Figure 4 . Spectral signature at varying moisture saturation
levels measured as a fraction of the saturated water mass of S.
capillifolium (left) and S. lenense (right). Spectra for both
species were collected at regular intervals using a SVC - HR-1024i with
light source at 100% under a progressive drying experiment. Fraction
calculated as mass of water in samples divided by total water mass (g
H2O at interval * g total
H2O-1). Sentinel-2 bandpasses are
indicated with vertical bars to illustrate the advantage of imaging
spectrometers with contiguous bands over multispectral instruments.
In addition to spectral changes, metabolic activity of bryophytes is
also significantly influenced by moisture content with primary
production decreasing as moisture decreases (Green & Lange, 1995);
however, decoupling of reflectance and productivity has been noted inSphagnum spp. and Pleurocarpous mosses, such as Hylocomium
splendens and Pleurozium schreberi with spectral indices such as
NDVI returning to near-initial values within minutes after rehydration,
but primary production response lagging for more than 24 hours (May et
al., 2018).
Given the generally low canopy cover across the arctic, bryophytes are
likely driving spectral reflectance of mixed pixels across large
extents, making timing of data collection and awareness of moisture
content crucial for interpreting IS observations. For this reason, early
and late summer provide opportunities for IS of bryophytes. Though there
are many lab studies of bryophyte physiology (Green & Lange, 1995), the
few studies scaling bryophyte spectral signatures for classification and
chemical analysis show promise for estimating water, N, C, and P
(Thomson et al., 2021). Translating bryophyte spectra to trait maps
using remote sensing is an important opportunity to better constrain
ecosystem models (Wullschleger et al., 2014).
Bryophyte reflectance spectra differ from vascular vegetation by
exhibiting a wider and taller peak in the green to yellow, a gentler red
edge, and a greater variability in the NIR (Figure 3). Additionally, the
SWIR region is very responsive to moisture content with large increases
in reflectance under drier conditions. Bryophytes also produce
photoprotective compounds that influence the spectral profiles. For
example, many Sphagnum species under high light conditions may
develop photoprotective pigments that will affect their reflectance.
Studies of open-growing Sphagnum have shown that they are
photo-inhibited in full sun and exhibit faster vertical growth under
lower (e.g., shaded) illumination (Harley et al., 1989; Murray et al.,
1993) Little is known about the variability of pigments among bryophytes
species across the extent of the Arctic. Reflectance measurementsin situ indicate broad diversity both within and among bryophyte
species that will be further complicated by the impact of variable
hydration status.
Though short in stature, bryophytes can form small but highly visible
homogeneous patches, carpets and hummocks. Bryophyte mixtures are very
commonly intermixed with vascular plants (dwarf shrubs and grass-like
plants) and lichens, in the understory living-mat matrix. The mixtures
of patch sizes of each species and degree of heterogeneity combined with
vascular plant canopy cover make it challenging to separate them
spectrally. Similar to lichens, classification accuracy of bryophytes
can be high if pixels are small (< 1 m) and there are
sufficient and appropriate bandpasses (Räsänen & Virtanen, 2019). For
context, researchers found that increasing to 20 m pixels reduced the
absolute accuracy of their plant classification of remotely sensed
spectra by 50% compared to 2 m pixels (Thomson et al., 2021) . Like
lichens, small patch sizes of bryophytes present a challenge for SBG
that will need to be met with scaling studies to understand within-pixel
variation.
Bryophytes generally do not display strong seasonal patterns in their
reflectance, although there are few studies of pure bryophyte patch
phenology. Vegetation classes with high fractional cover by bryophytes
do show some phenological variability but this is likely primarily due
to the non-bryophyte fraction in the vegetation class (Rautiainen et
al., 2011). In the spring, following snowmelt, bryophytes are green and
photosynthetically active well before the deciduous vascular plants
begin greening up (Huemmrich et al., 2010). New annual growth of many
bryophyte species appears much lighter green than older growth.
Bryophytes in shaded vs open areas also show different chlorophyll and
other pigment concentrations (Niinemets & Tobias, 2014). Bryophyte
reproductive structures develop annually in many species and these
tissues display apparent coloration distinct from the vegetative tissue.
Bryophyte phenological variation may occur at scales at which IS could
be useful in detecting physiological changes relevant to ecosystem
processes.
