Anda di halaman 1dari 11

2600 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 7, NO.

6, JUNE 2014

Continuous Fields From Imaging Spectrometer Data


for Ecosystem Parameter Mapping and Their Potential
for Animal Habitat Assessment in Alpine Regions
Mathias Kneubühler, Alexander Damm, Anna-Katharina Schweiger, Anita C. Risch, Martin Schütz, and
Michael E. Schaepman, Senior Member, IEEE

Abstract—Remote sensing offers an objective and efficient way to quality (P, N). Nevertheless, multitemporal CF maps derived from
monitor ecosystem properties including their spatial variability APEX IS data were found to provide spatially explicit and fine-
across different land cover types. Especially, the representation of scaled information for analyses of an ecosystem’s total surface water
gradients of biochemical and structural properties of ecosystems content. The combination of multitemporal CF maps of a wide
using continuous fields (CF) approaches bears advantages com- range of ecosystem parameters and more accurate and extensive
pared to discrete land cover classification schemes. This paper observations of animal habitat use will contribute to ongoing and
presents a concept to synergistically generate CF maps of an alpine future vegetation-ungulates research in the SNP.
ecosystem parameter, i.e., total surface water content, from imaging
spectrometer (IS) data. Further, the potential of linking such maps Index Terms—Airborne imaging spectroscopy, continuous field
to ecological patterns, i.e., the spatial distribution of large ungulates mapping, ecosystem, land cover, vegetation, water content.
is being assessed. In vegetated areas, total surface water content is
considered as a surrogate of plant physiological status. Water is, I. INTRODUCTION
besides temperature, light, or nutrients, an important limiting
growth factor determining biomass production and therefore
potential animal forage quantity in alpine grasslands. Resource
ecology interested in trophic interactions between large ungulates
R EMOTE sensing (RS) provides spatial and temporal data
to efficiently monitor the actual status of ecosystems, in
particular, habitat extent and condition [1]. The development of
and their forage requires spatial and temporal information on
ecosystem properties and processes. The study area is located in
spatially accurate and high-resolution RS techniques allows
the upper Trupchun Valley (Val Trupchun) in the Swiss National mapping of ecological properties like terrain properties, exposi-
Park (SNP). The valley is famous for its high densities of chamois tion, land cover, and could, in turn, help explain ecological
(Rupicapra rupicapra L.), ibex (Capra ibex L.), and red deer (Cervus processes. Environmental variables can be treated as data layers
elaphus L.). CF maps of total surface water content were derived and, in combination with respective models, can be used to define
from Airborne Prism EXperiment (APEX) IS data collected over
the SNP in June 2010 and 2011. Abundance maps of predominant
an “envelope” of suitable conditions for specific plant or animal
land cover classes were derived from linear spectral mixture analy- species. Nowadays, RS consequently allows for habitat model-
sis (SMA). They were then combined with water content informa- ing and environmental impact assessment [2], [3]. While RS data
tion of the respective land cover originating from either empirically allow derivation of abiotic factors (e.g., elevation, exposition,
or physically based approaches. The resulting CF maps depicted a and terrain features) in high detail, it remains a demanding task to
spatially continuous representation of relative total surface water
content. APEX IS data from two consecutive seasons revealed
obtain relevant information regarding the distribution of biotic
differences in total surface water content in June 2010 and 2011, resources such as forage quality and quantity [4], [5]. Indeed,
predominantly related to an advanced phenological development in imaging spectroscopy (IS) has been shown to increasingly
spring 2011 and to considerable differences in snow cover between support monitoring abundance and distribution of plant species
the 2 years. Linking total surface water content of grasslands to [1], [6]. In recent years, IS has, in addition, proven to successfully
observed ungulates spatial distributions did not reveal any statisti-
cally significant patterns of habitat use. We conclude that water
provide quantitative information on biochemical properties, i.e.,
availability in Val Trupchun may not be the dominant limiting plant nutrient contents [6]–[8], which then might be used for
factor for potential forage quantity (biomass), or that ungulates assessing forage quality for consumers [9], [10] or investigating
choose their grazing sites based on other criteria, i.e., high nutritious ecosystem services [11], [12].
Information on prevalent land cover is typically required in
Manuscript received August 14, 2013; revised April 15, 2014; accepted May many RS approaches used to estimate either qualitative or
07, 2014. Date of publication May 28, 2014; date of current version August 01, quantitative landscape properties. Land cover information is
2014. mainly applied to optimize the parameterization of algorithms
M. Kneubühler, A. Damm, and M. E. Schaepman are with the Department of
Geography, University of Zurich, Zürich 8057, Switzerland (e-mail: kneub@geo. or to incorporate a priori knowledge to constrain the retrieval of
uzh.ch; alexander.damm@geo.uzh.ch; michael.schaepman@geo.uzh.ch). landscape properties. Generally, RS concepts to map land cover
A.-K. Schweiger is with the Research Department, Swiss National Park, follow either discrete land cover classes or continuous fields
Zernez 7530, Switzerland (e-mail: anna.schweiger@nationalpark.ch).
A. C. Risch and M. Schütz are with the Community Ecology, Swiss Federal (CF). Since discrete land cover mapping represents landscapes as
Institute for Forest, Snow and Landscape Research WSL, Birmensdorf 8903, a spatial mosaic of classified entities [13], [14], such classes
Switzerland (e-mail: anita.risch@wsl.ch; martin.schuetz@wsl.ch). cannot reproduce the full range and variability of land surface
Color versions of one or more of the figures in this paper are available online at
http://ieeexplore.ieee.org. properties that often are necessary to adequately quantify and
Digital Object Identifier 10.1109/JSTARS.2014.2323574 manage landscape patterns and processes [14], [15]. In contrast,

1939-1404 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
KNEUBÜHLER et al.: CF FROM IMAGING SPECTROMETER DATA FOR ECOSYSTEM PARAMETER MAPPING 2601

