Reflectance-Based Vegetation Index Assessment of Four Plant Species Exposed to Lithium Chloride

This study considers whether a relationship exists between response to lithium (Li) exposure and select vegetation indices (VI) determined from reflectance spectra in each of four plant species: Arabidopsis thaliana, Helianthus annuus (sunflower), Brassica napus (rape), and Zea mays (corn). Reflectance spectra were collected every week for three weeks using an ASD FieldSpec Pro spectroradiometer with both a contact probe (CP) and a field of view probe (FOV) for plants treated twice weekly in a laboratory setting with 0 mM (control) or 15 mM of lithium chloride (LiCl) solution. Plants were harvested each week after spectra collection for determination of relevant physical endpoints such as relative water content and chlorophyll content. Mixed effects analyses were conducted on selected endpoints and vegetation indices (VI) to determine the significance of the effects of treatment level and length of treatment as well as to determine which VI would be appropriate predictors of treatment-dependent endpoints. Of the species considered, A. thaliana exhibited the most significant effects and corresponding shifts in reflectance spectra. Depending on the species and endpoint, the most relevant VIs in this study were NDVI, PSND, YI, R1676/R1933, R750/R550, and R950/R750.


Introduction
Anthropogenic activities can result in the release of a wide array of contaminants, particularly metals, to the environment. Responsible environmental stewardship involves the management and remediation of such releases. Numerous remediation strategies exist, depending on the circumstance, but the technique considered here is reflectance spectroscopy. Reflectance spectroscopy has potential for use as a cost-effective, non-destructive analytical technique for detecting and assessing plant stress, specifically metal stress [1][2][3].
Chemometric mathematical methods are often used to analyze reflectance spectra when extensive samples are available; a few hundred are required to develop a robust model. For smaller sample sizes, as in this study, vegetation indices (VI; mathematical combinations of different reflectance spectral bands) can be used to provide rapid and convenient semi-analytical measures of vegetation activity, which in turn can provide indication of plant health [4].
The ability to detect metal exposure in plants is particularly relevant when employing phytoremediation strategies. Phytoremediation is defined, for our purposes, as the use of green plants for environmental clean-up, or the use of plants to remove or neutralize pollutants in the biosphere [5]. Specific applications for reflectance spectroscopy as a tool complimentary to phytoremediation could include: (1) early recognition of contaminated areas through identification of plant stress indicative of metal exposure, (2) surveillance of sites with existing contamination or sites with the potential to become contaminated, (3) assessment of phytoremediation efforts, (4) assessment of risk to human health and the environment through coarse quantification of contamination, etc.
The contaminant of concern in this study is lithium. Lithium (Li) is widely used in the USA, which is the leading producer and consumer of Li materials, finding utility in ceramics and glass, aluminum production, the medical industry, certain batteries and greases, nuclear reactor coolant, radiation dosimeters, and historically, in nuclear weapon development [6][7][8][9]. Although Li is not a radioactive concern, it is an anthropogenic contaminant related to the nuclear fuel cycle and to legacy waste and contamination from nuclear weapons development [8,10,11]. For example, historical waste-disposal activities at the Department of Energy's Oak Ridge Y-12 Plant resulted in the release of Li to the groundwater [8]. Numerous studies have shown that there are shifts in plant reflectance spectra due to metal stress (or simulated metal stress) [2,[12][13][14][15][16][17][18][19][20][21][22][23][24] but to the authors' knowledge none have considered response to Li exposure.
To further investigate the use of reflectance spectroscopy as a useful method for assessing metal stress in plants, reflectance spectra for four species of plants were collected every week for three weeks using an ASD FieldSpec Pro spectroradiometer with both a contact probe and a field of view (FOV) probe. These plants were treated twice weekly in a laboratory setting with 0 mM (control) or 15 mM of lithium chloride (LiCl) solution and harvested weekly, immediately after spectra collection, for assessment of physical endpoints such as relative water content and chlorophyll content. The specific objectives of this exploratory study are to: (1) identify changes in plant status due to Li exposure in multiple plant species and (2) determine if reflectance is useful in identifying these changes through the utilization of both previously defined and newly determined vegetation indices.

