Spectral behavior of Persian oak under compound stress of water deficit and dust storm

Persian oak (Quercus Brantii Lindl.) which is the most widely distributed tree in the Zagros Mountain forests is affected by western dust storms, mostly originating in Iraq, and harsh water stress as well. The objective of this research is to analyze the spectral behavior of Persian oak under water and dust stress scenarios, aiming to pave the way for modeling the stresses of drought and dust storms on oak trees using remote sensing images. Experiments were carried out on 54 two-year old oak tree seedlings, using a portable wind tunnel in greenhouse conditions. Water stress was induced on seedlings by means of changes in irrigation practices, i.e. well-watered (100 % field capacity), medium water deficit condition (40 % field capacity), and severe water deficit condition (20 % field capacity) treatments. Dust stress is also investigated by using three different dust particle concentrations, i.e. 350, 750 and 1500 (μg/m3). The spectrometry experiments were carried out at leaf and canopy levels in dark room by Fieldspec-3-ASD spectrometer. Spectral analysis was conducted using four procedures: (i) narrow-band spectral indices analysis, (ii) geometric indicators extraction from absorption features, (iii) Partial Least Squares Regression (PLSR), and SVM classifier. Results show that water stress could be modeled much better using PLSR statistic (R2= 0.87, RMSE=0.12), narrow-band indices analysis (Rcv= 0.75, RMSEcv= 0.17), and continuum removal (R2= 0.71, RMSE=0.20), respectively. For dust stress, PLSR (R2=0.83, RMSE=0.14) and narrow-band indices (R2 cv= 0.7, RMSE cv= 0.30) showed the best results, respectively. SVM could successfully separate stressed and not-stressed samples and also the stress types at both leaf and canopy levels, but it could not distinguish the different levels of stresses.


Introduction
The Zagros Mountains are the longest mountain range in Iran, where the climax vegetation is an open, xerophytic, cold-resistant, deciduous, semi-arid oak forest with an area of about six million hectare, accounting for almost 44 % of Iranian forests (Sagheb-Talebi and Pourhashemi, 2014). Persian oak (Quercus brantii Lindl.) is the most widely distributed tree in the Zagros Mountain forests (Noori et al., 2015). This region has experienced climate change with warmer temperatures, reduced rainfall as well as severe dust storm events that had adverse ramifications for Persian oak trees. The forest ecosystem, especially Brant's (or Persian) oak, has been affected by widespread tree death in the last decades (Ahmadi et al., 2014).
Environmental stresses constitutes the major factor limiting plant productivity (Spieß et al., 2012), wherein dust particle is a multi-influential stressor that impacts physical, chemical and physiological characteristics of leaves (Thompson et al., 1984) and dry biomass (Farmer, 1993), reduces photosynthesis (Armbrust, 1986), changes transpiration (Eveling, 1969) and leaf surface temperature (Eller, 1977). Drought is one the most threatening natural phenomena for Persian oaks native to Iran, Iraq, Syria, and Turkey in the southwest Asia region (Ghanbary et al., 2017). Albeit many factors have been mentioned for oak decline, drought is perhaps one of the most common stress factors in different types of forests (Bigler et al., 2006). These forests are located in the west Asia dust corridor, with the starting point being the Syrian Desert and end point at the Indian Ocean. These blowing dust particles are another major stressor for trees in the region.
During the past several decades, tools for vegetation remote sensing have evolved significantly (Asner, 1998). Several remote sensing and spectrometry methodologies and techniques have been developed for modeling the behavior of various trees under different stresses. Imaging and non-imaging hyperspectral data are used for measuring the chemical-physical properties of plants and the application of field spectrometry offers opportunities to identify biochemical characteristics of plants (Asner et al., 2000;Darvishzadeh, 2008a). Changes that take place as a result of stress influence the amount and direction of radiation reflected and emitted from plants (Jackson, 1986). The lack of spectral information about oak trees and the stresses has induced limitations on the usage of time series remote sensing images to make a synoptic estimation of damages caused by dust storms and water stress  A. Darvishi Boloorani, et al. Int J Appl Earth Obs Geoinformation 88 (2020) 102082 in this region. Measuring spectral signatures of dust and water stresses by spectrometers will help model the type and level of stress by remote sensing. Remote sensing methods for plant parameters estimation are mainly based on statistical and physical models and their accompanied advantages and disadvantages (Darvishzadeh, 2008b). Sensitivity analysis of plant indices using single and multiple regression analysis is very common in plants stress modeling with remote sensing data. In recent years, the geometric-spectral characteristics of absorption features of spectrum such as area, depth, shape, and position are becoming very customary characteristics in plant parameters estimation and plant stress modeling and prediction (Chávez et al., 2013;Del et al., 2014;Majeke et al., 2008;Stimson et al., 2005). Gamon et al. (1992) introduced physiological reflectance index (PRI) using (R550 − R530 / R550 + R530) for modeling the Xanthophyll cycle of sunflower. They modeled the stress-induced changes in photosynthesis and nitrogen of the plants. Alexander et al. (2015) investigated the fluorescence chlorophyll potential of water, heat, and nitrogen stresses on leaf and canopy cover levels. Results were indicative of a meaningful relation between water stress and the rate of fluorescence signal in red and NIR spectrum.
The objective of this experimental research is to perform a thorough analysis of Persian oak spectral response to water and dust stress conditions by exploring the capabilities of narrow band vegetation indices, PLSR, spectrum continuum removal (CR), and SVM classification.

