SSSAJ Journal of Natural Resources and Life Sciences Education
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online 16 May 2007
Published in Soil Sci Soc Am J 71:1029-1037 (2007)
DOI: 10.2136/sssaj2006.0187
© 2007 Soil Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Guerrero, C.
Right arrow Articles by Gómez, I.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Guerrero, C.
Right arrow Articles by Gómez, I.
Agricola
Right arrow Articles by Guerrero, C.
Right arrow Articles by Gómez, I.
Related Collections
Right arrow Fire
Right arrow Other Models
Right arrow Soil Analysis

LANDSCAPE MANAGEMENT

Near-Infrared Spectroscopy to Estimate the Maximum Temperatures Reached on Burned Soils

César Guerrero*, Jorge Mataix-Solera, Victoria Arcenegui, Jorge Mataix-Beneyto and Ignacio Gómez

GEA (Grupo de Edafología Ambiental), Dep. de Agroquímica y Medio Ambiente, Universidad Miguel Hernández. Avda. de la, Universidad s/n, E-03202 Elche, Spain

* Corresponding author (cesar.guerrero{at}umh.es).


   ABSTRACT
TOP
 NOTES
 ABSTRACT
INTRODUCTION
 MATERIAL AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
We studied the use of near-infrared (NIR) reflectance spectroscopy as a potential method to estimate a posteriori the maximum temperatures reached (MTR) on burned soils. When soils are heated, the NIR spectra change in accordance with the MTR. Thus, after calibrating, these patterns of NIR could be used as a fingerprint to estimate the MTR in burned soils. Successful validations of the models relating NIR spectra with MTR were obtained in each of the five soils studied (local models), with r2 values ranging from 97.47 to 98.56%. A global model constructed with samples from the five soils studied obtained a similar accuracy, suggesting the presence in soils of some NIR-detectable compounds with similar thermal sensitivity. The influence of the variability caused by the soil type and the duration of heating during model constructions is also evaluated and discussed. The use of NIR presents interesting advantages, such as low cost, low time consumption, minimal pretreatment of samples, no need for chemicals, and accuracy. The results indicate that the MTR could be estimated in burned soils with NIR, offering a new perspective on studies of wildfire effects on soils.

Abbreviations: MIR, mid-infrared • MTR, maximum temperature reached • NIR, near-infrared • PLS, partial least squares • RMSECV, root mean square error of cross validation • RMSEE, root mean square error of the estimation • RPD, residual predictive deviation


   INTRODUCTION
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
MATERIAL AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
During the past few decades, near-infrared (NIR) reflectance spectroscopy has developed rapidly as a fast and robust analytical method for many agricultural, pharmaceutical, and food products (Blanco and Villarroya, 2002). This technique obtains the reflectance spectra of a sample in the range of the NIR region (780–2500 nm, in wavelengths {approx}12820–4000 cm–1, in wavenumbers). In the NIR region, the radiation is absorbed by different chemical bonds, such as C–H, N–H, S–H, C=O, and O–H, of compounds present in the sample. The radiation is absorbed in accordance with the concentration of these compounds. As a consequence, NIR reflectance spectra contain basic information about the organic composition of a sample. The NIR spectra are the result of the overtones and combinations of fundamental vibrational bands for these bonds, which are more strongly absorbed in the mid-infrared (MIR) region (Burns and Ciurczak, 2001). As a consequence of overlapped bands, NIR information must be extracted and could not be directly interpreted, as could occur in a MIR spectrum. Advances in computer capacity have led to an increase in powerful multivariate statistics software that is able to extract this information and perform sophisticated statistical procedures (the so-called "chemometrics") (Martens and Næs, 1989; Burns and Ciurczak, 2001). In this sense, NIR spectra coupled with chemometrics techniques has been used to estimate some parameters related to the chemical composition of a broad spectrum of materials, including soils (Ben-Dor and Banin, 1995; Burns and Ciurczak, 2001; Blanco and Villarroya, 2002). One advantage of NIR over MIR is the simplicity of the sample pretreatment (sieving in soils or grinding in other materials). Generally, in MIR analysis, as a consequence of the stronger absorption, a dilution of the sample must be done (performing a small disk where the sample is diluted at 1% with KBr), increasing time-consuming sample pretreatment. In the case of soils, several parameters related to soil quality, such as organic carbon (C) and nitrogen (N), cation exchange capacity, nutrients, microbial biomass, potentially mineralizable C and N, or soil respiration, have been predicted with NIR with more or less accuracy (Fritze et al., 1994; Ben-Dor and Banin, 1995; Chang et al., 2001; McCarty et al., 2002; Viscarra Rossel et al., 2006).

