ارزیابی قابلیت طیف‌سنجی بازتابی در پیش‌بینی کربنات‌های خاک (مطالعه موردی: منطقه جونقان در استان چهارمحال و بختیاری)

در سال‌های اخیر طیف‌سنجی به‌عنوان یکی از تکنیک‌های ممکن در جایگزینی روش‌های مرسوم آزمایشگاهی در علوم خاک معرفی شده است. هدف از پژوهش حاضر، ارزیابی قابلیت این تکنیک در دو محدوده‌ی مرئیمادون قرمز نزدیک (visnir) و مادون قرمز میانی (midir) در پیش‌بینی کربنات‌های خاک در منطقه جونقان استان چهارمحال و بختیاری می‌باشد. به این منظور 272 نمونه‌ی خاک سطحی از عمق 10-0 سانتی‌متری جمع‌آوری و میزان کربنات هر یک با روش تیتراسیون برگشتی تعیین شد. اطلاعات طیفی خاک‌ها در گستره  visnirبا استفاده از اسپکترورادیومتر زمینیfieldspec 3, asdanalytical spectral devices, boulder colorado, usa))، در محدوده‌ی 2500-350 نانومتر با تفکیک طیفی 1 نانومتر و در محدوده midir با اسپکترورادیومتر تبدیل فوریه مادون قرمز ftir (thermo fisher scientific inc., waltham, ma)، در گستره  cm^-14000-400 (25000-2500 نانومتر) با تفکیک طیفی 1.2 نانومتر استخراج شد. سپس انواع مختلف روش‌های پیش‌پردازش بر اطلاعات طیفی، اعمال شده، داده‌ها به دو گروه واسنجی (70%) و اعتبارسنجی (30%) تقسیم و چهار مدل، رگرسیون حداقل مربعات جزئی (plsr)، ماشین بردار پشتیبان (svm)، جنگل تصادفی rf)) و رگرسیون فرآیند گاوسی (gpr) برای پیش‌بینی کربنات‌ها از این اطلاعات، برازش یافت. نتایج نشان داد که ترکیب مدل svm با داده‌های خام طیفی در محدوده‌ی visnir و ترکیب مدل plsr با روش پیش‌پردازش منحنی حذف پیوستار (cr) در گستره midir، به ترتیب با 0.81=r^2 و 0.86=r^2 بهترین عملکرد را در پیش‌بینی کربنات‌ها داشته‌اند. همچنین نتایج نشان داد که عملکردگستره midir در برآورد کربنات‌ها نسبت به visnir بالاتر بوده است. در مجموع می‌توان تکنیک طیف‌سنجی را به‌عنوان روشی سریع و البته دقیق در تخمین کربنات‌های خاک مطرح و موردارزیابی‌های بیشتر قرار داد.

Evaluation of reflectance spectroscopy for assessment of soil carbonates (case study: Juneqan district in Chaharmohal and Bakhtiari Province)

Introduction Carbonates are an essential and prominent constituent of soil chemical properties particularly in arid and semiarid regions, in regards with soil productivity and conservation. The conventional techniques for assessing soil properties rely on direct laboratory measurements which are expensive, time consuming and labor intensive. Hence, it is required to develop fast and costefficient techniques for evaluation of mentioned parameters. The Koppen climatic classification generally categorizes Iran among the arid and semiarid climates. About 90 % of its lands are arid or semiarid. According to Soil Survey Staff (2014), calcareous soils contain 5% or more volumes of inorganic carbon (or carbonate calcium equivalent), which are the prevailing formation in arid and semiarid areas. These soils are typical of areas where minerals cannot be leached away from the soil profile due to low precipitation. Based on the reports of FAO.UNDP (1972), approximately 12% of soils all over the world and 65% in Iran are calcareous. Therefore, carbonate is a key component that physically and chemically influences soil properties, as well as its fertility and productivity. One of the fast, easytouse, costeffective and nondestructive methods of soil analysis is the visible to nearinfrared (VisNIR) and midinfrared (midIR) spectroscopy, that can partly be employed for the optimization of traditional techniques. Therefore, the reflectance spectroscopy is considered as one of relatively inexpensive and fast techniques to evaluate these features. The purpose of the present study was to evaluate the capability of the reflectance spectroscopy technique in VisNIR (2502500 nm) and midIR (400400 cm1) ranges to estimate soil carbonates content as one of the key components affecting the physical and chemical properties of soils (especially in arid and semiarid regions). Materials and Methods The study area is located in Juneqan District, Chaharmohal and Bakhtiari Province, southwest of Iran. 272 soil samples were collected from a depth of 010 cm, air dried and passed through a 2 mm sieve. The carbonates value of each sample was determined by standard laboratory method. The spectral reflectance of soil samples was extracted in the VisNIR (2502500 nm) and midIR (400400 cm1) ranges using a spectroradiometer FieldSpec 3 (ASDAnalytical Spectral Devices, Boulder Colorado, USA) and Nicolet 6700 Fourier Transform Infrared (FTIR) (Thermo Fisher Scientific Inc., Waltham, MA), respectively. In the next step, seven preprocessing methods included absorbance transformation (log [1/reflectance]) (Abs), multiplicative scatter correction (MSC), standard normal variate transformation (SNV), SavitzskyGolay derivation (SGD), Continuum removal transformation (CR), Normalization in range <1,>1 (Nor) and Detrend (Det), were performed over original spectra for correcting light scattering in reflectance measurements and data improvement before using data in calibration models. Afterward, The dataset (272 samples) for each spectra range was randomly divided in calibration (70%) and validation (30%) datasets. Four different calibration models were fitted over VisNIR and midIR spectra to develop carbonates prediction models including: Partial Least Squares Regression (PLSR), Support Vector Machine (SVM), Random Forest (RF) and Gaussian Process Regression (GPR). The evaluation of soil predicting models was done according to the value of R2, RMSE and RPD. According to some researches, RPD values more than 2 shows that the models provide precise predictions, values of RPD between 1.4 and 2 are considered to be reasonably representative, and values less than 1.4 indicate poor predictive value. Results and Discussion The carbonates content in studied samples ranged from 1 to 76% with an average value of 24.7%. Overall, carbonates content promoted increase of spectral reflectance intensity on several region of spectrum in both spectral ranges. The specific absorption wavelength in VisNIR spectra used to indicate the presence of soil carbonates was 2338 nm and in the midIR range were 714, 850, 870, 1796, and 2510 cm1. The results showed that the best performance of the used models in the VisNIR spectral range was related to the SVM model (R2=0.81, RMSE=5.36) and in the midIR range allocated to PLSR model (R2=0.86, RMSE=4.5). Both of these models showed great accuracy in carbonates estimating (RPD>2). Besides, the results showed that the midIR spectral range in the prediction of carbonates provided better performance than the VisNIR range. This can explained by the fact that the fundamental molecular vibrations of soil components occur in the midIR range, while only their overtones and combinations are detected in the VisNIR range. Conclusion It seems that the reflectance spectroscopy technique can be considered as a precise substitute for the conventional methods of measuring carbonates, which are sometimes costly, time consuming and destructive. However, due to the spatial and temporal variability of soil properties as well as the huge variety of models and spectral preprocessing methods, it is necessary to examine the capability of this technique in other areas with other preprocessing methods and regression models.