Feasibility of hyperspectral vegetation indices for the detection of chlorophyll concentration in three high Arctic plants : Salix polaris , Bistorta vivipara , and Dryas octopetala

Remote sensing, which is based on a reflected electromagnetic spectrum, offers a wide range of research methods. It allows for the identification of plant properties, e.g., chlorophyll, but a registered signal not only comes from green parts but also from dry shoots, soil, and other objects located next to the plants. It is, thus, important to identify the most applicable remote-acquired indices for chlorophyll detection in polar regions, which play a primary role in global monitoring systems but consist of areas with high and low accessibility. This study focuses on an analysis of in situ-acquired hyperspectral properties, which was verified by simultaneously measuring the chlorophyll concentration in three representative arctic plant species, i.e., the prostrate deciduous shrub Salix polaris, the herb Bistorta vivipara, and the prostrate semievergreen shrub Dryas octopetala. This study was conducted at the high Arctic archipelago of Svalbard, Norway. Of the 23 analyzed candidate vegetation and chlorophyll indices, the following showed the best statistical correlations with the optical measurements of chlorophyll concentration: Vogelmann red edge index 1, 2, 3 (VOG 1, 2, 3), Zarco-Tejada and Miller index (ZMI), modified normalized difference vegetation index 705 (mNDVI 705), modified normalized difference index (mND), red edge normalized difference vegetation index (NDVI 705), and Gitelson and Merzlyak index 2 (GM 2). An assessment of the results from this analysis indicates that S. polaris and B. vivipara were in good health, while the health status of D. octopetala was reduced. This is consistent with other studies from the same area. There were also differences between study sites, probably as a result of local variation in environmental conditions. All these indices may be extracted from future satellite missions like EnMAP (Environmental Mapping and Analysis Program) and FLEX (Fluorescence Explorer), thus, enabling the efficient monitoring of vegetation condition in vast and inaccessible polar areas.


Introduction
Imaging spectroscopy, which consists of recording electromagnetic radiation in hundreds of narrow bands (2-5 nm), makes it possible to analyze how electromagnetic radiation interacts with the analyzed matter [1].The selective absorption, reflection, or transmission of various wavelengths allows a detailed analysis of the spectral properties of individual plants and vegetation communities [2].Spectrometer measurements are carried out in both the visible range (VIS, 400-700 nm), near infrared range (NIR, 700-1,500 nm), short-wave infrared range (SWIR, 1,500-2,500 nm), thermal infrared range (TIR, 8,000-15,000 nm), and finally in the microwave range (1 cm-1 m) [3].A basic measure is spectral reflectance, which indicates the quotient of energy reflected from the incident energy of a given electromagnetic spectrum [4].Spectral properties of plants depend on their anatomical structure, morphology, and physiological processes [5].Visible radiation that reaches the plant is absorbed and reflected.The absorbed radiation is used in photosynthesis and fluorescence processes and is emitted as heat.Chlorophyll, carotenoids, and anthocyanins absorb photons of light in the visible range.In infrared, reflection depends on the plant's cellular structures [6], its water concentration [7], its chemical components [8], leaf thickness [9], roughness of leaf surface and canopy [10], the physiological age and arrangement of leaves [11], habitat exposure, solar radiation, phonological period, and various types of diseases and vegetation damage [12].VIS and NIR wavelengths play an important role in the identification of pigments, e.g., chlorophylls, xanthophylls, and carotenoids.Spectral characteristics (reflectance in selected wavelengths) are used to calculate remote sensing vegetation indices [12], which use various mathematical combinations of relevant coefficients to identify the analyzed properties of plants.Vegetation indices can be divided into many groups and can be used to conduct assessments of different vegetation features, e.g., its general condition, the concentration of photosynthetically active pigments, the amount of light used in photosynthesis, levels of nitrogen in the plants, dry biomass, organic carbon, and water content.Quantitative measurements of chlorophyll, the main photosynthetically active pigment, and protective pigments allow for the assessment of plant condition and potential different stress factors [13].An analysis performed by Gitelson [14] demonstrates that the relationship between the chlorophyll concentration and the amount of accumulated light is nonlinear, as the absorption per unit of chlorophyll decreases at high chlorophyll concentrations [14].Some plants can modulate light absorption during the process of photosynthesis through the use of mechanisms such as leaf movement, leaf angle adjustments [15], covering the leaves with substances such as wax [16], and changing the concentration of certain pigments, e.g., anthocyanin [17].Photochemical processes use most of the energy absorbed by chlorophyll [18].Some of the energy that reaches the plants is used to produce sugars.The rest is lost in the form of heat or re-emitted as fluorescence.These processes complement each other, and the level of each of them depends on the levels of the other two [19].
Chlorophylls are responsible for about 65% of the photosynthesis process, while the share of carotenoids (with xanthophylls) is approximately 35%; although these proportions vary depending on the phenological period of the analyzed plant species [20,21].The ratio of chlorophylls to carotenoids is an indicator of plant health and constitutes a characteristic value for individual species [22].High concentrations of chlorophyll are characteristic of plants in good condition, which translates into high reflectance in the green range of the electromagnetic spectrum and absorption in the blue and red ranges [23].Quantitative studies of vegetation are related to plant indices [24]; they require calibration with biophysical variables [25].Currently, they are used to assess crops, e.g., soil property modeling [26,27], irrigation efficiency [28], adaptation to stress factors [29], and the identification of dominant types of meadows [30] or forests [31,32].Narrowband indices acquired from spectrometric measurements are applicable for assessing vegetation vigor, e.g., chlorophyll concentration or change in pigment concentration during the growing season [33].Hyperspectral data are commonly used for carrying out vegetation assessments in many valuable ecosystems, e.g., Yellowstone National Park [34], spectral properties of high Arctic plants [35], or modeling of heterogeneous grasslands [36].org/10.5586/asbp.3604

