Silva Balcanica, 20(3)/2019

Spatial Distribution of High-Mountain Ecosystems – Application of Remote Sensing and GIS: A Case Study in South-Western Mountains ()

Kostadin Katrandzhiev, Svetla Bratanova-Doncheva Institute of Biodiversity and Ecosystem Research – , Bulgarian Academy of Sciences

Abstract

The dynamics in spatial distribution of high-mountain ecosystems (HME), located in South-Western (SW) Rila Mountains, Bulgaria, were studied for a period of around 40 years (1977 – 2018). The results were obtained related to these dynamics using satellite data in GIS environment. Normalized Difference Vegetation Index (NDVI) based on satellite data from Landsat and Sentinel 2 sensors was used for analysis and assessment of ecosystem condition. Some climate components (rainfall, solar radiation, temperature, evapotranspiration) were analysed in order to reveal existing dependency between the NDVI and meteorological parameters. To verify the obtained results, a Normalized Difference Greenness Index (NDGI) was also computed. A clear connection between the studied climate components, i.e. meteorological elements and NDVI values was found for the last decade. The results are presented as thematic maps, providing quantitative and qualitative assessment of the NDVI spatial distribution in the selected study area. Persistence of forested areas’ functionality was observed and the seasonal variation in functional condition of shrubs and grasslands was established. These variations remained seasonally cyclic and did not change significantly the functional condition of the HME vegetation in the long term.

Key words: Ecosystem condition, satellite data, GIS, NDVI, NDGI

Introduction

High-mountain ecosystems (HME) are dynamic, sensitive and vulnerable to climate changes (Grace et al., 2002). This article aims to characterise the relation between ecosystem condition and meteorological component changes. Long-term HME spatial distribution depends on changes in their condition. In 2017, a methodology for assessment of woodland and forest ecosystems was developed which provides an indicator-based assessment guide for forest ecosystems’ condition (Kostov et al., 2017). The forest ecosystem condition also can be determined by dint of vegetation indices (VI) based on remote sensing data along with analyses of climate characteristics. Remote sensing data have proven to be a reliable approach in environmental research activities like ecosystem monitoring and risk assessment (Radeva et al., 2018; Radeva, 2018). It is important to mention that remote sensing methods does not confront field work but just the opposite, they amplify them as proven by the authors Tuominen et al. (2009) and Lawley et al. (2016) . The most applied method for vegetation condition assessment,

57 based on remote sensing data, includes computation and analysis of vegetation indices. The Normalized Difference Vegetation Index (NDVI) is the most frequently used index for vegetation condition and vitality assessment (Tucker, 1979). It shows vegetation’s capability of energy absorption and reflection, photosynthetic capacity and concentration of ecosystem’s leaf biomass (Pavlova, Nedkov, 2005). The NDVI is the most widely used index for establishing of presence or absence of vegetation. Moreover, with respect to time, the NDVI provides an estimate of gross primary production (Tucker, Sellers, 1986). The disadvantage of this index appears to be that in some cases the NDVI is not sensitive enough to detect small changes in vegetation cover (e.g. after wildfire or other catastrophic events). On the other hand, the Normalized Difference Greenness Index (NDGI) is a very sensitive index capable of detecting even small increase of vegetation growth (Nedkov, 2017a). In the current research, we offer assessment of HME condition through combining remote sensing via VIs (NDVI) and meteorological components analyses which represents innovative technic for HME research. Respectively, the objectives of the current article were computing of the NDVI based on satellite data from Landsat and Sentinel 2 sensors and using it for quantitative and qualitative assessment of a HMEs’ condition for a period of around 41 years (1977 – 2018). Further, a verification of the NDVI results via NDGI computation was performed. Additionally, analyses of some climate components (rainfall, solar radiation, temperature, evapotranspiration) were carried out in order to reveal existing dependency between the NDVI and climate parameters.

