the transect line. It is used for the survey and subsequent estimation of the number and areal density of selected point objects in a landscape.

The main objective of this survey is to record scattered vegetation in a landscape that does not fall under any of the FAO FRA land categories (a minimum required width of 20 m and a stand area of 0.5 ha). Typically, these are alleys, bank stands (up to 20 m in width), solitary trees and small groups of trees (an area under 0.04 ha), small area stands of woody species (an area under 0.5 ha) and other specific cases.

The start of the photogrammetric interpretation of transects in 2014 was a bit delayed compared to the launch of the NFI2 photogrammetric interpretation of sample points in 2010. Because of the increased workload of operators and because of the anticipated slower trends in the respective populations, it was decided to extend the transect survey to ten years (unlike the five years cycle of the field survey).


[1] Barabesi, L., Marcheselli, M. (2008): Improved strategies for coverage estimation by using replicated line-intercept sampling. Environ Ecol Stat 15:215-239, DOI 10.1007/s10651-007-0048-6.

[2] Gregoire, T. G., & Valentine, H. T. (2008): Sampling strategies for natural resources and the environment. Chapman and Hall/CRC (chapter 9), 474 pp.

[3] Hájek, F., Adolt, R., Tomančák, O., Kantorová, M., Kučera, M. & Čech, Z. (2016): Metodika a pracovní postupy fotogrammetrického šetření NIL2. Brandýs nad Labem: Ústav pro hospodářskou úpravu lesů Brandýs nad Labem (ÚHÚL), 161 pp., ISBN 978-80-905995-8-1.

[4] Penman, J. et al., (2003): Good Practice Guidance for Land Use, Land-Use Change and Forestry. Institute for Global Environmental Strategies (IGES) for Intergovernmental Panel on Climate Change (IPPC), 590 pp., ISBN 4-88788-003-0.