approximating land cover information in a continuous way using competition and facilitation in such multispecies ungulate
the concept of CF allows representing gradients of biochemical assemblages in the European Alps.
and structural ecosystem properties rather than classified entities MacKey and Lindenmayer [39] mention the complexity of
of them. Consequently, CF representations are increasingly factors that potentially influence patterns of animal distribution
being preferred over discrete land cover classification and conclude on the need to examine spatially distributed
approaches [13], [16]–[18] and considered as a more realistic patterns and processes and the role of environmental constraints
approach to assess land surface properties compared to discrete and resource availability. Austin [40], [41] discusses three types
classes with “hard” boundaries. of ecological gradients (predictors), namely resource (nutrients,
Environmental scientists have applied species distribution water, and light), direct gradients (e.g., temperature and pH) and
modeling (SDM) approaches to link species location information indirect gradients (e.g., geology, topography, and climate).
with environmental data for quite some time [19]. SDM has also Temperature, light, water, and mineral nutrients directly influ-
been referred to as environmental, or species niche modeling, ence both animal physiology and the productivity and structure
habitat suitability modeling or predictive mapping [19]. Franklin of vegetation upon which animals depend for shelter, nutrients,
[19] defines predictive vegetation mapping as predicting the and food [39], [42]. The primary resource water exerts funda-
geographic distribution of vegetation composition across a mental control on biological processes, since it influences the rate
landscape from mapped environmental variables. of biochemical reactions, comprises a basic resource needed to
Predictive vegetation mapping is founded in vegetation sustain physiological functions, and is indispensable for biolog-
gradient analysis and requires available maps of environmental ical growth [39]. Spatial and temporal patterns of movement
variables [19]. Kessel [20] originally described gradient model- relate to an animal’s need for forage (among others), being
ing as a new approach to resource management in Glacier largely driven by primary environmental resources such as
National Park. This work together with studies by Hoffer [21] available water [39], [43].
and Strahler [22], where digital environmental data layers (e.g., Gradient modeling or predictive parameter mapping as out-
topographic variables) were incorporated as ancillary data to lined above is the original concept behind a number of studies
develop forest maps from Landsat data, are considered as first based on RS data that aim to represent ecosystem properties in a
examples of predictive vegetation mapping [19]. Gradient continuous way. A few ecological studies applied CF to assess
modeling or predictive vegetation mapping establishes a rela- plant functional traits [17], [44], floristic gradients [45], forest
tionship between environmental variables that are correlated diversity [46], or peatland patterns [47]. One example of
with environmental or resource gradients and species distribu- CF-based products is the suite of annual land cover products
tion patterns [19]. originating from the moderate resolution imaging spectrometer
Franklin [19] states that predictive vegetation mapping is (MODIS) providing global layers of vegetation CF (VCF) in a
directly and methodologically related to animal habitat modeling 500-m subpixel resolution. Considered land cover classes are
to produce a predictive map of habitat suitability [19]. Environ- bare ground, herbaceous, and tree cover [48], with the latter
mental factors that affect animals include vegetation composition being further split into percentage abundances of evergreen,
and structure. Vegetation type is often the primary variable deciduous, coniferous, and broadleaf cover. The MODIS VCF
driving an animal habitat model because of its direct importance characterization approach has then been used to produce 30 m
for food [19]. VCF layers of the United States [49], to develop a tree cover
Foraging strategies of consumers are the central processes validation data set in Zambia [50], or to complement an approach
studied in resource ecology, the ecology of trophic interactions to analyze fire incidences based on land cover types in South
between consumers and their resources [23]. Trophic interac- America [51].
tions between large ungulates and their primary food resource, In this study, we applied an RS concept to continuously map
the vegetation, are of particular relevance because large ungu- an ecosystem’s total surface water content, indicative for forage
lates 1) spend most of their time feeding; 2) consume high quantity in vegetated areas, and assessed the suitability of these
amounts of biomass; 3) have relatively accurate spatial memory; maps to increase understanding on the habitat use of large
4) are highly mobile and make choices when to feed whereon ungulates. Water content is an indicative proxy to characterize
various spatiotemporal scales; and 5) have a large impact on potential forage quantity because plant growth or biomass
quantitative and qualitative spatiotemporal pattern of vegetation production, respectively, is strongly related to water availability
[4], [23], [24]. In contrast to a subset of African (e.g., [25]–[27]) in ecosystems worldwide [52], [53]. Additionally, water avail-
and North American ecosystems (e.g., [28]–[30]) knowledge ability also determines forage quality since both soil microbial
about niche differentiation of ungulate communities in the activity that mobilizes nutrients and nutrient uptake by the roots
European Alps is still very limited. Although there have been are being reduced when water content in the soil is low [54], [55].
studies investigating either habitat choice [31], [32], diet choice More generally, water content maps may be directly used to
[33]–[35], or population dynamics [36]–[38] of ungulates in the monitor plant water stress caused by droughts e.g., in agriculture
European Alps, a comprehensive picture is missing. In particular, [6], [56]–[58], and to prevent wildfires in forestry [59]–[61].
it remains unclear if and how competition or ecological facilita- Further, such maps might indirectly serve as surrogates for the
tion shapes ungulate communities and foraging strategies in physiological status of plants and the vegetation, respectively
these ecosystems. Consequently, understanding the variability [58], [62], [63]. Photosynthetic activity [64], [65] and plant shoot
or potential limitation of forage quality and quantity in space and quantity (aboveground biomass) [65]–[68] and quality (nutrient
time is essential to explain habitat selection and the role of content) [67], [69], [70] have all been shown to be closely and
2602 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 7, NO. 6, JUNE 2014

TABLE I
APEX PERFORMANCE AND TECHNICAL SPECIFICATIONS FOR THE VNIR AND
SWIR DETECTOR

represent gradients of ecosystem parameters at the landscape


level, e.g., related to the water cycle, carbon cycle (biomass,
NPP), or nutrient cycles (N, P, fiber content, etc.).

II. STUDY AREA AND DATA


A. Swiss National Park (SNP)

Fig. 1. True color composite of a mosaicked APEX scene of upper Val Trupchun The SNP is located in the south-eastern part of Switzerland and
acquired on June 24, 2010. covers an area of . About are covered by
vegetation, with forests occupying , alpine grasslands
significantly related to water content. Thus, water content re- and subalpine grasslands [72], [73]. Elevation
presents a measurable parameter that can be used as a surrogate ranges from 1350 to 3170 m above sea level (asl). Land cover in
the upper Trupchun Valley (Val Trupchun), our specific study
for other plant-related factors that are more difficult to measure.
area (Fig. 1), is composed of alpine grassland communities
Such quantitative and qualitative vegetation characteristics are,
distributed over a large gradient of altitude, rocks, and bare soil,
in addition, important factors in affecting trophic interactions
snow fields depending on the season and open forest stands of
between vegetation as food resource and herbivores as the
Swiss stone pine and larch (Pinus cembra L./Larix decidua Mill.)
primary consumers, particularly influencing foraging patterns
of the latter. up to roughly 2000 m asl.
We used data of the Airborne Prism EXperiment (APEX) The SNP is the largest protected area in Switzerland and the
imaging spectrometer [71], collected over the Central European country’s only national park. It offers a great potential to study
ecosystem processes in the absence of human intervention. The
Alps region of the Swiss National Park (SNP), to map the total
Val Trupchun is famous for its exceptionally high densities of
surface water content of this environment. The specific objec-
chamois (Rupicapra rupicapra L.), ibex (Capra ibex L.) and red
tives of this study were 1) to define and apply a methodological
deer (Cervus elaphus L.).
framework to assess the ecosystem’s total surface water content
on a per-pixel basis; 2) to generate CF maps of two consecutive
B. APEX IS Data
years to assess temporal variations of total surface water content
in the study area; and 3) to assess the potential of linking CF maps APEX IS data sets of two consecutive years were collected
to spatial patterns of habitat use of a multispecies ungulate over the upper Val Trupchun on June 24, 2010 and June 26, 2011.
community. These data sets were part of an attempt to cover large areas of the
Results of this study are considered to serve as a guideline on SNP with APEX IS data on a recurring basis to assess ecosystem
how IS data, in combination with a method of subpixel assess- processes over time. APEX measures the solar reflected radiance
ment of ecosystem properties can support research relying on in the wavelength range from roughly 380 to 2500 nm with up to
quantitative information of spatiotemporal dynamics. More spe- 334 reconfigurable spectral bands in the visible/near infrared
cifically, IS allows the simultaneous acquisition of a large (VNIR) and 198 spectral bands in the shortwave infrared (SWIR)
number of spectrally continuous bands and therefore provides spectral region [71]. The pushbroom scanner’s 1000 spatial
a wealth of data for optimal land cover class specific assessment across track pixels covered a field of view (FOV) of 28 (Table I).
of an ecosystem parameter (i.e., total surface water content). We The study site was imaged with two flight lines at a flight level
claim to look at several CF simultaneously by separating their of 6500 m asl, resulting in across track pixel sizes ranging from
information content using spectral mixture analysis (SMA). 1.5 to 2.4 m, depending on actual terrain height. The pixel size
Attribution of abundances of multiple CF is only possible at was resampled to 2 m. Individual flight lines of APEX IS data
high quality with the advent of having high-dimensional data, were geometrically and radiometrically corrected using the
such as presented. The approach accounts for the complexity of PARGE [74] and ATCOR-4 [75] standard approaches. The used
the landscape and incorporates subpixel information into a CF parametric geocoding approach reconstructs the scanning geom-
approach. Therefore, it may be useful for habitat assessment to etry of each pixel considering various input data, i.e., boresight
monitor ecosystem responses to environmental change or to information, flight positioning data, and terrain elevation data.
KNEUBÜHLER et al.: CF FROM IMAGING SPECTROMETER DATA FOR ECOSYSTEM PARAMETER MAPPING 2603