Materials and Methods

The current research focuses on HME. The study was carried out on a territory which was selected based on pre- set criteria, i.e. to encompass more than one vegetation type; to cover a territory corresponding to the definition of a high-mountain habitat, including treeline ecotone which marks the transition from closed tall forest to the alpine zone and to be easily accessible. The selected study area is located in the south-western parts of the Rila Mountains and includes parts of the lands of seven populated places (Bistritsa, Gorno and Dolno Osenovo, Bachevo, , and Belitsa) and also of the Parangalitsa Reserve with an overall area of 9296 ha. The highest point of the study area is the Ezernik Peak (2485 m a.s.l.). Three main vegetation types were presented: (i) woodland and forests, (ii) shrubs and (iii) grasslands (Uzunov, Gussev, 2003). It includes three structurally and functionally interrelated zones (Fig. 1): closed tall forest, groups of trees (>3m) and a zone with isolated tree individuals which forms the transition from montane forest to the alpine zone (Korner, Paulsen, 2004). Dominating is the conifer forest vegetation, followed by grasslands and shrubs which form the physiognomy of the entire HME. In an attempt to assess the functional condition of the HME for a long time period from 1977 to 2018, a methodology was used based on remote sensing data (Katrandzhiev, 2018), such as multispectral satellite data from sensors Sentinel 2

58 Table 1. Data sources used for indices calculation

№ Date Sensor

1 22.08.1977 Landsat

2 23.05.1984 Landsat

3 26.07.1984 Landsat

4 27.06.1985 Landsat

5 22.05.1986 Landsat

6 03.07.1987 Landsat

7 11.07.1990 Landsat

8 02.09.1992 Landsat

9 29.06.1994 Landsat

10 28.06.2000 Landsat

11 13.06.2009 Landsat

12 15.07.2009 Landsat

13 23.09.2011 Landsat

14 09.09.2012 Landsat

15 02.05.2017 Landsat

16 26.10.2017 Sentinel 2

Fig. 1. HME_zones- (a) Scheme of the HME zone; 17 03.04.2018 Landsat (b) Aero photo image from study area

(Copernicus – ESA) and Landsat with high spatial, spectral and radiometric resolution. The data used are presented in Table1. The Landsat database provides access to past time satellite images which is fundamental for our research. It allows us to go back in time and track condition changes in HME up to the present. As far as the Sentinel 2 sensor, it is a newer generation remote sensing tool which provides more qualitative images from space and offers free access to its database. The data from both sensors were used for computing both the NDVI and NDGI. Satellite data processing The most important in the data processing phase was the quality data extraction. A selection of appropriate scenes from a database containing the selected study area and covering the whole period of 41 years was collated. The satellite data were required to

59 Fig. 2. Maps of the NDVI for the period 1977- 2018 meet some criteria, the most important of which were to cover the vegetation period (from the end of March to the end of October) and to fulfill quality characteristics like clearness and cloudlessness. These were essential for further data processing, including elaboration of composite images, raster file creation of the selected research area and the NDVI processing of this raster which were applied to each of the selected satellite images or a total of 17. For these technical operations, specialised software for satellite data processing was used. But more work was necessary in order to make the NDVI data suitable for analysing and interpretation. Using another software product in GIS environment, we were able

60 Fig. 3. Distribution of correlation coefficients for the period 1977- 2018

Fig. 4. NDVI regression models for the period 2000-2018

61 Fig. 5. Spatial distribution of the NDVI for the period of 41 years (1977-2018) to convert the obtained NDVI raster file to shape file containing digital (numerical) data with the NDVI values and coordinates of each value. Thematic maps were also elaborated and presented, in order to illustrate the distribution of the values of the NDVI (for each satellite image) over the selected study area (Fig. 2). Thus, statistical processing of the data was possible. By means of the Sigma Plot 11.0 statistical tools, the NDVI data were visualized as 2D and 3D graphics, correlation graphic (Fig. 3) and regression models (Fig. 4). Based on them, results were obtained about the spatial distribution of the NDVI values (Fig. 5), correlation coefficients, etc. To ensure the objectivity of the NDVI results, the NDGI was computed for their verification, using the same satellite data. This index shows changes in the vegetation cover that have occurred over a specific time period. The NDGI can be used for determination of both negative and positive changes in vegetation functioning (Nedkov, 2017b). Analogically, using the same computer software, the NDGI was calculated and thematic maps were elaborated in order to verify the NDVI results (as shown in results section, Fig. 7).