3.4. Vascular
plants
Living vascular plant tissue shows remarkable similarities as a group in
the general shape of spectral response, specifically characterized by a
modest increase in reflectance in the green (relative to blue and red)
and a steep “red edge,” followed by a plateau across the NIR (Figure
3). Variation in spectral profiles amongst vascular plants is often most
notable in the inflection point of the NIR and features of the SWIR,
which in turn inform the derivation of many important functional
attributes (e.g., phenology, photoprotective pigmentation, water
content, disease). The long history of research into the relationships
between leaf-level biochemistry, traits, and spectra (discussed in
Section 4.2) points to science just scratching the surface of the
potential spectra to inform plant science. Spectral profiles are
evolutionarily conserved (Meireles et al., 2020), which provides a basis
for assuming the ability to separate species using spectra. Reflectance
profiles have recently been used to separate species and even genotypes
among co-occurring plants (e.g., Dryas sp., one of the most
common vascular plant genera in the Arctic) (Stasinski et al., 2021).
This level of distinction is likely beyond the capacity of SBG but
points to the profoundly strong linkage between vascular plants and
their reflectance profiles.
Vegetation in the Arctic occurs largely in confluent mixtures, where the
boundary between an individual and group blurs. Viewing this problem in
terms of pure patches of a single species helps describe the challenge
for remote sensing. Patch size varies by species across several orders
of magnitude, from individual plants (cm scale) to confluent forest or
shrub canopies (10 m scale) or continuous patches of a single type (km
scale) such as tussock tundra dominated by Eriophorum vaginatum .
Snow, wind and ice scour the landscape and force shrubs to form thickets
that can cover very large areas but change in size and shape across
species of dominant shrubs, like Salix spp. (willows) orAlnus spp. (Alder). The sparse distribution of trees presents
unique challenges to spectral remote sensing, particularly for coarse
spatial resolution imagery where tree crowns may be widely spaced and
collectively constitute on average 30% of a 30 m pixel (Paul Mannix
Montesano et al., 2016). In contrast, some regions of the TTE are
characterized by clumps of dense tree cover with minimal spacing between
crowns across otherwise open tundra vegetation. As with non-vascular
plants, many vascular plant patches are smaller than the likely pixel
size of SBG (30 m). This underscores the need to measure features at
high spatial and spectral resolution with coordinated field campaigns to
validate SBG pixels and fully utilize the spectral resolution of SBG to
estimate vegetation composition and function.
Vascular plants exhibit strong variation in phenology across groups,
from fully dormant species such as forbs that are absent aboveground or
buried under snow in the winter to persistent year-round tissues of
evergreen trees and shrubs. The brief growing season results in very
rapid progression of plant phenological stages, which elicits the common
perception by observers that changes in reflectance are visually
apparent at a daily time scale. Most studies have focused on summer
reflectance for peak photosynthetic activity, but imaging at other times
of year provides opportunities to characterize the important features of
green up and senescence. For most plants snowmelt defines the onset of
annual growth and initiation of myriad phenological processes including
flowering and leaf-out. Characterizing differences in phenology among
plant functional types may help separate co-occurring plant groups with
similar reflectance profiles during peak summer (Beamish et al., 2017).
Spatial variation in onset of green up (earlier at lower latitudes,
south facing aspects, and lower elevations) and senescence (earlier at
higher latitudes, north-facing aspects, and higher elevations) provides
both a challenge and an opportunity for SBG to capture the important
spectral information about the biophysical changes in tundra vegetation.
In shoulder seasons when understory vegetation is buried under snow but
tree crowns protrude above the snowpack, lower albedo distinguishes
these patches from surrounding snow-covered tundra. However, to detect
phenological events in sparsely treed regions, indices that can account
for background effects - namely the coincidence of snow with vegetation
phenology - are critical (see Section 4.4).
Vascular plants generally become increasingly important, more diverse,
and larger with decreasing latitude and altitude in the Arctic. By
subzone C and south vascular plants become more prevalent than
nonvascular plants, with increasing diversity of growth forms/functional
types, graminoids, forbs, cushion plants, and deciduous and evergreen
shrubs and trees that represent general life history strategies
characterized by specific traits (with many exceptions) that influence
ecosystem and spectral properties. For example, evergreen shrubs and
trees are characterized by long-lived leaves, low photosynthetic rates,
low leaf nitrogen but high leaf mass per unit leaf area (LMA), and
tolerance to water stress. Forbs and deciduous shrubs tend to have
short-lived leaves, high photosynthetic rates and leaf nitrogen
contents, and low LMA. Graminoids may span the entire spectrum.
