Integrating Hyperspectral Imaging, Plant Functional Diversity, and Soil-Lithology to Uncover Mountainscape Disturbance Dynamics Induced by Landsliding

Abstract

The global biodiversity crisis has emphasized the unique contribution of functional diversity to ecosystem function, stability, and resilience. In this regard, the increasing availability of remotely sensed data together with the development of new sensors and approaches has the potential to improve our ability to quantify and monitor ecosystem traits and functions at unprecedented spatial, temporal, and spectral scales. In particular, air- and spaceborne hyperspectral data are making possible the measurement of plant-level functional traits to investigate ecosystem function and functional diversity in novel ways. In this review, we posit that these developments, together with similar ones on soils and lithologies, can help us understand relationships between functional diversity, ecosystem processes, and landsliding, and more broadly the disturbance dynamics of mountainscapes. Acknowledging the challenges associated with mountainous regions, this review aims to (1) synthesize broad established methods to retrieve functional traits from remotely sensed data, (2) summarize approaches to characterize functional diversity derived from remotely sensed functional traits, (3) review work addressing functional diversity, ecosystem functioning, and dynamics of mountainscapes, and (4) elaborate on how these methods and approaches can help develop a needed “ecosystem-centered” view of landslides. This view acknowledges that ecosystem diversity influences both slope resistance and susceptibility to failure and landslide recovery, that interactions between ecosystem and geomorphic processes drive the dynamics of mountainscapes mediated by landslides at multiple scales, and that the variability in landslide size represents a source of diversity while also playing a fundamental role in landslide recovery and landscape memory.

1. Introduction

The global biodiversity crisis has emphasized the contribution of biodiversity to ecosystem stability and resilience and highlighted monitoring needs at multiple scales [1,2]. In this context, functional diversity has emerged as an essential dimension of biodiversity due to its potential to link ecosystem processes and function with taxonomic, phenotypic, and genomic diversity [3,4]. Remotely sensed multispectral data collected along the visible (VIS) through the long-wave infrared (LWIR) regions has already made possible the characterization of ecosystem traits and functions [5,6,7,8,9,10,11] and helped monitor the Earth’s “pulse”. Simultaneously, important developments in imaging spectroscopy are making possible the establishment of relationships between field-based leaf- and plant-level functional trait measurements with air- and spaceborne hyperspectral data to investigate functional diversity and ecosystem function at unprecedented spectral scales [12,13,14,15,16]. Here, we posit that these developments, together with similar ones focusing on soils and lithologies [17,18,19,20,21,22], can help understand relationships between functional diversity, ecosystem processes, landsliding, and, more broadly, disturbance dynamics of mountainscapes (Figure 1).

Investigating the aforementioned relationships will greatly benefit from a new generation of spaceborne hyperspectral missions (Files S1 and S2: Table S1) [22,23,24,25,26,27,28,29,30,31,32,33], yet there are two sets of challenges that need to be recognized. The first set stems from the nature of mountain environments and the landslides themselves. Atmospheric, topographic, and bidirectional reflectance distribution function (BRDF) effects are ubiquitous in mountainous areas. This may require extensive image pre-processing to correct for water vapor, aerosol content, multiple scattering, cloud and terrain shadows, and mask clouds to ultimately retrieve sun-lit pixels and evaluate surface reflectance [34,35,36] and also the inclusion of topographic effects in radiative transfer models [37]. The latter may improve simulations of canopy reflectance and emissivity along the optical to thermal regions. In addition, landslides, depending upon the stage of development, may contain a mixture of vegetated pixels and bare soil that will influence the spectral characteristics of vegetation and substrate. Vegetation studies based on hyperspectral data recommend the removal of bare soil pixels to eliminate background reflectance [38], whereas soil/rock studies recommend the removal of vegetated pixels [39]. In the context of this work, both sets of pixels provide critical information about functional diversity, ecosystem and geomorphic processes in landslides, and more broadly the dynamics of mountainscapes.

