Comparing different multivariate calibration methods for the determination of soil organic carbon pools with visible to near infrared spectroscopy

A successful determination of spectrally active soil components with visible and near infrared reflectance spectroscopy (VIS-NIRS, 400-2500 nm) depends on the selection of an adequate multivariate calibration technique. In this study, the contents of thermolabile organic carbon (C_375 °C), the inert organic C fraction (C_inert) and the sum of both (total soil organic carbon, OC_tot) were estimated with three different methods: partial least squares regression (PLSR) as common standard tool, a combination of PLSR with a genetic algorithm (GA-PLSR) for spectral feature selection, and support vector machine regression (SVMR) with non-linear fitting capacities. The objective was to explore whether these methods show differences concerning their ability to predict soil organic carbon pools from VIS-NIR data. For this analysis, we used both measured spectra and also spectra successively blurred with uniformly distributed white noise. Soil sampling was performed in a floodplain (grassland plots) near Osnabrück (Germany) and comprised a total of 149 samples (109 calibration samples, 40 validation samples); spectral readings were taken in the laboratory with a fibre-optics ASD FieldSpec II Pro FR spectroradiometer.In the external validation, differences between the calibration methods were rather small, none of the applied techniques emerged to be the fittest with superior prediction accuracies. For C_375 °C and OC_tot, all approaches provided reliable estimates with r² (coefficient of determination) greater than 0.85 and RPD values (defined as ratio of standard deviation of measurements to standard error of prediction) greater than 2.5. For C_inert, accuracies dropped to r² < 0.50 and RPD < 1.5; after the removal of two extreme values (n = 38) results improved at best (GA-PLSR) to r² = 0.80 and RPD = 1.98. The noise experiment revealed different responses of the studied approaches. For PLSR and GA-PLSR, increasing spectral noise resulted in successively reduced r² and RPD values. By contrast, SVMR kept high coefficients of determination even at high levels of noise, but increasing noise caused severely biased estimates, so that regression models were less accurate than those of PLSR and GA-PLSR.     

高光谱图像结合特征变量筛选定量检测羊肉中狐狸肉掺假

为了探讨快速无损检测羊肉糜中狐狸肉掺假含量的可行性,该研究利用高光谱技术结合特征变量筛选方法开展了其定量检测研究。利用遗传算法、竞争性自适应重加权算法和二维相关光谱分析（Two-Dimensional Correlation Spectroscopy,2D-COS）3种方法分别对代表性样品全部846个波长进行特征波长筛选,得到207、34和14个特征波长;基于全部波长和特征波长建立羊肉糜中狐狸肉掺假含量的偏最小二乘回归（Partial Least Squares Regression,PLSR）和支持向量回归（Support Vector Regression,SVR）模型并进行比较。研究结果表明,基于全部波长和特征波长建立的SVR模型性能均优于PLSR模型。其中,利用2D-COS方法提取的14个特征波长建立的SVR模型（即2D-COS-SVR模型）性能最优,其预测集决定系数和均方根误差分别为0.928和3.00%,相对分析误差为4.85,表明高光谱结合2D-COS-SVR模型可以有效实现羊肉糜中狐狸肉掺假的定量检测。该研究结果为开发低成本肉类掺假检测系统提供技术支持和参考依据。     

Quantitative analysis of soil nitrogen and carbon at a farm scale using visible and near infrared spectroscopy coupled with wavelength reduction

