Keywords: satellite images, old-growth forests
Abstract. The largest areas of old-growth, or virgin forests from the temperate region of the European Union can be found in the Carpathian Mountains of Romania (Jan Knorn et al., 2012). A lot of these forests are endangered by logging activities, which are intensifying. Cataloging a forest as virgin or quasi-virgin can only be done in the field studies, however an estimation of the location and extent of these forests can be done using GIS and remote sensing techniques. This kind of estimation could be very useful when planning the future field studies. Therefore, this paper proposes a methodology for the identification of the potential old-growth forests in the Făgăras Mountains.
This paper proposes a methodology for the automated identification of virgin and quasi-virgin forests in Făgăras Mountains.
H. Leibundgut defines the virgin forests as forests that were formed under the influence of the natural factors and in which biological processes are unfolding without any direct or indirect anthropogenic influence. In Romania, virgin and quasi-virgin forests are defined by the ministerial order no. 3397 from 10/09/2012. Within this order, a criteria list for the identification of virgin and quasi-virgin forests is presented, some of which are to be used as reference for the potential virgin forests identification in the Făgăraș Mountains.
The Făgăraș Mountains are included in one of the 237 Special Areas of Conservation (SACs) of the Natura 2000 network in Romania. This site will represent our study area. This is the second largest site in Romania after the Danube Delta, having an area of1986,1km2.
According to Corine Land Cover 2012, the forests are covering 1420km2, representing about 71% of the total area of the site. Coniferous and mixed forests have a similar spatial extension, representing 39%, respectively 37% of the total forest cover area. Broad leafed forests are not so developed, mainly because of the high altitudes, representing about 24%. Because of the terrain inaccessibility, the probability of finding virgin and quasi-virgin forests within this area is pretty high. This paper will propose a methodology for identifying the areas where virgin and quasi-virgin forests could be found.
The European Space Agency (ESA) is consideringwithin the Copernicus programme, between 2014 and 2020, seven Sentinel missions. Sentinel 2 has a multispectral sensor (MSI), capable of recording the electromagnetic radiation from the visible, near infrared and shortwave infrared spectrum. The spatial resolution of the images differs, depending on the spectral channel. The multispectral instrument aboard Sentinel 2 satellite is dividing the visible and infrared spectrum in 13 bands, four of those having a 10m resolution, six of 20m and three of 60m.
The spectral signature of a forest is dependent, mostly, of the interaction between the electromagnetic radiation and the crown cover. In this case, the reflected electromagnetic radiation will be analyzed taking into account the technical characteristics of the sensors present on board of the Sentinel 2 satellite. Thus, the processing level and the quality of the images will influence the perceived spectral response of the forests, this being more or less accurate depending on this factors. For example, after the radiometric correction of the images, which is reducing the influence of the atmosphere and minimalizing the shadow effect which appears due to different illumination conditions, the spectral signature of the forest will be much more realistic. In general, the spectral signature of the crown cover is dependent of a series of factors, like species composition, vegetation period, health or age of the trees. In the case of a satellite image, the spectral signature of a forest can be influenced by shadows caused by different tree heights, or crown cover inconsistencies. This factors are influencing the appearance of the forests, older forests having a more textured look, younger forests being more homogenous. This differences will be quantified by combining the different spectral bands of the Sentinel 2 satellite and also by using vegetation indexes, like the Leaf Area Index (LAI) or the Normalized Difference Vegetation Index (NDVI).
Materials and methods
In the identification process of the old-growth forests in the Făgăraș Mountains, Sentinel 2 images will be analyzed, using three forests that were charted in the field as reference. In August 2016, following a field study conducted by Greenpeace Romania, three forests from the Făgăraș Mountains were charted, respectively Știubeaua, ObârșiaCumpănitei and Mușeteica. Subsequent to the field studies, using the criteria of the ministerial order 3397/2012 as reference, those three forests were proposed for including in the “National catalogue of virgin and quasi-virgin forests of Romania”. In the Mușeteica forest, because of the higher altitudes, coniferous species are dominant, while in the other two forests broad leafed species are prevailing.
This method is based on the different spectral response of the Sentinel 2 bands. The spectral response is represented by different pixel values which are at the base of the raster data corresponding to each spectral band. Thus, in the case of each band, the pixels corresponding to the virgin forests charted in the field will be extracted and the most common values will be identified. Therefore, the first step is represented by the extraction of the most common pixel values corresponding to the reference forests polygons. More precisely, the pixels with a frequency of the values above the average will be extracted. The next step consists in the identification and extraction of the pixels with similar values across the entire study area. This operation will be ran on all the used bands, including the LAI and NDVI rasters, resulting a raster dataset whose pixel values are representing the most common values (above average) of the pixels corresponding to the virgin forests charted in the field. Each resulting raster will be reclassified, assigning the value 1 to each pixel. The reclassified raster dataset will be mosaicked using the “SUM” function.