Plant Species Considered
The species considered were Arabidopsis thaliana, Helianthus annuus (sunflower), Brassica napus (rape), and Zea mays (corn). A. thaliana is a member of the mustard family that is closely related to various crop plants. It has been the subject of intense study over the past several decades and is considered a model organism and ideal for use in the laboratory setting for biological research [25]. Several species of the Brassica family, which are vegetable and oilseed crops, have been identified as metal accumulators and are considered potential phytoremediation candidates [26][27][28]. H. annuus is an ornamental flower as well as an important environmental crop, and it has been shown to be an effective phytoremediation crop [29][30][31]. Z. mays has also been shown to have metal phytoremediation potential [31]. It is the major feed grain (>95%) in the United States and is also processed into a broad assortment of food stuffs, from cereals to sweeteners. Additionally, Z. mays has industrial utility as a component in the fabrication of fuel ethanol. The United States is currently the world's largest producer and exporter of corn [32]. Several studies have considered the reflectance spectra of corn and sunflower, from assessing pigment concentrations to nutrient/water status and photosynthetic efficiency at both leaf and canopy scales [33][34][35][36][37][38][39][40][41][42][43][44][45].
Once pots were arranged for treatment, spike solution was evenly applied to the top of each pot as 100 mL (25 mL delivered to each quadrant) of 15 mM LiCl in 1/64 strength nutrient solution twice weekly, with control plants receiving 100 mL 1/64 nutrient solution only. Two pots were randomly selected from each treatment group for weekly spectra collection and harvest. After each application of nutrient solution, the plants were rotated within the tubs and the tubs were rearranged among the growth shelves to account for potential variation in lighting or other environmental conditions.

Equipment Setup and Spectra Collection
Reflectance spectra were collected using a FieldSpec Pro (FSP 350-2500P; Analytical Spectral Devices (ASD), Boulder, CO, USA) which is a full range (350 nm to 2500 nm) portable spectroradiometer (with sampling intervals/spectral resolutions of 1.4 nm/3 nm and 2 nm/10 nm for 350 to 1000 nm and 1000 to 2500 nm respectively) [59]. Contact probe (CP) spectra were collected using a leaf clip attachment on individual leaves. The CP provides light (3.825 V, 4.05 W low intensity bulb) and collects reflectance spectra. The leaf clip attachment has both a white (for white reference) and black (to minimize back scatter) background. Multiple reflectance readings were taken on the leaves of each species of plant to obtain an overall representation of reflectance.
FOV spectra were collected using an 8 • probe (i.e., a viewing angle of 8 • ). Incident light was provided by two halogen lamps (Pro Lamp, 14.5 V, 50 W, P/N 145378, ASD) angled at 30 degrees from horizontal. The lights were 180 • apart at 30.5 cm from the center of pot on the horizontal and 76.2 cm above the table surface. The fore optics probe was centered between the lights at 66.7 cm above the plane of the pot surface, resulting in a spot size diameter of 9.32 cm. Reflective surfaces were covered with light-absorbent material to minimize noise and thus variability in spectra, and dark room conditions were approximated by surrounding the lights and fore optics with a black felt canopy. Tripod surfaces were also wrapped in black felt and the table surface was lined with light-absorbent black rubber. The white reference was a calibrated Spectralon (25.4 × 25.4 cm, LabSphere, North Sutton, NH, USA) panel of 99% reflectance that was elevated to a height equivalent to a grow pot. Four spectra, each collected at a different arbitrary rotation of the pot, were acquired and then averaged to get an overall assessment of the reflectance of the sample. FOV spectra were always acquired prior to CP because it is possible for the CP to injure the plant and therefore affect subsequent FOV readings.