Graminoids (mainly sedges) form a large component of boreal and tundra
herbaceous vegetation, ranging from dry ridges to wet areas and standing
water. Reflectance profiles of graminoids are broadly similar to other
vascular plants with some distinctive features in the SWIR and overall
lower green values. However, fine spatial mixtures of living and dead
tissue in graminoid end members present a different spectral challenge
for remote sensing. Collecting clean graminoid spectral signatures in
the field under controlled light conditions is difficult due to the
shape and size of the leaves. For this reason, most measurements of
graminoids in the field are taken with a larger FOV under ambient and
therefore often have dead leaves and stems that remain mixed in with
living graminoid tissue thereby creating the mixture of living and dead
tissue in the spectral profiles for this group.
The tussock-forming sedge Eriophorum vaginatum (cottongrass) is a
dominant species over very large areas throughout the Arctic (0.9x 106 km2; (Oechel et al.,
1993)). Its unique tussock growth form provides an unusual surface
topography that introduces shadows and at low observation angles may
obscure vegetation on the opposite side. Cottongrass and many other
graminoid species also have predominantly vertically-oriented leaves
that present a challenge for top-down, nadir remote sensing because most
of the leaf area is not apparent to the sensor; further. Again, a key
challenge for remote sensing of graminoids will be accounting the amount
of dead material in spectral profiles of these plants.
Forbs are the dominant vascular plants in snow banks and snow beds,
where the annual growing season is brief but water and nutrient supplies
are high and present in dry to semi-aquatic habitats throughout the
Arctic. They are non-woody non-graminoids that typically present only
leaves and flowering stalks above the soil surface during the growing
season. Forbs show broad similarity to shrubs in their spectral
profiles, but with more variability in the visible range and more
symmetrical variation about the median in the SWIR (Figure 3).
Separating forbs from other vascular vegetation may be a challenge for
SBG but one opportunity may be during the autumn, when the spectacular
variation in pigments of Arctic tundra forbs and dwarf shrubs becomes
strikingly apparent.
The expansion of deciduous shrubs is one of the most apparent responses
of tundra ecosystems to climate warming. Deciduous shrub species have
high environmental plasticity and are unique among tundra plant
functional types in the Low Arctic, because they can achieve canopy
heights of 2 m or more and greatly overtop other vascular plants.
Therefore, the development of upright, woody canopies in tundra
landscapes strongly influences biophysical processes throughout the
year. Shrubs promote a strong positive winter feedback by trapping
drifting snow in the winter that insulates the soil; subsequently warmer
soils allow faster decomposition; decomposition releases nutrients that
promote further shrub growth (Sturm et al., 2005). In warmer parts of
the Low Arctic, the large size attained by individual deciduous shrubs,
and their tendency to develop dense canopy patches in favorable
landscape positions provides opportunities for IS to sample a relatively
pure spectral signal, which is otherwise not possible in most tundra
landscapes dominated by small, intermixed, low-statured plants.
Deciduous shrubs exhibit limited variation in the visible range and a
notable plateau in the NIR (Figure 3).
Evergreen shrubs present a different set of challenges and opportunities
for IS. In moist acidic and dry tundra, dwarf evergreen shrubs are a
major component of the vegetation, often as an understory layer above
bryophyte species (e.g., Vaccinium vitis-idaea L.). The evergreen
growth form is associated with low nutrient habitats where conservative
use of nutrients is favored. Evergreen shrubs retain leaves for 1-5 or
more years (Shaver, 1981) and thus have the potential to photosynthesize
whenever conditions are able to sustain it, even under snow (Starr &
Oberbauer, 2003), especially during the shoulder seasons. Most
evergreens produce photoprotective pigments that protect the leaves
during the cold season and strongly affect spectral reflectance of these
plants (explored further in section 4.2).