As demonstrated in some studies (R. Pokorný, S. Stojnič, 2012), the Leaf Area Index (LAI) is in a direct correlation with the forest stand age.
Because of the different spectral response of the coniferous and broad leafed species, a separate analysis was necessary. The purpose was to obtain homogenous forest samples in terms of species composition. Finally, the results of the two analysis will be combined.
For the coniferous species analysis, April images were utilized. In April broad-leafed species are lacking the foliar apparatus and the alpine meadows are covered in snow. For this analysis, bands 4, 8a, 11, 12 and the LAI and NDVI vegetation indexes were used. Band 4 is an indicator of photosynthesis activity, a process which is mostly using the red portion of the electromagnetic spectrum. Band 8a was preferred because of the small bandwidth, coniferous species having a high reflectance values in the middle of this band. This is ideal for the differentiation of the coniferous forests from other vegetation types. Shortwave infrared bands (11 and 12) are also very useful.
For the analysis of the mixed and broad-leafed species, images dated from July 2016 were used. As in the case of the coniferous species, a number of six rasters were used, respectively bands 3, 8, 11, 12 and also LAI and NDVI vegetation indexes. In this case, a major problem is represented by the spectral similarity between some pixels belonging to mixed forests and alpine meadows. This is the reason for using band 3. Band 3 was preferred instead of band 4 which was used in the analysis of the coniferous species, because the reflectance of the alpine meadows in the green region of the spectrum is higher when compared to the red channel, which this time has lower values in the case of broad-leafed forests, because of the intense photosynthetic activity compared to the cold season. Shortwave infrared bands, along with the near infrared ones, are useful for the elimination of the coniferous species and other elements, like bare soil.
Following the separate analysis of the broad-leafed and coniferous species, two rasters were obtained. The two rasters were mosaicked using the „Mosaic to New Raster” tool in ArcMap. The final raster was obtained using the “Maximum” function, thus being possible to select only the highest value from two overlapping pixels with different values. From the six pixel values resulted from the analysis, only the highest three were displayed, lower values being specific to non-forest areas, such as meadows, bare soil, snow or water bodies.
To minimize the errors, the resulting forest polygons were filtered using a set of criteria. Subsequent to the transformation of the raster dataset into a vector layer, potential virgin forest polygons corresponding to medium value pixels were filtered according to the spatial relationship with the polygons corresponding to the higher value pixels, using the “intersecting” function. The same filtering process was applied to the polygons corresponding to the lower pixel values. In this case, only the polygons with a minimum area of 100m2 were used. One of the factors which are contributing to the fragmentation of the virgin and quasi-virgin forests is the road network. Thus, roads were used for the clipping of the areas corresponding to them and also for the fragmentation of the resulted forest polygons.
Finally, only the polygons with a minimum area of 0,5ha were selected. It is important to mention that the minimum surface of a virgin forest, according to the Romanian law, is 6ha. Because of the low homogeneity of the resulted forest layer, erasing the polygons smaller than 6ha would be inadequate, many of the potential virgin forests being eliminated, including some polygons belonging to the forests charted in the field. The low homogeneity of the resulted forest layer could be explained by the small samples of virgin forests used as reference, or by the existence of errors in the satellite images, like shadows, overcorrection, or by the existence of clouds and their shadows. The degree of homogeneity was calculated for each polygon, by dividing the perimeter to the polygon area, low values being specific to very fragmented polygons or polygons with a very small area. Polygons with a value lower than 10 were erased. The resulted polygons were filtered considering the spatial relationship with the remaining polygons.
The anthropogenic influence represents the main factor which we must take into consideration in the identification process of the virgin and quasi-virgin forests. In order to have a better estimation of the areas were virgin and quasi-virgin forest might be, a series of parameters can be taken into consideration, like the road network or terrain inaccessibility. These two parameters can be quantified by calculating the geodesic distance. Therefore, this distance will be proportional to the chance of finding virgin forests, in the following field studies.
Results and discussions
The chances of finding virgin and quasi-virgin forests are proportional with the altitude, these forests being very common, most probably, to the coniferous forests, especially to high altitude ones. This can be explained by the remoteness that is characteristic to those areas. This remoteness can be expressed by calculating the geodesic distance, using the road layer and a digital elevation model. This situation can be observed in the case of the Mușeteica forest, where the geodesic distance values are very high, especially in the upper part of the watershed. In the case of the Știubeaua and ObârșiaCumpăniței forests, the geodesic distance will have lower values, because they are located in an area where the road network density is higher. Other potential virgin forests, with a significant size, can be localized in the upper parts of the watersheds of the rivers BoiaMică, Boia Mare, IzvorulCăldărușei, Berivoi, LuțeleMici, Frăcea, Urlea, Brezicioara or Arpășel. This forests should be the first options when planning the field studies.