TABLE II endmember resulted in neglectable impact, most probably due to


COUNTS OF LARGE UNGULATE SPECIES GROUPS IN VAL TRUPCHUN IN SUMMER 2010
AND 2011
a high solar zenith angle of 32 during both data acquisitions in
2010 and 2011, and was therefore abandoned. Due to the lack of
ground truth in our study, an analysis of the statistics of the
residual error image of the SMA provided the only possibility to
assess the quality of the resulting land cover abundance maps.
Based on discrete classes obtained from an additional maximum
likelihood classification of forest stands, grassland, rock/soil,
and snow, the class specific RMSE within the SMA error image
Noncovered areas were filled by data interpolation using bilinear was assessed. The mean band-wise RMSE values were 1.18%
interpolated gaps. This interpolation method allows preserving reflectance ( reflectance) for forest areas (dominant
the original radiometry in mapped areas and only interpolates abundance), 1.66% reflectance ( reflectance) for grass-
data in data gaps. The final pixel size was resampled to 2 m and an land areas (dominant abundance), 1.36% reflectance (
evaluation of the resulting geometric accuracy using 15 ground reflectance) for soil/rock areas (dominant abundance), and 2.03%
control points revealed a total uncertainty of , which reflectance ( reflectance) for snow-covered areas,
corresponds to a total pixel shift of 1.3–1.9 pixels, respectively (2011 data). Water content estimates for the individ-
pixel. The atmospheric correction approach is based on a pre- ual land cover types were retrieved from APEX IS data using
calculated look-up-table, modeled with the atmospheric radiative empirical and physically based approaches (see next section).
transfer (RT) code MODTRAN5. The definition of an atmo- 1) Water Content Estimation: The water content of the four
sphere type and an aerosol model (in our case mid latitude main land cover types present in Val Trupchun (forest, grassland,
summer, and rural aerosol model) in addition with image based rock/soil, and snow) was estimated using spectral indices
estimates of atmospheric water vapor and visibility allows identified from literature, and a physical approach for the
selecting respective LUT entries to compensate atmospheric grassland. The main emphasis in this study was to
effects in measured radiance data and to provide top-of-canopy demonstrate the potential of using continuous information
hemispherical conical reflectance factor (HCRF) data. An synergistically derived from an IS in ecosystem studies rather
evaluation of the radiometric performance using three in situ than fundamentally improving the accuracy of parameter
measured surface reflectances of invariant targets revealed an derivation (i.e., water content) from such data itself.
average RMSE of 15% considering the entire wavelength range Consequently, the choice of indices to estimate the water
(data not shown). Subsequent analyses were performed on content of some land cover types might not yet represent the
mosaicked HCRF data (for terminology see [76]). best possible solution, but was considered accurate for our
purpose. The identification of more robust approaches to
C. Spatial Distribution of Ungulates increase the accuracy would need further assessment.
The SNP’s research department monitored ungulate popula- The specific index to predict water content of snow-covered
tion sizes and distributions within the park for decades using the areas was the Normalized Difference Snow Index (NDSI) [79].
same method. On a day with good visibility, the SNP’s park NDSI was originally developed for the broad spectral bands of
rangers simultaneously mapped the position of all ungulate LANDSAT TM as [80]
groups in Val Trupchun, always using the same observation
points that overview the entire area. We used animal data
collected in summer 2010 and 2011 (Table II) to analyze if where TM2 and TM5 are LANDSAT TM bands 2
there were differences in total surface water content between the ( μ ) and 5 ( μ ). Accordingly, respec-
grassland areas used by the animals from 1 year to the other. tive APEX reflectance data were spectrally convolved to match
the LANDSAT broadband configuration (refer to [81] for a
description of the convolution) and the NDSI was afterwards
III. METHODS applied. For our purpose, we assumed that high NDSI values
A. Continuous Field Mapping indicate high snow cover, which is in turn related to high water
content. In literature, the mean error of subpixel snow cover
The ecosystem parameter total surface water content was
estimation of NDSI is reported to be less than 10% over the entire
mapped using the proposed CF approach. The CF maps combine
range of 0.0–1.0 [79].
subpixel abundance estimates of predominant land cover classes
For forest stands, water content was related to the Normalized
(both vegetated and nonvegetated) in the upper Trupchun Valley
Difference Infrared Index (NDII) [82], which was found to be
(i.e., forest stands, grassland, and rock/soil and snow) and
sensitive to changes in water content of plant canopies. NDII
quantitative ecosystem parameters, in this particular case, total
comprises NIR and SWIR spectral information and is defined
surface water content. Land cover abundances of user defined
as [83]
endmembers of forest stands, grassland, rock/soil, and snow
were derived using constraint linear SMA [7], [77] over the full
set of available APEX bands, and requiring the abundances
within a pixel to sum to unity while still allowing abundances where and are the reflectances at 0.85 and μ ,
to lie below 0 or above 1 [78]. The introduction of a shade respectively. Wavelengths slightly differ in literature. However,
2604 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 7, NO. 6, JUNE 2014

in this study, APEX spectral bands closest to the indicated NDII


bands were used. NDII is reported to be linearly related to canopy
moisture with an [82], and linearly related to canopy
equivalent water thickness for a large range of land cover classes
with an [84].
The estimation of surface soil water content (and potentially
on rocks) was based on the Normalized Soil Moisture Index
(NSMI) [85] as

where and are the reflectances at 1.80 and


μ , respectively. In our case, APEX spectral bands
closest to the indicated NSMI bands were used. A high linear
correlation ( , ) between NSMI de-
rived surface soil moisture and field reference data of upper 5 cm
soil column is reported in literature [85].
The canopy water content of grassland was retrieved by
applying an inversion of the combined one-dimension (1-D)
RT models PROSPECT [86] and SAILh [87]. The estimation of
vegetation variables using model inversion was based on a look-
up-table approach. The leaf level variables chlorophyll content,
dry matter content, and water content as well as the canopy
parameter LAI were kept free within prior defined ranges and
25 000 combinations were randomly sampled assuming uni-
form distributions within the defined parameter range. The
quasi-inversion minimizes a cost function, which corresponds
in our case to the relative RMSE between the measured and the
simulated surface reflectance signal, and considers the entire
wavelength range from 400 to 2500 nm. The canopy water Fig. 2. Generation of CF map of total surface water content.
content was finally calculated as product of the leaf level water
content from PROSPECT and the leaf area index estimate from
SAILh. In literature, an uncertainty of space based canopy respectively. The spectral indices maps were subsequently
water content estimates of various agricultural crops using multiplied with the corresponding land cover abundance
an LUT approach is reported as (15%) and maps. For grassland, the canopy water content map derived
[88]. A second study on canopy water content of from the physical model was multiplied with the grassland
pioneer vegetation and grassland derived from PROSAIL abundance values. The final integration of the individual
simulations and subsequent validation with in situ field spec- water content maps into a combined product resulted in a
trometer data reports uncertainties of (13.6%) map of CF values of total surface water content. This total
and for a homogeneous site and uncertainties of surface water content map represents water content values in
(14%) and for a heterogeneous site, relative units on a per-pixel basis.
respectively [89]. Given the consistency of reported uncer-
B. Animal Positions and Water Content
tainties, we assume similar accuracies in grassland canopy
water content in our study. To evaluate the spatial distributions of ungulates on grasslands
2) Generation of CF Maps of Total Surface Water in Val Trupchun and its relation to total surface water content, CF
Content: The combination of quantitative ecosystem para- values for total surface water content were extracted for every
meters (i.e., total surface water content) and land cover position of an ungulate group mapped in Val Trupchun in
abundance information is the key step to generate the summer 2010 and summer 2011. We investigated the differences
envisaged CF maps (Fig. 2). Essential steps that have to be in total surface water content between both years for all areas
applied before combining all input data are 1) to scale the water used by the single species (chamois, ibex, and red deer), by the
indices values to a common basis and 2) to adjust the SMA two subfamilies Caprinae Gray (chamois and ibex) and Cervinae
derived land cover abundances to be between 0 and 1. The maps Goldfuss (red deer), and by all three species. We also tested for
of spectral indices (i.e., NDSI, NDII, and NSMI) were linearly differences in total surface water content between areas used by
scaled to ranges between 0 and 1 [79], [82], [85], thus depicting ungulates (all three species combined) in 2010 and areas they
low to high water content of the respective indices. Land cover would have used if they were in the positions of 2011, and vice
abundances with values below 0 or above 1 (as a result of versa. Further, we investigated the differences in total surface
constraint linear SMA with per-pixel abundances summing to water content of areas used by the single species and the two
unity while allowing values below 0 or above 1) were set to 0 or 1, subfamilies when data from the 2 years were combined. We
KNEUBÜHLER et al.: CF FROM IMAGING SPECTROMETER DATA FOR ECOSYSTEM PARAMETER MAPPING 2605

Fig. 4. CF map of relative total surface water content for June 24, 2010.
Fig. 3. Subset of a valley slope in Val Trupchun, depicting a true color
representation (left), relative total surface water content based on discrete land
cover classes (middle) and relative total surface water content based on a CF
mapping approach including subpixel abundances of land cover classes (right).

applied the nonparametric Kruskal-Wallis test for all compar-


isons. For spatial data analysis, we used ArcGIS (version 10.0,
Environmental Systems Research Institute, Redlands, CA, US)
and for statistical analysis the freeware R (version 2.15.1, R
Development Core Team, Vienna, AT).