Even in otherwise tundra-dominated landscapes, trees can persist in
sparse numbers across the tundra domain. The primary example of this is
along the tundra-taiga ecotone (TTE), which is an often diffuse (rather
than abrupt) transition between denser boreal forest tree cover to
tundra-dominated plant cover. Common tree genera of the TTE include a
mix of evergreen needleleaf (e.g., Picea and Pinus ),
deciduous needleleaf (e.g., Larix ), and deciduous broadleaf
(e.g., Betula and Populus ). Having more structural complexity
than forbs, bryophytes, or lichens, trees exhibit different effects on
radiative transfer within canopies, particularly affecting multiple
scattering in the NIR and SWIR regions. For example, conifer needles
have similar reflectance to deciduous in the VNIR, but their IR
reflectances are lower than deciduous due to morphological
characteristics of needles (Hovi et al., 2017). Observed and simulated
radiative transfer of conifer needles infer that part of the spectral
differences between deciduous leaves may be due to variation in leaf
angle with both convex and flat leaf sides to their needles (J. Wang et
al., 2020). Conifer arrangement in shoots, and the presence of woody
material in twigs and boles that alter multiple scattering likely also
differs between deciduous trees. Evergreen needleleaf trees in the TTE
tend to have exceptionally narrow crowns (maximum 1-2 m in diameter),
and black spruce (Picea mariana ) can often have sparse foliage
clustered at the top of the crown, especially in regions where fire had
caused non-lethal disturbance. Due to their upright structure and
tendency to be widely spaced in much of the TTE, the interaction of high
solar zenith angles with tree stems and canopies cast extended shadows
on surrounding tundra vegetation. The vertical distribution of foliage
along narrow crowns causes problems for nadir viewing of trees in the
TTE to characterize gradients in foliar properties (Moorthy et al.,
2008). In addition to the structural complexity of trees, deciduous vs.
evergreen species experience strong phenological differences which may
complicate interpretation of spectral information in mixed-forest stands
(Pierrat et al., 2021) (Section 4.4). As with shrub-dominated
landscapes, understory tundra vegetation may be obscured from
measurement by nadir-viewing sensors in regions with denser tree cover.
Similar to the case of shrubs, encroachment of trees into tundra
landscapes influences biophysical processes such as snow distribution,
wind patterns, and soil active layer depth (F. K. Holtmeier & Broll,
2007). Characterization of geographic position, composition, and
condition of the TTE is important for detecting expansion or retreat of
tree species across the tundra domain (F.-K. Holtmeier & Broll, 2019;
Paul M Montesano et al., 2020; Stumberg et al., 2014).
Though lidar is often the primary tool for delineating the TTE and
characterizing the structure of trees in this zone, spectroscopy can
provide valuable information on phenology, physiological state, and
heterogeneity among trees (Montesano et al., 2016a; Montesano et al.,
2016b). Spectroscopy is particularly useful for characterizing
photosynthetic dynamics of trees in the tundra domain since these
individuals tend to be especially slow growing at the northern range
limit for their species distribution (hence limited structural change
detectable by repeated lidar campaigns) but contribute a substantial
amount to landscape-scale carbon flux.
3.5. Non-vegetated Surfaces
The Arctic tundra is characterized by low leaf area and sparse
vegetation cover, resulting in other materials, such as snow, water,
bare ground, and dead or burned material comprising significant portions
of the landscape. Each of these materials have unique spectral
characteristics which can confound retrievals of vegetation. Remote
sensing instruments with fine to moderately sized pixels (e.g.,
AVIRIS-NG ~ 5 m2) can capture multiple
landscape components within a single pixel, producing a mixed spectral
signal that can be difficult to interpret. Our ability to tease apart
vegetation signals from these non-vegetated tundra landscape components
is important not only for understanding vegetation, but also for
understanding the condition of the landscape itself and its feedbacks on
vegetation dynamics. An additional complication is that many surfaces
are non-vegetated for only part of the year due to snow pack, snow melt,
or flooding; at other seasons understory vegetation in the form of
lichens, bryophytes, or biocrusts becomes visible from above making the
timing of signal retrievals an important component of Arctic vegetation
dynamics.
Remote sensing of the cryosphere has been a key focus of Arctic remote
sensing. Snow, ice, and permafrost are important drivers of tundra
ecosystem structure and function, impacting components such as the depth
of the soil active layer, freshwater availability, and the formation of
important landscape features such as thermokarst lakes. Fresh snow has
very high reflectivity in the visible and near-infrared (>
80% between 400 - 900 nm, with values > 50% between 900
– 1200 nm), while clean ice, as from a glacier, has relatively high
reflectivity (> 60% between 400 – 600nm, steadily
decreasing to < 10% for 1000 – 1200 nm) (Tedesco, 2015). The
reflectivity of ice and snow is reduced over time as dirt accumulates
and darkens the surface. Snow can interfere significantly with
vegetation spectral retrieval as snow can both accumulate over
vegetation canopies (i.e., obscuring direct visibility) and snow
reflectance can saturate any vegetative signal. The timing of snowmelt,
a key driver of tundra phenology, can vary drastically throughout the
tundra (Kelsey et al., 2021), making snow dynamics both an important
process to study but also a confounding factor in vegetation remote
sensing (further explored in Section 4.4).