Digital signature
This PDF has been certified using digital signature with a trusted timestamp to assure its origin and integrity.A verification trust dialog appears on the PDF document when it is opened in a compatible PDF reader.Certificate properties provide further details such as certification time and a signing reason in case any alterations made to the final content.If the certificate is missing or invalid it is recommended to verify the article on the journal website.
The aim of this study is to assess how suitable narrow band vegetation indices are in assessing the chlorophyll concentration of three widespread and abundant high Arctic plants.A selection of indices allows us to limit the number of local surface measurements in favor of remote sensing-based monitoring, which can cover large areas.This is an important issue in the context of the approaching launch of the next two satellite missions, EnMAP [37] and FLEX [38].The German spaceborne EnMAP mission will provide hyperspectral images in 230 bands (420-2,450 nm) with a ground resolution of 30 × 30 m.The swath width covers 30 × 5,000 km with a revisit of 4 days.One of the main goals of the EnMAP is to acquire biophysical and geochemical properties of the biogeosphere worldwide.ESA's FLEX mission in 2022 is for vegetation fluorescence mapping and analyzing.The system will orbit with a Copernicus Sentinel-3 satellite, which offers a swath width of 1,270-1,420 km, to integrate the optical and thermal properties of plants.

Research area and targets
The study area covers selected sites in the surroundings of Longyearbyen, which is the major settlement in the archipelago of Svalbard (78.2°N and 15.6° E).Longyearbyen is a former mining town.Hence, the landscape is contaminated with dust containing heavy metals [39][40][41].The area is characterized by nutrient-rich soils and a maritimebuffered high Arctic climate with an average temperature for July and February of +5.9°C and −16.2°C, respectively [42].The vegetation is dominated by moss-rich ferns and marshes in the lower parts of Adventdalen and by wind-exposed, dry dwarf shrub tundra in elevated areas (Fig. 1).The dominant species of the tundra are mountain avens (Dryas octopetala L.; Fig. 2), white arctic bell-heather [Cassiope tetragona (L.) D. Don], and polar willows (Salix polaris Wahlenb.; Fig. 3).Alpine bistort [Bistorta vivipara (L.) Delarbre] is one of the most abundant herbs in this tundra system.These species are pan-Arctic in distribution and widespread at Svalbard [42,43].
Dryas octopetala is a prostrate woody species (Fig. 2) forming a distinct heath community that grows in dry localities characterized by gravel and rocky barrens and where snow melts early [43,44].It is semievergreen, meaning that chlorophyll in active leaves breaks down, the leaves become brown but overwinter attached to the shoots, and chlorophyll is reproduced in the same leaves the next spring [45].Salix polaris is a prostrate mat-forming willow growing among gravel or in moss carpets (Fig. 2, Fig. 3).Fall leaf senescence generally starts early [43,44].Bistorta vivipara is a perennial herb,  which in Svalbard, can be up to 15 cm tall under optimal conditions; however, it can grow on almost any substrate in Svalbard [43,44] and is often sterile (i.e., only producing basal leaves but no stalk).