Reducing large spectral datasets to parsimonious representations of wavelengths is of value for efficient storage and easing analysis, in addition to the potential to use a simpler and cheaper spectrophotometer. This study evaluated the potential of calibrating visible and near infrared (vis‐NIR) spectra to total nitrogen (N), total carbon (C), organic C and inorganic C in soil on a 15‐ha farm, with the aim of comparing several wavelength reduction algorithms and rates in terms of model prediction accuracy. We explored the uninformative variables elimination (UVE), UVE coupled with successive projections algorithm (SPA) and two uniform‐interval wavelength reduction approaches (UWR‐I and UWR‐II) with successive wavelength reduction rates (WRRs) of 2, 5, 10, 20, 50, 100, 200, 500 and 1000. The standard normal variate (SNV)‐transformed absorbance spectra of soil samples recorded from 400 to 2499 nm at 1‐nm intervals were used. The calibration sets were subjected to a partial least squares regression (PLSR) with leave‐one‐out cross‐validation. Prediction results showed that UVE can reduce wavelength variables significantly while retaining good model prediction accuracy. The UVE‐SPA produced only three or four wavelengths, with which PLSR models achieved competitive prediction performance, compared with those based on all 2100 wavelengths, with coefficient of determination (R²) of 0.91, 0.89, 0.91 and 0.53 and residual prediction deviation (RPD) of 3.53, 2.95, 3.27 and 1.53 for soil total N, total C, organic C and inorganic C, respectively. The UWR tests showed that PLSR models responded insensitively to various WRRs from 2 to 100. The models calibrated for the 100‐nm interval spectra (21 remaining wavelengths) performed almost as well as those for the 1‐nm interval spectra. Although these findings might be valid only at the farm scale, it is recommended that the proposed wavelength reduction algorithms for more soil types and soils originated from larger areas should be examined.     

Near infrared reflectance spectroscopy (NIRS) could be used for characterization of soil nematode community

Studying soil nematofauna provides useful information on soil status and functioning but requires high taxonomic expertise. Near infrared reflectance (NIR) spectroscopy (NIRS) has been reported to allow fast and inexpensive determination of numerous soil attributes. Thus the present study aimed at assessing the potential of NIRS for determining the abundance and diversity of soil nematodes in a set of 103 clayey topsoil samples collected in 2005 and 2006 from agricultural soils in the highlands of Madagascar.The morphological characterization of soil nematofauna involved extraction through elutriation then counting under binoculars and identification at family or genus level using microscopy, on ca. 150-g fresh soil samples. Taxa were assigned to five trophic groups, namely bacterial feeders, fungal feeders, obligate plant feeders, facultative plant feeders, and omnivores and predators (together). In addition, four ecological indexes were calculated: the Enrichment index, Structure index, Maturity index, and Plant parasitic index.Oven-dried (40 °C) < 2-mm sieved 5-g soil subsamples were scanned in the NIR range (1100-2500 nm), then spectra were fitted to nematofauna data using partial least square regression. Depending on the sample set considered (year 2005, year 2006, or both years), NIRS prediction of total nematode abundance was accurate (ratio of standard deviation to standard error of cross validation, i.e. RPD ≥ 2) or acceptable (RPD ≥ 1.6). Predictions were accurate, acceptable, or quasi-acceptable (RPD ≥ 1.4) for several of the six most abundant taxa, and to a larger extent, for most trophic groups (except facultative plant feeders); but they could not be made for taxa present in a small number of samples or at low abundance. By contrast, NIRS prediction of relative abundances (in proportion of total abundance) was poor in general, as was also the prediction of ecological indexes (except for the 2006 set). On the whole, these results were less accurate than NIRS predictions of soil attributes often reported in the literature. However, though not very accurate, NIRS predictions were worthwhile considering the labor-intensity of the morphological characterization. Most of all, NIRS analyses were carried out on subsamples that were probably too small (5 g) to allow representative sampling of nematofauna. Using larger samples for NIRS (e.g. 100 g) would likely result in more accurate predictions, and is therefore recommended. Scanning un-dried samples could also help improve prediction accuracy, as morphological characterization was carried out on samples not dried after sampling.Examining wavelengths that contributed most to NIRS predictions, and chemical groups they have been assigned to, suggested that NIRS predictions regarding nematofauna depended on constituents of both nematodes and preys’ food. Predictions were thus based on both nematofauna and soil organic properties reflected by nematofauna.     

基于不同光谱变换的土壤盐含量光谱特征分析

跟踪初生盐渍土壤的微生物修复实验,采用同步实测得土壤盐含量和光谱数据,详细分析了基于34种光谱变换,修复过程中盐渍土的光谱特征。对于选取的6种光谱变换,采用全波段(400~1650 nm)和分析获得的最佳敏感波段分别建立了土壤盐含量的光谱反演PLSR(partial least squares regression)模型。研究表明,光谱变换处理使土壤盐含量与平滑后的光谱反射数据的相关性明显增强,且最佳敏感波段范围进一步聚焦。本研究得到最佳光谱变换为导数变换,基于全波段的土壤盐含量预测模型以SGSD变换效果最好,与原始光谱相比,模型的r、RMSEP分别从0.537和1.928改善到0.823和1.256。而SGSD(Log R)是基于最佳波段所建立的盐含量预测模型的有效光谱变换方法,该研究为进一步实现盐渍土中盐含量快速定量分析提供了方法和数据参考。     

