Figure 1 . Examples of heterogeneous vegetation and landforms in tundra landscapes. (a) Close-up of ground lichens in upland tundra, Izaviknek Hills, Alaska; (b) mosaic of shrublands, wetlands, and waterbodies, Yukon Delta, Alaska; (c) mosaic of tall deciduous shrubs and open tundra, Seward Peninsula, Alaska; (d) intermixed sedges and low shrubs, Alaska North Slope; (e) polygonal ground, Alaska North Slope; (f) High Arctic tundra dominated by mosses and cryptogamic crust, Franz Josef Land, Russia; (g) thaw slump and exposed ground-ice, Yugorskiy Peninsula, Russia; (h) frost boils in forest-tundra ecotone, northwestern Siberia. The extent of the Arctic tundra biome is shown in red in the central map based on the Circumpolar Arctic Vegetation Map (CAVM Team, 2003).
The composition of tundra includes significant coverage by both nonvascular and vascular vegetation. Nonvascular vegetation types pose unique challenges, in that they have different spectral signals than vascular plants (Hope & Stow, 1996; Stow et al., 1993) their spectra are highly influenced by their moisture content (Bubier et al., 1997; A. Harris et al., 2005; Vogelmann & Moss, 1993), and physiologically they behave differently than vascular plants (Green & Lange, 1995; Tenhunen et al., 1995). Relationships between remotely-sensed spectra and plant function have not yet been developed at spatial scales adequate to capture nonvascular plant physiological responses and the mixed composition of vascular and nonvascular plants within spectral footprints complicates interpretation of observations. Collectively, these issues suggest a need for new methodologies for assessing the composition of tundra systems. One approach is to collect colocated ground vegetation composition data and remotely sensed spectral observations at varying spatial scales, and utilize their relationships to enable subpixel vegetation cover retrieval (Thomson et al., 2021). Alternatively, spectral unmixing algorithms parameterized by fine-scale observations can be used to disentangle the sub-pixel contributions to a spatially integrated observation (Beamish et al., 2017; Bratsch et al., 2016; Huemmrich et al., 2013). Such work will be critical to interpret compositional effects on imaging spectroscopy observations from SBG - but present a major opportunity for future work.
Meteorological conditions inherent to Arctic regions, such as high frequency cloud occurrence, seasonal snow cover, and ephemeral surface water often preclude high quality spatially contiguous or temporally continuous observations (Walther et al., 2016, 2018). The limited snow- and ice-free period (including episodic snowfall events in the middle of the growing season) constrains the number of clear observations of vegetation. Additionally, rapid transitions and highly variable shoulder season weather restrict the utility of even high frequency spaceborne observations to detect important phenological events (e.g., start-of-season and end-of-season) (Karlsen et al., 2021; Parazoo et al., 2018; Vickers et al., 2020). Smoke from frequent and extensive wildfires in the neighboring boreal forest biome can drift over the tundra biome for substantial periods during the growing season of a given year, making interannual comparisons challenging.
Illumination geometry at high latitudes also complicates remote sensing of Arctic tundra (Buchhorn et al., 2016). High latitude regions experience extremes in daylength, from continuous daylight in midsummer to continuous darkness in midwinter, the latter of which limits the capacity for reflectance-based observations on the winter edge of shoulder seasons. The effects of the continuous daily photoperiod of midsummer challenge assumptions established in the temperate regions about the connections between spectral imaging observations and dynamic physiological processes (e.g., accumulated stress). Overall, surface radiation is lower due to high solar zenith angles and consequent scattering due to atmospheric path length, and photon scattering at such angles complicates radiative transfer.
Existing IS data over the Arctic is sporadic in space and time. For example, since 2017 ABoVE (Miller et al., 2019) has collected a large amount of airborne IS data over a broad Arctic region in North America using NASA’s Next Generation Airborne Visible Infrared Imaging Spectrometer (AVIRIS-NG). While these data are of high value for characterizing vegetation function, stress, and mapping functional traits (Gamon et al., 2019), the discontinuous coverage (non-overlapping flight lines collected over a larger region) and the volume of data (several gigabytes in size for an individual flight line) mean that, at present, an individual researcher is often required to identify and download a number of different scenes, and therefore a large data volume, to carry out a study. Some of these challenges will be exacerbated with upcoming satellite IS missions such as SBG (Cawse-Nicholson et al., 2021) which will provide voluminous datasets. More efficient usage of IS datasets for Arctic research will require new data hosting and access methods to find, extract, and apply IS data without large bandwidth or local storage requirements.
Here we present a technical perspective - informed by empirical observations of spectral variability - of the numerous ecological, geographic, and technical challenges associated with spectroscopic observation of Arctic tundra ecosystems. We discuss how we may leverage our understanding of spectral dynamics and characteristics to understand tundra ecology. We delimit our region of interest based on the Circumpolar Arctic Vegetation Map (CAVM Team, 2003) (see Fig. 1). First, we provide context for the degree of spectral complexity of the tundra biome by quantifying the intrinsic spectral dimensionality from a series of observations from airborne IS (Section 2). Next, we describe how attributes of the land surface in the tundra biome (e.g., plant functional type and vegetation-substrate composition) impose particular challenges for interpreting spectroscopy (Section 3). We then elaborate on how IS enables an opportunity to achieve several common goals for advancing our understanding of the Arctic tundra biome: long-term change detection, land cover and vegetation classification, retrieval of biophysical properties, and phenological and diurnal change (Section 4). We conclude by providing recommendations for Arctic tundra spectroscopy research (Section 5) by addressing the following key questions:
How can we use spectral observations at a variety of spatiotemporal resolutions (e.g., from spaceborne, airborne, and surface-based instruments) to address inherent challenges associated with IS and better understand Arctic tundra ecosystems?
How can our understanding of Arctic tundra ecology advise further research and the development of new instruments and sampling designs?