Collection of Physical Measures
As metal stress is known to mimic drought stress [3], plants were harvested after spectra collection each week to determine chlorophyll content and relative water content. The concentrations of chlorophyll a (Chl a) and chlorophyll b (Chl b) were determined for each replicate (i.e., pot) [26,60,61]. Four circular leaf subsamples were collected from representative leaves of the plants in a pot using a #3 cork borer (Fisher Scientific, Pittsburgh, PA, USA). Leaf samples were stored in the dark at 4 • C in capped 20 mL vials (KG-33 borosilicate glass; Kimble Chase, Vineland, NJ, USA) containing 2 mL 100% ethanol for three days before absorbance (A) at 665 nm, 649 nm, 629 nm, and 696 nm, with an offset at 750 nm, was determined for 1.5 mL subsamples for each vial using a NanoDrop 2000c UV-Vis spectrophotometer (Thermo Scientific, Wilmington, DE, USA). Disposable methacrylate cuvettes with transmission from 300 to 800 nm >80% were used with the 1.5 mL subsample (Cole Palmer, Vernon Hills, IL, USA). Chlorophyll content was determined using appropriate, previously published [61] equations where A x is absorbance at x nm: After leaves were sampled for chlorophyll content, additional sufficient leaves were removed to obtain between 1000 and 2000 mg of fresh mass for each replicate (i.e., pot) to determine relative water content. Samples were placed in weigh boats, fresh mass was obtained, samples were dried in an oven to a constant mass, and dry mass was obtained. A sample's relative water content (RWC) was then expressed as Equation (3): Lithium uptake has been shown to have a stimulatory effect at low levels, and potentially cause necrosis at high levels depending on the plant species [47,48], so remaining plant material from above was similarly retained, dried, and weighed to determine each sample's overall dry biomass. Note that dry biomass was determined for all experiments except the first A. thaliana experiment, as these samples were initially used elsewhere for teaching purposes.
A visual assessment of the proportion of a pot covered by plant material was performed by overlaying a 6 × 6 (18 mm × 18 mm) grid on top-down photos of each group of plants at each of three timepoints, forming 36 squares with 49 evenly-spaced points (grid intersections). Photographs were taken immediately prior to spectra collection, directly above each six-pot treatment group and six-pot control group in the same manner each week. However, to account for any potential change in magnification or alignment, gridlines were laid based on pot dimensions, which were definitively consistent. Using the grid intersections, an additional endpoint, Coarse Leaf Area Index (CLAI), was defined as: where N total is the total number of points in the grid and N leaf is the number of points on leaf material. CLAI provides an approximate indication of how much of the pot surface is covered by plant material. Finally, heights of plants above the top of the pot were measured. The only plants with noticeable variation in heights were H. annuus and thus are the only results reported for plant height.
For each species at each view, a mixed-effects model analysis was conducted for each vegetation index and each plant endpoint (RWC, Chl a + b, dry biomass, CLAI, and/or height) to consider the fixed effects of week (1, 2, 3), treatment group (control or 15 mM LiCl treatments), and week-by-treatment interaction with a random effect for sampling unit. The Kenward-Rogers approximation of degrees of freedom was used to account for variation among the week-by-treatment combinations. When a treatment effect or week-by-treatment interaction was significant for a plant stress endpoint (RWC, Chl a + b, biomass, CLAI, and/or height), subsequent linear mixed-effects analyses were conducted to consider the endpoint measure as a dependent variable with vegetation indices as predictors that also had significant treatment or week-by-treatment interaction effects from the mixed-effects model analysis as predictors. SAS Software v. 9.4 (SAS Institute Inc., Cary, NC, USA) was used for all statistical analyses and a significance level of 0.05 was used for all tests of significance.

Spectra
Average relative reflectance spectra of treatment compared to control are shown by week and technique (CP or FOV) for each species in Figure 1 (CP) and Figure 2 (FOV). Average reflectance spectra for each experimental group (i.e., control and treatment) are shown in Appendix A ( Figures A1-A4). Average reflectance data (from which the figures were developed) is included in the supplemental online material. Notice that the vertical axes in Figures 1 and 2

Phenotypic Observations
A. thaliana had visible symptoms of Li exposure starting in week 1, from a slight yellowing (chlorosis) at the leaf tips in week 1 to significant necrosis in week 3 ( Figure 3).

Phenotypic Observations
A. thaliana had visible symptoms of Li exposure starting in week 1, from a slight yellowing (chlorosis) at the leaf tips in week 1 to significant necrosis in week 3 ( Figure 3).

Phenotypic Observations
A. thaliana had visible symptoms of Li exposure starting in week 1, from a slight yellowing (chlorosis) at the leaf tips in week 1 to significant necrosis in week 3 ( Figure 3).   B. napus plants showed some symptoms of toxicity around leaf edges in weeks 2 and 3 ( Figure 4a). A few H. annuus plants began exhibiting slight symptoms of lithium toxicity between weeks 2 and 3, as slightly mottled leaves and occasional necrotic spots (Figure 4b). Z. mays showed no symptoms of toxicity. Observed symptoms are typical of lithium toxicity; different species of plants are known to have varying tolerances to lithium exposure [48,72].