Partial least squares regression (PLSR) associated with spectral response to predict soil attributes in transitional lithologies

New techniques and improvements are required to quantify soil’s chemical and physical properties on production environment, reducing environmental impacts and minimizing soil analysis time. The aim of this study is to evaluate the possibility to estimate the content of silt, sand, clay, total iron and organic matter in soils formed by different lithologies in Parana State, Brazil, using VIS-NIR spectrum associated with Partial Least Square Regression (PLSR). 200 soil samples were collected in an area formed by Lixisols, Cambisols, Ferralsols, Arenosols and Nitisols in a depths of 0–0.2 and 0.2–0.8 m. Spectral readings were obtained in laboratory by FieldSpec 3 JR sensor. The spectral curves of the samples were correlated to the attributes through PLSR. The results obtained for sand in prediction were better when compared to the other attributes, presenting R² = 0.90, r = 0.95 and RPD = 2.3. Clay and total iron presented satisfactory results, mainly for RPD values, which were above 2.4. Based on the results, it can be concluded that the PLSR technique associated with the spectral response of the soils, was able to estimate sand, clay and total iron with accuracy in a region formed by reworked materials, derived from several lithologies.     

Statistical and geostatistical techniques for geospatial modeling of soil cation exchange capacity

Soil cation exchange capacity (CEC) is a main criterion of soil quality and pollutant sequestration capacity. This research was carried out to evaluate cokriging prediction map of soil CEC spatial variability with the principal components derived from soil physical and chemical properties. Two hundred and forty-seven soil samples were collected that 75% of them were used for training soil CEC and 25% for testing of prediction. The first principal component (PC1) was highly correlated with soil CEC (r = 0.81, P < 0.01), whiles there was no significant correlation between CEC and PC2 (r = -0.19). Then, the PC1 was used as an auxiliary variable for the prediction of soil CEC in cokriging method. The determination coefficient (R²) of cross-validation for the test dataset was 0.47 for kriging and 0.71 for cokriging. Therefore, according to the results, principal components that have the highest positive and significant correlation with the dependent variable have the most potential for cokriging prediction.     

基于高光谱成像技术的猪肉新鲜度评价

该文研究利用高光谱成像技术预测猪肉新鲜度参数,挥发性盐基氮(TVB-N)和pH值。在470～1000nm波长范围内,从高光谱图像中提取的反射光谱,分别经过2次Savitzky-Golay(S-G)平滑、多元散射校正(MSC)处理后,建立PLSR(偏最小二乘法)的预测模型。对TVB-N的预测,使用2次S-G平滑处理、MSC光谱建立的PLSR预测模型相关系数分别为0.90和0.89,预测模型标准差分别为7.80和8.05。对pH值的预测,经过MSC处理比2次S-G平滑处理的结果好,相关系数为0.79,预测模型标准差为0.37。同时综合2个参数利用MSC处理后的预测模型对猪肉新鲜度进行评定,准确率达91%。研究结果表明,高光谱成像技术可以用于猪肉新鲜度快速无损检测。     

Prediction of soil organic matter using a spatially constrained local partial least squares regression and the Chinese vis–NIR spectral library

We need to determine the best use of soil vis–NIR spectral libraries that are being developed at regional, national and global scales to predict soil properties from new spectral readings. To reduce the complexity of a calibration dataset derived from the Chinese vis–NIR soil spectral library (CSSL), we tested a local regression method that combined geographical sub‐setting with a local partial least squares regression (local‐PLSR) that uses a limited number of similar vis–NIR spectra (k‐nearest neighbours). The central idea of the local regression, and of other local statistical approaches, is to derive a local prediction model by identifying samples in the calibration dataset that are similar, in spectral variable space, to the samples used for prediction. Here, to derive our local regressions we used Euclidean distance in spectral space between the calibration dataset and prediction samples, and we also used soil geographical zoning to account for similarities in soil‐forming conditions. We tested this approach with the CSSL, which comprised 2732 soil samples collected from 20 provinces in the People's Republic of China to predict soil organic matter (SOM). Results showed that the prediction accuracy of our spatially constrained local‐PLSR method (R² = 0.74, RPIQ = 2.6) was better than that from local‐PLSR (R² = 0.69, RPIQ = 2.3) and PLSR alone (R² = 0.50, RPIQ = 1.5). The coupling of a local‐PLSR regression with soil geographical zoning can improve the accuracy of local SOM predictions using large, complex soil spectral libraries. The approach might be embedded into vis–NIR sensors for laboratory analysis or field estimation.     