By using only the pixels with they’re frequency values above the average, important portions of potential virgin forest can be eliminated. This problem can be observed by analyzing the Știubeaua forest, where the resulted forest polygon is fragmented. Also, in the case of Știubeaua forest the geodesic distance values are much lower compared to the Mușeteica forest, Știubeaua forest being a quasi-virgin one, while Mușeteica having the characteristics of a virgin forest. Therefore, we could conclude that high values of geodesic distance are indicating the presence of virgin forests.
Nebulosity must be taken into account when analyzing the results, clouds being present on the July images. Thus, from a total of 1424km2 of forest covered areas within the Natura2000 Făgăraș site, about 31km2 (2,1%) are representing areas with cloud cover and cloud shadows.
The spectral bands which are recording the electromagnetic radiation from the visible, near infrared and shortwave infrared spectrum, along with the high spatial and temporal resolution, are making from Sentinel 2 a valuable instrument for the analysis of the vegetation, including the forest cover.
This paper must be considered, firstly, a methodological exercise, the results being influenced by the small area of the sample forests and also by the small number of the satellite images that were used. Therefore, utilizing a multi-temporal series of satellite images would greatly increase the accuracy of the result, finding images with an acceptable amount of cloud cover being the biggest issue.
Also, by utilizing multiple factors that could influence the spatial distribution of the virgin and quasi-virgin forests, a more accurate result could be obtained. These factors can be, for example, information regarding the spatial extension of the forest loggings or structures used in forest logging activities, like skidding cables.
Some of the essential information needed in the process of the virgin forests identification can’t be derived using satellite images. An example would be the common presence of standing or lying deadwood in the old-growth forests
7. Delegido J., Verrelst J., Alonso L., Moreno J. (2011) – Evaluation of Sentinel-2 Red-Edge Bands for EmpiricalEstimation of Green LAI and Chlorophyll Content, Sensors 2011, 11, 7063-7081
8. Fernandes R., Omari K., Canisius F., Lethbridge U., Rochdi N., Baret F. (2010) Robust LAI Retrieval from theSentinel-2 Red-Edge Bands, CCRS Technical Report Series
9. Frampton W., Dash J., Watmough G., Milton E. (2013) – Evaluating the capabilities of Sentinel-2 for quantitativeestimation of biophysical variables in vegetation, ISPRS Journal of Photogrammetry and Remote Sensing
10. Giurgiu V., Doniță N., Bândiu C., Radu S., Cenușă R., Dissescu R., Stoiculescu C., BirișIovu-Andrian(2001) – Pădurilevirgine din Romania, asblForetwallone
11. Jan Knorn, Tobias Kuemmerle, Volker C. Radeloff, William S. Keeton, Vladimir Gancz, Iovu-Adrian Biris, Miroslav Svoboda, Patrick Giffiths, Adrian Hagatis, Patrick Hostert (2012) – Environmental Conservation 40 (2): 182–193
12. Motohka T., Nasahara K., Oguma H., Tsuchida S. (2010) – Applicability of Green-Red Vegetation Index for RemoteSensing of Vegetation Phenology, Remote Sensing 2010, 2, 2369-2387
13. Mueller-Wilm U. (2016) – Sen2Cor Configuration and User Manual
14. Norby R., Sholtis J., Gunderson C., Jawdy S. (2003) – Leaf dynamics of a deciduous forest canopy: no response to elevated CO2, Oecologia (2003) 136:574–584
15. Richter R., Kellenberger T., Kaufmann H. (2009) – Comparison of Topographic Correction Methods, Remote Sens. 2009, 1, 184-196
16. Richter R., Louis L., Uwe Müller-Wilm (2012) – Sentinel-2 MSI – Level 2A Products Algorithm Theoretical Basis Document
17. Slaughter J. (2014) – The Sentinel Satellites and Copernicus Contributing Missions
18. Ursu A., Nicoară M., Grădinaru I. (2013) – Ghidulsiturilor Natura 2000, Iași, Ed. StudIS
19. Wang Q., Adiku S., Tenhunen J., Granier A. (2004) – On the relationship of NDVI with leaf area index in a deciduous forest site, Remote Sensing of Environment 94 (2005) 244–255
20. Zheng G., Moskal M. (2009) – Retrieving Leaf Area Index (LAI) Using Remote Sensing: Theories, Methods and Sensors, Sensors 2009, 9, 2719-2745