Applying fingerprint FTIR spectroscopy and chemometrics to assess soil ecosystem disturbance and recovery

Abstract

The assessment and monitoring of soil disturbance and its effect on soil quality (i.e., ability to support a range of ecosystem services) has been hindered due to the shortcomings of many traditional analytical techniques (e.g., soil enzyme activities, microbial incubations), including: high cost, long-term time investment and difficulties with data interpretation. Consequently, there is a critical need to develop a rapid and repeatable approach for quantifying changes in soil quality that will provide an assessment of the current status, condition and trend of natural and managed ecosystems. Here we report on a rapid, high-throughput approach to develop an ecological ‘fingerprint’ of a soil using Fourier transformed infrared (FTIR) spectroscopy and chemometric modeling, and its application to assess soil ecosystem status and trend. This methodology was applied in a highly disturbed forest ecosystem over a 19-year sampling period to detect changes in soil quality (detected via changes in spectral properties), resulting from changes in dynamic soil properties (e.g., soil organic matter, reactive mineralogy). Two chemometric statistical techniques (i.e., hierarchical clustering analysis and discriminate analysis of principal components) were evaluated for interpreting and quantifying similarities/dissimilarities between samples utilizing the entire FTIR spectra (i.e., fingerprint) from each sample. We found that this approach provided a means for clearly discriminating between degraded soils, soils in recovery and reference soils. Results from fingerprint FTIR analysis illustrate its power and potential for the monitoring and assessment of soil quality and soil landscape change.

Human activities are impacting the Earth’s ecosystems at an unprecedented rate.

Globally, 83% of the Earth’s land surface has been influenced by human activities, with 43% of the land surface experiencing some form of human induced degradation (Daily 1995, Sanderson et al. 2002). With the growing recognition of the goods and services soils provide for human well-being, there is a critical need to develop rapid, economical and repeatable measures of soil quality that will provide an assessment of the current status, condition and trend of natural and managed ecosystems. Soil quality is defined as the functional capacity of a soil to support important ecosystem services (e.g., produce food, feed, and fiber; enhance environmental quality), which includes both inherent (governed by the soil’s pedogenic properties), and dynamic (governed by soil management practices) soil ecosystem functions (Idowu et al. 2009). A soil’s inherent function is determined by static properties (e.g., mineralogy, texture, porosity, structure) that form in response to the particular combination of soil forming factors present in an ecosystem (i.e., climate, organism, topography, parent material, and time). In contrast, a soil’s dynamic function is determined by those properties and processes that can change (i.e., deteriorate) over relatively short time periods in response to human use and management (e.g., bulk density, soil organic matter content, soil microbial populations). Since soil quality results from the integration of physical, chemical, and biological soil properties, any practical assessment of soil quality requires identifying and integrating those key static and dynamic properties that control soil function.

Recent studies have proposed several conceptual frameworks for monitoring soil quality, including the use of soil biochemical properties/indices (Anderson 2003, Schloter et al. 2003, Gil-Sotres et al. 2005, Bastida et al. 2008) and the development of more general soil quality indices (SQI) based on key soil properties (Andrews et al. 2002; Andrews et al. 2004; Karlen and Stott 1994; Mohanty et al. 2007; Qi et al. 2009; Viscarra Rossel et al. 2006). Many studies investigating soil quality have focused on microbial based indicators due to its central role in driving soil processes and its rapid and dynamic response to different disturbance events. Microbial based indicators commonly include biochemical properties that reflect either the activity of microbial processes (i.e., microbial biomass C, dehydrogenase activity, soil respiration, etc.) or that of hydrolytic soil enzymes (i.e., urease and phosphatase activities) (Gil-Sotres et al. 2005). However, biochemical properties have produced contradictory results as indicators of soil quality due to a lack of reference values, contradictory behavior when degraded, and regional variations in expression levels Consequently, the assessment and monitoring of soil quality has been hindered due to the shortcomings of current analytical techniques, including: high cost, long-term time investment and difficulties with data interpretation.