The second set of challenges to investigating functional traits and functional diversity is associated with the sources of remotely sensed data and the distinctive trade-offs among spatial, spectral, and temporal resolutions [40,41,42]. First, landslides vary greatly in size [43,44] thus, sensor spatial resolution can strongly influence landslide detection [45]. Second, downscaling of spaceborne multi-, e.g., [5,46,47,48], and hyperspectral [49,50,51,52] data may be necessary to characterize functional diversity and habitat diversity derived from functional traits and functional types and substrates, respectively. Used methods include band-sharpening or bilinear interpolation to match the smallest field sampling unit and modeling fractional cover using spectral mixture analyses (SMA). In both instances, subsequent upscaling using environmental data allows examination of diversity patterns at landscape and regional scales. Third, upscaling of functional traits derived from airborne, hyperspectral data may be necessary to characterize functional diversity from image scenes to regions; this may require modeling upscaled functional traits with environmental data [12,53,54]. Finally, an improvement of the temporal resolution of data offered by the new generation of spaceborne hyperspectral missions (Files S1 and S2: Table S1) can be extremely valuable not only to capture landslide populations triggered by precipitation or earthquake events but also the development of individual landslides within any population or assemblage.

Leveraging studies that have overcome some of the challenges outlined above, this review brings together spectroscopy work conducted in different fields to develop the ecosystem “centered” view of landslides outlined in Figure 1. This view recognizes the critical influence of biodiversity both on slope resistance and susceptibility to failure and landslide recovery and, more broadly, of interactions between ecosystem and geomorphic processes driving the dynamics of mountainscapes mediated by landslides at multiple scales. More specifically, this review aims to (1) synthesize broad established methods to retrieve functional traits from remotely sensed data, (2) summarize approaches to characterize functional diversity derived from remotely sensed functional traits, (3) review work addressing functional diversity, ecosystem functioning, and dynamics of mountainscapes, and (4) elaborate on how these methods and approaches can help develop a needed holistic view of landslides. On the one hand, (1) and (2) hinge on the selection of relevant functional traits at leaf and plant levels, developing models associating these traits with their optical properties, scaling functional traits with air- and spaceborne data, and assessing functional groups and functional diversity at multiple scales ([15,16,53,55,56]). On the other hand, (3) acknowledges the need to integrate diverse approaches and fields, and (4) the importance of landslide size and scale to understand interactions between ecosystem and geomorphic processes from individual landslides and sites to entire landslide populations, assemblages, and landscapes (Figure 1).

2. Functional Traits—From Field to Remotely Sensed Observations

In a broad sense, functional traits refer to attributes that play a role in the establishment, growth, survival, reproduction, and ultimately fitness of an organism [57]. In a narrow sense, functional traits represent a subset of biochemical, physiological, morphological, and behavioral attributes that mediate organisms’ responses to environmental stressors, including disturbance, and have a marked effect on ecosystem functions and ecosystem stability at multiple scales [58,59,60,61,62]. Thus, upscaling functional trait measurements to regions, e.g., [63,64], can help understand underlying causes of the observed variability and consequences under global changes. In this regard, imaging spectroscopy is providing unique opportunities to identify spectral features [65,66] representative of leaf- and plant-level traits [55,67,68] that mediate ecosystem functions (Figure 2; [12,15]) and explore their variability with regard to geomorphic [69] and environmental [70,71] gradients, landscape heterogeneity [72], and spread of invasive species [73,74]. In addition, this work may help understand the consequences that changes in the underlying drivers may have on trait distributions and ultimately ecosystem functions, which becomes important for monitoring and forecasting purposes [75,76,77] as well as for calibrating ecosystem and biosphere models [50,78,79,80].

The selection of leaf- and plant-level traits (Figure 2a, Stage 1a, 1b) that collectively translate into critical ecosystem processes and functions, the retrieval of this information from optical data (Figure 2a, Stage 2), and scaling from leaves and plants to pixels and plots (Figure 2a, Stage 3) are central in the definition of spectral features (Figure 2b Stage 4). Functional traits at the leaf level include pigments (chlorophyll, carotenoids, anthocyanins), nutrients (nitrogen, phosphorus, calcium, potassium, magnesium, iron), structural compounds (lignin, cellulose, polyphenols), water content (equivalent water thickness, EWT), and mass per unit area (LMA), whereas at the plant level, tree height and canopy characteristics (e.g., size and shape, leaf area index, leaf angle distribution, fractional cover) [15,38,67,68,80,81,82,83]. At the individual level, both sets of traits are related to light capture and growth, photoprotection, stress resistance and defense, maintenance, and metabolism. At the ecosystem level, these traits are informative of photosynthesis, primary production, carbon storage, nutrient cycling, decomposition, allocation and growth, stress resistance, and hydraulic regulation.