(Methyl)Mercury,Arsenic, and Lead Contamination of the World’s Largest Wastewater Irrigation System: the Mezquital Valley (Hidalgo State—Mexico)

We investigate the potential of near-infrared (NIR) spectroscopy to predict some heavy metals content (Zn, Cu, Pb, Cr and Ni) in several soil types in Stara Zagora Region, South Bulgaria, as affected by the size of calibration set using partial least squares (PLS) regression models. A total of 124 soil samples from the 0–20 and 20–40 cm layers were collected from fields with different cropping systems. Total Zn, Cu, Pb, Cr and Ni concentrations were determined by Atomic Absorption Spectrometry. Spectra of air dried soil samples were obtained using an FT-NIR Spectrometer (spectral range 700–2,500 nm). PLS calibration models were developed with full-cross-validation using calibration sets of 90 %, 80 %, 70 % and 60 % of the 124 samples. These models were validated with the same prediction set of 12 samples. The validation of the NIR models showed Cu to be best predicted with NIR spectroscopy. Less accurate prediction was observed for Zn, Pb and Ni, which was classified as possible to distinguish between high and low concentrations and as approximate quantitative. The worst model performance in cross-validation and prediction was for Cr. Results also showed that values of root mean square error in cross-validation (RMSE_cv) increased with decreasing number of samples in calibration sets, which was particularly clear for Cu, Pb, Ni and Cr content. A similar tendency was observed in the prediction sets, where RMSEP values increased with a decrease in the number of samples, particularly for Pb, Ni and Cr content. This tendency was not clear for Zn, while even an increase in RMSEP for Cu with the sample size was observed. It can be concluded that NIR spectroscopy can be used to measure heavy metals in a sample set with different soil type, when sufficient number of soil samples (depending on variability) is used in the calibration set.     

基于相似光谱匹配预测土壤有机质和阳离子交换量

土壤可见光-近红外波段光谱(350~2 500 nm)包含了大量的土壤属性信息,相同类型的土壤具有相似的光谱曲线特征,但相似光谱曲线是否具有相似的属性含量?探讨此问题可为土壤光谱库的应用提供依据,从而最终服务于快速获取土壤信息技术体系的构建。该研究以安徽宣城为研究区,根据母质、地形特征和土地利用等信息,采集91个典型土壤剖面,共含400个土壤发生层样品,测定了有机质(soil organic matter,SOM)和阳离子交换量(cation exchange capacity,CEC)含量,同时采用VARIAN公司的Cary 5000分光光度计测定了土壤光谱,并将光谱数据变换为反射率(R)、反射率一阶导数(FDR)和吸收度(Log(1/R))3种形式。该文采用光谱角(spectral angle mapper,SAM)、偏最小二乘回归(partial least square regression,PLSR)和SAM-PLSR(spectral angle mapper-partial least square regression,SAM-PLSR)3种方法预测土壤SOM和CEC。SAM方法是通过对测试集104个光谱曲线与参考集的296个光谱曲线进行相似性计算,并以此实现土壤SOM和CEC含量的预测。SAM-PLSR方法以SAM算法下的匹配结果作为建模样本建立PLSR模型和进行预测分析。结果表明,具有相似光谱曲线的土壤具有相似的SOM和CEC含量,SAM算法下相似光谱匹配可直接预测SOM(R2=0.78,RPD=2.17)和CEC(R2=0.82,RPD=2.41)。PLSR方法可很好地预测SOM(R2=0.87,RPD=2.77)和CEC(R2=0.87,RPD=2.59);相较之下,SAM-PLSR方法不仅可以更加准确预测SOM(R2=0.89,RPD=3.00)和CEC(R2=0.91,RPD=3.06),而且大大减少了建模样本的数量。该研究使可见光-近红外光谱可更加高效地用于土壤属性分析,并为土壤光谱数据库的建设及应用提供技术参考。     