The development of rapid, repeatable, and quantitative indicators of soil quality is necessary to overcome the financial and logistical constraints imposed by current analytical techniques. Rapid, high throughput approaches will enable more extensive spatial and temporal sampling of soils across a broad range of ecosystems, thus allowing more frequent assessments of landscape condition and change. While rapid quantitative methods for measuring soil quality are emerging (e.g., thermal analysis, pyrolysis mass-spectroscopy, infrared spectroscopy), results from many of these techniques can be difficult to interpret (Doran and Jones 1996). Infrared spectroscopy, however, has been used extensively to measure a wide range of soil properties using the mid-infrared (MIR) and near-infrared (NIR) wavelengths of the electromagnetic spectrum and the techniques for interpreting soil spectra are well defined. (Stenberg et al. 2010, Reeves 2010). FTIR spectroscopy (MIR wavelengths: 2.5 to 25μm), coupled with chemometic modeling, has become a popular technique for predicting individual soil properties and many studies have found strong correlations between specific MIR absorption bands and soil properties including soil carbon (McCarty and Reeves 2006, Ludwig et al. 2008, Ladoni et al. 2009, Reeves 2010, Bellon-Maurel and McBratney 2011), texture (McCarty and Reeves 2006, Viscarra Rossel et al. 2006, Soriano-Disla et al. 2014), nutrients (Janik et al. 1998, McCarty and Reeves 2006, Viscarra Rossel et al. 2006, Reeves and Smith 2009, Djuuna et al. 2011), and physical properties (e.g., bulk density, shrink-swell potential) (Minasny et al. 2008). However, focusing on an individual soil property filters out spectral information relating to other dynamic soil properties within the sample.

An alternative approach is to include the entire mid-infrared (MIR) spectrum of the soil in the analysis, which incorporates all information embedded within that spectrum, thus producing a biogeochemical or ecological ‘fingerprint’ of the sample. Fingerprint FTIR has been used successfully in a wide range of applications, including: differentiating bacterial and fungal species (Helm et al. 1991; Oberreuter et al. 2002; Goodacre et al. 1998); examination of salinity effects on tomato fruit (Johnson et al. 2003); detecting plant biochemical responses to N deposition (Gidman et al. 2005); differentiating between complex biological samples, including leaf litter and worm casts at different stages of microbial colonization and decomposition (Scullion et al. 2003); and degree of soil degradation from opencast mining (Elliott et al. 2007). This ‘fingerprinting’ methodology has broad applicability in assessing soil ecosystem disturbance due to its ability to detect changes in dynamic soil properties related to disturbance events, as well as track restoration trajectories as ecosystems recover (figure 1). In theory, fingerprint FTIR allows us to assess the degree of disturbance and associated loss of soil ecosystem function associated with a disturbance. This would be accomplished based upon the chemometric assessment of spectral similarity/dissimilarity within the FTIR spectra, given the availability of endmember spectra representing both the undisturbed reference and degraded states. In a study examining soils degraded from opencast mining, Elliott et al. (2007) were the first to demonstrate the potential of analyzing the complete FTIR soil spectra using chemometrics to differentiate soils at different stages of recovery. Despite the tremendous potential of this approach, no other studies have applied fingerprint FTIR and chemometrics as a rapid screening procedure for both assessing the current status and monitoring future changes in soil ecosystem health and function.

Figure 1.

Conceptual diagram illustrating the relationship between soil quality and degree of ecosystem degradation. Diagram further illustrates how ‘fingerprint’ FTIR can be used to chemometrically assess the spectral similarity/dissimilarity between ecosystem states along a restoration trajectory, where key absorption peaks (corresponding to changes in key dynamic soil properties) change relative to the degree of disturbance.