Methods
The field campaign was carried out from August 4th to 7th in 2015 [35].The following devices were used during the field research: (i) ASD FieldSpec 3 spectrometer connected with a fiber optic contact probe (ASD PlantProbe, which records the reflected electromagnetic radiation in the 350-2,500 nm range from the built-in lamp, providing stable conditions for all tested plants [46]) and (ii) Force-A Dualex Scientific sensor, which allows us to perform real-time and nondestructive measurements of chlorophyll (Chl), anthocyanins (Anth), flavonoids (Flav), and nitrogen (NBI) indices [47].The chlorophyll index is highly correlated with the chlorophyll extracts measured in laboratory conditions (R 2 oscillated around 0.88-0.96 in hundreds of samples, and errors were not higher than 16%).Therefore, the presented chlorophyll index values are expressed as µg cm −2 [47].
The ASD FieldSpec 3 and Force-A Dualex Scientific have different sizes of sensors for measuring leaves; ASD PlantProbe measures a signal from an area of 2 cm in diameters and Force-A Dualex Scientific from an area of 5 mm in diameter.The recorded index values were not taken in the same leaves or plants, and the health status of the leaves was not documented in the field.So, data was collected from 10 different plants from the same sites with both instruments.For the spectrometric measurements, the size of the ASD PlantProbe detector required us to place smaller leaves next to each other, thus, chlorophyll concentrations were not tested in different parts of the leaves.This issue is not important in the case of upscaling field data to airborne or satellite levels.Each device was run 10 times at each site, with 10 spectrometric measurements consisting of 25 independent scans that were later averaged to one measurement, which in total gives 250 independent measurements for each species per site (Tab.1).The field-acquired spectrometric data was transferred to the ASD ViewSpec Pro software and exported as an ASCI file into the Statistica 13 software (StatSoft, Poland) to calculate the remote sensing indices (Tab.2) by assessing the concentration of photosynthetic pigments in the vegetation and the amount of light used in the photosynthesis process (Tab.3).At the same time, data from the chlorophyll measurements were imported into the Statistica software.We then used the Shapiro-Wilk test [48] to analyze the normality of the distributions of the calculated data.After that, the Levene test [49] to analyze the homogeneity of the variance was employed.Subsequently, the Kruskal-Wallis one-way analysis of variance by ranks (verification of the hypothesis about the irrelevance of differences between the medians of the tested variable in the populations, no normal distribution of the analyzed data [50]) was carried out for each species to check the statistical differences of the indices in individual sites at the significance level of p < 0.05.Then, a Tukey's test was used, which means that the sites were analyzed to verify which of them are different from the others in a statistically significant manner.Due to the nonparametric character of the data (no normal distribution), the Spearman's rank correlation coefficient [51] was determined for the vegetation indices, which was calculated from the spectral reflection curves that were correlated with the data from the measurements of the biophysical variables (pigment concentration), and the statistical significant was determined at the level of p < 0.05.This allowed for the determination of the remote sensing vegetation indices, which reflects the variability of the chlorophyll concentration of the analyzed species in a statistically significant manner.