基于VNIR和机器学习算法的原状土剖面Cu含量预测

快速测量土壤剖面重金属含量是评估土壤重金属污染状况并选择相应修复技术的关键。为了探讨可见光-近红外光谱法（Visible and Near-Infrared Reflectance Spectroscopy,VNIR）预测原状土壤剖面重金属含量的潜力,以江西省两个典型工矿厂周边农田土壤为研究对象,共采集了19个深度约100 cm的完整土壤剖面样品,分别测定土壤剖面样品的VNIR数据及其Cu含量。采用偏最小二乘回归法（Partial Least Squares Regression,PLSR）、Cubist混合线性回归决策树（Cubist Regression Tree,Cubist）、高斯过程回归（Gaussian Process Regression,GPR）和支持向量机（Support Vector Machine Regression,SVM）方法研究不同光谱预处理方法对土壤Cu含量预测精度的影响。结果显示,Cubist、GPR和SVM这三种机器学习算法的预测精度普遍高于PLSR,其中一阶导数（First-Order Derivative,FD）预处理的SVM模型预测精度最高（R2=0.95,均方根误差为7.94 mg/kg,相对分析误差为4.34）。这表明利用VNIR和机器学习可以对原状土壤剖面Cu含量进行有效预测,为快速监测Cu及其他重金属含量的相关研究提供参考。     

The potential of hyperspectral images and partial least square regression for predicting total carbon,total nitrogen and their isotope composition in forest litterfall samples

Purpose

The main objective of this study was to examine the potential of using hyperspectral image analysis for prediction of total carbon (TC), total nitrogen (TN) and their isotope composition (δ¹³C and δ¹⁵N) in forest leaf litterfall samples.

Materials and methods

Hyperspectral images were captured from ground litterfall samples of a natural forest in the spectral range of 400–1700 nm. A partial least-square regression model (PLSR) was used to correlate the relative reflectance spectra with TC, TN, δ¹³C and δ¹⁵N in the litterfall samples. The most important wavelengths were selected using β coefficient, and the final models were developed using the most important wavelengths. The models were, then, tested using an external validation set.

Results and discussion

The results showed that the data of TC and δ¹³C could not be fitted to the PLSR model, possibly due to small variations observed in the TC and δ¹³C data. The model, however, was fitted well to TN and δ¹⁵N. The cross-validation R² _cv of the models for TN and δ¹⁵N were 0.74 and 0.67 with the RMSE_cv of 0.53% and 1.07‰, respectively. The external validation R² _ex of the prediction was 0.64 and 0.67, and the RMSE_ex was 0.53% and 1.19 ‰, for TN and δ¹⁵N, respectively. The ratio of performance to deviation (RPD) of the predictions was 1.48 and 1.53, respectively, for TN and δ¹⁵N, showing that the models were reliable for the prediction of TN and δ¹⁵N in new forest leaf litterfall samples.

Conclusions

The PLSR model was not successful in predicting TC and δ¹³C in forest leaf litterfall samples using hyperspectral data. The predictions of TN and δ¹⁵N values in the external litterfall samples were reliable, and PLSR can be used for future prediction.

     

Near-Infrared Reflectance Spectroscopy as a Method for Determining Organic Carbon Concentrations in Soil

Near-infrared reflectance spectroscopy (NIRS) has the potential to be a reliable method for accurately quantifying soil organic carbon (SOC). The objective of this study was to evaluate NIRS as a method for predicting SOC. Partial least squares (PLS) regression was used to predict SOC from soil reflectance values or the first derivative of the reflectance values. Two model validation techniques were evaluated: One was a full cross-validation and in the other 30 percent of the samples were removed from the calibration data set and then tested using the calibrated model. Significant relationships were observed for predicted SOC when compared to laboratory-measured SOC for all models evaluated, regardless of validation technique. The prediction models using the first derivative of the reflectance values outperformed prediction models using the reflectance values alone. In conclusion, NIRS can be used as a quick and accurate method for measuring SOC.     