The retrieval of leaf- and plant-level functional traits from optical data is based on the development of relationships between trait measurements and individual bands, band indices, or the whole spectra via parametric and non-parametric regression, physically based, and hybrid modeling approaches (Figure 2a, Stage 2; [16,77,84,85,86,87,88]). Parametric approaches include models that establish relationships between a trait of interest and individual bands, spectral indices [78,89,90,91], metrics derived from the shape of spectra [92,93], or uncorrelated, synthetic variables derived through principal component analysis (PCA) [65,94]. Non-parametric approaches often use the entire spectrum and include linear (e.g., partial least squares regression—PSLR), e.g., [80,95,96,97], and non-linear, including machine learning models, e.g., [98,99]. Physical-based approaches make use of coupled leaf-canopy radiative transfer models (RTM) to estimate canopy reflectance via forward mode or leaf traits by model inversion; these models require look-up tables as well as trait data sets calibrated both in the lab and field [100,101,102,103]. Finally, hybrid models combine both approaches [104,105,106].

Scaling and mapping plant functional traits (Figure 2a,b, Stages 3–4) are often part of large studies aimed at the development of new sensors, complex instruments, and applications (Figure 2a, Stages 1–2; [79,107,108,109]), yet they may also take place in other contexts [110]. These studies consider retrieval methods, conduct simulations, collect in situ field and sensor calibration and validation measurements, and pre-process the imagery to different degrees. Central to scaling and mapping are in situ measurements of canopy reflectance at relatively homogenous sampling points using field spectroradiometers or complex airborne instruments and/or estimates derived from RTMs (Figure 2a, Stage 1–2). Sampling points may represent elementary sampling units (ESU) within plots matching pixels of a given airborne or spaceborne sensor, tree crown pixels or full tree crowns, and field subplots within plots whose size and shape may represent pixels of a given sensor [46,80,82,109,111]. Extensive pre-processing of air- and spaceborne multispectral and hyperspectral data may involve data fusion, estimating fractional vegetation cover, masking out clouds, bare ground, and tree-canopy shade, delineating tree canopies, and characterizing canopy geometry (Figure 2a, Stage 1b; [15,107,112]). At this stage, the models linking optical data with leaf- and plant-level traits are applied to the pre-processed remotely sensed data to scale up predictions and map these traits at pixel, plant, image scene, and region levels (Figure 2a Stage 3; [16,38,82]). Scaling may involve averaging pixel-level traits for individual tree crowns [82,113] or averaging the spectra of multiple pixels within sampling plots [12,38] or weighing vegetation indices or traits by the corresponding plant abundance, e.g., tree crown area [80,114]. At scene and region scales, upscaling leaf- and plant-level traits may involve additional modeling using ancillary environmental data and resampling at resolutions lower than the original hyperspectral data (Figure 2, Stages 3–5; [38,95,115]).

3. Functional Diversity and Ecological Function

Functional diversity is broadly defined as the value and range of functional traits in a community [116,117], and its importance cannot be overstated. In practical terms, functional diversity facilitates comparisons at multiple scales. In conceptual terms, there is increasing evidence that functional diversity greatly contributes to the stability of ecological function [117,118,119,120,121] and to ecosystem resilience within and across scales [121,122], that functional diversity is influenced by the intensity of disturbance [123,124], and that functional diversity and species diversity are related [125,126,127]. This may explain efforts aimed at developing qualitative and quantitative measures [128,129] to characterize within (local or alpha), between (landscape or beta), and overall (regional or gamma) [130,131,132] functional diversity and understand their scale dependency in space [120,133,134,135] and time [136,137] in response to changing environmental conditions. It is important to emphasize that functional diversity is a community property [128,138] and therefore, when using spatially explicit remotely sensed data, functional diversity is estimated in communities or neighborhoods of pixels [15,56,66]. In high-resolution airborne images (e.g., 1–5 m), single or few pixels represent individual trees, whereas the size of the neighborhoods, field sampling units, or plots capture community structure and processes in any given region.

The development of metrics to quantify functional diversity mirrors similar efforts aimed at characterizing spectral diversity, i.e., the variability in spectral reflectance, a dimension of diversity linked to plant taxonomic and phylogenetic diversity [139,140,141,142,143]. Therefore, the metrics described below, although focusing mainly on functional diversity, will also include references to spectral diversity (Figure 2b, Stage 5; Files S1 and S2: Table S2). Quantifying functional diversity may use single or multiple functional traits to identify functional groups, i.e., groups of species sharing similar traits and most likely performing similar functions or exhibiting similar ecological strategies within ecosystems [121,144,145,146,147,148,149]. One large-scale functional grouping of plants uses categorical traits such as life form, leaf type, and leaf phenology to identify plant functional types (PFT); these traits are currently retrievable from remotely sensed data [150]. Another functional grouping of plants used continuous traits, more specifically canopy functional traits derived from airborne hyperspectral data and cluster analyses to identify functional forest classes (FFC) and functional forest groups (FFG) that were subsequently scaled at a country level (Figure 2b Stage 5; Files S1 and S2: Table S2; [12]). Identifying these classes is a prerequisite to characterize alpha and beta diversity, as shown by work on spectral diversity [56,151].