Endpoints
Plots of chlorophyll content, relative water content, dry biomass, and CLAI by treatment level and time are shown in Figures 5-8 respectively for each species. A plot of H. annuus height is shown in Figure 9. Summary statistics for these endpoints are provided in Appendix B. Notice in Figure  B. napus plants showed some symptoms of toxicity around leaf edges in weeks 2 and 3 ( Figure 4a). A few H. annuus plants began exhibiting slight symptoms of lithium toxicity between weeks 2 and 3, as slightly mottled leaves and occasional necrotic spots (Figure 4b). Z. mays showed no symptoms of toxicity. Observed symptoms are typical of lithium toxicity; different species of plants are known to have varying tolerances to lithium exposure [48,72].

Endpoints
Plots of chlorophyll content, relative water content, dry biomass, and CLAI by treatment level and time are shown in         Results (p-values) of the statistical analysis for relative water content, chlorophyll content, dry biomass, CLAI, and height are shown in Table 1. F(n,d) in table headers indicates the degrees of freedom with which the F test statistic and associated p-value were calculated, where n is numerator degrees of freedom, and d denominator degrees of freedom. Note that denominator degrees of freedom vary due to differing numbers of observations and Kenward-Rogers degrees of freedom approximation. There are no significant time, treatment, or time by treatment interaction effects for B. napus chlorophyll content, nor for Z. mays CLAI. All other endpoints had significant differences in time for each species. There was a significant treatment effect on Chl a + b and a significant time by treatment interaction effect on height for H. annuus.   Results (p-values) of the statistical analysis for relative water content, chlorophyll content, dry biomass, CLAI, and height are shown in Table 1. F(n,d) in table headers indicates the degrees of freedom with which the F test statistic and associated p-value were calculated, where n is numerator degrees of freedom, and d denominator degrees of freedom. Note that denominator degrees of freedom vary due to differing numbers of observations and Kenward-Rogers degrees of freedom approximation. There are no significant time, treatment, or time by treatment interaction effects for B. napus chlorophyll content, nor for Z. mays CLAI. All other endpoints had significant differences in time for each species. There was a significant treatment effect on Chl a + b and a significant time by treatment interaction effect on height for H. annuus. Results (p-values) of the statistical analysis for relative water content, chlorophyll content, dry biomass, CLAI, and height are shown in Table 1. F(n,d) in table headers indicates the degrees of freedom with which the F test statistic and associated p-value were calculated, where n is numerator degrees of freedom, and d denominator degrees of freedom. Note that denominator degrees of freedom vary due to differing numbers of observations and Kenward-Rogers degrees of freedom approximation. There are no significant time, treatment, or time by treatment interaction effects for B. napus chlorophyll content, nor for Z. mays CLAI. All other endpoints had significant differences in time for each species. There was a significant treatment effect on Chl a + b and a significant time by treatment interaction effect on height for H. annuus.

Vegetation Indices and Overall Response
Detailed results (p-values) from the initial mixed-effects analysis of the vegetation indices are contained in Appendix C. Significant p-values (p < 0.05; bolded) indicate a statistically significant difference between the treatment, week, or treatment-by-week interaction outcome means. As above, denominator degrees of freedom vary. From these values we see that methods of spectra acquisition as well as the results between species are generally varied. However, this is not wholly unexpected as four fairly different species were considered; these species vary in leaf size, height, structure, etc.
Nearly all A. thaliana VIs had significant treatment, time, and interaction effects between treatment and time, for both spectra acquisition techniques (CP and FOV). For other species, however, results differ between VI, spectra acquisition techniques, endpoints, and time/treatment. Every species exhibited significant time-dependent differences in VI. Corn only had significant treatment effects for YI acquired by FOV. H. annuus had significant treatment or interaction differences in PSND, YI, R 750 /R 550 , and R 1636 /R 1933 . B. napus had significant treatment or interaction differences in WI, YI, R 750/R550 , and R 1636/1933 . However, only A. thaliana and H. annuus had significant treatment effects with respect to measured endpoints (RWC, Chl a + b, biomass, CLAI, and/or height) ( Table 1), so only these species were considered in secondary (i.e., follow-up) analysis to determine which VIs were significant predictors of these endpoints. Results of the corresponding mixed effects model are shown in Table 2 below. The only index with no predictive power was WI; all other VI were significant predictors of at least two endpoints. Interestingly, R 750 /R 550 was a significant predictor in at least one view (CP or FOV) of all endpoints considered, as further discussed below. This index is therefore deemed the most promising of those considered for prediction of lithium exposure.
-Not included in secondary analysis.