62 Data processing of climate elements In addition to the indices data analyses, an analysis of climate characteristics, such as rainfall, evapotranspiration, solar radiation and mean temperature for the last decade (2009- 2018), was carried out for the territory of the study area based on the European climate database (Table 2). Using GIS software (in this case ArcMap 10.1), national (Bulgarian) climate database was derived from the European climate database for the period of 2009-2018. It is important to note that the derived data have to match the date of the satellite image in order to achieve objectivity of the results. Thus, in this case for the period 2009- 2018, seven images were selected to correspond to the above- mentioned quality criteria. Therefore, six diagrams (Fig. 6) were elaborated (first in line covers two images), describing the dynamics of the selected climate parameters for the ten- year period.

Table 2. Climate component values

Date R, l/m2 Evpt, l/m2 SR, kJ/m2 t, C0

11.06.2009 28 37.3 25.6 16.3 21.06.2009 70 29 21.8 13.3 11.07.2009 26 38.8 25.3 17 21.07.2009 1.5 43.5 27 19.8 01.09.2011 23 26 19 17.7 11.09.2011 19 22 17.2 18 21.09.2011 41 11 13.7 13 01.10.2011 11 17 15.6 10.6 01.09.2012 2 25.8 20.8 17.4 11.09.2012 67 16 13.9 14 21.09.2012 0 17 18 15.7 01.10.2012 3 12.8 15.4 13.1 01.05.2017 45 29 20.3 9.9 11.05.2017 11 36 24.1 11.5 21.05.2017 72 32 20.1 10.8 01.06.2017 45 34 20.1 14.6 01.10.2017 37 16 13.1 7.9 11.10.2017 0 17 14.7 10.8 21.10.2017 42 13 11.4 5.5 01.11.2017 10 7.5 9.6 3.4 01.04.2018 17 25 19.6 7.3 11.04.2018 13 27 19.1 10.2 21.04.2018 6 34 22.8 12.4 01.05.2018 33.7 33 20.8 12.8

63 Fig. 6. Dynamics of the studied meteorological parameters (rainfall, solar radiation, evapotranspiration and temperature) for the period 2009-2018

Fig. 7. Spatial distribution of the NDGI for the periods 11.07.1990- 03.07.1987 (a) and 23.09.2011- 15.07.2009 (b)

64 Originally, the climate data represented raster files (for each rain, temperature, solar radiation and evapotranspiration) which were necessary to be converted in order to be usable. And again, using the above- mentioned software product, the raster files were converted to shape files (polygons or point) containing digital (numerical) data. Through the statistical – application we built graphic models using these datasets. Finally, in addition to the applied methods, a simple statistic approach was implemented based on the count of positive and negative values of the NDGI in order to consolidate the results.

Results and Discussion

The results from the current research demonstrated that for a period of 41 years the NDVI of the studied area showed seasonal fluctuations related to changes in the studied climate parameters (solar radiation, mean temperature and evapotranspiration). The clear connection between the studied climate elements and the NDVI values established by Pavlova, Nedkov (2005) was confirmed with the current study for the last decade. The observed long-term spatial and temporal dynamics of the NDVI indicated persistence of the vegetation functionality of the forested areas. Regarding dynamics of the NDVI in the other territories (shrubs and grasslands), our results showed variation of their condition according to the seasonal changes in climate parameters. These results were confirmed via the NDGI computation (Fig. 7) and correlation analyses of the NDVI (Fig. 8). The lack of climate data for the studied parameters before 2008 made impossible the establishment of direct relation between the NDVI values and changes of climate components for the period before 2008. But as shown in Fig. 2, there were fluctuations in the NDVI values depending on the season in which the image had been captured. In the spring (April and May), the NDVI values closer to negative prevailed in the territories above the timberline towards the alpine zone. On the other territories, where the forests prevailed, the NDVI values were positive but close to 0, which we interpreted as an early phase of the vegetation process after the winter season. When we analysed the maps from the summer (June, July and August), we found that the NDVI values were positive almost everywhere: closer to +1 in forested areas and 0 – 0.4 for the areas towards the alpine zone. That we could explain with the advanced vegetation phase. The situation changed again in the autumn (September and October) when the NDVI values in the high elevated parts became lower (close to 0) again and remained high (0.6 – 0.8) but lower than in the summer for forested areas, which showed seasonal fluctuations of the NDVI values. The results from the count of the positive and negative values of the NDGI demonstrated prevalence of the positive values for the examined time intervals. For the first period (1990-1987) were established 15 271 positive and 10 261 negative NDGI values. And for the second period (2011-2009) were established 70 555 positive and 31 509 negative NDGI values.