Quantitative measurements of chlorophyll concentration
The chlorophyll concentrations of the three species varied from site to site along the gradient from Ytre Bjørndalen to Bolterdalen.The highest average chlorophyll concentration was found in S. polaris with an average Chl index of 34.1, expressed as µg cm −2 [47].The lowest average chlorophyll concentration was found in D. octopetala (average Chl index = 24.6).The lowest chlorophyll concentration for D. octopetala (Fig. 4) was found in the SVH site (13.0), while the highest was found in the BOL site (47.0).The lowest chlorophyll concentration for B. vivipara was obtained in the SVH site (11.6), while the highest was found in the SVH (47.7) and BOL (48.0) sites.For the species S. polaris, the lowest chlorophyll concentration was in the SVH site (14.0), while the highest was found in the YBJ site (51.0).The highest variability amongst the chlorophyll concentrations was found in the SVH site (Fig. 4).There were not any significant intraspecific differences concerning chlorophyll content along the gradient taking into account the locations of the sites; the lowest average value for the Chl index was obtained in the SVH site (26.6), while the highest was found in the ISD site (31.4).

Vegetation indices
The vegetation indices extracted and calculated from the hyperspectral measurements were verified using Chl indices from the Dualex instrument.The calculated correlation (Spearman's rank correlation, R values) for the analyzed species was different at each research site.However, in the case of the analyzed species, high correlations between Tab. 2 Overview of the vegetation indices that were applied to the collected data.RGR Red/green ratio; anthocyanins/chlorophyll (R600 − R699) / (R500 − R599) [70] RARSa Ratio analysis of reflectance spectra algorithm chlorophyll a R675 / R700 [61] parameters of the indices were confirmed (statistically significant values at the significance level of p <0.05 are marked in bold in Tab. 4).

Index
As the sites were located in different places and thus characterized by different environmental conditions, statistically significant differences (at the significance level of p < 0.05) were observed among the index values.Out of all the research sites containing the analyzed species, the most numerously represented indices that showed statistically significant changes were: VOG 1, ZMI, mNDVI, mND, VOG 2, VOG 3, and NDVI 705.This allowed us to narrow down the analyzes to those indices that best represent the condition of the vegetation (highest percentage) and to confirm it with the values and strong correlations of the chlorophyll measurements (Tab.4).
Tab. 3 Application of the selected wavelengths for the plant pigment absorption analyzes via the hyperspectral analysis [71].

Salix polaris
Better indices for selection are NDVI 705 and GNDVI, GM 1, or GRVI (which are statistically significant, but the chlorophyll concentration information does not overlap).A proper analysis should be based on a sensor filter range of full width at half maximum (FWHM; technical data including the values of used wavelengths are presented on sensor web pages) because not all sensors have the same ranges as the used bands, thus, not all presented indices in Fig. 5 could be calculated from all sensors.This problem is solved during airborne hyperspectral missions via the satellite's Hyperion or oncoming EnMAP scanners.
One of the most commonly used indices from the narrowband group is the NDVI 705 index.The present research confirms that S. polaris and B. vivipara had the highest values and the highest amount of chlorophyll of the species in the analyzed sites (0.49 and 0.48, respectively), while D. octopetala had the lowest results (0.35) in the same   sites (Fig. 6).There is a significant difference between D. octopetala and B. vivipara for the sites YBJ, ISD, and SVH, while there was a significant difference between D. octopetala and S. polaris for the site YBJ.However, there were no intraspecific significant differences between sites.The mNDVI 705 index, which is a modification of the NDVI (a broadband index) or NDVI 705 (a narrowband index), also allowed us to indicate the level of chlorophyll concentration in S. polaris and B. vivipara.Both species had the highest values (averages = 0.56) and the highest amount of chlorophyll of the species in the analyzed sites, while D. octopetala had the lowest results (0.42) in the same sites (Fig. 7).There is a significant difference between D. octopetala and B. vivipara for the YBJ, ISD, and SVH sites, while there was a significant difference between D. octopetala and S. polaris for the YBJ site.However, there were no intraspecific significant differences between sites.
The VOG 1, 2, 3 indices (there should be a negative relationship between the VOG 1, VOG 2, and VOG 3 indices; Fig. 5) indicate a similar relationship for the three species studied in the different sites.Hence, each of the analyzed indices confirms in a statistically significant way the chlorophyll concentration levels.The VOG 1 index (Fig. 8-Fig.10) also had the highest values for S. polaris and the lowest for D. octopetala, similar to what was observed for NDVI 705.The other two indices, VOG 2 and VOG 3, confirm the data reported above.Because they are indices that are interpreted inversely, Salix polaris had the lowest values on the graph, while Dryas octopetala had the highest value (Fig. 9, Fig. 10).
As the mND index was developed for the wavelengths of 750-900 nm, 660-720 nm, and 445 nm, it is very good at illustrating the radiation absorption by pigments and, therefore, also determining the chlorophyll concentration levels.This confirms the above-described observations, i.e., the highest values of the index were for S. polaris, the medium-level values were for B. vivipara, and the lowest values were for D. octopetala in the YBJ, ISD, and SVH sites.In the BOL site, the highest values were for B. vivipara (Fig. 11).
As the ZMI index was developed for the wavelengths of 750 and 710 nm, it is very good at illustrating the radiation absorption by pigments and, therefore, also determining the chlorophyll concentration levels and cellular structures.This confirms the above-described observations, i.e., the highest values of the index were for S. polaris, the medium-level values were for B. vivipara, and the lowest values were for D. octopetala in the YBJ, ISD, and SVH sites.In the BOL site, the highest values were for B. vivipara (Fig. 12).