Greenhous experiments
For experimental purposes, the study used two-year old oak seedlings within plastic pots. The experiment was carried out using a total of 58 seedlings in March 2015 (before blossom bud season) and lasted about 6 months under greenhouse conditions. The water and dust stresses were induced in a plastic greenhouse (Fig. 1). During the intervention of stresses, greenhouse temperature, humidity, and air condition were controlled and measured. Soil samples were collected from the sources of dust storms. Then dust storms were accordingly simulated using a portable wind tunnel. The greenhouse was divided into four sections ( Fig. 1-A, B, C, and D). Three levels of dust concentration

Table 1
Vegetation indices adapted in this research. ρ is the reflectance, λ is wavelength, a and b are the soil line coefficients between λ 1 and λ 2 . A. Darvishi Boloorani, et al. Int J Appl Earth Obs Geoinformation 88 (2020) 102082 i.e. 1500, 750, and 350 μg/m³ was created ( Fig. 1-A, B, and C, respectively). Dust concentration measurement was measured using 176000A Microdust Pro Dust Monito ( Fig. 1-E). Dust storms were injected in three sections of the greenhouse, using a portable wind tunnel, while one section left out for control samples. Greenhouse drainage system was installed and adapted in the four sections to investigate tree water stress, i.e. zero, 60 and 80 percent of deficiency field capacity (FC) measured using Time Domain Reflectometer, TDR ( Fig. 1-F). Among the total number of 58 samples, 45 seedlings were put into three sections (15 for each). Water and dust stresses were then induced in each section. As a result, 4 seedlings were totally dried out and lost their leaves, therefore, they were not considered in the spectrometry phase. Fig. 2 shows single samples from each section at the end of stress experiments.

Laboratory spectrometry
Upon termination of the experimental phase in the greenhouse, samples were moved to the darkroom of the spectrometry laboratory. Fieldspec 3 was used for both leaves and canopy cover of the trees. For canopy a 150-watt halogen lamp was used as the light source. The pots central-vertical axis and the spectrometer IFOV central point were coaxial, therefore IFOV was applied to fully cover the pots (Fig. 3). The fiber optic, with a field of view (FOV) of°25 , was placed in a pistol and mounted on the tripod and positioned 50 cm above pot at nadir position. In the setting, the spectrometer had IFOV with 22 cm diameter on the soil surface, with the nadir point being the center of the plastic pot.
The illumination was the same for all samples for canopy measurements and the differences of spectral signatures were obtained as functions of leaf surface, leaf shape, and other chemical-physical characteristics of leafs and canopy. In order to obtain a proper reflectance, after each measurement, the pots were rotated 45 degrees and subsequent spectrometry measurements were done. Consequently, 8 spectral measurements were obtained for each sample at canopy level. Instrument calibration was carried out every 5 measurements. Contact probe was used for leaf level experiments.

Leafs chlorophyll measurement
Directly after spectrometry, the chlorophyll concentrations in three leaves from each sample were measured using SPAD. The average SPAD for each sample was converted into absolute units of chlorophyll concentration using Eq. (1). This equation is precisely developed for absolute chlorophyll concentration of oak by Percival et al. (2008). (1)

Leaves moisture content measurement
In order to measure leaf moisture contents, they were weighted before and after drying process. Leaves were dried at 70°C for 72 h via oven drying. Leaves moisture content (MC) was calculated using Eq. (2) (Noomen et al., 2008).
Where, F shows fresh leaf weight and D is the dried leaf weight.