Discussion
In this study, four species of plants were treated with lithium chloride over three weeks to discern if symptoms of the exposure could be identified from their reflectance spectra through the use of VI. Plants were harvested each week, with reflectance spectra collected at the whole plant and leaf scales. Various physical endpoints were measured immediately after spectra acquisition. Controls were maintained in parallel with the treatment plants at the same sample size, as matching controls are important to be able to discern if temporal changes in reflectance result from contaminant exposure or from the natural growth pattern or inherent biological variability of the plants [17,22,69].
The progressive Li exposure had slightly different effects, as well as severity of effects depending on the species. All species considered exhibited significant changes in time, but only physical measures of H. annuus and A. thaliana exhibited significant treatment effects. Although Z. mays did not demonstrate any obvious symptoms of Li toxicity, B. napus did show slight symptoms of toxicity around the leaf edges starting in week 1. Uptake of Li is expected to be higher in dicots (e.g., A. thaliana, H. annuus, and B. napus) than in monocots (e.g., Z. mays) [7], which might explain why no symptoms were seen in Z. mays. Similarly, the low biomass of A. thaliana as compared to plants with much higher biomass for an equivalent soil concentration may have led to the more severe toxicity symptoms.
Observed symptoms in A. thaliana, H. annuus, and B. napus were consistent with Li toxicity, which in turn are similar to those resulting from Mg deficiency [56]. Mg is essential for photosynthesis as it is the central element of the chlorophyll molecule [73]. It is possible that the competitive binding of Li results in Mg deficiency, exacerbating the toxicity symptoms. Of the initial 15 VI considered, seven were found to be significant predictors of these treatment-dependent endpoints. Comparisons between species and the associated relationships with the acquired spectra and calculated VIs are discussed below.

Chlorophyll Content
Excess metal exposure negatively affects photosynthetic processes and typically induces a general stress reaction in plants [74]. Photosynthetic pigments typically decrease with metal exposure, which has obvious consequences for photosynthesis and plant growth. Inhibition of photosynthesis is one effect that most metals have in common when present at toxic concentrations; reduction in photosynthetic efficiency will be seen as an increase in reflectance in the visible range, as less light is being utilized for photosynthesis and chlorophyll production [3].
A. thaliana and H. annuus showed significant changes in chlorophyll content by treatment level, which corresponds to the increase in reflectance in the visible region (see Figures 1 and 2; notice differences in H. annuus are subtle compared to A. thaliana). Interestingly, for H. annuus, chlorophyll content increased from week 1 to week 2, but remained essentially constant from week 2 to week 3, whereas Z. mays chlorophyll content remained the same from week 1 to week 2 and decreased from week 2 to week 3 ( Figure 5). Also, although not statistically significant, chlorophyll content in treatment plants appears to increase slightly above that in control plants in Z. mays in week 3 ( Figure 5), which corresponds to the increase in absorbance (and thus decrease in reflectance) seen in the Z. mays spectra in Figures 1 and 2. There were no significant differences in B. napus chlorophyll content by week or treatment (Table 1), and B. napus also generally had lower chlorophyll content than the other species ( Figure 5 and Table A1); Z. mays and H. annuus had the highest chlorophyll contents.
The vegetation indices that proved to be significant predictors of chlorophyll content for H. annuus and A. thaliana were PSND, YI, R 750 /R 550, and R 1636 /R 1933 . PSND, which is determined by the normalized difference (R 800 − R 680 )/(R 800 + R 680 ), has been shown to be correlated with pigment concentration per unit area [70]. Here, PSND was found to be a significant predictor of chlorophyll content only for spectra acquired by CP, not FOV (Table 2). However, for A. thaliana, PSND as determined using FOV was a significant predictor of CLAI. Taken together, this is consistent with previous findings. Note that CLAI did not have significant treatment differences for H. annuus, and PSND also did not have significant treatment differences for H. annuus spectra acquired via FOV. Similarly, R 1636 /R 1933 , which has been previously suggested to be associated with cesium (Cs, also an alkali metal) exposure in A. thaliana [69], was also a significant predictor of chlorophyll content when assessed using CP spectra, and CLAI when assessed utilizing FOV spectra.
YI is an approximated second derivative of the spectra at about 600 nm and is intended to provide indication of chlorosis, i.e., leaf yellowing [67]. In this study, YI was found to be a significant predictor of chlorophyll content with FOV spectra for H. annuus and with CP spectra for A. thaliana. As the YI was originally developed at the CP scale, it is likely that soil background will influence this VI, particularly in the instances like that of A. thaliana week 3 treatment plants where much of the soil is exposed to the fore optics. H. annuus leaves are large and broad and, being closer to the fore optics than A. thaliana plants (and much closer than the soil), leaf reflectance dominates the spectra. Also, FOV may potentially capture the mottled nature of the yellowing of H. annuus leaves better than CP. R 750 /R 550 was originally developed for remote sensing of chlorophyll [68,71] but has also been found to be correlated with metal content [15]. Here, R 750 /R 550 was found to be a significant predictor of chlorophyll content at both CP and FOV scales for both species considered.