The main objective of this study was to evaluate the use of fingerprint FTIR and chemometric modeling to characterize soil disturbance effects across a 19-year sampling period in a mature second-growth Douglas-fir forest. The 19-year temporal sampling covered a wide range of ecosystem states and disturbance events: starting as a mature second-growth Douglas-fir forest; then subjected to clear-cutting; followed by intensive site preparation (mixing and tilling) and replanting; and finally post-disturbance successional recovery. Chemometric modeling using fingerprint FTIR spectroscopy incorporates all of the information contained within the spectra, and thus encapsulates information on a range of dynamic soil properties (e.g., soil carbon). Specific objectives were to (i) evaluate two chemometric statistical techniques (i.e., hierarchical clustering analysis and discriminate analysis of principal components) for interpreting and quantifying similarities/dissimilarities between samples utilizing fingerprnt FTIR; and (ii) evaluate the ecological significance of differences detected via fingerprint FTIR by evaluating changes in soil organic carbon (a master variable in assessing soil quality) across our sampling period.

Material and Methods

Study site and soil sampling.

The study was conducted in a mature (210 yrs.) Douglas-fir forest (P. menziesii) located on the west slope of the Cascade Mountain range in western Oregon (44.425°N, 122.037°W). The study site has an elevation of 1219 m, a mean annual air temperature (MAT) of 7.6°C, and a mean annual precipitation (MAP) of 204 cm. Soils in the study area are classified as Medial-skeletal Typic Fulvicryands (U.S. Soil Taxonomy) and are derived from glacially worked volcanic ejecta (volcanic ash/glacial till). Surface soils have a loamy sand texture and contain appreciable amounts allophane, imogolite, and undifferentiated amorphous minerals. The soil has a 2 cm thick organic horizon (Oi/Oa) underlain by a 0–13 cm mineral horizon (A) that contains approximately 5 to 8% organic matter.

The study site is located in an area of active timber harvesting, where clear-cutting and replanting are common land-use practices. The study area was clear-cut in 1991, followed by intensive site preparation and replanting in 1993 within a 6 × 13 m fenced study plot located within the clear-cut (figure 2a). Intensive site preparation consisted of mechanically tilling the surface organic and mineral horizons before planting Douglas-fir seedlings. Surface soil samples (0–13 cm) were collected periodically over a 19 year period (1991to 2010) within the fenced study plot to assess temporal changes in forest soil properties. A single surface soil sample was collected in 1991 (pre-clearcut) and 1993 (pre site preparation), while three replicate samples were randomly sampled within the fenced study plot in 1994 (1-yr post site preparation) and 1996 (3-yrs post site preparation); and four replicates in 1998 (5-yrs post site preparation), and 2010 (17-yrs post site preparation) (figure 2). Additionally, in 2010, four replicate samples were randomly sampled within an adjacent mature forest (contemporary reference) (figure 2). Soil samples were collected using a 4.74 cm diameter hand operated soil corer. In the field the collected soil samples were stored on ice during transport to the lab, where they were lyophilized and passed through a 2 mm sieve to remove coarse fragments and roots. The dry, sieved soils were mixed and a sub-sample was finely ground (80 mesh or finer) using a Retsch MM-400 or equivalent mixer mill. Total carbon (TC) analysis was performed by high temperature dry combustion on a Costech Model ESC4010 elemental analyzer. Sodium pyrophosphate extractable carbon (PP-C) was extracted by mixing 0.5 g of sieved (<2mm) soil samples with 30 mls of 0.1 M Na₄P₂O₇ and shaken overnight. The solution was then allowed to settle overnight before centrifuging and filtering (0.45 μm) to obtain a clear extract (Soil Survey Staff 2014). The organic C in the sodium pyrophosphate extract was analyzed using a Dohrmann UV enhanced-persulfate TOC analyzer (Phoenix 8000) with a limit of detection (LOD) of ~0.10 mg L⁻¹.

Figure 2.

Schematic of a) study site location with clear-cut and forest reference plot locations and b) timeline of forest ecosystem disturbance over a 19-year sampling period.

FTIR spectroscopy.