Functional diversity is more commonly expressed by metrics or indices that use as input single or multiple functional traits that can take discrete and/or continuous values [135,152,153,154]. A common metric based on single traits is the community weighted mean of trait values (CWM) that expresses the importance of a trait based on species abundances; when estimated for many communities, it can be used to understand trait-environment relationships. In the context of this review, CWM has been used to scale leaf- and plant-level traits derived from field, laboratory, and spectroscopy measurements to canopy or plot scales (Figure 2b, Stage 5; Files S1 and S2: Table S2; [5,47,80]). Subsequently, the plot-level traits are used with multispectral or hyperspectral data to develop predictive models for trait mapping purposes. Metrics based on multiple functional traits use distance-based measures, geometric properties of the trait space, or the variability observed in trait probability densities (TPD) to estimate the various components of functional diversity [128,129,130,152,154,155,156]. Studies aimed at examining the potential of remotely sensed data to scale up functional diversity vary in terms of sources of field-based leaf and plant trait data, sources of remotely sensed data, sources of spectral features, derived spectral features, scaling methods, and metrics used (Figure 2b, Stage 5; Files S1 and S2: Table S2). One set of studies used pixel- and tree-level band indices and point-cloud data to derive variables representative of leaf physiological and tree morphological traits from which to calculate functional diversity metrics using the geometric properties of the trait space [15,40,47,114,157] or the variability of trait probability densities [158,159]. A second set of studies used plot-level CWMs of field-based leaf and tree traits in combination with spectral features from multispectral, spaceborne sensors to estimate functional diversity metrics based on the geometric properties of the trait space [46]. These metrics provide a measure of the range of trait values in a community based on the area of a convex hull encompassing pixels in a neighborhood (functional richness; FRic), the variability of trait values based on the mean distance of the pixels to the trait space centroid (functional dispersion; FDis), the variability of trait values based on the mean distance to the convex hull centroid (functional divergence; FDiv), and the distribution of pixels in trait space (functional evenness; FEve). Further work based on these metrics has explored their sensitivity to changes in sources of trait data [157], spatial (changes in neighborhood size) and spectral (convolving airborne hyperspectral data) resolution [15,40,114], as well as their relationship with ecosystem-level processes [47] and spectral diversity [157].

The metrics just described provide a quantitative measure of functional diversity at local (within or alpha diversity) scales. Yet differences between communities, i.e., beta diversity, can help understand the role of environmental factors in structuring diversity at landscape and regional scales (Figure 2b, Stage 5; Files S1 and S2: Table S2). Using three structural variables derived from spaceborne and airborne LiDAR, Schneider et al. [159] used a TPD approach to investigate possible effects of data sources on estimates of alpha and beta functional diversity. Using airborne hyperspectral data and a subset of principal components, Robertson et al. [160] applied methods described in Féret and De Boissieu [151] to estimate beta diversity and evaluate the extent to which image resolution, window size, and extent influenced this metric. Finally, Rossi et al. [66] and Laliberté and collaborators [65,161] developed different approaches to partition the total spectral diversity of a region (gamma; γSD) into additive alpha (within community spectral diversity; αSD) and beta (between community spectral diversity; βSD) components. The first author used band indices from a temporal series of Sentinel 2 and the Rao Q metric to calculate γSD, whereas the second used a subset of principal components and the total variance to calculate SDγ (Figure 2b, Stage 5; Files S1 and S2: Table S2). In sum, the development of functional and spectral diversity metrics from remotely sensed hyperspectral remains an active area of research.