Spectra preprocessing
16 spectrometry measurements were obtained at canopy and leaf A. Darvishi Boloorani, et al. Int J Appl Earth Obs Geoinformation 88 (2020) 102082 levels for each sample, i.e. 8 for each level. Due to the high level of noises, spectra less than 400 and more than 2400 nm were removed. In order to reduce the instrument and illumination errors from the spectrum, a moving Savitzky-Golay smoothing filter (Savitzky, 1964) with a frame size of 15 data points (second degree polynomial) was adapted to smooth out all averaged reflectance spectra. In the next step, noisy and useless data were eliminated using Principal Component Analysis (PCA) and visual investigations.

Narrow band indices
Mathematical (experimental) models for creating certain statistical relations between spectral features and plants characteristics are mainly implemented using vegetation indices (Chávez et al., 2013). Here, five previously developed vegetation indices were adapted. These indices have proven their potential in modeling certain relations between biophysical-biochemical characteristics of plants and stresses. To find the answer to: "What is the best bands combination for highlighting the water and dust stress of Persian Oak?" the indices' equations (Table 1) were subsequently adapted. All possible two-bands combinations in the entire spectral range of 400-2400 nm were investigated (4 × 10 6 indices). After that, correlation between all calculated indices and stresses were obtained for two data sets, i.e. spectral measurements at leaf and canopy levels.

Absorption Features, AF
AFs are specific parts of wavelength where the incident light more absorbed than surrounding wavelengths. These spectral features are related to the chemical-physical characteristics of objects. Spectral measurements contain both AFs and continuum data. Subsequent to Continuum Removal (CR) using Eq. (3), the geometric characteristics of AFs were calculated (Kokaly et al., 2007) (Eq.s 4,5).
Where, Rc is continuum removed spectra, Ro is the original spectra, and RL is the continuum curve. The relative depth, D, of AFs is defined as the reflectance value at the shoulders minus the reflectance value at the absorption-band minimum. D is defined relative to the continuum, Rc, Fig. 7. Area, asymmetry, depth, position and width indices extracted from the mean of Oak seedling's CR reflectance for different water stress levels (control, medium and severe stress conditions).
A. Darvishi Boloorani, et al. Int J Appl Earth Obs Geoinformation 88 (2020) 102082 Fig. 8. Area, asymmetry, depth, position and width indices extracted from mean of oak seedling's CR reflectance for different dust stress levels (control, low, medium and severe dust stress). A. Darvishi Boloorani, et al. Int J Appl Earth Obs Geoinformation 88 (2020) 102082 as: Where, Rb is the reflectance at the bottom wavelength and Rc is the reflectance of the continuum at the same wavelength as Rb. The absorption-band position is defined as the band having the minimum reflectance value over the wavelength range of the AF. The asymmetry factor, S, of the AF is defined as: In which, A left is the area of the absorption from starting point to maximum point and A right is the area of the absorption from maximum absorption point to the end point of the AF.
Position or location of the maximum width wavelength (LAMDAmax) and width (Full width at half maximum) are the other indices that extracted for each AFs.

Support vector machine, SVM
SVM classification was carried out on 54 sample seedlings in three procedures. (i) classification of samples into two categories of control (Co) and stressed (St), regardless the type and intensity of the corresponding stresses. (ii) samples were divided into 4 subclasses including: control (Co), water stress (WrS), dust stress (DuS), and both water and dust storm stresses (WrDuS). (iii) samples were categorized into 12 granular subclasses of control (Co), low water stress (WrS L ), moderate water stress (WrS M ), intense water stress (WrS I ), low dust stress (DuS L ), moderate dust stress (DuS M ), intense dust stress (DuS I ), combined low and moderate dust stresses and moderate water stress (DuS L WrS M ), combined low dust stress and intense water stress (DuS L WrS I ), combined moderate dust and water stress (DuS M WrS M ), combined moderate dust stress and intense water stress (DuS M WrS I ), combined intense dust stress and moderate water stress (DuS I WrS M ), and combined intense dust and water stress (DuS I WrS I ). In each procedure, the classification accuracy for all classes was calculated. 60 percent of data were used for training and the rest (40 percent) for test the results of classification. Fig. 4 shows the workflow of the this study. Table 2 shows the chlorophyll and moisture contents of leaves that are measured for test and control samples. The minimum amount of chlorophyll (53.26 mg/g) was obtained from 1500 (mg/m 3 ) of dust concentration. This is due to the reduction of photosynthesis activities with high density of dust particles. Leaves moisture contents did not show meaningful relations with different dust stress levels. Table 3 shows the correlation of chlorophyll and moisture with dust and water stresses. Results show that both chlorophyll and moisture content have inverse relations to (water stress) WrS and (dust stress) DuS. In nearly all cases, the leaves moisture content showed higher correlation to WrS in comparison with chlorophyll.