IV. RESULTS
A. CF Maps of Total Surface Water Content
The advantage of using a CF approach that includes subpixel
land cover abundances to map gradients of ecosystem parameters
instead of discrete classification approaches is exemplarily
shown in Fig. 3 for a subset of a steep valley slope in Val
Trupchun. A true color representation (APEX IS data of June 24,
2010) of an area composed of grasslands and soils of varying
cover density, as well as rock and snow patches is given in Fig. 3
(left). Relative total surface water content values derived from Fig. 5. CF map of relative total surface water content for June 26, 2011.
1) a mapping approach based on discrete land cover classes and
2) the CF approach as described in Section III-A are shown in
Fig. 3 (middle and right), respectively. The discrete approach total surface water content amounted to 0.170 for 2010 and to
(Fig. 3, middle) no longer incorporates the subpixel abundance 0.074 for 2011.
estimates of all present land cover classes but only the dominant The availability of multitemporal data sets allowed assessing
class. As a consequence, the respective total surface water spatial differences in derived total surface water content between
content map is subject to a blocky pattern of water content the two dates. Fig. 6 shows the percentage difference in relative
values and continuous gradients are no longer present over total surface water content between the two CF maps of 2010
distinct land cover borders. The CF map (Fig. 3, right), in (Fig. 4) and 2011 (Fig. 5). Roughly 66% of the study area
contrary, contains smooth gradients across land cover classes changed less than 5% in relative total surface water content
and therefore represents the actual situation more realistically. (blueish to reddish colors). Dark blue and dark red denote areas
CF maps of relative total surface water content of the upper Val with positive ( > ) and negative differences ( > ) between
Trupchun for June 24, 2010 and June 26, 2011, respectively (data both years. Large negative differences (red) were mainly found at
values between and high altitudes, whereas large positive differences (blue) were
) are shown in Figs. 4 and 5. The mean relative value of predominantly present below the tree line.
2606 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 7, NO. 6, JUNE 2014

Fig. 8. Boxplots for the relative total surface water content of areas used by the
three ungulate species (chamois, ibex, red deer) combined in Val Trupchun in
summer 2010 and summer 2011. Next to the boxplots for the years (2010, 2011),
the boxplots for the water content of the areas the ungulates would have used if
they were in the same positions as in the other year are shown (2010_2011: water
content of 2010 if the animals were in the 2011 positions, 2011_2010: water
content of 2011 if the animals were in the 2010 positions). Horizontal bars
represent the median, box heights the interquartile range, and whiskers span
interquartile range. Outliers > interquartile range) are not shown. Differ-
ent letters above the whiskers indicate significant differences ( < ) between
the groups.
Fig. 6. Difference map (%) of relative total surface water content between values
of June 24, 2010 and June 26, 2011, respectively.

Fig. 9. Boxplots for the relative total surface water content of areas used by
(a) three ungulate species (chamois, ibex, red deer) and (b) two ungulate
subfamilies (Caprinae, Cervinae) in Val Trupchun in summer 2010 and 2011
combined. Horizontal bars represent the median, box heights the interquartile
range, and whiskers span interquartile range. Outliers ( > interquartile
Fig. 7. Boxplots for the relative total surface water content of areas used by range) are not shown. Different letters above the whiskers indicate significant
(a) three ungulate species (chamois, ibex, red deer), (b) two ungulate subfamilies differences ( < ) between the groups.
(Caprinae, Cervinae) in Val Trupchun in summer 2010 (i.e., 10) and summer
2011 (i.e., 11), respectively. Horizontal bars represent the median, box heights the
interquartile range, and whiskers span interquartile range. Outliers ( > (Kruskal-Wallis , , >
interquartile range) are not shown. Different letters above the whiskers indicate , Fig. 8: compare “2011” with “2011_2010”). When we
significant differences ( < ) between the groups.
combined the total surface water content data from 2010 to 2011,
we neither found differences between the areas used by the
B. Linking the Spatial Distribution of Ungulates to CF Maps
three species [Kruskal-Wallis , ,
No significant yearly differences in total surface water content > , Fig. 9(a)], nor between the areas used by two
were found for grassland areas used by the three species subfamilies [Kruskal-Wallis , ,
[Kruskal-Wallis , , > > , Fig. 9(b)].
, Fig. 7(a)] or the two subfamilies [Kruskal-Wallis
, , > , Fig. 7(b)].
V. DISCUSSION
The yearly difference in total surface water content was signifi-
cant for areas used by all three species (Kruskal-Wallis The concept of using CF allows a realistic representation of
, , < , Fig. 8: ecosystem parameters including their gradients and overcomes
compare “2010” and “2011”). Considering areas used by all limitations of common mapping approaches based on a discre-
ungulate species in 2010 and potentially used if they were in the tization of land surface properties in classes [16]–[18]. A meth-
same positions as in 2011, no significant differences were odological framework was presented to synergistically derive
obvious (Kruskal-Wallis , , continuous relative information of total surface water content
> , Fig. 7 compare “2010” with “2010_2011”). in an alpine ecosystem dominated by forest stands, grassland,
The same applies to areas used by all ungulate species in 2011 soil/rock, and snow patches from IS data. In vegetated areas, total
and potentially used if they were in the same positions as in 2010 surface water content was considered as a surrogate of plant
KNEUBÜHLER et al.: CF FROM IMAGING SPECTROMETER DATA FOR ECOSYSTEM PARAMETER MAPPING 2607