Permafrost thaw in the Arctic tundra is one of the most concerning
results of climate change due to the biogeochemical feedbacks which
drive increased greenhouse gas emissions. The spatial dynamics of
permafrost thaw are complex, involving interactions between multiple
processes including biogeochemical cycles, hydrology, and climate
(Grosse et al., 2013). Vegetation cover can provide insulation from
summer warming, with different types of cover providing varying levels
of protection against thaw, which makes vegetation change detection an
important component of understanding permafrost thaw changes (Anderson
et al., 2019). Vegetation feedbacks between the permafrost and
vegetation provide a key geophysical connection for SBG in studying the
Arctic because the high spectral resolution will allow separation of
more kinds of tundra attributes. However, permafrost features have
highly patterned features, often considerably finer scale than the 30 m
resolution of SBG, requiring field campaigns to describe patterns in the
permafrost at higher spatial resolution. Permafrost thaw can also impact
vegetation cover through landscape transformation.
One of the most rapid and noticeable landscape features of permafrost
thaw is the development of thermokarst lakes (Grosse et al., 2013).
Thermokarst lakes form from the degradation of ice wedges in continuous
permafrost areas, leaving standing water and unfrozen ground, called
taliks, underneath the lake. The presence of thermokarst lakes, which
have been forming in the Arctic since the Last Glacial Maximum, have
been increasing and existing lakes have been expanding. Thermokarst
lakes increase the amount of standing water that is present in the
Arctic tundra. Standing water poses a challenge for tundra vegetation
remote sensing. The tundra is studded with thermokarst lake depressions
that form due to the freeze-thaw cycle of permafrost, and in the summer
much of the tundra is covered with standing water. Water most strongly
interferes with the retrieval of vegetation reflectance in the visible
range (400 – 700 nm), but it can also cause a reduction across the
entire spectrum. This can potentially influence vegetation signals
retrieved from vegetation indices such as NDVI which use red reflectance
(~650 nm although this varies by sensor), or PRI which
uses green (~ 531 nm). Liquid water absorbs light in the
NIR, reducing vegetation reflectance in that region, thus dampening
vegetation signals in pixels with standing water (Lang et al., 2015).
Remote sensing instruments with finer spatial resolution can help to
solve this problem by improving pixel purity.
The amount of vegetation cover varies significantly across the Arctic
tundra due to differences in topography and soil properties such as
nutrient content (Liu et al., 2017). Exposed bedrock and bare soil are
common and bare soil can be intermixed with sparse vegetation cover.
Soil and rock spectra vary depending on the type and color of the
substrate and moisture content. Most dark colored soils are more
strongly absorptive in the visible range than vegetation, but the
vegetation signal is more strongly reflective in the NIR than soil. As
with water, interference in the red and NIR can confound commonly used
vegetation indices such as NDVI. Another complication is senescent or
dry vegetation, which can have a spectral signal similar to bare soil
(Liu et al., 2017). In the high Arctic, tundra vegetation can have a
very brief growing season, so it is important that remote sensing
measurements have short revisit times to capture phenological changes
appropriately, and tease apart vegetation from bare soil or litter.
Tundra fires have a sparse historical record, but recent data and model
projections indicate that tundra fires will increase in frequency and
severity under climate change (French et al., 2015). Fire has become a
growing concern as a source of tundra change. Spectrally, burned
vegetation reflectance is high in the shortwave NIR which can help
distinguish it from green vegetation, but bare soil which is exposed
during burning can interfere with vegetation retrieval (Boelman et al.,
2011). Alternative vegetation indices have been proposed to assess
burned vegetation areas, but full spectral data will help to tease apart
burned areas from green vegetation.
3.6. Mixed Composition
Observations
Although many tundra vegetation communities can often contain both
vascular and non-vascular species, the combined spectral signature can
be distinct enough to allow for separability among communities. For
example, (Davidson et al., 2016) successfully distinguished among eight
different tundra vegetation communities including bryophyte-shrub,
bryophyte-lichen, and tussock-shrub utilizing the Blue (450-510 nm), Red
(640-692 nm) and Red Edge (705-745 nm) regions (Figure 5). (Bratsch et
al., 2016) distinguished among four tundra plant communities at Ivotuk,
Alaska (particularly early in the growing season), using Blue, Red, and
Near-Infrared hyperspectral bands.