滩涂土壤有机质含量的反射光谱估算

Rapid determination of soil organic matter (SOM) using regression models based on soil reflectance spectral data serves an important function in precision agriculture. “deviation of arch”(DOA)-based regression and partial least squares regression (PLSR) are two popular modeling approaches to predict SOM. However, few studies have explored the accuracy of the DOA-based regression and PLSR models. Therefore, the DOA-based regression and PLSR were applied to the visible near-infrared (VNIR) spectra to estimate SOM content in the case of various dataset divisions. A two-fold cross-validation scheme was adopted and repeated 10 000 times for rigorous evaluation of the DOA-based models in comparison with the widely used PLSR model. Soil samples were collected for SOM analysis in the coastal area of northern Jiangsu Province, China. The results indicated that both modelling methods provided reasonable estimates of SOM, with PLSR outperforming DOA-based regression in general. However, the performance of PLSR for the validation dataset decreased more noticeably. Among the four DOA-based models, the linear model of the DOA provided the best estimation of SOM and a cutoff of SOM content (19.76 g kg^-1), and the performance for calibration and validation datasets was consistent. As the SOM content exceeded 19.76 g kg^-1, SOM became more effective in masking the spectral features of other soil properties to a certain extent. This work confirmed that reflectance spectroscopy combined with PLSR could serve as a non-destructive and cost-efficient way for rapid determination of SOM when hyperspectral data were available. The DOA-based model, which requires only 3 bands in the visible spectra, also provided SOM estimation with acceptable accuracy.     

支持向量机在土壤镁含量高光谱估算中的应用

研究利用土壤样本实验反射光谱,分析了土壤镁(Mg)含量与土壤反射光谱的关系,比较了主成分回归分析(PCR)、偏最小二乘回归分析(PLSR)和支持向量机回归分析(SVMR)等方法,以及土壤反射光谱及其变换光谱与土壤Mg含量之间的估算模型,为土壤Mg含量高光谱估算提供依据。结果表明:PCR、PLSR、SVMR 3种建模方法在Mg含量的估算中,SVMR的估算精度相对较高,估算精度平均达到80.96%,分别比PCR和PLSR提高了6.16%、4.20%;对于不同的数学变换处理方法,一阶微分变换相对较好,估算精度平均为80.76%,分别比反射率、倒数对数变换提高了4.95%、4.61%。因此,运用土壤反射光谱一阶微分变换的SVMR进行建模,可以相对较好地估算全Mg含量,精度达84.04%。     

Soil carbon mapping using on-the-go near infrared spectroscopy, topography and aerial photographs

Efficient tools for accurate soil carbon (SC) mapping are imperative for large scale assessment of total SC stocks and their changes in time as well as for site-specific tailoring of agricultural management practices. On-the-go near infrared (NIR) reflectance spectroscopy has been used recently in aid to the conventional, laborious and expensive soil analyses, since NIR measurements are taken in-situ quickly and non-destructively. However, NIR spectrum data need to be effectively calibrated with conventionally measured SC. Our objectives are to compare calibration approaches, including pre-processing transformations (Savitzky-Golay derivatives, standard normal variate and mean centering) and multivariate statistical methods (principal component regression, partial least squares, partial least squares leaving one-outlier-out) for using NIR spectra data as SC predictor, to evaluate NIR reliability in predicting SC under low carbon contents typical for Midwest Alfisols; and finally to compare predictions of SC by using three sources of auxiliary information (NIR spectral data, visible-NIR reflectance obtained from aerial photographs and topographical features). No improvements in calibration accuracy were observed when using pre-processing transformations. Partial least squares (RMSE = 1.90) tended to perform better than principal component regression (RMSE = 1.96); especially when spectral-NIR outliers are dropped (RMSE = 1.68). Our results suggested that visible-NIR data from aerial photographs used along with topographical attributes outperformed on-the-go spectral NIR data. Topographical data improved prediction in the models with aerial photograph visible-NIR data; however no improvement was noticed when used with spectral-NIR data. Though, NIR spectral data is frequently used as a proxy for SC prediction, we notice that this auxiliary information is not well suited under all scenarios. Particularly, when SC levels are low and the range of SC data is narrow, as in this study, NIR was only moderately successful in predicting SC.     