65 Correlation analyses 8. Correlation Fig.

66 The NDGI was calculated to confirm the results from NDVI. The difficulties we confronted in index calculation emanated from differences in the image series as a result from sensors’ distinction. The solution was computing the NDGI for two periods (Fig. 7), depending on the correlation between the late and early date. As shown on Fig. 3 and 8, the highest correlations for the entire period were recorded between the dates 11.07.1990 – 03.07.1987 and 23.09.2011 – 15.07.2009. Therefore the NDGI was calculated for these two periods. Considering the climate elements dynamics (Fig. 6) for the last decade and the findings revealed from Nedkov (2017b) for the existing correlation between climate parameters and the NDVI, it was confirmed that there was a dependency between them (Katrandzhiev, 2018). Analysing table 2 (Climate elements values), we revealed the essence of this connection. We found that changes in solar radiation, temperature and evapotranspiration occurred independently of rainfalls which probably were accumulated naturally in the soil and participated in the water balance of the HME. Eventually, the rain component affected the NDVI distribution indirectly. However, we also found out that increased temperature and solar radiation values lead to increase of evapotranspiration values (table 2). When comparing these changes to the NDVI values (Fig. 2) for the period from 2009 to 2018, a clear connection was found revealing that increased values of climate parameters temperature, solar radiation and evapotranspiration lead to increased NDVI values. Respectively, decreased values of the same climate components corresponded to decreased NDVI. As mentioned above, our results on the NDGI confirmed the established changes in the NDVI. For the studied time intervals as presented in Fig. 7, the values of the NDGI around 0 indicated remaining of vegetation processes (unchanged), positive values (above 0) showed improving of the vegetation (photosynthetic) processes, i.e. vegetation increment, and negative values (beneath 0) showed loss of functional activity (i.e. lack of vegetation process). Consequently, the NDGI results indicated persistency of the vegetation activity for the first time interval (1990-1987) in the study area which corresponded to the NDVI correlation analysis applied for the same period (Fig. 8). A similar persistency of the vegetation functionality was observed for the second time interval (23.09.2011 – 15.07.2009). Those results also corresponded to the NDVI correlation analyses for this period. This long-term persistency of the vegetation was consolidated by the results from the count of the positive and negative values of the NDGI. We considered the predominant positive values of the index as an additional proof for the endurance of the HME.

Conclusions

Characteristic long-term spatial dynamics of vegetation functionality (using the NDVI) was observed in this case study. Our analyses suggested that the prevailing forested areas retained a continuously steady condition over the study period (around

67 Fig. 9. Map of the grasslands and shrubs in the study area

40 years). The areas covered with shrubs and grasslands (Fig. 9) appeared to be more susceptible to changes, in particular to seasonal changes. Their functional condition varied from good (spring and early summer) to poor (late summer and autumn). We hypothesised that these variations remained seasonally cyclic and did not change significantly the functional condition of the HME vegetation in the long term. Finally, the used methodological approach, based on remote sensing data, proved suitable to assess both short- and long-term changes in condition of the HME and, hence the capacity to provide ecosystem services. In order to assess more accurately tendencies in HME condition, further studies are necessary and they should include more detailed climate data for the years after 1977.