Spectral responses to water stress
The spectral reflectance of oak in relation to different levels of water stress are shown in Fig. 5-(A and B) at leaf and canopy levels, respectively. Results shows that the increase of water stress will result in the increase of red and 1300−2200 nm, whereas it decreases the 800−1300 nm of spectrum. The spectral CR reflectance of oak in relation to different levels of water stress are shown in Fig. 5-(C and D) at leaf and canopy levels, respectively, wherein the spectral changes are clearly observable. As can be seen from these figures, changes in absorption are more evident at leaf level compared to canopy level.

Spectral responses to dust stress
The spectral reflectance of oak in relation to different levels of dust stress are shown in Fig. 6-(A and B) at leaf and canopy levels, respectively. The spectral CR reflectance of oak in relation to different levels of dust stress are shown in Fig. 6-(C and D) at leaf and canopy levels, respectively. Results show that spectral changes are higher at canopy level compared to leaf level at the red, 1400 and 1900 nm AF. Nonetheless the changes are similar at leaf level, i.e. in 1870-2170, 550-750, 400−550 nm of spectrum.

Geometric parameters of AFs
The geometric parameters (area, depth, width, position and skewness) of absorption features in the CR spectra were used to measure different levels of water and dust stresses. Results showed that the area and depth have the highest correlation with WrS at both leaf and canopy levels. As illustrated in Fig. 5, effects of water stress on AFs are rather similar at both leaf and canopy levels. For a better description, Fig. 7 illustrates the effects of water stress on CR elimination as well as the area, depth, skewness, width, and maximum depth wavelength on three main absorption features of the seedlings (400-750, 1350-1550, and 1850−2150 nm) for mean spectrums of each stress level. Fig. 8 shows the effects of dust on CR elimination, area, asymmetry, depth, position and width indices extracted from the spectrums of each stress level. This figure demonstrates the effects of dust on AFs that shows similar behaviors for both leaf and canopy levels.

Single band correlation
The coefficient of determination (R 2 ) between different spectral bands with WrS and DuS are shown in Fig. 9.
As can be seen in Fig. 9-(A), there is a high correlation between water stress and spectral reflectance for almost all wavelengths at leaf level. Despite differences in correlation, a similar behavior is observed for both the leaf and canopy levels. Also, a positive correlation is observed between reflectance and water stress for absorption features located in the red edge area and the spectrum of 1400−1900 nm A. Darvishi Boloorani, et al. Int J Appl Earth Obs Geoinformation 88 (2020)  A. Darvishi Boloorani, et al. Int J Appl Earth Obs Geoinformation 88 (2020)  A. Darvishi Boloorani, et al. Int J Appl Earth Obs Geoinformation 88 (2020) 102082 region. Fig. 9-(B) indicates, a similar pattern of change in wavelength as a result of dust stress at both the leaf and canopy levels, with the highest negative correlation observed for NIR and red edge area.

Narrow bands vegetation indices
In aiming to reduce processing time, spectral resampling was performed (at 3 nm threshold) on pre-processed spectra. All possible combinations of band pairs were created using vegetation indices equations (Table 1) and spectral narrow bands within 400−2400 nm of wavelength. The coefficient of determination (R 2 ) between the created indices and stress levels for linear regression were calculated and plotted in 667 × 667 matrices and shown in Fig. 10. These matrices were used to find the optimal two-band (pair-wise) combinations. Results showed that NDI (λ 2108 , λ 2258 ) and RI (λ 2327 , λ 2348 ) have the highest R 2 at the leaf level water stress. As shown in Table 4, for NDI, DI and RI indices, the SWIR portion of the spectrum has a higher R 2 in comparison to other parts of spectrum. At canopy level, vegetation indices are adapted that uses the soil line concept. Therefore 5 vegetation indices equations were subsequently adapted. Table 4 show the vegetation indices with the best spectral position and two-band combinations for WrS. Furthermore, Fig. 11 depicts a graphical model of the correlations between observed values and predicted values for different water stress levels.
As shown in Fig. 12, R 2 between ratio-based vegetation indices: RVI, DVI and NDVI and dust stresses have almost the same results. Results of 5 vegetation indices for dust stress at the canopy level are shown in Fig. 12. Maximum R 2 were obtained for SAVI 2 (λ 635 , λ 680 ), TSAVI(λ 635 , λ 683 ) in 780-810, 1200-1400,1850-1950 and 2230−2290 nm of the spectrum (Table 5). Fig. 13 also depicts a graphical model of the correlation between observed and estimated values of dust stress levels.