4. Diversity, Landslide Dynamics, and Mountainscapes

Earlier, we posited that remotely sensed hyperspectral data may help understand relationships between functional diversity, ecosystem processes, landsliding, and, more broadly, disturbance dynamics of mountainscapes (Figure 1). The ecological and geomorphic significance of these relationships remains poorly understood, but for certain, a focus on areas disturbed by landslide activity as well as undisturbed areas [162,163,164,165] is key to understanding the dynamics of mountainscapes. First, landslides expose soil horizons exhumed all the way down to altered bedrock or even minimally weathered bedrock upon which communities and ecosystems reorganize and recover [163,164]. Second, landsliding may contribute to the diversity [166,167,168,169,170] and functioning [171,172,173,174] of montane ecosystems through the creation of unique habitats and landscape configurations that favor subsets of unique species and the alteration of abiotic conditions that directly influence ecosystem processes such as carbon and nutrient cycling and rock weathering. Third, the newly exposed substrates are heterogeneous at local (individual landslides), regional (landslide populations), and landscape (landslide assemblages) scales. Lastly, landslides leave long-lasting legacies that may influence landscape memory and therefore landsliding itself [175,176]. Studying undisturbed slopes becomes equally important because ecosystems may influence slope resistance and susceptibility to failure in complex ways given the dynamic nature of vegetation–soil–saprolite interactions during ecosystem development [163]. Advances in hyperspectral sensors and platforms (Files S1 and S2: Table S1) as well as data analytical approaches (Figure 2) may help characterize the vegetation based on functional traits and functional diversity (Figure 3), and the underlying soils and lithology based on physical and geochemical attributes. Together this has the potential to inform about the diversity and functioning of montane ecosystems at scales commensurate with landslide activity (Figure 1).

4.1. Plant Traits, Ecosystem Function, Montane Ecosystems, and Landsliding

In mountainous environments, multi- and hyperspectral remote sensing has contributed to the characterization of plant communities [177,178,179,180], land cover/land use [181,182,183,184], tree line dynamics [34], and disturbance regimes including deforestation, forest die-off, mining, and landsliding [185,186,187,188,189,190]. In particular, the efforts towards landslide mapping have been substantial not only in terms of the number of papers but also the use of diverse machine learning methods (Figure 1; [45]). In contrast, the use of multi- and hyperspectral remote sensing to characterize functional traits, functional diversity, and ecosystem function in mountainous environments, particularly in the context of landslide studies, has been limited.

The first set of studies has used vegetation indices and canopy chemical traits derived from whole spectra (Figure 2; Files S1 and S2: Table S2) to study responses of montane ecosystems to climate variability and to characterize components of the carbon and, to a lesser extent, nutrient cycles. For example, phenological metrics derived from the Normalized Vegetation Index (NDVI) have been used to infer temperature sensitivity of mountain vegetation [191]. Similarly, NDVI and the Enhanced Vegetation Index (EVI) have been used in carbon cycling studies to estimate above-ground biomass (AGB) and leaf area index (LAI) in forest and non-forest montane ecosystems [9,192,193,194,195]. The quantification of canopy nitrogen and lignin at landscape- and regional-scales through air- and spaceborne hyperspectral data collected over temperate and tropical hilly and mountainous ecosystems has improved our understanding of nutrient dynamics [70,76,196,197]. The characterization of leaf chemistry through airborne hyperspectral data in the eastern Andes of Peru and Mount Kinabalu in Malaysia has provided unique insights into the cycling of rock-derived nutrients [53,82,113,198]. Altogether, the aforementioned work has improved our understanding of the contribution of abiotic (temperature, geomorphic, and topographic settings) and biotic (species composition) factors driving trait variability and ecosystem, soil, geomorphic, and hydrologic processes.

The second set—Biodiversity-Ecosystem Function (BEF) studies—has estimated functional diversity metrics from remotely sensed traits to derive large-scale relationships between functional diversity and ecosystem functions (Files S1 and S2: Table S2). Work conducted in the Laegern Mountain of Switzerland estimated functional richness (FRic), functional evenness (FEve), and functional divergence (FDiv) (Figure 2b, Stage 5; Files S1 and S2: Table S2) from physiological (chlorophyll, leaf carotenoids, and equivalent water thickness) and morphological (canopy height, plant area index, and foliage height diversity) traits and showed that FRic responded more strongly to environmental conditions than FEve and FDiv [15,40]. More specifically, FRic was highest on the south aspect and lowest at the mountain ridge. At this same site, Schneider and collaborators [199] simulated gross primary productivity (GPP) and found that it was positively related to FRic; yet there were large differences among sites differing with elevation. A fourth study conducted in the southern aspect of Mt. Shennongjia in China using similar traits and functional diversity indices showed that FRic was highest at mid and high elevations [47]. Perhaps more importantly, this work revealed a humped-back relationship between aboveground biomass (AGB) and FRic derived from morphological traits and a more complex, non-linear relationship between productivity (as reflected by NDVI) and FRic derived from physiological traits. Using a different approach in the eastern Andes of Peru, Duran and collaborators [200] found that the CWMs for three leaf traits (LMA, non-structural carbohydrates, and percent water) were positively related to elevation, whereas a fourth one (chlorophyll) and FRic were negatively related. This work further showed that FRic was positively related to GPP and net primary productivity (NPP).