Water Content
Metals can disrupt the plant-water balance [3], which can be seen as an increase in reflectance in the mid-infrared region, as water absorbs fairly strongly at 1450 nm, 1940 nm, and 2500 nm, with slight increase in the near infrared region, associated with slight water absorption [75,76]. A. thaliana was the only species that exhibited significant treatment effects associated with RWC (Table 1), although all species exhibited significant temporal differences, with RWC decreasing in time ( Figure 6). The difference in treatment versus control reflectance for A. thaliana is clear in Figures 1 and 2 as spikes in relative reflectance at the aforementioned water absorption wavelengths that get larger each week. Although difficult to see in the relative spectra, temporal differences in the remaining species can be seen in the average (i.e., raw) reflectance spectra as an increase in reflectance at these same wavelengths in both treatment and control plants ( Figures A2-A4). Differences for each species except corn are greater between weeks 2 and 3 than between weeks 1 and 2, although the overall range of RWC is narrow for control plants, varying by only a few percent. It is interesting to note that the 1940 nm band in Figure 2 appears to give a qualitatively more consistent relative response between weeks than the other water bands. Although not statistically significant, B. napus RWC of treatment plants appears to decrease in time more than the control plants. This species, along with H. annuus had increasing variability (i.e., standard deviation) in RWC with time (Table A1), which explains why the apparent difference is not statistically significant.
The vegetation indices that proved to be significant predictors of RWC for A. thaliana were R 750 /R 550 with FOV, R 950 /R 750 with CP, and (R 950 − R 750 )/(R 950 + R 750 ) with CP. The ratio (R 950 − R 750 )/(R 950 + R 750 ) was considered because normalized differences are frequently utilized to improve upon a simple ratio by accounting for background reflectance [71]. In this instance, the normalized ratio has similar significance to its corresponding simple ratio, R 950 /R 750 , providing good indication that background reflectance is not a confounding factor. Similar to R 1636 /R 1933 discussed above, R 950 /R 750 was previously found to be associated with Cs exposure in A. thaliana, being correlated with RWC and Chl a + b at the CP scale and CLAI at the FOV scale in Cs contaminated plants [69]. Here, R 950 /R 750 was not found to be significant for Chl a + b or CLAI (although p = 0.0661 [CP] and p = 0.0663 [FOV], which would be significant at a 90% confidence level), but it is found to be significant for RWC with CP. Also consistent with this study is that the Cs study also found R 750 /R 550 at the FOV scale to be a significant predictor of RWC. This suggests that, as mentioned above, R 750 /R 550 along with R 950 /R 750 , may be good predictors of plants exposed to alkali metals.