Acknowledgements: This study was carried out with the support of National program “Young scientists and Postdoctoral candidates”, Project BG161PO003-1.2.04-0053-C0001 from OP “Development of the Competitiveness of the Bulgarian Economy 2007-2013“ and Programme “COPERNICUS“ (ESA) and courtesy of Roumen Nedkov from “Aerospace Information Department”, Space Research and Technology Institute at the Bulgarian Academy of Sciences.

68 References

Grace, J., F. Berninger, L. Nagy. 2002. Impacts of Climate Change on the Tree Line. – Annals of Botany, 90, 537-544. DOI:10.1093/aob/mcf222. Katrandzhiev, K. 2018. Application of Remote Sensing for High Mountain Forest Ecosystem Condition Assessment (South West Rila Mountain- Bulgaria). – Ecological Engineering and Environment Protection, 2, 35-40; e-ISSN 2367-8429. Körner, Ch., J. Paulsen. 2004. A world-wide study of high altitude treeline temperatures. – Journal of Biogeography, 31, 713-732. Kostov, G., E. Rafailova, V. Vassilev, S. Bratanova- Doncheva, K. Gocheva, N. Chipev. 2017. Methodology for assessment and mapping of woodland and forest ecosystems condition and their services in Bulgaria. ISBN 978-619-7379-07-5. Lawley, V., M. Lewis, K. Clarke, B. Ostendorf. 2016. Site-based and remote sensing methods for monitoring indicators of vegetation condition: An Australian review. – Ecological Indicators, 60, 1273–1283, http://dx.doi.org/10.1016/j.ecolind.2015.03.021. Nedkov, R. 2017a. Orthogonal transformation of segmented images from the satellite Sentinel-2. – Comptes rendus de l’Acad´emie bulgare des Sciences, 70, 5, 687-692 Nedkov, R. 2017b. “Normalized Differential Greenness Index for Vegetation Dynamics Assessment”. – Comptes rendus de l’Acad´emie bulgare des Sciences, 70, 8; Prof. Marin Drinov Publishing House of Bulgarian Academy of Sciences, ISSN: 1310–1331, 1143-1146. Pavlova, A., R. Nedkov. 2005. Application of the Different Vegetation Indexes Regarding to Forest Physiology and Climatic Seasons. – Scientific Conference “Space, Ecology, Safety” with International Participation, 10–13 June 2005, Varna, Bulgaria, S E S, pp. 263-268 Radeva, K. 2018. Aspects and perspective of Interim Ecological Monitoring application on ecosystems by means of Remote sensing; Ecological engineering and environment protection, 2, pp. 26-34; e-ISSN 2367-8429. Radeva, K., I. Ivanova, D. Borisova. 2018. Application of remote sensing for ecosystems monitoring and risk assessment. – Proc. SPIE 10773, Sixth International Conference on Remote Sensing and Geoinformation of the Environment (RSCy2018), 107731Q; DOI: 10.1117/12.2325854 Tucker, C. J. 1979. Red and Photographic Infrared Linear Combinations for Monitoring Vegetation. – Remote Sensing of Environment, 8, pp. 127-150. DOI: 10.1016/0034-4257(79)90013-0 Tucker, C. J., P. J. Sellers. 1986. Satellite remote sensing of primary production. – Int. Journal of Remote Sensing, 7, 11. Tuominen, J., T. Lipping, V. Kuosmanen, R. Haapanen. 2009. Remote Sensing of Forest Health. – Geoscience and Remote Sensing, Pei-Gee Peter Ho. (Ed.), ISBN: 978-953-307-003-2. Uzunov, U., Ch. Gussev. 2003. High mountain flora of Bulgaria. High mountain flora of Bulgaria- Statistics, ecological characteristics and phytogeography. – Bocconea, 16(2), 763-770. ISSN 1120-4060. World Wide Web electronic publication. Last update: 14 August 2019. URL: http://spirits.jrc.ec.europa. eu/?page_id=184.

E-mail: [email protected]

69