PLSR analysis
PLSR is used to model the relations between WrS and spectral signatures of leafs and canopies. Similarly, the fif th component with RMSE and variance equal to 0.22 and 0.80 respectively, was selected as the most suitable component for canopy level under WrS. Following the same procedure as WrS, for DuS at leaf level the six th component of PLSR was selected as the most suitable component (Table 6). Results showed that the spectral intervals of 414-418, 658-688, 710-725, 778-745, 1400-1425, and 1848−1858 nm had the best results at leaf level. Results for the DuS at canopy level showed that the four th component of PLSR had the most reliable results at spectral intervals of 640-693, 705-735, and 1985−1900 nm.

SVM classification
A SVM-based hierarchy of classifiers was adapted to discriminate seedlings under both water and dust stress scenarios at all levels. The classification accuracy was calculated at each level independently. The first classification was conducted to separate stressed from non-stressed samples. At leaf level, results showed that the under-stress seedlings are discriminable from non-stressed samples with an accuracy of 87 percent. Second classification carried out to separate the samples of water stress, dust stress, combined stress, and control group (4-class classification), the results show an accuracy of 52 percent. In the third classification, i.e. the separation of samples into 12 classes based on their stress levels, results show an accuracy of 30 percent. The same SVMbased hierarchy of classifiers was conducted at canopy level, results are shown in Fig. 14