Remaining Endpoints: Size and Shape
Although there are no significant treatment effects, Z. mays, B. napus, and H. annuus share similar magnitudes of and temporal changes in dry biomass (Figure 7). By week 3, plants have more than doubled their week 1 biomass, and B. napus has the highest average mass followed by Z. mays, then H. annuus, although Z. mays has the largest variation in biomass; the maximum Z. mays biomass was about 1700 mg, followed by a maximum of about 1200 mg for both H. annuus and B. napus. The temporal increase in FOV acquired reflectance for these species (for both treatment and control groups) in the near-IR is partially due to the increase in biomass of the plants (Figures A2-A4). For H. annuus, not only do the plants get taller (Figure 7), and therefore become closer to the fore optics, but there is also an increase in number and size of leaves. H. annuus leaves are broad and grow parallel to horizontal, resulting in a large reflective surface perpendicular to the fore optics. There will be an increase in light scattering within a plant containing a greater proportion of cell surfaces exposed to intercellular air space, due to different indices of refraction of these materials. As near infrared light is not used for photosynthesis, increasing light scatter in the infrared region means less transmission of light through the plant and more reflection back to the fore optics. Differences in plant structure will therefore affect reflectance of light in the near-IR; larger leaf areas will result in higher reflectance, whereas cell degradation or reduction in leaf thickness will result in lower reflectance, as there will be less light scattering within the plant leaf [77].
Although the mean CLAI for H. annuus is comparable to that for B. napus (~50%-60% coverage), the variability is much greater with some plants at almost 100% coverage by week 3. Also, B. napus grew less than 10 cm above the top of the pot so are much further away from the fore optics than H. annuus. Although there were no differences in Z. mays heights between groups (details not shown), plants grew to~45 cm in week 3, about double the H. annuus control plant mean (Table A2). However, the leaves are narrow and grow at an angle vice horizontal, meaning there is less leaf surface directly exposed to the fore optics compared to H. annuus. In fact, there are no significant differences in time or treatment for Z. mays CLAI, which remains essentially constant at~35% coverage. These points together can be seen as differences between species in the magnitude of reflectance in the near-IR of FOV spectra in Figures A2-A4. For Z. mays, reflectance in the near-IR varies between~0.18 and 0.25, for B. napus between~0.32 to 0.50, and for H. annuus between~0.48 to 0.80 from weeks 1 to 3.
Contrasting with the other three species, A. thaliana had significant treatment and temporal differences in biomass. A. thaliana plants also had much less biomass than the other species considered, as they are small, squatty plants prior to flowering. The dry biomass maximum for A. thaliana was less than 300 mg. Field-of-view control reflectance for A. thaliana increased in the near-IR each week ( Figure A1), corresponding to the increase in biomass of the control plants. Treatment reflectance in the near-IR was lower than the control reflectance for both the CP and FOV in weeks 2 and 3 (Figures 1 and 2); the lower CP reflectance indicates a difference in leaf structure. Lower FOV reflectance indicates a lower biomass comparatively, and a shift in the shape also indicates a structural difference. A. thaliana also has significant treatment differences for CLAI, with the control plant mean exceeding the H. annuus mean CLAI. However, prior to flowering A. thaliana does not grow vertical as H. annuus and Z. mays, so reflectance in the near-IR of~0.3-0.4 in the FOV spectra relates more to biomass.
All species showed slightly increased reflectance in the FOV treatment group as compared to the control in week 1 (Figure 2). This implies a possible initial stimulatory effect at these time points, which is commonly seen response to low-level lithium exposure [7,47]. However, it also appears that there may be structural degradation at week 2 for all four species as treatment reflectance in FOV spectra is lower than control reflectance (Figure 2). In week 3 these differences remain in A. thaliana and Z. mays spectra but are no longer apparent in H. annuus or B. napus spectra. It has been suggested that plants could simultaneously experience both stimulatory effects and toxicity symptoms from Li exposure, due to different concentrations of Li within the plant [50,51]. A separate study saw stimulated growth along with slight chlorosis in leaves of snap bean (Phaseolus vulgaris) exposed to 4 ppm lithium nitrate (LiNO 3 ) [47].
The vegetation indices that proved to be significant predictors of biomass for A. thaliana were YI with FOV and R 750 /R 550 with both CP and FOV. Significant predictors of CLAI were NDVI, PSND, R 950 /R 750 , R 1636 /R 1933 , and (R 950 − R 750 )/(R 950 + R 750 ) with FOV, and R 750 /R 550 with both CP and FOV. NDVI has been well-associated with green biomass and leaf area [63]; here NDVI is not associated with A. thaliana biomass, but treatment-induced changes in biomass were also associated with necrosis and browning of the leaves. Biomass and CLAI are plant scale endpoints as opposed to leaf scale endpoints, however, R 750 /R 550 was also a significant predictor of biomass at the leaf scale. One explanation for this is that biomass is related to leaf thickness (i.e., thinner leaves would result in less biomass), which would be seen in CP acquisition. Also, as statistically significant treatment effects were observed, A. thaliana biomass is related to lithium exposure. It is therefore highly likely that R 750 /R 550 is responding to lithium uptake, particularly as it is a significant predictor of all other treatment-dependent endpoints considered. The remaining VI have been previously discussed.
The significant predictors of H. annuus height were PSND, YI, and R 750 /R 550 , which are the same indices as were significant for chlorophyll content in H. annuus (Table 2). YI was significant with FOV spectra; as H. annuus leaves mottle slightly with time, the leaves also become broader and closer to the fore optics which could also be an increase in yellowness. PSND was significant with CP spectra, which is at first counterintuitive. However, as the plant grows taller, the internal structure of the plant changes as well. As discussed above, PSND has been associated with areal pigment concentration which likely changes as the plant grows. R 750 /R 550 was significant at both the leaf and plant scales. H. annuus height has significant interaction effects, which essentially indicates that height differences depend on treatment duration. Lithium uptake into plants follows the dose given, although the specific relationship is species dependent [7]. Thus, height is likely associated (negatively) with lithium uptake. R 750 /R 550 , as above, is likely more indicative of lithium uptake than height specifically, as also discussed above for other endpoints.