Discussion
The selection of vegetation indices in this study confirmed that the most applicable indices for all species were VOG 1, ZMI, mNDVI, mND, VOG 2, VOG 3, and NDVI 705.There were also less, but strong, relationships between the chlorophyll measurements and the vegetation indices of chlorophyll levels above 0.5 for the following indices: GM 2, LIC 2, SIPI, GARI, and PSRI.It should also be mentioned that for the plants in the alpine zone of the Tatras, the NDVI 705 and VOG 1, 2, 3 indices were statistically significant in 87.5% of cases [52], while the GM 2 index was statistically significant in 69% of cases, and RARSc in 71% [52].Other important vegetation indices of the alpine zone included: RARS (80%), GI (77%), and LIC 2, mNDVI 705, RGR, and SRPI in 70% of analyzed species [53].
For the best mND, the average correlation coefficient for all species was 0.41 (Tab.4), while S. polaris showed the best result in all individual species: 0.59.For B. vivipara, the value was 0.51.Slightly better results were obtained in laboratory conditions by Sims and Gamon [54] as the mND index correlated with the chlorophyll concentration (0.6 mmol m −2 ) at the R 2 level of 0.62-0.66(nonlinear models).On the other hand, the results for the NDVI 705 index (not including the 445 nm wavelength) ranged from 0.50 to 0.58 (excluding 0.19 of D. octopetala) with the above-mentioned chlorophyll concentration level [54].In all of the above cases (concerning both this study and the cited publications), the relationships were statistically significant.The above-referenced studies [54] were used for the assessment of vegetation greenness and ecosystem CO 2 exchange in response to a drought in the Southern California chaparral ecosystem [55].Analyzes based on surface and airborne ADAR imager measurements (captures the relationship between CO 2 accumulation and chlorophyll activity) in the field (chlorophyll content index -CCI) were calculated from the image NDVI 705 at R 2 = 0.85 [55].
Hope et al. [56] achieved similar results measuring Alaskan tundra plants (three vegetation communities); hand-held radiometer data of 5 × 5 m point-quadrat estimates of the photosynthetic active biomass sites correlated with NDVI vegetation indices, which were adapted to the parameters of satellite images of Landsat and SPOT.The nonlinear correlation force (R 2 ) of the indices ranged from 0.48 to 0.52 [56].Significantly stronger relationships between NDVI indices (e.g., NDVI 705 and GNDVI) can be observed in the case of agricultural crops [57]; correlations between the vegetation indices and the surface chlorophyll measurements reached R 2 = 0.95 [55], e.g., for NDVI 705, chlorophyll = 0.86 (soybean) and 0.94 (maize).In this study, the values are lower by almost a half (0.41-0.63).In the case of the GNDVI index, the correlation values were 0.88 for soybean and 0.85 for maize (in this study, the correlation coefficients were 0.46-0.57).It should be remembered that polar plants are covered with waxes and other layers that protect them against winter warming events that may lead to drying and freezing.These elements significantly reduce the possibilities of chlorophyll identification [58].Further assessment of the results from the analysis indicated that the species S. polaris and B. vivipara were in very good to good condition, while D. octopetala was in medium condition.There were also differences between the sites, and the variability within sites was high, especially in the site BOL, indicating that the environmental conditions differed.These results are in accordance with results by Zagajewski et al. [35] and Bjerke et al. [39], which showed that the health condition was lower for D. octopetala compared to that of S. polaris and B. vivipara due to different climatic and contaminant conditions (mining) within the study area [41].