Most of these parameters are modified by the effect of fire and generally in a manner that is dependent on the temperature reached (Raison, 1979; DeBano et al., 1998; Neary et al., 1999; Pietikäinen et al., 2000; Certini, 2005; Guerrero et al., 2005). Other important factors related to the temperatures reached in soils during wildfires are water repellence (i.e., the enhancement or disappearance of hydrophobicity and the depth of formation), aggregation-disruption of soil particles, enzyme denaturation, changes in the microbial structure community, nutrient volatilization-mineralization processes, and seed mortality and dormancy disruption (Raison, 1979; Bradstock et al., 1992; Saa et al., 1993; Neary et al., 1999; Pietikäinen et al., 2000; Certini, 2005; Shakesby and Doerr, 2006). Thus, it is desirable for post-fire maps to contain information about temperatures reached in soils (Lewis et al., 2006).

Burn severity maps could be obtained through satellite imagery (van Wagtendonk et al., 2004), but the spatial resolution would be inadequate for spatially heterogeneous ecosystems, such as those in semiarid Mediterranean regions. Moreover, most of the indexes used to assess burn severity with remote sensing are based on changes in vegetation cover and soil exposed and are not exclusively based on soil characteristics (Lewis et al., 2006). Thus, these indexes are more related to the degree of site alteration than to changes in soil properties (van Wagtendonk et al., 2004). At other scales (close to the point scale), visual estimations, such as ash color or the minimum diameter size of remaining branches (DeBano et al., 1998; Pérez and Moreno, 1998), could provide information about heat intensity, but the precision of these measurements is poor. Thus, the development of new tools to estimate with precision and accuracy the temperature reached on soils during wildfires is of great interest (Lewis et al., 2006; Shakesby and Doerr, 2006). Under Mediterranean conditions, wildfires can start soil degradation processes and erosion, especially in association with a high-severity fire (Cerdà, 1998; Neary et al., 1999; Guerrero et al., 2005; Lewis et al., 2006). It is important to carry out rehabilitation practices before the start of the erosion processes. For improved planning of rehabilitation strategies, it is necessary to obtain maps of severity with better resolution than those obtained by remote sensing and within a short time after the fire. These tasks need large numbers of samples, so an efficient, precise, and accurate technique is desirable for the post-fire assessment of temperatures reached in soils. NIR spectroscopy addresses these characteristics. Moreover, the NIR spectra are sensitive to compounds (mainly organics) that are usually affected by heat during fires (Almendros et al., 1984; González-Pérez et al., 2004). For this study, we hypothesized that soil's NIR spectra are related to the maximum temperature reached (MTR) and thus could be used as a tool for the assessment of the MTR in soils after forest fires.

The objective of this study was to check the ability of models using NIR reflectance spectroscopy to estimate the MTR in burned soils. Specific objectives were to (i) evaluate the ability of local models (one model for each soil type studied) and of a global model (one unique model for the five soils studied) to estimate MTR of burned soils and (ii) to compare the modeling approaches.


   MATERIAL AND METHODS
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIAL AND METHODS
RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Experimental Design
In this experiment, three factors were considered: eight furnace-oven temperatures (70–700°C), six durations of heating (10–60 min), and five soil types. Thus, samples of the five soils were heated in a furnace-oven at 70, 160, 250, 340, 430, 520, 610, and 700°C for 10-, 20-, 30-, 40-, 50-, and 60-min periods. Moreover, in the set of samples heated for 10 min, more temperatures were used (70–700°C in 30°C increments).