The finely ground soil samples were re-dried overnight (60°C) prior to measurement. FTIR spectra of each sample (approx. 30 mg) were collected in the diffuse reflectance mode using a Thermo Nicolet Nexus 670 FT-IR spectrophotometer equipped with a sixty sample auto-sampler PIKE AutoDiff accessory (PIKE Technologies, Madison, WI) with KBr serving as the background reference sample as recommended by the manufacturer. The samples were run neat (i.e., no KBr was added) and the FTIR spectra were collected over a wavenumber range from 4000 to 600 cm⁻¹ and are the result of 128 co-added scans recorded at 4cm^–1 resolution, resulting in 850 spectral bands per sample. Prior to analysis, the FTIR spectra were transformed from reflectance units to absorbance units (log [1/reflectance]).

Spectral processing and chemometric analysis.

All FTIR spectra were smoothed using locally weighted smoothing (LOESS) in order to increase the signal to noise ratio. The smoothing procedure resulted in a reduction of the spectral resolution from 4 to 8 cm^-1. FTIR spectra were then baseline corrected using a third order polynomial baseline correction function. The baseline correction function automatically finds appropriate spectral regions for baseline fitting by iteratively fitting a polynomial to the spectrum until a fit is obtained where no points from the spectrum fall below the final polynomial. As a final step, the spectra were normalized using area normalization. Processed spectra were then analyzed using two different chemometric techniques: (1) hierarchical clustering analysis (HCA), and (2) discriminate analysis of principal components (DAPC). HCA performs an agglomerative hierarchical cluster analysis of the FTIR spectra. In agglomerative clustering each observation starts in its own cluster and pairs of clusters are merged as you move up the hierarchy. Clusters are merged based on some measure of dissimilarity between groups of observations. We applied Ward’s minimum variance method as our dissimilarity metric, which minimizes the total within-cluster variance. DAPC identifies grouping within the data structure by minimizing within group variation, while optimizing the variance between groups (Jombart et al. 2010). DAPC identifies the optimal number of groups by using k-means clustering (maximizes the variation between groups) of PCA transformed data, with k-means run sequentially with increasing numbers of clusters and the clustering solution identified using the lowest Bayesian Information Criterion (BIC) (Jombart 2013). Once the optimal number of clusters has been identified, DAPC transforms the data using PCA and performs a discriminate analysis on the retained principal components. The variable contribution of each wavenumber in discriminating between each group or cluster can be visualized for each discriminant function, thus providing insight into the important biochemical differences between groups. All chemometric data processing and statistical analysis was performed using R statistical software. Spectral processing and HCA were performed using the ‘hyperSpec’ package (Beleites and Sergo 2015). DAPC was performed using the ‘adegenet’ package (Jombart et al. 2010).

Results and Discussion

Soil properties and spectral features.

Results from our analysis of TC concentration, showed there was no appreciable change in TC between 1991 and 1993, indicating that tree harvesting had a minimal effect on soil C (figure 3a). However, dramatic increases in TC (5-fold increase) occurred after site preparation (i.e., 1994), followed by a gradual decrease with time towards pre-disturbance levels by 2010 (figure 3a). The 5-fold increase in soil carbon in 1994 is not surprising given that the top 15 cm of the soil was mechanically tilled, incorporating the 2 cm Oi horizon (~38% carbon) into the top mineral horizon (~2 to 4% carbon). A similar pattern can be seen with PP-C, corresponding to organically complexed Fe and Al, which increased dramatically in 1994 followed by a gradual decrease towards pre-disturbance levels by 2010 (figure 3b). FTIR spectra across the 19-year sampling period show a similar trend of changing absorption peaks (figure 4). While the spectra share common spectral features across all sampling years, subtle differences in several peaks can be visually identified. In particular the aliphatic (C-H stretching; ~2910 cm⁻¹), phenolic (C-O stretching; ~1240 cm⁻¹) and aromatic (C=C stretching; ~1490 and 1650 cm⁻¹) regions were identified, where increased aliphatic (fats, lipids) and decreased phenolic/aromatic (lignin) absorption occurs following site preparation (1993) (figure 4, table 1). By 2010, however, the aliphatic and phenolic/aromatic peaks return to pre-disturbance absorption values (figure 4). These changes can be attributed to a large input of particulate organic matter (high aliphatic, low phenolic/aromatic) in 1993 during site preparation when the O and A soil horizons were mechanically mixed. Soils within the study plot contain appreciable amounts of allophane (25%), imogolite (10%), and undifferentiated amorphous minerals (43%) (Unpublished data), which provides reactive surfaces for the complexation and stabilization of SOM. The pyrophosphate extract dissolved organic carbon that represents the microbial decomposed soil carbon fraction that is stabilized by reactive mineral surfaces (e.g., oxides, clays, allophane/imogolite). This reactive minerology imparts a high degree of ecological resilience by preventing the loss of SOM by sorption processes during disturbance events, as seen by the short-term stabilization of carbon in these soils (figure 3b).