Conclusions
The analyzes of the dominant Svalbard species showed that their chlorophyll concentration levels were within the optimal range for S. polaris and B. vivipara; attention was paid to the variability of these values, depending on the species and sites.
Simultaneous spectral and chlorophyll concentration-focused analyzes confirmed the statistical significance of individual narrow-band vegetation indices.The most optimal indices were: VOG 1, ZMI, mNDVI, mND, VOG 2, VOG 3, and NDVI 705.This is an important factor in the context of new satellite missions, e.g., Sentinel, EnMAP, or FLEX.Although the observed relationships were not very strong, the use of hyperspectral data for the monitoring of vast areas of the Arctic will allow for the observation of trends regarding changes in vegetation and the continuous monitoring of the Arctic greening process.Use of satellite remote sensing supported by periodic biometric surface measurements conducted in permanent sites will be valuable for vegetation monitoring in the high Arctic.

Fig. 1
Fig. 1 View of one of the sites in Bjørndalen near Longyearbyen, Svalbard (photo: Z. Bochenek).

Fig. 5 A
Fig. 5 A correlation matrix between different remote sensing indices derived from ASD FieldSpec measurements (N = 160 averages of 4,000 independent measurements).Full names of the indices are presented in the Tab. 2.

Fig. 6
Fig. 6 The values of the NDVI 705 index (maximum, minimum, median, and upper and lower percentiles) for the species D. octopetala, B. vivipara, and S. polaris in the YBJ, ISD, SVH, and BOL sites.

Fig. 7
Fig. 7 The values of the mNDVI 705 index (maximum, minimum, median, and upper and lower percentiles) for the species D. octopetala, B. vivipara and S. polaris in the YBJ, ISD, SVH, and BOL sites.

Fig. 8
Fig. 8 The values of the VOG 1 index (maximum, minimum, median, and upper and lower percentiles) for the species D. octopetala, B. vivipara, and S. polaris in the YBJ, ISD, SVH, and BOL sites.

Fig. 9
Fig.9 The values of the VOG 2 index (maximum, minimum, median, and upper and lower percentiles) for the species D. octopetala, B. vivipara, and S. polaris in the YBJ, ISD, SVH, and BOL sites.

Fig. 10
Fig.10 The values of the VOG 3 index (maximum, minimum, median, and upper and lower percentiles) for the species D. octopetala, B. vivipara, and S. polaris in the YBJ, ISD, SVH, and BOL sites.

Fig. 11
Fig. 11 The values of the mND index (maximum, minimum, median, and upper and lower percentiles) for the species D. octopetala, B. vivipara, and S. polaris in the YBJ, ISD, SVH, and BOL sites.

Fig. 12
Fig.12 The values of the ZMI index (maximum, minimum, median, and upper and lower percentiles) for the species D. octopetala, B. vivipara, and S. polaris in the YBJ, ISD, SVH, and BOL sites.
Location of the research sites and patterns (UTM, 33rd zone).In all sites, the following species were measured: Salix polaris, Dryas octopetala, and Bistorta vivipara. 1ab.1