Conclusions
This study explored the feasibility of using reflectance spectroscopy as a quick and easy-to-use technique in supplementing phytoremediation efforts. Certain vegetation indices seem promising for selected endpoints and species but the variable responses of plants to similar Li concentrations makes applying VI across species less reliable. Treating species with various concentrations of Li to induce a similar level of toxicity may be a more appropriate assessment of vegetation indices than assessing a certain level of Li across all species. For A. thaliana, R 750 /R 550 was the best indicator of plant response to Li exposure, as it was a significant predictor of all endpoints considered, and it also has been previously associated with alkali metal (i.e., Cs) exposure [15,69]. Similarly, although with less predictive power, PSND, R 950 /R 750 , and R 1636 /R 1933 were also previously associated with metal exposure [69], and would be reasonable VI to consider for identifying metal stress in the future. It is likely that a combination of VI and spectra collection techniques (i.e., CP and FOV) could provide the overall best approximation of plant stress status by accounting for both whole plant and leaf optical properties; multi-index use should be given future consideration in studies utilizing both CP and FOV. Although we found some success in the laboratory for identifying relationships between symptoms induced by low-level Li exposure and reflectance spectra, environmental and sampling conditions were controlled; therefore, care should be given if the intent is to extrapolate to field studies. Measurements taken in the field may not be as consistent or informative as measurements taken in the laboratory due to extraneous and potentially unknown environmental factors. However, this study did find that the normalized difference (R 950 − R 750 )/(R 950 + R 750 ) provided comparable results to its corresponding simple ratio in a situation with low background reflection. In another setting, utilization of normalized differences where simple ratios were useful here may still be fitting.
Reflectance spectra provide useful information, but they can be expected to be different for distinct species. However, when treatment reflectance spectra are considered relative to a control, stress may be able to be quantified across species. The general difference in metal toxicity symptoms between all four species explains why reflectance spectra did not always shift in similar ways. Knowledge of the mechanisms involved in plant species uptake and response to a desired metal is necessary to appropriately apply vegetation indices as predictors of stress. Consideration of individual reflectance spectra as well as treatment spectra relative to control can provide additional insight into understanding results.

Conflicts of Interest:
The authors declare no conflict of interest. The results were reviewed and approved for publication by funding sponsor, but the sponsor had no role in the design of the study; in the collection, analyses, or interpretation of data; or in the writing of the manuscript.

Conflicts of Interest:
The authors declare no conflict of interest. The results were reviewed and approved for publication by funding sponsor, but the sponsor had no role in the design of the study; in the collection, analyses, or interpretation of data; or in the writing of the manuscript.
Appendix A Figure A1. Control and treatment spectra for A. thaliana as acquired by CP and FOV. Figure A1. Control and treatment spectra for A. thaliana as acquired by CP and FOV.        Figure A4. Control and treatment spectra for B. napus as acquired by CP and FOV.
Appendix B Table A1. Summary statistics for sample endpoints by week and treatment group. Cells contain the mean ± the standard deviation of the mean. For each cell n = 12, except dry biomass for A. thaliana for which n = 6 for each cell, and Chl a + b for A. thaliana in Week 3 for which n = 11 (insufficient plant material available for one sample).