Figure 3.

Boxplot of a) soil carbon and b) sodium pyrophosphate extractable C in the surface horizon (0–13 cm) over a 19-year sampling period. The middle of each boxplot indicates the median value. The upper and lower edges of each boxplot indicate the 75th and 25th percentiles, respectively. The ends of the vertical lines indicate the minimum and maximum data values. Single samples were collected in 1991 and 1993, while replicate samples were collected in 1994 (n=3), 1996 (n=3), 1998 (n=4), and 2010 (n=4). 2010-cc: samples collected within the clear-cut in 2010. 2010-ref: samples collected within the reference forest in 2010.

Figure 4.

Fourier transform infrared (FTIR) spectra of the surface soil (0–13 cm) over 19 years. Grey shaded rectangles show the wavenumber region surrounding the aliphatic (C-H stretching at ~2910–2700 cm⁻¹), aromatic (C=C at ~1680–1640 cm⁻¹ and ~1440 cm⁻¹), and the phenolic (C-O stretching at ~1240 cm⁻¹) absorption peaks. 2010-cc: samples collected within the clear-cut in 2010. 2010-ref: samples collected within the reference forest in 2010.

Table 1.

Assignment of the main IR absorption bands (wavenumber cm⁻¹) in the surface forest soil samples.

Wavenumber (cm−1)	Assignment	Characterization†
2830–2695	Asymmetric and symmetric C-H stretch	Fats, waxes, lipids
1820–1860	C=O stretch	Carboxylic acids
1730	C=O stretch	Carboxylic acids
1680–1640	C=C stretch	Aromatic structures
1600–1585	C-C (in-ring)	Aromatic structures
1460	C-H deformations	Phenolic (lignin) and aliphatic structures
1440	C=C stretch	Aromatic carbon, indicative of lignin
1270–1210	C-O	Lignin backbone
1160	C-O-C	Polysaccharides
1050–1080	Si-O stretch	Silica/clay minerals
1030–1080	C-O stretch and O-H deformation	Polysaccharides

^†Reference absorption bands taken from: Artz et al. (2008), Wander and Traina (1996) and references therein.

Chemometric analysis.