The closest reference to functional traits retrieved from remotely sensed data in landslide studies makes use of the green-red vegetation (GRVI) and normalized vegetation (NDVI) indices (Files S1 and S2: Table S3) and tree height [201]. Li et al. [202] used GRVI in combination with back estimations of lateral apparent cohesion contributed by roots to understand slope resistance/susceptibility to intense rainfall events in China. In contrast, NDVI and tree height have been used to characterize vegetation recovery and its role in stabilizing sediment fluxes and capturing carbon, respectively. A common metric used in these studies is the Vegetation Recovery Ratio (VRR; Files S1 and S2: Table S3), representing the vegetation gained over time divided by the vegetation lost after a major landslide-triggering event [203,204,205,206,207,208,209,210,211,212,213,214,215,216,217,218]. Here, vegetation gains and losses are estimated at pixel scales based on NDVI. Some interesting observations, as well as limitations, emerge from this work. First, the variability of NDVI within [210] and among studies (Files S1 and S2: Table S3) can be substantial. Second, the presence of active and inactive landslides, together with the variability of VRR in any given population, suggests that successional trajectories (Figure 1c) are highly variable, which may explain the within-study NDVI variability, e.g., [210]. Third, NDVI₀, VRR, and annual VRR varied among studies, most likely reflecting differences among sites and the time over which recovery was monitored. Fourth, VRR decreases with elevation and slope [206]. Fifth, rainfall-triggered landslides may recover faster than those triggered by earthquakes [213]. Finally, the limited use of ecological principles precludes a deeper understanding of ecosystem-geomorphic interactions. For example, NDVI is not modeled to characterize ecosystem development, nor is it related to variables such as aboveground biomass or tree height, two variables informative of important ecosystem processes. Similarly, the variability of NDVI at the scale of individual landslides may be used to characterize functional diversity; integrated over time and across landslides within a population, it may help understand the contribution of landslides to the diversity of montane ecosystems.

In contrast to the aforementioned studies, Freund and collaborators [201] focused on a landslide assemblage to examine changes in vegetation height and aboveground biomass in landslides of different ages along a broad elevation gradient in Peru’s Kosñipata Valley. These authors combined a 1 m-resolution top-of-canopy height (TCH) raster derived from high-resolution full waveform airborne LiDAR data [219] with mapped landslides. The latter were characterized by TCH’s mean (TCH_mean) and standard deviation (TCH_sd), and TCH was subsequently used to estimate biomass at a 30 m resolution. Over a period of 25 years, TCH_mean increased linearly with time; residual vegetation on landslides translated on average into larger TCH_mean compared to landslides without residual vegetation. In contrast, in landslides without residual vegetation, the linear relationship between TCH_sd and time exhibited a steeper slope than that of landslides with residual vegetation. Aboveground biomass increased linearly with time, and the rate of increase was larger at low elevations, followed by mid and high elevations. Three independent studies in Japan that used orthoimages and digital surface models generated from data collected by UAVs found that residual vegetation on landsliding greatly contributed to their recovery [215,220,221].

4.2. Soil and Lithology Attributes, Montane Ecosystems, and Landsliding

In mountainous environments, parent material strongly influences soil [222,223,224,225] and ecosystem [226,227,228] properties with likely consequences on slope stability and ultimately landsliding (Figure 1a). Yet, it is also likely that the exposure of fresh rock resulting from landslide activity sets in motion geochemical, biogeochemical, and biological processes poorly linked but that may have profound consequences for the biosphere. Addressing these possibilities, however, is difficult given the poor characterization of parent materials in many mountainous regions around the world. In this regard, multi- and hyperspectral remotely sensed data have been invaluable for lithological mapping of hilly and mountainous regions [17,39,229,230]. These data have also been used, albeit to a lesser extent, to quantify rock mineralogy, geochemistry, and degree of weathering [231,232,233,234,235,236] and characterize soil properties [237,238,239,240,241,242,243]. One common thread among most of this work is its focus on bare rock and soil or ecosystems with sparse vegetation, such as deserts, grasslands, or croplands. Yet, landslides often occur in densely vegetated mountains. Here, landslide activity exposes bare substrates, and this may open opportunities to study rock and soil processes, borrowing work conducted beyond these environments [244,245,246,247,248].