TABLE III Tussilago farfara L.), and grasses (e.g., Dactylis glomerata
PHENOLOGICAL INDICATORS RECORDED IN VAL TRUPCHUN FOR THE YEARS 2010
AND 2011
L.) could be observed (Table III).
Besides snow cover and phenological development of trees
also shadow effects as a result of different illumination condi-
tions during IS data acquisition might determine differences in
total surface water content values between the 2 years. Indeed, IS
data were acquired between 09:30 and 10:00 UTC in 2010 and
between 12:30 and 13:00 UTC in 2011, causing differences in
sun position of sun zenith and sun azimuth
within 1 day, and sun zenith and sun azimuth between
both data acquisitions. The content of high spatial resolution RS
data, including the shadow effect, is known to gain increasing
physiological status. It therefore serves as a proxy to determine importance in data analysis [91]–[93] and needs further assess-
potential forage quantity of grasslands, and its predictive power ment when deriving ecosystem parameters.
to trace movement patterns of large ungulates was assessed. Maps of total surface water content shown in this study are
given in relative units. This is caused by the absence of reference
A. CF of Total Surface Water Content data of absolute water content, which hinders to relate the current
Multitemporal CF maps of total surface water content revealed relative values to absolute surface water content values. The
changes in water content over time. The presented difference relative characterization of total surface water content, however,
map (Fig. 6) shows small-scale, subtle changes of max. 5% is considered to be sufficient for this first exploration on the
mainly in grasslands and, on the other hand, large variations in applicability of the presented CF concept to represent ecosystem
total surface water content between June 24, 2010 and June 26, properties synergistically derived from IS data. Further, only a
2011 for forested and high elevation areas. For snow-covered visual validation of the results was possible to ensure that
areas, large differences in total surface water content between the the algorithm did not lead to potentially unrealistic results.
2 years are associated with the presence or absence of snow A quantitative validation requires extensive efforts considering
patches in respective areas. Although IS data were acquired on the water content of all predominant land cover types including a
almost identical days of the year (DOY, i.e., June 24 and June 26) quantitative assessment of their respective abundances. In future
in two consecutive years, considerably less snow was present work, where maps are intended to be assimilated in e.g., process
in the high altitudes in 2011 compared to 2010 (dark red areas in models, absolute quantifications, and validations are required.
Fig. 6). Further, Fig. 6 depicts large positive differences (blue) in
B. Relation Between CF Maps and Animal Distribution
total surface water content for forest-covered areas. These dif-
ferences can be associated with interannual phenological differ- This study also assessed the suitability of CF maps of ecosys-
ences in tree development, mainly visible in the development tem parameters—in our case total surface water content—to
stages of larch (Larix decidua Mill.). Indeed, phenological understand animal distribution and grassland habitat use in the
indicators recorded in Val Trupchun on a regular basis confirm SNP. The underlying assumption was that forage quantity or
an earlier onset of needle development in both, larch and pine biomass production of grassland can be characterized by the
in 2011 (Table III). Further, meteorological data from the growth limiting factor water. Our results indicate that the spatial
MeteoSwiss weather station Buffalora (1980 m asl). i.e., growing distributions of ungulates in summer 2010 and 2011 were not
degree days with temperatures above 5 C (GDD5) [90], daily related to the total surface water content of the grassland areas as
total sunshine duration, and daily total precipitation for the represented in the respective CF maps. Distinct spatial patterns of
period between January to June also suggest differences in the habitat use by single species (chamois, ibex, and red deer) and
phenological development in both years. GDD5 amounted to total surface water content in grassland could neither be found
116.5 in 2010 (starting on DOY 114) and to 159.4 in 2011 within a summer nor between the two summers. The same holds
(starting on DOY 97) from January until the respective dates of true for ungulate subfamilies (Caprinae and Cervinae). Analysis
APEX data acquisition. The number of days with more than 6 of both single species and subfamilies data for summer 2010 and
hours of direct sunlight from the first GDD5 to the date of APEX 2011 in combination did not reveal specific patterns in habitat
data collection amounted to 19 days (185 h) in 2010 and to 41 use, either. Only the comparison of ungulate positions of all three
days (395 h) in 2011. Total precipitation between January 01 and species were related to significant differences in total surface
the date of APEX data acquisition accumulated to 259 mm in water content of used grassland patches between 2010 and 2011.
2010 and to 292 mm in 2011. The higher number of GDD5, A hypothetical combination of summer 2011 animals positions
higher amounts of direct irradiance, and the increased availabili- and 2010 total surface water content data (and vice versa)
ty of water all suggest that climatic conditions allowed advanced revealed, however, that the same total surface water content
phenological development in spring 2011 compared to 2010. (predominantly contained in forage) could have been achieved
The early onset of the vegetation period in 2011 was probably given the animals had visited the same areas in both years. Given
also a main factor influencing the development of the vegetation the limited number of observations, these findings could indicate
later in the season, when an earlier bloom of shrubs (e.g., Rowan that 1) water availability of the grasslands in Val Trupchun is not
Berry, Sorbus aucuparia Michx.), herbs (e.g., Coltsfoot, a dominant limiting factor for potential forage quantity (biomass)
2608 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 7, NO. 6, JUNE 2014

compared to e.g., temperature, light, and nutrients, or that and [3] A. K. Skidmore and J. G. Ferwerda, “Resource distribution and dynamics:
2) ungulates choose their grazing sites based on other criteria. Mapping herbivore resources,” in Resource Ecology, Spatial and Temporal
Dynamics of Foraging, H. H. T. Prins and F. Van Langevelde, Eds. New
Several studies on plant-herbivore interactions in the SNP have York, NY, USA: Springer-Verlag, 2008, pp. 57–77.
shown competition on forage of highest nutritious quality [73], [4] D. W. Bailey et al., “Mechanisms that result in large herbivore grazing
[94]. With phosphorus (P) being the limiting nutrient in these distribution patterns,” J. Range Manage., vol. 49, pp. 386–400, 1996.
[5] J. M. Fryxell, J. F. Wilmshurst, and A. R. E. Sinclair, “Predictive models of
grasslands, red deer prefer the most P-rich parts for grazing, movement by Serengeti grazers,” Ecology, vol. 85, pp. 2429–2435, 2004.