While visual inspection of FTIR spectra can detect prominent spectral differences between samples, it cannot account for more subtle changes occurring simultaneously across the FTIR spectra. Consequently, chemometric techniques are required to evaluate these complex multivariate datasets to identify meaningful biochemical or ecological differences. The first chemometric technique used to analyze similarities/differences in FTIR spectra across our temporal sampling was HCA (figure 5). HCA produced two main groupings of spectra (1991, 1993, 2010-ref, vs. 1994, 1996, 1998), indicating its ability to separate the soil spectra between highly disturbed years vs. pre-disturbance years and the contemporary reference. Samples from the 19-year post clear-cut in 2010, however, were split across the two groups indicating that disturbance effects can still be seen 17-years post intensive site disturbance. The physical disturbance of the soil in 1994 resulted in a high degree of variability in soil properties and processes. This is illustrated by the large inter-quartile range for both TC and PP-C in 1994, which steadily decreased by 2010 but remained considerably larger than the contemporary forest reference (2010-ref) (figure 3). Additionally, as the dendrogram is further subdivided, the subgroupings of the right side branch (disturbance years) are highly mixed in terms of years since disturbance, further illustrating the high degree of variability in soil properties resulting from the physical mixing of the soil. In contrast, the subgroupings of the left side branch show much clearer distinction between sampling year. Although 1991 and 1993 only had a single sample, the contemporary forest reference (2010-ref) showed tight grouping within the dendrogram as well as a narrow inter-quantile range for both soil C and PP-C (Figs. 3 & 5).Our second chemometric technique was DAPC analysis (figure 6). DAPC requires prior group identification through k-means clustering of PCA transformed data. The optimal number of clusters was identified by running k-means clustering sequentially with increasing values of k, and choosing the clustering solution with the lowest BIC. The optimal number of clusters in our relatively simple dataset was two. DAPC then transformed the data using PCA and performed a discriminate analysis on the 7 retained principal components (explaining ~96% of the overall variance). For this dataset, there was good separation between the two clusters with only minimal overlap in the density of samples across discriminant function 1 (figure 6a). Group membership within the DAPC shows close to 100% membership probability for each soil sample within each sampling year between the two clusters (figure 6b). Unlike PCA, which summarizes the overall variability among soil spectra (both between-group and within-group variability), DAPC summarizes the spectral differentiation between groups, while overlooking within-group variation. Consequently, this technique has the potential to detect significant shifts or transitions of the soil ecosystem to alternate states. These results match those from HCA, supporting the utility of these techniques in deciphering disturbance patterns and trajectories using soil spectral information coupled with chemometric analysis.

Figure 5.

Cluster dendrogram of surface soil FTIR spectra over a 19-year sampling period. 2010-cc: samples collected within the clear-cut in 2010. 2010-ref: samples collected within the reference forest in 2010.

Figure 6.

Discriminant analysis of principal components (DAPC) of surface soil FTIR spectra showing, (a) scatterplot of discriminant functions producing two spectrally distinct clusters, (b) the percent group membership between the two clusters for each sampling year over the 19-year sampling period, and (c) component loading for each FTIR spectra across discriminant function 1. 2010-cc: samples collected within the clear-cut in 2010. 2010-ref: samples collected within the reference forest in 2010.

In addition to grouping soil samples by their spectral similarity/dissimilarity, the DAPC technique allows us to compute the loading or contribution of each spectral band, thus allowing us to identify regions of the FTIR spectra that contributed strongly to differentiation between groups. The variable contribution for each spectral wavenumber across discriminant function 1 is shown in figure 6c. The FTIR spectral regions that contribute the strongest to cluster differentiation include: C-O (1180 to 1260 cm⁻¹; phenolic), C=C (1640 to 1700 cm⁻¹; aromatic), C-C (1585–1630 cm⁻¹; aromatic), and C=O (1800 to 1880 cm⁻¹; carboxylic acids). Smaller contributions can be seen by C=C (1400–1460 cm⁻¹; aromatic/phenolic), and C-H (2750 to 2830 cm⁻¹; aliphatic) (figure 6c, table 1). These results support our visual interpretation of the FTIR spectra, confirming the importance of the phenolic/aromatic absorption bands in differentiating soil groups. The contribution of the aliphatic absorption bands (2750 to 2830 cm⁻¹) was smaller than expected from our visual interpretation (Figs. 3 and 6c). However, in general these results confirm our initial interpretation of changing soil organo-mineral dynamics following intensive site disturbance in 1993, resulting in high aliphatic and low phenolic/aromatic absorption peaks and the gradual reversal of these trends during the 17-year successional recovery.

Potential application for monitoring soil ecosystem disturbance and recovery.