Soils
Five soils were sampled from five different sites in the province of Alicante (southeastern Spain). Vegetation of these sites is dominated by shrubs and dispersed trees (Pinus halepensis P. Mill.). The most common species of shrubs are Quercus coccifera L., Rosmarinus officinalis L., Cistus albidus L., Pistacia lentiscus L., Globularia alypum L., Thymus spp., Ulex parviflorus Pourr., and Brachypodium retusum (Pers.) P. Beav. as dominant grass. The main characteristics of sites and soils are described in Table 1. In each site, around 50 soil samples (0–5 cm depth) of ~1 kg were randomly collected and mixed. After air-drying for 2 wk, each soil was sieved (<2 mm) and thoroughly mixed to obtain a homogeneous mixed sample.


View this table:
[in this window]
[in a new window]

  		Table 1. Main characteristics of soils (0–5 cm) from Alicante (Southeastern Spain) used in this study.

 
Soil Sample Heating
Approximately 80 g of air-dried, sieved soil were placed in a ceramic cup and placed in a furnace-oven that was preheated to the desired temperature. At the same time, a thermocouple (k-type, NiCr-Ni; Testo SA, Barcelona, Spain) was inserted inside the soil (2-cm depth). The thermocouple temperature was recorded every minute. In each sample, the MTR during the heating was registered. Each sample was heated separately.

Near-Infrared Spectra
After cooling, subsamples of around 50 g of the burned soil samples were placed in glass Petri dishes and scanned on reflectance mode from 12000 to 3800 cm–1 (approximately equivalent from 830 to 2630 nm). For these measurements, we used a Fourier-Transform near-infrared spectrophotometer (MPA; Bruker Optik GmbH, Germany), equipped with a quartz beamsplitter and PbS detector. The instrument is also equipped with an integrating macrosample sphere and rotating sample cup, allowing the scanning of large areas of the samples. In each of the reflectance measurements, 64 scans were averaged. Samples were measured in duplicate, increasing the surface of soil sample scanned. After this, they were averaged again. The resolution used for spectral analysis was 8 cm–1. Background corrections were made before each sample scan. Each spectrum was composed of more than 1000 values of absorbance (obtained from reflectance) at the different wavenumbers between 12000 and 3800 cm–1. The x-scale of each NIR spectrum was transformed from wavenumber to wavelength, obtaining a 1000-absorbance point's spectrum between 830 and 2630 nm. The time used for the spectral measurement was approximately 1 min per sample. No chemical or hazardous reagents were needed to obtain the NIR spectra.

Models Construction
Several empirical models or calibration functions were made, based in this general structure:

Formula 1[1]
where Y is the target parameter (MTR), b the calibration function, and X is the NIR spectra, resulting in:

Formula 2[2]

Thus, two matrixes were constructed previously:

For model construction (empirical calibration functions) relating the spectral information (NIR-spectra matrix) with MTR, we used partial least squares (PLS) regression. PLS regression is widely used as a chemometric method in NIR analysis (Martens and Næs, 1989; Burns and Ciurczak, 2001; McCarty et al., 2002; Viscarra Rossel et al., 2006). Briefly, PLS reduces the NIR matrix to a few components, such as in a principal component analysis. During component extraction in PLS, the data of the target parameter (i.e., MTR) that will be estimated is taken into account. The number of PLS components (the so-called "PLS-vectors") used is the "rank" of the PLS regression (the rank of the model). The first PLS-vectors are those that provide more information about the target parameter. In general terms, models with low ranks are preferred because the higher the rank used, the higher the noise included. According to this, Eq. [2] can be described more accurately in Eq. [3]:

Formula 3[3]
where PLS-v are PLS-vectors, NIR{lambda}x-{lambda}y is the region of the NIR spectra related to the MTR, and k is the rank of the model.

Typical spectroscopic preprocessing of the spectra were tested: no data preprocessing, first derivative, second derivative, linear offset subtraction, straight line subtraction, multiplicative scatter correction, vector normalization, min-max normalization, and combinations of the above. More than 1000 models were tested using the spectroscopic software OPUS 5.5 (Bruker Optik GmbH, 2004) during each calibration. These procedures (preprocessing) were made with the aim of reducing optical interference not related to the chemical composition of the sample, such as variations caused by different sample particle size (Blanco and Villarroya, 2002). Derivative treatment not only reduces scattering effects but also increases the resolution of spectral peaks (Burns and Ciurczak, 2001).