In hilly and mountainous humid environments with dense canopies, two complementary approaches may help characterize underlying parent material and soil properties, landslides undergoing succession, and/or ecosystem-geomorphic interactions. Although both focus on the vegetation, their emphasis varies. The first approach uses remote sensing to establish relationships between leaf/canopy traits retrieved from multi- and hyperspectral data (see Section 3) and soil fertility. In planted and natural forests, a reduction in leaf area and foliar concentration of chlorophyll (Chl), nitrogen (N), and phosphorus (P) is associated with nutrient deficiencies that, in the case of rock-derived nutrients (e.g., P), are inherited from parent material (Figure 1; [228]). In the archipelago of Hawaii, three studies that used airborne LiDAR and hyperspectral data to obtain tree heights, including derived estimates of aboveground biomass and foliar N and P have contributed different insights into leaf traits-ecosystem function relationships. Across a substrate age gradient—a surrogate for degree of weathering and nutrient limitation—mean foliar N was slightly greater at the two intermediate-age sites, whereas the variability of tree height, but not its mean, increased with substrate age [249]. In Kaua’i, the oldest island of the archipelago, remotely sensed foliar P resembled spatial predictions of foliar P based on analytical measurements, elevation, and rainfall [250]. More specifically, the concentration of foliar P was highest in lower slopes and depositional areas and lowest on ridges and the stable shield surfaces. A negative correlation between foliar P and strontium isotopes indicated that increased fertility was the result of inputs from rock weathering. Finally, in Hawai’i, the youngest of the islands, remotely sensed foliar N did not change with substrate characteristics, but the same was not true for aboveground biomass [76]. In Costa Rica, low concentrations of remotely sensed foliar N were observed on slopes, whereas high concentrations on ridges [70]; several soil N metrics followed this pattern [251]. In the Andean foothills of eastern Peru, remotely sensed foliar concentrations of N and rock-derived nutrients (P, calcium (Ca), magnesium (Mg), and potassium (K)) increased non-linearly with incision depth; the opposite was true for LMA [198,252]. Furthermore, remotely sensed foliar Ca and K were significantly and positively correlated with the corresponding available nutrients in soil, whereas foliar and soil P were negatively correlated. In Mt. Kinabalu, Malaysian Borneo, a study centered on two lithologies (sedimentary and ultramafic lithologies) and 32 catchments (16 per lithology) found that remotely sensed foliar P and N were significantly and positively correlated with a measure of hillslope disequilibrium (HD; decaying to steepening) in areas underlain by both lithologies; for remotely sensed foliar Ca, this was true but only in areas underlain by sedimentary lithologies [69]. Finally, LMA was negatively correlated with HD. This work not only has informed about underlying rock and soil properties but also suggested the role of erosional processes in supplying rock-derived nutrients to ecosystems, otherwise referred to as soil rejuvenation.

The second approach to characterize underlying substrates from remote sensing data in hilly and mountainous humid environments with dense canopies is based on Geobotanical Remote Sensing (GbRS) [253,254]. Deeply rooted in phyto/bio geochemistry, geobotany, and phytoremediation [99,255,256,257,258], GbRS uses remotely sensed data to identify vegetation anomalies, indicator species, and plant traits associated with heavy metal pollution, mineralization, and geo-environments for the purpose of phytoremediation, mineral exploration, and geologic delineation [242,259,260,261,262,263,264,265,266,267,268,269,270,271]. The principle behind GbRS is that plants growing at sites with anomalous concentrations of metals have altered physiological states that can be observed in the chemical and spectral signatures of leaves, the height and phenology of individuals and populations, and the composition and structure of communities; soil properties include concentration of elements and minerals (Figure 1). These altered physiological states reflect stress or hyperaccumulation of metals. At any given site targeted for mineral exploration, the within-species responses to different minerals as well as the among-species responses to a single mineral can be highly variable, and this has the potential to identify indicator species. In the context of ecosystem-geomorphic interaction studies, this work can potentially inform about changes in the redistribution of biomass between the ecosystem’s above- and belowground compartments and their influence on slope stability and landslide regeneration (Figure 1). In addition, it can inform about plant-microbial associations that may help plants survive under stressful conditions.

5. Integrating Plant, Soil, and Rock Hyperspectral Remote Sensing Studies to Understand the Functional Significance of Landslides in Mountainscapes

The “ecosystem-centered” view of landslides outlined in Figure 1 recognizes the critical influence of biodiversity both on slope resistance and susceptibility to failure and landslide recovery, and more broadly of interactions between ecosystem and geomorphic processes driving the dynamics of mountainscapes mediated by landslides [162,163,164,165]. In addition, this view acknowledges the importance of landslide size [43,44] and scale [163] to understand interactions between ecosystem and geomorphic processes from individual landslides and sites to entire landslide populations, assemblages, and landscapes.