Discussion
Spectral behavior of Persian oak under water and dust stress scenarios at leaf and canopy levels were analyzed by different methods. Spectral changes of oak seedling under water stress were observed in the shape of three absorption features: the blue chlorophyll in 400−550 nm, red chlorophyll in 550−750 nm, and water absorption in 1870−2170 nm of the spectrum (Fig. 5). Reflectance growth in water absorption feature for dust stressed seedlings could be related to water loss in leaves due to dust particles (Armbrust, 1986). For water stress the area and depth geometric parameters decreased in all three absorption features at both leaf and canopy levels. As stress levels increased, the maximum width wavelength (400−750 nm at the leaf level, 1350-1550 nm at both the leaf and canopy level, and 1850−2150 nm at the canopy level) tends towards higher wavelengths (Fig. 7). Peñuelas and Inoue (1999) showed that the increas of water content will lead to decrease of reflectance in the whole spectrum (350−2500 nm) and the AFs located in 1430 and 1950 nm will have greatest change, that is in agreement with our resultes. In Ullah et al. (2014) study, the result is also the same, where the biggest negative correlation was achieved in 1400−2400 nm, but in red edge they did not find negative correlation.
Spectral changes of dust stress showed that the changes are more noticeable in water absorption in 1870−2170 nm, red chlorophyll absorption in 550−750 nm and blue chlorophyll absorption in 400−550 nm. Also, from 700 to 900 nm (NIR) experianced a decreaseing bahaviour by the increas of dust concentration (Fig. 6). As can be seen in Fig. 6, area, depth, skewness and width have all decreased in all three absorption features at both the leaf and canopy levels. No significant behavior is observed regarding LAMDAmax (Fig. 8).
Narrow band vegetation indices analysis for water stress revealed that RI and NDVI (R cv 2 = 0.75) are the best indices at leaf level. While, RVI (R cv 2 = 0.60) and SAVI 2 (R cv 2 = 0.50) are the best indices at canopy level (Table 4). These two-bands indices are in SWIR and the second water absorption feature of plants. This behavior shows a noticeable change in the slop of this part of the spectrum due to water stress. RVI and NDVI (R cv 2 = 0.42) at leaf level and SAVI 2 (R cv 2 = 0.70) and TSAVI (R cv 2 = 0.68) at canopy level are the most robust indices for dust stress measurement using spectral response of oak seedlings under stress ( Table 5). The soil adjusted vegetation indices showed higher potential for measuring different levels of dust stress at leaf and canopy levels, while, the ratio indices showed better correlation with water stress. The best two-bands indices for dust stress modeling were red region and the second absorption features ( 1900 nm) of water.  A. Darvishi Boloorani, et al. Int J Appl Earth Obs Geoinformation 88 (2020) 102082 The geometric characteristics of absorption features are indicative of their robustness for quantifying water and dust stress. Depth (R 2 = 0.71) and area (R 2 = 0.7) of water absorption feature in 1837−2200 nm at leaf level and area (R 2 = 0.39) and depth (R 2 = 0.38) in 1837−2234 nm at canopy level are the most useful geometric characteristics of absorption features for measuring water stress. As mentioned, this portion of spectrum has shown the highest correlation with water stress as well. Depth (R 2 = 0.35) in 900−1114 nm of NIR and depth (R 2 = 0.31) at 1309−1678 nm of water absorption feature at leaf level, depth (R 2 = 0.51) and area (R 2 = 0.47) in 1837−2234 nm at canopy level are the most useful geometric characteristics of absorption features for measuring dust stress of Persian oak seedlings. Despite the fact that the correlation between dust stress and geometric indicators are generally not high, but the second water absorption feature of spectrum in SWIR show the best results.
Red (715−725 nm), green (540−657 nm) and chlorophyll absorption features (665−685 nm) (R 2 = 0.87) at leaf level; green (500−560 nm) and red edge (715−735 nm) and NIR (770−850 nm) (R 2 = 0.80) at canopy level are the dominant areas of the spectrum to measure water stress using PLSR, respectively. 639−640 nm in the visible and red edge area (705−735 nm) were assigned maximum weights (R 2 = 0.78) in PLSR model to measure dust stress at the leaf level. Similarly, for the canopy level, bands within the SWIR  and red edge regions were assigned maximum weights (R 2 = 0.83) indicating their high correlation with dust stress (Table 6). Fig. 15 illustrates spectral reflectance regions with the highest R 2 in water and dust stress measurement using different approaches. As shown in Fig. 15-(A), PLSR at red and red edge, narrowband indices at SWIR (Table 4) and geometric characteristics of absorption features at SWIR had the highest performances in water stress measurement, respectively. In conclusion, SWIR and red edge are the best portions of the spectrum for water stress measurement. Moreover, PLSR showed optimal results for dust stress measurement at both the leaf and canopy levels for SWIR, visible, and NIR spectra (Fig. 15-B). Results of Gopal et al., 2019 for 10 rice samples under various water stresses were also indicative of the superiority of PLSR in water stress measurement compared to other statistical models.
Considering results for 54 samples, the SVM classification accuracy of 80 and 87 percent in separating stressed from not-stressed samples and the accuracy of 52-60 percent in separating stress types at both leaf and canopy levels were obtained. The findings could be useful for the applications of hyperspectral imagery for water and dust stresses modeling of Persian oak.  A. Darvishi Boloorani, et al. Int J Appl Earth Obs Geoinformation 88 (2020) 102082

Conclusion
Spectral analysis of Persian oak, for the first time, demonstrates that compound stresses of water deficit and dust storm can be studied by spectrometry techniques as reliable tool with promising results. Combined stress scenarios over oak seedling, including water and dust stresses, have been developed. In this framework the spectral response of the leaves and canopy were investigated. Water and chlorophyll absorption spectral features at SWIR (1870−2170 nm) and visible (550−750 nm) were the most sensitive areas of the spectrum to both stresses, respectively. These two areas of the spectrum had showen direct relations with both stresses. Increase in both stressors will result in the decrease of chlorophyll and water in the seedlings. Such results were expected considering prior studies on water stress applied to vegetation. In conclusion, from the investigated techniques the PLSR had the best performance for both stresses. SVM model could separate stressed and not-stressed samples very well, while, for separating stress types at both leaf and canopy levels, the results were acceptable, but it couldn't recognize the levels of stresses. As a general conclusion, it can be deduced that SWIR (1900−2000 nm), red edge (705−735 nm) and NIR (770−850 nm) are the areas of the spectrum with high potentials to model water and dust stresses, whereas water stress alone can be measured more precisely than dust stress using spectrometry analysis.