In this study, two methods of model construction (and validation) were used: "cross validation" and "test set validation." Briefly, in the cross validation method, one sample of the calibration set is excluded during the calibration. Thus, the excluded sample is estimated (and validated) with the others. This exclusion step is repeated until all samples have been validated with calibrations performed by the other. In the test set validation method, the data are split randomly into two sets: One data set is used in the calibration, and the other is used in the validation. In all the cases, the validation method used is indicated. All of these procedures were made using the specialized spectroscopic software OPUS 5.5 (Bruker Optik GmbH, 2004).

Selection of Models and Statistics
Best models were selected that presented lower values of RMSE of estimation (RMSEE), low ranks, and higher r2. Depending on the validation method used, the RMSE of cross validation (RMSECV for cross validation) or RMSE of prediction (RMSEP for test set validation) was calculated. The RMSECV and RMSEP are based on predicted values (validation), and RMSEE is based on fitted values (calibration). With the aim of comparing the accuracy of our models with other NIR models cited in the literature, we calculated the residual predictive deviation (RPD), which is the ratio of the SD of measured data to RMSEP (for test set validation) or to RMSECV (for cross validation). In each of the models, outliers were detected when the residuals (Resi or Differi) were out of the confidence interval (at 99%) constructed with the residuals from all samples. These outliers were not removed. A description of statistics used follows:

Formula 4[4]
where

Formula 5[5]

Formula 6[6]
where

Formula 7[7]

Formula 8[8]

Groups of Models Constructed
According to objectives proposed and the variability in the samples (from different soil types and duration of heating), some combinations or groups of models were made with the aim of evaluating the possible influence of this variability (soil type and duration of heating) on the precision and accuracy of the models. Before describing the groups of models, we defined some combinations: (i) local model: model calibrated and validated with a unique soil type; (ii) nonlocal model: model calibrated with several soils but validated with a soil not used in calibration; and (iii) global model: model calibrated and validated with samples from all the soils type.

Four different groups of models were made:

Group of models A (five local models): Five models were constructed separately for each of the five soils studied. The cross validation method was used for the construction of models.
Group of models B (five nonlocal models): Each of these models was constructed (calibrated) with samples from four soils type (calibration set) and validated with samples from the soil type not used in calibration (validation set). These five models were constructed changing the soil type that is used to validate but not to calibrate. The test set validation method was used.
Group of models C (two global models): Two global models (pooling all the experimental samples) were constructed using cross validation and test set validation methods, respectively.
Group of models D (effect of duration of heating): In this experiment, the samples were heated during different periods of time (10–60 min). The influence of the different duration of heating on the MTR estimation was evaluated. For this, one set of samples heated during the same duration was used as a validation set in models constructed (calibrated) without this set. This procedure was performed with all the sets heated for the same duration (six models, one per duration of heating). For these models, we chose the test set validation method.


   RESULTS
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIAL AND METHODS
 RESULTS
DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Changes in Near-Infrared Spectra of Soil as Consequence of Heating
The heat caused changes in the NIR spectra of soils. The effects of heat on the NIR spectra of some soil samples from Relleu site are shown in Fig. 1. The patterns of changes observed in samples of this soil were similar in the others four soils studied (spectra not shown), and thus the next results are also valid for them. The changes in the NIR spectra of soils were different depending on the spectral region considered. The baselines of the absorbance spectra were increased according to the increase in MTR (Fig. 1a). Similar increases were observed by Pietikäinen and Fritze (1995). First derivative of spectra coupled with other preprocessing methods was able to remove undesirable effects on baseline caused by the darkening of the samples. Moreover, first derivative enhanced the resolution of spectrum peaks (Fig. 1b) and was one of the best pre-processing methods for models construction. The decreases in the absorbance peaks on first derivative spectra followed the increase of MTR in most of the spectral range studied. The arrows in Fig. 1b denote regions where the successive increase in MTR caused a progressive decrease in absorbance.