Prediction of carbon mineralization rates from different soil physical fractions using diffuse reflectance spectroscopy

Soil carbon (C) mineralization rate is a key indicator of soil functional capacity but it is time consuming to measure using conventional laboratory incubation methods. Recent studies have demonstrated the ability of visible-near infrared spectroscopy (NIRS) for rapid non-destructive determination of soil organic carbon (SOC) and nitrogen (N) concentration. We investigated whether NIRS (350-2500 nm) can predict C mineralization rates in physically fractionated soil aggregates (bulk soil and 6 size fractions, n=108) and free organic matter (2 size fractions, n=27) in aerobically incubated samples from a clayey soil (Ferralsol) and a sandy soil (Arenosol). Incubation reference values were calibrated to first derivative reflectance spectra using partial least-squares regression. Prediction accuracy was assessed by comparing laboratory reference values with NIRS values predicted using full hold-out-one cross-validation. Cross-validated prediction for C respired (500 days) in soil aggregate fractions had an R² of 0.82 while that of C mineralized (300 days) in organic matter fractions was 0.71. Major soil aggregate fractions could be perfectly spectrally discriminated using a 50% random holdout validation sample. NIRS is a promising technique for rapid characterization of potential C mineralization in soils and aggregate fractions. Further work should test the robustness of NIRS prediction of mineralization rates of aggregate fractions across a wide range of soils and spectral mixture models for predicting mass fractions of aggregate size classes.     

Near-Infrared Spectroscopy Coupled with Chemometrics Tools: A Rapid and Non-Destructive Alternative on Soil Evaluation

Near-infrared (NIR) spectroscopy is a rapid, non-destructive and accurate technique for analyzing a wide variety of samples, thus, the growing interest of using this technique in soil science. The objective of this study was to evaluate the potential of NIR spectroscopy to predict organic carbon (OC), total nitrogen (TN), available phosphorus (P) and available potassium (K) in the soil. NIR spectra from 20 cm³ of soil samples were acquired on the range of 750 to 2500 nm in diffuse reflectance mode, resolution of 16 cm^?1 and 64 scans. Eight models of calibration/validation were constructed. Calibration and validation models showed that the predictive potential of NIR varied with the specific soil property (OC, TN, P and K) under evaluation and according to the methodology employed in the model construction (cross-validation or test set). Good prediction models were obtained for OC and TN content based on the statistical parameters. Test set methodology was able to predict soil OC, TN, P, and K better than cross-validation methodology.     

Use of different chemometric approaches for an estimation of soil properties at field scale with near infrared spectroscopy

Visible and near infrared spectroscopy (vis‐NIRS) may be useful for an estimation of soil properties in arable fields, but the quality of results are often variable depending on the applied chemometric approach. Partial least squares regression (PLSR) may be replaced by approaches which employ supervised learning methods or variable selection procedures in order to increase the proportion of informative wavelengths used in the estimation procedure, to reduce the noise of the spectra and to find the best fitting solution. Objectives were (1) to compare the usefulness of PLSR with either PLSR combined with a genetic algorithm (GA‐PLSR) or support vector machine regression (SVMR) for an estimation of soil organic carbon (SOC), total nitrogen (N), pH, cation exchange capacity (CEC) and soil texture for surface soils (0–5 cm, n = 144) of an arable field in Bangalore (India) and (2) to test and optimize different calibration strategies for GA‐PLSR for an improved estimation of soil properties. PLSR was useful for an estimation of SOC, N, sand and clay. In the cross‐validation (n = 96), accuracies of estimated soil properties generally decreased in the order GA‐PLSR > SVMR > PLSR. However, the order of estimation accuracies for the random validation sample (n = 48) changed to SVMR > GA‐PLSR > PLSR for SOC, N, pH, and CEC, whereas for clay the order changed to SVMR > PLSR > GA‐PLSR. A sequential procedure, which used the most frequently selected wavelengths of the GA‐PLSR runs, proved to be useful for an improved estimation of SOC and N. Overall, SVMR especially improved estimations of SOC and clay, whereas GA‐PLSR was particularly useful for SOC and N and it was the only approach which successfully estimated CEC in cross‐validation and validation.