leading to grazing patterns closely related to patterns of soil-P [6] S. Ustin, D. A. Roberts, J. A. Gamon, G. P. Asner, and R. O. Green, “Using
[94]. Grazing adapted vegetation was found to depict a positive imaging spectroscopy to study ecosystem processes and properties,”
Bioscience, vol. 54, pp. 523–534, 2004.
relationship between soil-P and the P concentration in leaf tissues [7] J. W. Boardman, “Automated spectral unmixing of AVIRIS data using
[95]. Consequently, herbivores repeatedly visit specific grazing convex geometry concepts,” in Proc. Summaries 4th JPL Airborne Geosci.
sites of high nutrients quality rather than forage quantity. This may Workshop, Washington, DC, USA, JPL Publication 93-26 V.1, 1993,
pp. 11–14.
explain why we could not find differences in grazing patterns [8] M. E. Schaepman et al., “Earth system science related imaging
between the two summers. Besides, the issue of species competi- spectroscopy—An assessment,” Remote Sens. Environ., vol. 113,
tion on popular grazing sites [73] needs further investigation. pp. S123–S137, 2009.
[9] B. O. Oindo, A. K. Skidmore, and P. de Salvo, “Mapping habitat and
Since the multitemporal CF maps derived from APEX IS data biological diversity in the Maasai Mara ecosystem,” Int. J. Remote Sens.,
were found to realistically represent fine-scaled spatiotemporal vol. 24, pp. 1053–1069, Jan. 1, 2003.
pattern of ecosystem parameters, it appears promising to contin- [10] E. Leyequien et al., “Capturing the fugitive: Applying remote sensing to
terrestrial animal distribution and diversity,” Int. J. Appl. Earth Observ.
ue our analyses of animal distribution patterns in the SNP. We Geoinf., vol. 9, pp. 1–20, 2007.
plan to further develop CF maps from IS data for plant biomass, [11] X. Feng, B. Fu, X. Yang, and Y. Lü, “Remote sensing of ecosystem services:
adapt the methodology to other traits, e.g., plant tissue N-content, An opportunity for spatially explicit assessment,” Chin. Geogr. Sci., vol. 20,
pp. 522–535, 2010.
and acquire ungulate position data on a monthly basis. CF maps [12] B. Burkhard, F. Kroll, S. Nedkov, and F. Müller, “Mapping ecosystem
of biomass and N-content will enable to characterize the ecosys- service supply, demand and budgets,” Ecol. Indic., vol. 21, pp. 17–29, 2012.
tem in a more detailed way and to test whether grassland areas of [13] R. S. DeFries, J. R. G. Townsend, and M. C. Hansen, “Continuous fields of
vegetation characteristics at the global scale at 1-km resolution,” JGR,
high total surface water content do offer high forage quantity and vol. 104, pp. 16911–16923, 1999.
quality. Monthly ungulate positioning data will provide a more [14] L. Mathys, A. Guisan, T. W. Kellenberger, and N. E. Zimmermann,
robust comparison between the IS data and the spatial distribu- “Evaluating effects of spectral training data distribution on continuous field
mapping performance,” ISPRS J. Photogramm. Remote Sens., vol. 64,
tion of animals. pp. 665–673, 2009.
[15] R. S. DeFries et al., “Mapping the land surface for global atmosphere-
biosphere models: Toward continuous distributions of vegetation’s func-
tional properties,” J. Geophys. Res., vol. 100, pp. 20867–20882, 1995.
VI. CONCLUSION [16] M. Schwarz and N. E. Zimmermann, “A new GLM-based method for
mapping tree cover continuous fields using regional MODIS reflectance
In this study, we applied a concept of CF maps synergistically data,” Remote Sens. Environ., vol. 95, pp. 428–443, 2005.
derived from IS data to describe ecosystem properties including [17] S. L. Ustin and J. A. Gamon, “Remote sensing of plant functional types,”
their gradients across different land cover types. The continuous New Phytol., vol. 186, pp. 795–816, 2010.
[18] M. C. Levin et al., “Moving from discrete to continuous landscape metrics
representation of total surface water content in a high mountain using remote sensing: The ecological significance of paddock trees for bird
environment bears the potential for improved assessment and richness,” in Proc. Int. Congr. Model. Simul. (MODSIM’07), Dec. 2007,
understanding of spatiotemporal dynamics using change-vector pp. 1342–1348.
[19] J. Franklin, “Predictive vegetation mapping: Geographic modelling of
analysis. The detailed information contained in CF maps can serve biospatial patterns in relation to environmental gradients,” Progress Phys.
as a spatially explicit input for habitat assessment in ecology. Geogr., vol. 19, pp. 474–499, 1995.
Today, the quality of IS data and the methods for subsequent [20] S. Kessell, “Gradient modeling: A new approach to fire modeling and
wilderness resource management,” Environ. Manage., vol. 1, pp. 39–48,
parameter retrieval do no longer set the limits of vegetation- 1976.
ungulates research on plant ecology, but rather on the scarcity [21] R. M. E. A. Hoffer, “Natural resource mapping in mountainous terrain by
of accurate and extensive data on animal habitat use. IS data computer analysis of ERTS-1 satellite data,” Purdue Univ., West Lafayette,
IN, USA, Agricultural Experimental Station Research Bulletin 919 and
contain a wealth of spectral information and allow retrieving a LARS Contract Rep. 061575, 1975.
variety of biochemical and structural vegetation properties. Such [22] A. H. Strahler, “Stratification of natural vegetation for forest and rangeland
vegetation characteristics can be combined in CF maps to repre- inventory using Landsat digital imagery and collateral data,” Int. J. Remote
Sens., vol. 2, pp. 15–41, 1981.
sent a wide range of ecosystem parameters, e.g., related to the [23] F. Van Langevelde and H. H. Prins, “Structuring herbivore communities:
water cycle, carbon (biomass, NPP), or nutrients cycles (N, P, The role of habitat and diet,” in Resource Ecology, Spatial and Temporal
etc.). Longer-term multitemporal studies and spatially and tem- Dynamics of Foraging, H. H. T. Prins and F. Van Langevelde, Eds. New
York, NY, USA: Springer-Verlag, 2008, pp. 1–6.
porally improved animal positioning data will help to evaluate the [24] N. De Jager and J. Pastor, “Declines in moose population density at Isle
robustness of correlations between ecosystem parameters repre- Royale National Park, MI, USA and accompanied changes in landscape
sented by CF maps and animal distribution patterns in the SNP. patterns,” Landscape Ecol., vol. 24, pp. 1389–1403, 2009.
[25] J. P. G. M. Cromsigt, H. H. T. Prins, and H. Olff, “Habitat heterogeneity as a
driver of ungulate diversity and distribution patterns: Interaction of body
mass and digestive strategy,” Diversity Distrib., vol. 15, pp. 513–522, 2009.
REFERENCES [26] M. M. Voeten and H. H. T. Prins, “Resource partitioning between sympatric
wild and domestic herbivores in the Tarangire region of Tanzania,”
[1] D. M. Muchoney, “Earth observations for terrestrial biodiversity and Oecologia, vol. 120, pp. 287–294, 1999.
ecosystems,” Remote Sens. Environ., vol. 112, pp. 1909–1911, 2008. [27] T. M. Anderson et al., “Landscape-scale analyses suggest both nutrient and
[2] A. Guisan and N. E. Zimmermann, “Predictive habitat distribution models antipredator advantages to Serengeti herbivore hotspots,” Ecology, vol. 91,
in ecology,” Ecol. Model., vol. 135, pp. 147–186, 2000. pp. 1519–1529, May 2010.
KNEUBÜHLER et al.: CF FROM IMAGING SPECTROMETER DATA FOR ECOSYSTEM PARAMETER MAPPING 2609