Figure 1
View larger version (32K):
[in this window]
[in a new window]

  		Fig. 1. Representative near-infrared (NIR) spectra of soil samples heated at different temperatures (from Relleu site). (a) NIR spectra of samples heated during 40 min and the maximum temperature reached (MTR) (without preprocessing). (b) NIR spectra of samples heated during 10 min (after first derivative preprocessing). Black line: spectra of the unheated soil. Gray lines: spectra of heated soils. Dotted line: spectra of samples with highest MTR. (b) Arrows denote regions where the successive increase in MTR caused a progressive decrease in absorbance.

 
Groups of Models A: Local Models
In each of the soils studied, the MTR was successfully estimated and predicted by NIR using local models. The values of the r2 (expressed in percentage) of the estimations (calibrations) ranged from 98.75 to 99.38, and the RMSEE ranged from 16.8 to 24.7°C (Table 2). The ranks of the PLS regression ranged from 7 to 9, and the numbers of outliers were from 1 to 2. With respect to the validation (by cross validation) of these models, the values of the r2 ranged from 97.47 to 98.56, and the RMSECV ranged from 25 to 32.5°C (Fig. 2). The numbers of outliers ranged from 1 to 2. The measured values against NIR-predicted (validations) are plotted in Fig. 2. The higher values of RPD, always above 6.2 (Fig. 2), indicate the high accuracy of the models. According to some authors (Chang et al., 2001; Dunn et al., 2002), RPD values higher than 2 can be considered acceptable models, but RPD values higher than 5 can be considered as excellent (Malley et al., 1999).


View this table:
[in this window]
[in a new window]

  		Table 2. Results of calibrations of the local models for the estimation of the maximum temperature reached on soils. Performed by cross validation method.

 

Figure 2
View larger version (24K):
[in this window]
[in a new window]

  		Fig. 2. Relationship between maximum temperature reached in soils measured with thermocouple and predicted by near-infrared spectroscopy in the validations of the local models. Stars denote outliers (not removed).

 
Group of Models B: Nonlocal Models
Results of the group of models B (nonlocal models) are shown in Table 3. For the predictions using nonlocal models, the r2 values ranged between 96.78 and 98.44, and the RMSEP from 26.0 to 37.4°C (Table 3). The high accuracy was also achieved based on RPD values, which were always above 5.7. Thus, the application of models for soils not used during model construction (calibration) is a successful and feasible option to predict MTR by NIR spectroscopy. The predictions of MTR using nonlocal models were good but slightly less accurate than those obtained using local models (~3.8°C on average for soils used in this study).


View this table:
[in this window]
[in a new window]

  		Table 3. Validation of models with a soil type not used in the calibration set (nonlocal models). Validation method: test set validation.

 
Group of Models C: Global Models
The global models constructed in this study, using the samples of five soil types and heated during different durations (the whole of experimental samples, n = 310), were successfully calibrated and validated for the prediction of MTR. Both methods of model construction (cross validation and test set validation) showed high levels of accuracy (r2 > 97); accuracy was slightly better for cross validation (Fig. 3 and 4). The ranks of the global models constructed using cross validation and test set validation methods were 15 and 14, respectively. The ranks of the global models were twofold higher than in local models, reflecting higher complexity. In both models, two and three outliers were detected in the calibration and validation, respectively. The values of the RPD were above 6.


Figure 3
View larger version (16K):
[in this window]
[in a new window]

  		Fig. 3. Relationship between the maximum temperature reached in soils measured with thermocouple and predicted by near-infrared spectroscopy in the global model constructed through cross validation method (n = 310). Closed symbols denote outliers (not removed).

 

Figure 4
View larger version (15K):
[in this window]
[in a new window]

  		Fig. 4. Relationship between maximum temperature reached in soils measured with thermocouple and predicted by near-infrared spectroscopy in the global model constructed with the test set validation method. The calibration set (n = 155) is composed by the 50% of samples (randomly selected), and the validation set (n = 155) is composed by the other 50% of samples. Closed symbols denote outliers (not removed).