The magnitude and rate of global landscape change necessitates the development of diagnostic screening tools for assessing soil condition at local to regional scales, and that will allow for spatially explicit targeting of management interventions. Soils are an inherently variable resource, both spatially and temporally, and exhibit a high degree of complexity in their expression of soil properties. This inherent complexity and variability has made it difficult to devise consistent and reproducible techniques for monitoring changes in soil quality. Therefore, it is critical to consider the whole soil ecosystem when assessing the recovery of disturbed soils, for some soil properties may recover within a few years while others may take decades to return to pre-disturbance conditions (Insam and Domsch 1988). For example, Hart et al. (1989) documented that after 10–25 years the earthworm and microbial populations and easily mineralizable nitrogen pool had almost fully recovered in mined soils, while total C and N were much slower to recover. Consequently, implementing approaches that utilize all relevant information embedded within the spectral signature, such as the fingerprint FTIR approach presented in this study, are essential for accurately assessing changes in soil quality.

A major strength of this approach is that potential differences in samples may be detected even when their prior nature is unknown (Elliott et al. 2007). When analyzed using chemometric approaches and paired to samples from known reference sites, soils may be classified according to their similarity to the reference sites (a qualitative measure of degree of degradation or recovery) and spectral differences corresponding to specific biochemical changes may be identified. This rapid screening approach provides information on where specific biochemical changes have occurred thus focusing future analysis for validation of biogeochemical changes caused by degradation processes. Thus, if desired, subsequent to the initial screening procedure, more time and labor intensive analyses can be performed on a subset of the dataset, representing the range of spectral variability across all samples.

While the results from this study, along with those from Elliott et al. (2007), confirm the tremendous potential of fingerprint FTIR as a rapid screening procedure for monitoring changes in soil ecosystem health and function, there are several issues that need to be considered when implementing this approach. The first involves establishing end member states, that is, a spectral signature representing the pristine reference condition and the degraded state. Conceptually, this can be visualized via a gradient of ecosystem states, as illustrated in figure 1. As a soil ecosystem moves from a pristine reference condition to a degraded state, its spectral signature will change in association with changes in dynamic soil properties. These changes can be quantified via changes in the soil spectral fingerprint (figure 1). However, in order to assess the degree of spectral similarity/dissimilarity between ecosystem states, both pristine and degraded reference conditions must be known. In this study, temporal sampling allowed us to quantify changes in soil spectral state across a wide range of ecosystem states. Additionally, our contemporary forest reference allowed us to evaluate the current condition of the recovering clear-cut site. Since it is difficult to find long-term datasets to evaluate soil disturbance and recovery, the use of contemporary reference sites will allow for broad scale assessments of soil condition within actively managed or impacted ecosystems. Another consideration has to do with the need for standardization of sampling, preprocessing, and analysis protocols. While FTIR analysis provides a standardized approach for sample analysis, standardized field sampling protocols, sample preprocessing steps (e.g., sample griding, KBr dilution) and analysis protocols (e.g., DRIFT vs ATR vs Transmission) need to be established so that FTIR spectra are comparable.

Summary and Conclusions

Soils represent the fundamental support system for terrestrial ecosystems, and thus any anthropogenic stressor that diminishes the soil’s ecological structure or function may potentially impact ecosystem health, productivity, species composition, hydrologic function, and ultimately the provision of ecosystem services (O’Neill et al. 2005). The development of rapid, repeatable, and quantitative techniques for monitoring changes in soil quality are necessary for understanding how both natural and anthropogenic change drivers are impacting soil ecosystem function. Fingerprint FTIR spectroscopy encapsulates information on the physical, chemical, and biological properties of the soil and thus directly relates to many dynamic soil properties that effect ecological processes. As we have shown, the method is highly sensitive to management practices and is capable of quantifying ecosystem status and condition compared to a reference condition, and thereby an assessment of ecosystem resilience and recovery. Finally, chemometric modeling of FTIR spectra produces easily interpretable indices that effectively discriminate between degraded vs. reference/restored states, thus allowing land managers and conservationists a rapid assessment tool for determining the current status, condition and trend of natural and managed ecosystems. Consequently, this is an effective approach for assessing the efficacy of restoration efforts and will assist in the long-term and large scale monitoring of soil and ultimately landscape condition.