[28] A. Kittle, J. Fryxell, G. Desy, and J. Hamr, “The scale-dependent impact of [55] J. Kreuzwieser and A. Gessler, “Global climate change and tree nutrition:
wolf predation risk on resource selection by three sympatric ungulates,” Influence of water availability,” Tree Physiol., vol. 30, pp. 1221–1234, 2010.
Oecologia, vol. 157, pp. 163–175, 2008. [56] T. J. Jackson et al., “Vegetation water content mapping using Landsat data
[29] K. J. Jenkins and R. G. Wright, “Resource partitioning and competition derived normalized difference water index for corn and soybeans,” Remote
among cervids in the Northern Rocky Mountains,” J. Appl. Ecol., vol. 25, Sens. Environ., vol. 92, pp. 475–482, 2004.
pp. 11–24, 1988. [57] S. S. Choudhary, P. K. Garg, and S. K. Ghosh, “Mapping of agriculture
[30] F. J. Singer and J. E. Norland, “Niche relationships within a guild of drought using remote sensing and GIS,” Int. J. Sci. Eng. Technol., vol. 1,
ungulate species in Yellowstone National Park, Wyoming, following pp. 149–157, 2012.
release from artificial controls,” Can. J. Zool., vol. 72, pp. 1383–1394, 1994. [58] J. Peñuelas and I. Filella, “Visible and near-infrared reflectance techniques
[31] S. Grignolio, I. Rossi, B. Bassano, F. Parrini, and M. Apollonio, “Seasonal for diagnosing plant physiological status,” Trends Plant Sci., vol. 3,
variations of spatial behaviour in female Alpine ibex (Capra ibex ibex) pp. 151–156, 1998.
in relation to climatic conditions and age,” Ethol. Ecol. Evol., vol. 16, [59] B. Koetz et al., “Radiative transfer modeling within a heterogeneous canopy
pp. 255–264, 2004. for estimation of forest fire fuel properties,” Remote Sens. Environ., vol. 92,
[32] S. Lovari, F. Sacconi, and G. Trivellini, “Do alternative strategies of space pp. 332–344, 2004.
use occur in male Alpine chamois?” Ethol. Ecol. Evol., vol. 18, pp. 221–231, [60] E. Chuvieco, D. Riaño, I. Aguado, and D. Cocero, “Estimation of fuel
2006. moisture content from multitemporal analysis of Landsat Thematic Mapper
[33] S. Bertolino, N. C. Di Montezemolo, and B. Bassano, “Food–niche relation- reflectance data: Applications in fire danger assessment,” Int. J. Remote
ships within a guild of alpine ungulates including an introduced species,” Sens., vol. 23, pp. 2145–2162, 2002.
J. Zool., vol. 277, pp. 63–69, 2009. [61] F. M. Danson and P. Bowyer, “Estimating live fuel moisture content from
[34] V. La Morgia and B. Bassano, “Feeding habits, forage selection, and diet remotely sensed reflectance,” Remote Sens. Environ., vol. 92, pp. 309–321,
overlap in Alpine chamois (Rupicapra rupicapra L.) and domestic sheep,” 2004.
Ecol. Res., vol. 24, pp. 1043–1050, 2009. [62] P. Ceccato, S. Flasse, S. Tarantola, S. Jacquemoud, and J.-M. Grégoire,
[35] J. Schröder and W. Schröder, “Niche breadth and overlap in red deer Cervus “Detecting vegetation leaf water content using reflectance in the optical
elaphus, roe deer Capreolus capreolus and chamois Rupicapra rupicapra,” domain,” Remote Sens. Environ., vol. 77, pp. 22–33, 2001.
Acta Zool. Fennica, vol. 172, pp. 85–86, 1984. [63] B. Datt, “Remote sensing of water content in Eucalyptus leaves,” Aust. J.
[36] R. Lande et al., “Estimating density dependence from population time series Bot., vol. 47, pp. 909–923, 1999.
using demographic theory and life-history data,” Amer. Nat., vol. 159, [64] H. G. Jones, “Monitoring plant and soil water status: Established and novel
pp. 321–337, 2002. methods revisited and their relevance to studies of drought tolerance,”
[37] B.-E. Saether et al., “Stochastic population dynamics of an introduced Swiss J. Exp. Bot., vol. 58, pp. 119–130, 2007.
population of the ibex,” Ecology, vol. 83, pp. 3457–3465, 2002. [65] M. L. Thomey et al., “Effect of precipitation variability on net primary
[38] B.-E. Saether, M. Lillegård, V. Grøtan, F. Filli, and S. Engen, “Predicting production and soil respiration in a Chihuahuan Desert grassland,” Global
fluctuations of reintroduced ibex populations: The importance of density Change Biol., vol. 17, pp. 1505–1515, 2011.
dependence, environmental stochasticity and uncertain population esti- [66] C. Körner, “Water, nutrient and carbon relations,” in Alpine Treelines.
mates,” J. Anim. Ecol., vol. 76, pp. 326–336, 2007. New York, NY, USA: Springer-Verlag, 2012, pp. 151–168.
[39] B. G. MacKey and D. B. Lindenmayer, “Towards a hierarchical framework [67] D. S. Wilson and D. A. Maguire, “Environmental basis of soil-site produc-
for modelling the spatial distribution of animals,” J. Biogeogr., vol. 28, tivity relationships in ponderosa pine,” Ecol. Monogr., vol. 79, pp. 595–617,
pp. 1147–1166, 2001. 2009.
[40] M. P. Austin, “Continuum concept, ordination methods, and niche theory,” [68] H. Yang et al., “Plant community responses to nitrogen addition and
Annu. Rev. Ecol. Syst., vol. 16, pp. 39–61, 1985. increased precipitation: The importance of water availability and species
[41] M. P. Austin, “Spatial prediction of species distribution: An interface traits,” Global Change Biol., vol. 17, pp. 2936–2944, 2011.
between ecological theory and statistical modelling,” Ecol. Model., vol. 157, [69] K. Djaman, S. Irmak, D. L. Martin, R. B. Ferguson, and M. L. Bernards,
pp. 101–118, 2002. “Plant nutrient uptake and soil nutrient dynamics under full and limited
[42] A. Guisan and N. E. Zimmermann, “Predictive habitat distribution models irrigation and rainfed maize production,” Agron. J., vol. 105, pp. 527–538,
in ecology,” Ecol. Modell., vol. 135, pp. 147–186, 2000. 2013.
[43] A. Guisan and W. Thuiller, “Predicting species distribution: Offering more [70] P. V. V. Prasad, S. A. Staggenborg, and Z. Ristic, “Impacts of drought and/
than simple habitat models,” Ecol. Lett., vol. 8, pp. 993–1009, 2005. or heat stress on physiologial, developmental, growth, and yield processes
[44] L. Homolová, Z. Malenovský, M. E. Schaepman, G. García-Santos, and of crop plants,” in Response of Crops to Limited Water, L. R. Ahuja, Ed.
J. G. P. W. Clevers, “Review of optical-based remote sensing for plant trait Madison, WI, USA: American Society of Agronomy, 2008, pp. 301–356.
mapping,” Ecol. Complexity, vol. 15, pp. 1–16, 2013. [71] M. Jehle et al., “APEX—Current status, performance and product genera-
[45] H. Feilhauer, U. Faude, and S. Schmidtlein, “Combining isomap ordination tion,” in Proc. IEEE Sens. Conf., Waikoloa, HI, USA, Nov. 1–4, 2010,
and imaging spectroscopy to map continuous floristic gradients in a hetero- pp. 533–537.
geneous landscape,” Remote Sens. Environ., vol. 115, pp. 2513–2524, 2011. [72] A. C. Risch, M. F. Jurgensen, D. S. Page-Dumroese, O. Wildi, and
[46] H. Feilhauer and S. Schmidtlein, “Mapping continuous fields of forest alpha M. Schütz, “Long-term development of above- and belowground carbon
and beta diversity,” Appl. Veg. Sci., vol. 12, pp. 429–439, 2009. stocks following land-use change in subalpine ecosystems of the Swiss
[47] O. N. Krankina et al., “Meeting the challenge of mapping peatlands with National Park,” Can. J. Forest Res., vol. 38, pp. 1590–1602, 2008.
remotely sensed data,” Biogeosci. Discuss., vol. 5, pp. 2075–2101, 2008. [73] M. Schütz, A. C. Risch, E. Leuzinger, B. O. Krüsi, and G. Achermann,
[48] M. C. Hansen et al., “Towards an operational MODIS continuous field of “Impact of red deer (Cervus elaphus L.) on patterns and processes on
percent tree cover algorithm: Examples using AVHRR and MODIS data,” subalpine grasslands in the Swiss National Park,” Forest Ecol. Manage.,
Remote Sens. Environ., vol. 83, pp. 303–319, 2002. vol. 181, pp. 177–188, 2003.
[49] M. C. Hansen et al., “Continuous fields of land cover for the conterminous [74] D. Schläpfer and R. Richter, “Geo-atmospheric processing of airborne
United States using Landsat data: First results from the Web-Enabled Landsat imaging spectrometry data. Part 1: Parametric orthorectification,” Int. J.
Data (WELD) project,” Remote Sens. Lett., vol. 2, pp. 279–288, 2010. Remote Sens., vol. 23, pp. 2609–2630, 2002.
[50] M. C. Hansen, R. S. DeFries, J. R. G. Townsend, L. Marufu, and R. Sohlberg, [75] R. Richter and D. Schläpfer, “Geo-atmospheric processing of airborne
“Development of a MODIS tree cover validation data set for Western imaging spectrometer data. Part 2: Atmospheric/topographic correction,”
Province, Zambia,” Remote Sens. Environ., vol. 83, pp. 320–335, 2002. Int. J. Remote Sens., vol. 23, pp. 2631–2649, 2002.
[51] F. D. S. Cardozo, Y. E. Shimabukuro, G. Pereira, and F. B. Silva, “Using [76] G. Schaepman-Strub, M. E. Schaepman, T. H. Painter, S. Dangel, and
remote sensing products for environmental analysis in South America,” J. V. Martonchik, “Reflectance quantities in optical remote sensing—
Remote Sens., vol. 3, pp. 2110–2127, 2011. Definitions and case studies,” Remote Sens. Environ., vol. 103, pp. 27–42,
[52] Z. Wu, P. Dijkstra, G. W. Koch, J. Peñuelas, and B. A. Hungate, “Responses 2006.
of terrestrial ecosystems to temperature and precipitation change: A meta- [77] J. W. Boardman, “Inversion of high spectral resolution data,” in Proc. SPIE,
analysis of experimental manipulation,” Global Change Biol., vol. 17, 1990, vol. 1298, pp. 222–233.
pp. 927–942, 2011. [78] F. Van Der Meer and S. M. De Jong, “Improving the results of spectral
[53] C. Boisvenue and S. W. Running, “Impacts of climate change on natural unmixing of Landsat Thematic Mapper imagery by enhancing the orthogo-
forest productivity—Evidence since the middle of the 20th century,” Global nality of end-members,” Int. J. Remote Sens., vol. 21, pp. 2781–2797, 2000.
Change Biol., vol. 12, pp. 862–882, 2006. [79] V. V. Salomonson and I. Appel, “Estimating fractional snow cover from
[54] H. Rennenberg et al., “Physiological responses of forest trees to heat and MODIS using the normalized difference snow index,” Remote Sens.
drought,” Plant Biol., vol. 8, pp. 556–571, 2006. Environ., vol. 89, pp. 351–360, 2004.
2610 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 7, NO. 6, JUNE 2014