 
Group of Models D: Effect of Duration of Heating
After a wildfire, some data are unknown, such as the MTR and the period of time that the soil has been above a certain temperature. The former unknown data could interfere in the prediction if the NIR bands related with the MTR are also modified by the duration of heating.

To address this question, the sets of samples heated during the same duration were used to validate models developed with other sets of samples heated during different duration. The results are shown in Table 4. The high accuracy of predictions (r2 > 97.21; RMSEP < 34.5°C; RPD > 5.9) suggests that the estimation of the MTR is not affected by the duration of heating. In other words, the MTR could be predicted through the NIR spectrum independently of the duration of heating, almost within these ranges of time and temperature.


View this table:
[in this window]
[in a new window]

  		Table 4. Validation of models with a set of duration of heating not used during the calibration. Validation method: test set validation.

 

   DISCUSSION
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIAL AND METHODS
 RESULTS
 DISCUSSION
CONCLUSIONS
 REFERENCES
 
The strongest absorbers in the NIR region are groups with bonds O–H, C–H, N–H, S–H, and C=O, characteristics of organic matter (Malley et al., 2002). Moreover, some of these groups are present in inorganic compounds of soils, such water, ammonium, hydroxides and some minerals. As some authors have noted (Raison, 1979; Neary et al., 1999; Certini, 2005), the heat affects physical, chemical, mineralogic, and biological soil properties. Depending on the compounds or properties considered, the effect of heating is more or less pronounced according to the temperature reached. It has been observed that the changes induced by heat in mineralogic properties take place at higher temperatures than in biological and chemical properties (Ketterings et al., 2000; Pietikäinen et al., 2000; Guerrero et al., 2005). The degree of organic matter oxidation during the pyrolysis depends on the temperature and duration (González-Pérez et al., 2004). Moreover, the thermal resistance of the compounds comprising soil organic matter is not the same. For instance, relative increases in lipid fraction and humin were found in burned soils (Almendros et al., 1988; Fernández et al., 2001). Conversely, holocelluloses and fulvic acids are the most thermolabile fractions (Fernández et al., 2001). Almendros et al. (1990), using an isothermal (350°C) heating (but increasing the duration of heating), found decreases in the atomic ratios H/C and O/C of humus fractions. They interpreted this response as increments in aromaticity, losses of oxygen-containing functional groups, dehydration, and decarboxylation. González-Pérez et al. (2004) offer an excellent review of the effect of fire on soil organic matter. As a consequence, the NIR spectra of soils should change with the effect of heat. In this sense, Pietikäinen and Fritze (1995) found changes in the NIR spectra of humus samples as a consequence of fire. Changes in the MIR spectra (from 4000 to 370 cm–1) of forest humus samples were also observed by Pietikäinen et al. (2000), but a clear relationship with the temperature of heating was not found by the authors. In another work (Fritze et al., 1994), NIR was applied in burned soils but with the aim of predicting microbial biomass and respiration. The MIR spectra of humic acids extracted from heated soils were clearly different according to the temperatures reached (Almendros et al., 1984).

No reports were found in the literature that described models relating the changes in the NIR spectra with temperatures reached in soils. In contrast, for different materials than soils (meat from fishes and beef), some authors (Ellekjaer and Isaksson, 1992; Thyholt et al., 1998; Uddin et al., 2002, 2006) have developed successful models relating the NIR spectra with the maximum temperature of cooking but in a lower range of temperatures than those used in this article.

In the soils used in this study, the heat caused changes in the NIR spectra. This implies that the thermosensible compounds in soils modify the NIR spectra. Most of the spectral regions were affected by the heat. The first derivative preprocessing of spectra enhances the resolution of absorbance bands and can help in the development of models. The direct interpretation of the changes observed in the NIR spectra of the soils can hardly be reached because the absorption bands of different molecules are overlapped in the same spectral regions. In contrast, a direct interpretation may be done if spectra were obtained in the MIR region. The changes induced in NIR spectra by heat take place in a stepwise manner related with the MTR reached. This pattern allowed the development of precise models relating NIR and MTR in each of the five soils studied. These local models suggest that this methodology could be applied for post-fire assessment of MTR in wildfire-affected soils once the calibration model (for the target soil; local model) has been constructed.