Fieldwork has greatly contributed to the characterization of soil and plant communities in landslides and nearby forests based on a very small fraction of sites and landslides therein, e.g., [174,272,273] that do not reflect the global importance of this process [274]. The scarcity of field studies may be partially explained by the difficulties in accessing landslides, including the perils involved. Bringing these studies and trait databases together has allowed a taxonomic and functional characterization of landslides locally [272,275] and globally [276]; yet, the databases are limited in terms of the traits and species for which these data have been recorded. Traits of interest in the context of this work should inform about the contribution of vegetation to slope stability [277], the ability of the vegetation to recover in areas affected by landslides [272], and the status of soil fertility that may suggest trade-offs in plant functions associated with slope stability [174,278]. Scaling this work (Figure 1 and Figure 3; File S3: Figure S1) [201,279] can greatly benefit from the integration of diverse approaches and fields discussed in this review and an increasing availability of air- and spaceborne hyperspectral data, including planned missions (Files S1 and S2: Table S1). Yet, this will require addressing the challenges outlined earlier and designing studies to improve the retrieval of functional trait information from hyperspectral data in mountainous regions.

The increasing availability of air- and spaceborne hyperspectral data can advance the “ecosystem-centered” view of landslides (Figure 1) along a second dimension. Mountainous regions around the world are often poorly characterized in terms of their underlying soils, lithology, and degree of weathering, i.e., geological and lithological maps are lacking or are too coarse. Hyperspectral data may help peek through the disturbed canopies to characterize soils and rocks, which may serve as a proxy for soil fertility, soil mechanical properties, and degree of weathering. Altogether, these data may contribute to the investigation of relationships between plant, soil, and rock attributes and landsliding at scales commensurate with landslide activity. Furthermore, understanding the variability in space and time of these relationships may help test a priori hypotheses as well as postulate new ones regarding ecosystem-geomorphic interactions. This variability may be related to biotic (e.g., species) and abiotic (climate, soil fertility, geologic substrate, degree of weathering) conditions.

Landsliding creates new substrates and new landscape configurations, and the reorganization and recovery of ecosystems in these areas has implications beyond stabilizing sediment fluxes. On the one hand, landslides alter abiotic conditions, carbon and nutrient cycling, rock weathering, and soil formation [171,172,173,174] while offering unique habitats that favor unique subsets of species [166,167,168,169,170]. On the other hand, landslide size can influence the rates of these processes, including developmental trajectories (Figure 1), recovery rates [163] (Files S1 and S2: Table S3), and landscape memory [176]. In this sense, landslides may play a fundamental role in the functioning of mountainscapes. We would argue that this would be a third dimension of the “ecosystem-centered” view of landsliding that can benefit from the increasing availability of air- and spaceborne hyperspectral data.

In sum, an “ecosystem-centered” or holistic view of landslides greatly aided by hyperspectral and point-cloud data can reveal unknown relationships between ecosystem and geomorphic processes from local to regional scales while informing about the diversity and functioning of montane ecosystems and, ultimately, the long-term functional significance of landslides in mountainscapes (Figure 1). This may have implications for restoration, management, and risk assessment at a time of rapid global change.

6. Concluding Remarks and Future Directions

This review aimed at synthesizing established methods to retrieve functional traits from remotely sensed hyperspectral and point-cloud data, summarizing approaches to characterize functional diversity from remotely sensed functional traits, and elaborating on how methods and approaches used in different fields, together with an increasing availability of hyperspectral data, could support an “ecosystem-centered” view of landsliding. We found that the extent to which hyperspectral and point-cloud data have been used to characterize functional traits, functional diversity, and rock attributes in mountainous regions has been limited. Moreover, there has been very little integration among the various sources of data to understand ecosystem-geomorphic interactions and, ultimately, the dynamics of mountainscapes mediated by landsliding. We identified three dimensions of the “ecosystem-centered” view of landsliding that can clearly benefit from new approaches and the availability of air- and spaceborne hyperspectral and point-cloud data. Improving our knowledge along these dimensions will help us to better understand mountainous regions and landsliding and, ultimately, re-assess management practices and risk assessments. Mountain biomes are among the most diverse on Earth. At the same time, they are very susceptible to numerous drivers of change, including climate. Thus, at a time of rapid global changes, it becomes imperative to focus our attention on mountainscapes.