[80] J. Dozier, “Spectral signature of alpine snow cover from the Landsat Anna-Katharina Schweiger received the B.Sc. de-
Thematic Mapper,” Remote Sens. Environ., vol. 28, pp. 9–22, 1989. gree in biodiversity and ecology from the University
[81] P. D’Odorico, A. Gonsamo, A. Damm, and M. E. Schaepman, “Experimental of Graz, Graz, Austria, in 2007, and the M.Sc. degree
evaluation of Sentinel-2 spectral response functions for NDVI time- in wildlife ecology from the University of Natural
series continuity,” IEEE Trans. Geosci. Remote Sens., vol. 51, no. 3, Resources and Life Sciences, Vienna, Austria, in
pp. 1336–1348, Mar. 2013. 2010. Currently, she is pursuing the Ph.D. degree in
[82] M. A. Hardisky, V. Klemas, and R. M. Smart, “The influences of soil grassland ecology at the Remote Sensing Laborato-
salinity, growth form and leaf moisture on the spectral reflectance of ries, University of Zurich, Zurich, Switzerland.
Spartina alterniflora canopies,” Photogramm. Eng. Remote Sens., vol. 49, She works in close collaboration with the Plant-
pp. 77–83, 1983. Herbivore Interactions Group, Swiss Federal Institute
[83] M. T. Yilmaz, E. R. Hunt, Jr., and T. J. Jackson, “Remote sensing of for Forest, Snow and Landscape Research WSL,
vegetation water content from equivalent water thickness using satellite Birmensdorf, Switzerland, and with the Swiss National Park, Zernez,
imagery,” Remote Sens. Environ., vol. 112, pp. 2514–2522, 2008. Switzerland, where she holds a position with the Research and Geoinformation
[84] M. T. Yilmaz et al., “Vegetation water content during SMEX04 from Department. Her research interest is on the analysis of animal tracking data with
ground data and Landsat 5 Thematic Mapper imagery,” Remote Sens. remote sensing techniques.
Environ., vol. 112, pp. 350–362, 2008.
[85] S.-N. Haubrock, S. Chabrillat, M. Kuhnert, P. Hostert, and H. Kaufmann,
“Surface soil moisture quantification and validation based on hyperspectral
data and field measurements,” J. Appl. Remote Sens., vol. 2, no. 1, 023552, Anita C. Risch received the M.Sc. degree in biology
2008, doi:10.1117/1.3059191. from the Swiss Federal Institute of Technology
[86] S. Jacquemoud and F. Baret, “PROSPECT: A model of leaf optical (ETHZ), Zurich, Switzerland, in 1999, the M.Sc.
properties spectra,” Remote Sens. Environ., vol. 34, pp. 75–91, 1990. degree in forest ecology from Michigan Technologi-
[87] W. Verhoef, “Light scattering by leaf layers with application to canopy cal University (MTU), Houghton, MI, USA, in 2000,
reflectance modeling: The SAIL model,” Remote Sens. Environ., vol. 16, and the Ph.D. degree in ecology, from ETHZ, in 2004.
pp. 125–141, 1984. From 2004 to 2006, she spent her Postdoctoral
[88] J. Cernicharo, A. Verger, and F. Camacho, “Empirical and physical time with Syracuse University, Syracuse, NY, USA.
estimation of canopy water content from CHRIS/PROBA data,” Remote In 2006, she became the Head of the Animal Ecology
Sens., vol. 5, pp. 5265–5284, 2013. Research Group, Swiss Federal Institute for Forest,
[89] J. G. P. W. Clevers, L. Kooistra, and M. E. Schaepman, “Estimating canopy Snow and Landscape Research (WSL), Birmensdorf,
water content using hyperspectral remote sensing data,” Int. J. Appl. Earth Switzerland. Since 2008, she has been a Graduate Faculty and Adjunct Associate
Observ. Geoinf., vol. 12, pp. 119–125, 2010. Professor at MTU and completed her habilitation at ETHZ in 2010. She currently
[90] L. Moser et al., “Timing and duration of European larch growing season leads the Plant-Animal-Interaction Research Group at WSL. Her research inter-
along altitudinal gradients in the Swiss Alps,” Tree Physiol., vol. 30, ests include how vertebrate and invertebrate animals affect above- and below-
pp. 225–233, 2010. ground ecosystem processes.
[91] C. E. Woodcock and A. H. Strahler, “The factor of scale in remote sensing,”
Remote Sens. Environ., vol. 21, pp. 311–332, 1987.
[92] G. P. Asner and A. S. Warner, “Canopy shadow in IKONOS satellite
observations of tropical forests and savannas,” Remote Sens. Environ., Martin Schütz received the M.Sc. and Ph.D. degrees
vol. 87, pp. 521–533, 2003. in biology from the Swiss Federal Institute of Tech-
[93] P. M. Dare, “Shadow analysis in high-resolution satellite imagery of urban nology (ETHZ), Zurich, Switzerland, in 1983 and
areas,” Photogramm. Eng. Remote Sens., vol. 71, pp. 169–177, 2005. 1988, respectively.
[94] M. Schütz et al., “Phosphorus translocation by red deer on a subalpine Since 1994, he was the Leader of several research
grassland in the central European Alps,” Ecosystems, vol. 9, pp. 624–633, groups with the Swiss Federal Institute for Forest,
2006. Snow and Landscape Research (WSL), Birmensdorf,
[95] C. Thiel-Egenter et al., “Response of a subalpine grassland to simulated Switzerland, and became the Senior Scientist in 2006.
grazing: Aboveground productivity along soil phosphorus gradients,” Since 2008, he has been the Adjunct Professor with
Community Ecol., vol. 8, pp. 111–117, 2007. Michigan Technological University (MTU), Houghton,
MI, USA, and a Member of the Board of Science of
the Swiss National Park. His research interests include assessing vegetation
Mathias Kneubühler received the M.S. degree in
dynamics and interactions between plants and animals.
geography and the Ph.D. degree (Dr. sc. nat.) in
remote sensing with emphasis on spectral assessment
of crop phenology, both from the University of Zürich,
Zürich, Switzerland, in 1996 and 2002, respectively. Michael E. Schaepman (M’05–SM’07) received the
He is currently the Head of the Spectroscopy M.Sc. and Ph.D. degrees in geography from the
Laboratory of the Remote Sensing Laboratories University of Zuürich (UZH), Zuürich, Switzerland,
(RSL), University of Zurich. His research interests in 1993 and 1998, respectively.
include spectro-directional data analysis and spectral In 1999, he spent his Postdoctoral time with the
assessment of phenological processes in vegetated Optical Sciences Center, The University of Arizona,
ecosystems. Tucson, AZ, USA. In 2000, he was appointed as a
Project Manager of the European Space Agency
Airborne Prism Experiment Spectrometer. In 2003,
Alexander Damm received the Diploma (M.Sc.) and
he accepted the position of Full Chair of Geoinforma-
Ph.D. degrees in geography from the Humboldt
tion Science and Remote Sensing, Wageningen Uni-
University, Berlin, Germany, in 2004 and 2008,
versity, Wageningen, The Netherlands. In 2009, he was appointed as the Full
respectively.
Chair of Remote Sensing, UZH, where he is currently heading the Remote
From December 2008 to 2012, he was as a Post-
Sensing Laboratories, Department of Geography. His research interests include
doctoral Researcher and since 2012, he has been a
computational Earth sciences using remote sensing and physical models, with
Lecturer with the Remote Sensing Laboratories, Uni-
particular focus on the land–atmosphere interface using imaging spectroscopy.
versity of Zürich, Zürich, Switzerland. He acted as the
responsible Project Manager for the largest APEX
(Airborne Prism Experiment) exploitation program
aiming at supporting advanced product development
using spectroscopy-based approaches. Currently, he is responsible for the Swiss
Earth Observatory Network (SEON), a competence center to monitor status and
functioning of ecosystems in a changing environment. His research interests
include Earth observation, with particular focus on biosphere-atmosphere inter-
actions and ecosystem functioning using imaging spectroscopy.

Anda mungkin juga menyukai