There may be interest in the development of models for wider (nonlocal) or global applicability. It was observed that most of the thermal changes that modify the NIR reflectance spectra are not exclusive to soil type. This observation suggests the possibility of the development of a nonlocal and a global model, such as that successfully obtained in this study. The term "global model" is not perfect if we take into account that the model was constructed empirically with only five soils. More types of soils must be evaluated and incorporated into a more realistic global model, improving the validity of this tool, and allowing the direct application irrespective of soil type to avoid the step of local model construction. The development of a perfect global model, taking into consideration all soil types, is almost impossible in accordance with the empirical method of model construction. The use of local models is recommended instead of nonlocal models, especially when the characteristics of the target soil will be largely different from those of soils used for the nonlocal models construction. Organic soils and fresh humus samples may need specific (local) models different from those models developed for mineral soils. Most of the spectroscopic software is able to detect outliers based on the Mahalanobis distance. The Mahalanobis distance offers a quantification of how similar (in multidimensional space) the target spectrum is to those used on model construction and thus alerts about the degree of applicability of the model to this spectrum. Thus, when predictions of low accuracy are suspected using nonlocal models, we recommended the development of a local model. Fernández et al. (2001) found differences in the patterns of the thermoresistant organic fractions of two soils heated at 350°C.

One unknown aspect of wildfire-affected soils is the period of time that the soil has been above a certain temperature. Our data suggest that those parts of the NIR spectrum related to the MTR are not affected by the duration of heating. Thus, the fingerprint in the NIR spectra caused by the MTR is independent of the duration of heating, considering the ranges of time and temperature used in this study. More experiments that increase the ranges of duration of heating must be considered for the development of a more valuable or realistic global model. Development of other NIR-based models focused on the estimation of the duration of heating will be of interest. If this is possible, a more valuable index to assess the fire severity in soils could be described after coupling with the models presented here.

A perfect global model, with its spectroscopic nature, it could be used to calibrate hyperspectral sensors such as AVIRIS (airborne) and Hyperion (boarded on Earth Observing-1 satellite) used in remote sensing (Jia et al., 2006; Kokaly et al., 2006), in contrast to the proximal sensing used in this study. Most of the indexes to assess the burn severity by remote sensing are based on a decrease in vegetation cover and the amount of soil exposed. Few burn severity indexes have been based on changes of soil characteristics (Lewis et al., 2006; Shakesby and Doerr, 2006). Thus, this kind of tool (models) could be integrated in remote sensing after corrections (especially for the strong absorbance of atmospheric water). More work is needed to link these tools.

Most of the direct modifications caused by fire on soils are related to temperature, being basic data needed on fire effects studies. In prescribed burning, the use of some techniques (e.g., thermosensible paints, pyrometers, calorimeters, thermocouples) could provide information about the MTR on soil (Pérez and Moreno, 1998; Kennard et al., 2005), but this kind of data is only available under experimental conditions. In the case of wildfires, the acquisition of these data is practically impossible (Shakesby and Doerr, 2006), and most of available methods are neither precise nor accurate; furthermore, some are subjective, being based on litter and soil appearance (DeBano et al., 1998). Near-infrared spectroscopy offers a new perspective for research focusing on fire effects on soils and fire ecology in general because the MTR can be easily estimated on soils a posteriori with precision and accuracy. Studies about the effects of wildfires could easily incorporate these fundamental and basic data (MTR), increasing the understanding and interpretation of fire effects and allowing the attainment of more powerful conclusions. More articles with conciliating results about direct wildfire effects on soils may become available.

The NIR spectroscopy offers a number of important advantages over other methods. It is a rapid, nondestructive method, requiring minimal pretreatment of samples (only air-drying and sieving), it is highly precise and accurate, and it is free of chemical reagents and toxic waste production. The only significant cost is the acquisition of the spectrometer, but economic spectrometers are available in the market. Highly specialized technicians are not required because NIR spectrometers are simple to operate. These characteristics and the recent development of powerful spectroscopic software, with sophisticated chemometrics tools included, have led to an exponential increase of the use of NIR in many applications, including soil analysis (Blanco and Villarroya, 2002). Once a model has been developed (i.e., calibrated and validated), the time needed per sample analysis is around 1 min. Thus, several hundred soil samples can be scanned (i.e., measured) per day. As a consequence, this tool meets the need for post-fire assessment of MTR. Moreover, once the soil spectrum has been obtained, it could be used in appropriate models to predict simultaneously other parameters in the soils, such as organic C, organic N, microbial biomass C, the cation exchange capacity, and available nutrients (Fritze et al., 1994; Ben-Dor and Banin, 1995; Chang et al., 2001; Malley et al., 1999, 2002). Similar or more powerful predictions of soil parameters have been also observed for MIR spectroscopy (McCarty et al., 2002; Viscarra Rossel et al., 2006).

This technique offers the possibility of conducting research that requires large numbers of samples (e.g., those concerning the spatial pattern of soil temperatures in wildfires [Gimeno-García et al., 2004]), thereby enhancing our understanding of those factors and influencing the spatial pattern at a different scale than does remote sensing. Moreover, the studies about relationships between the spatial pattern of temperatures and spatial changes of effects on site, such as soil hydrophobicity, C dynamics, microbial recolonization, vegetation recovery, or erosion, could be performed better (Shakesby and Doerr, 2006). Another important ecologic aspect related to temperature is post-fire germination and survival of seeds stored in the soil (Bradstock et al., 1992). In Mediterranean conditions, wildfires could start degradation processes of soils, such as erosion (Lewis et al., 2006; Shakesby and Doerr, 2006). Most of these processes are associated with high fire severity. It is important to conduct rehabilitation practices before the start of the degradation processes. For appropriate planning of rehabilitation strategies and location of priority intervention areas, land managers need maps of fire severity with better resolution than that provided by remote sensing and in a shorter time period after the fire. These tasks need large numbers of samples and could be rapidly performed (in a few days) by NIR at low cost. The development of local models is a task that can be made before wildfires.

Regarding the application of this tool on soil samples from field conditions (wildfire affected soils), several questions are still unanswered, such as the period of time that this finger print related to temperature remains in the NIR spectra of burned soils. This information could provide the maximum period of time available for applying this tool on fire affected soils. Moreover, the presence of nonpyrolyzed and semi-pyrolyzed debris and ashes from organic debris burned at different temperatures than the soil is a factor that is assumed to interfere with the MTR prediction. These questions must be evaluated in future studies.


   CONCLUSIONS
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIAL AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
REFERENCES
 
Heat causes changes in the spectra of soils; these changes are related to the MTR in burned soils. Some of these changes are not exclusive to the soil type studied, allowing the development of local and global models. The use of NIR spectroscopy offers the possibility of estimating the MTR accurately, rapidly, and at a low cost. Near-infrared spectroscopy opens new perspectives on studies of fire effects on soils, taking into account that generally the MTR data are not available with sufficient accuracy and precision for soils affected by wildfire.


   ACKNOWLEDGMENTS
 
This research was supported by the CICYT co-financed FEDER project REN2003-08424-C02-01 and the Generalitat Valenciana (Spain) project (GV05/018). NIR spectrometer was purchased from the Spanish Ministerio de Medio Ambiente project 4.2-054/2005/3-B. The authors thank Caja de Ahorros del Mediterráneo.


   NOTES
TOP
 NOTES
ABSTRACT
 INTRODUCTION
 MATERIAL AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Permission for printing and for reprinting the material contained herein has been obtained by the publisher.

Received for publication May 12, 2006.


   REFERENCES
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIAL AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 





This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Guerrero, C.
Right arrow Articles by Gómez, I.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Guerrero, C.
Right arrow Articles by Gómez, I.
Agricola
Right arrow Articles by Guerrero, C.
Right arrow Articles by Gómez, I.
Related Collections
Right arrow Fire
Right arrow Other Models
Right arrow Soil Analysis

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Agronomy Journal Crop Science
Journal of Natural Resources
and Life Sciences Education		 Vadose Zone Journal
Journal of Plant Registrations Journal of
Environmental Quality		 The Plant Genome