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。     

不同模型在渔业CPUE标准化中的比较分析

为了提高渔业数据单位捕捞努力量渔获量（catch per unite of effort,CPUE）标准化数据的质量和模型连续稳定预测能力,该文采用人工神经网络（artificial neural network,ANN）、回归树（regression trees,RT）、随机森林（random forest,RF）和支持向量机（support vector machine,SVM）等机器学习方法和传统的广义线性模型（generalized linear model,GLM）等方法,对2000－2013年大西洋大眼金枪鱼（Thunnus obesus）延绳钓CPUE数据进行标准化。采用平均绝对误差、平均均方误差、3种相关系数（Pearson's,Kendall's和Spearman's）和标准化均方误差等评价指标对不同模型标准化结果进行对比,寻找较优的标准化方法。研究结果表明,在验证数据集SVM方法得到的3种相关系数（0.596,0473和0.632）和RF（0.623,0.456,0.621）相似,高于RT(0.516,0.432和0.586)、ANN(0.428,0.249和0.365)和GLM(0.199,0.106和0.159)。SVM预测的均方误差（11.25）、平均绝对误差（2.107）和标准化均方误差（0.652）略低于RF（11.655,2.377和0.661）,明显低于RT（14.999,2.434和0.801）、ANN(16.692,2.883和0.823)和GLM（16.517,2.777和0.993）。各项指标揭示SVM方法要优于其他4种方法,RF次之,GLM计算结果在所有方法中最差,不适合渔业数据CPUE标准化。SVM和RF方法应该被优先考虑用于渔业数据CPUE标准化。研究结果为渔业资源管理和保护提供更好的支持。     

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.     

Chemical composition of cover crops and soil organic matter pools in no-tillage systems in the Cerrado

In no-tillage systems (NTS), cover crops are recommended to increase the productivity of agricultural systems. Furthermore, a greater diversity of cover crops in NTS favours an increase in soil carbon (C) stocks. However, there are scarce published data on the relationship between the chemical composition of cover crops and the accumulation of labile and stable fractions of SOM. We evaluated the relationship between the chemical composition of cover crops and SOM fractions, C stocks and maize yield. Hemicellulose, cellulose and lignin contents were determined for Urochloa ruziziensis, Canavalia brasiliensis, Cajanus cajan and Sorghum bicolor, cultivated in the off-season of maize. Canavalia brasiliensis had high N (20.96 g kg⁻¹) and hemicellulose (185.67 g kg⁻¹) contents, lower lignin content (39.50 g kg⁻¹) and high dry matter yield (3,251 kg ha⁻¹). All these characteristics resulted in a better SOM quality. Urochloa ruziziensis, with higher hemicellulose and lower lignin contents, and low lignin/N ratio, was associated with accumulation of TOC (19.95 and 18.33 g kg⁻¹ in 0- to 10-cm and 10- to 20-cm layers, respectively) and mineral-associated organic C (on average, 16.68 g kg⁻¹) in the soil. Cover plants with N:lignin ratio lower than 2.0 are fundamental for soil C sequestration. In conclusion, it is recommended the adoption of Urochloa ruziziensis and Canavalia brasiliensis as cover plants improve maize production, soil organic matter quality and C sequestration in the Cerrado region.     

Transferability and scalability of soil total carbon prediction models in Florida, USA

The applicability, transferability, and scalability of visible/near-infrared (VNIR)-derived soil total carbon (TC) models are still poorly understood. The objectives of this study were to: i) compare models of three multivariate statistical methods, partial least squares regression (PLSR), support vector machine (SVM), and random forest methods, to predict soil logarithm-transformed TC (logTC) using five fields (local scale) and a pooled (regional-scale) VNIR spectral dataset (a total of 560 TC spectral datasets), ii) assess the model transferability among fields, and iii) evaluate their up- and downscaling behaviors in Florida, USA. The transferability and up- and downscaling of the models were limited by the following factors: i) the spectral data domain, ii) soil attribute domain, iii) methods that describe the internal model structure of VNIR-TC relationships, and iv) environmental domain space of attributes that control soil carbon dynamics. All soil logTC models showed excellent performance based on all three methods with R² > 0.86, bias < 0.01%, root mean squared error (RMSE) = 0.09%, residual predication deviation (RPD) > 2.70%, and ratio of prediction error to interquartile range (RPIQ) > 4.54. The PLSR method performed substantially better than the SVM method to scale and transfer the TC models. This could be attributed to the tendency of SVM to overfit models, while the asset of the PLSR method was its robustness when the models were validated with independent datasets, transferred, and/or scaled. The upscaled soil TC models performed somewhat better in terms of model fit (R²), RPD, and RPIQ, whereas the downscaled models showed less bias and smaller RMSE based on PLSR. We found no universal trend indicating which of the four limiting factors mentioned above had the most impact that constrained the transferability and scalability of the models. Given that several factors can impinge on the empirically derived soil spectral prediction models, as demonstrated by this study, more focus on their applicability and scalability is needed.     

Evaluating a low‐cost portable NIR spectrometer for the prediction of soil organic and total carbon using different calibration models

This study aims to assess the performance of a low‐cost, micro‐electromechanical system‐based, near infrared spectrometer for soil organic carbon (OC) and total carbon (TC) estimation. TC was measured on 151 soil profiles up to the depth of 1 m in NSW, Australia, and from which a subset of 24 soil profiles were measured for OC. Two commercial spectrometers including the AgriSpec^TM (ASD) and NeoSpectra^TM (Neospectra) with spectral wavelength ranges of 350–2,500 and 1,300–2,500 nm, respectively, were used to scan the soil samples, according to the standard contact probe protocol. Savitzky–Golay smoothing filter and standard normal variate (SNV) transformation were performed on the spectral data for noise reduction and baseline correction. Three calibration models, including Cubist tree model, partial least squares regression (PLSR) and support vector machine (SVM), were assessed for the prediction of soil OC and TC using spectral data. A 10‐fold cross‐validation analysis was performed for evaluation of the models and devices accuracies. Results showed that Cubist model predicts OC and TC more accurately than PLSR and SVM. For OC prediction, Cubist showed R² = 0.89 (RMSE = 0.12%) and R² = 0.78 (RMSE = 0.16%) using ASD and NeoSpectra, respectively. For TC prediction, Cubist produced R² = 0.75 (RMSE = 0.45%) and R² = 0.70 (RMSE = 0.50%) using ASD and NeoSpectra, respectively. ASD performed better than NeoSpectra. However, the low‐cost NeoSpectra predictions were comparable to the ASD. These finding can be helpful for more efficient future spectroscopic prediction of soil OC and TC with less costly devices.     

Transfer of carbon incubation parameters to model the SOC and SON dynamics of a field trial with energy crops applying digestates as organic fertilizers

The fertilization with organic amendments and digestates from biogas plants is increasingly used to increase carbon stock and to improve the soil quality, but little is still known about their long-term effects. A common method to analyse organic amendments and their mineralization is incubation experiments, where amendments get incubated with soil while CO₂ release is measured over time. In a previous study, carbon models have been applied to model the carbon dynamics of incubation experiments. The derived parameters describing the carbon turnover of the CCB model (CANDY Carbon Balance) are used to simulate the SOC and SON dynamics of a long-term field trial. The trial was conducted in Berge (Germany) where organic amendments like slurry, farmyard manure or digestates were systematically applied. To grant a higher model flexibility, the amounts of crop residues were calculated for roots and stubble separately. Furthermore, the mineralization dynamics of roots and stubble are considered by the model parameters for each crop. The model performance is compared when using the dry matter and carbon content received from the field trial and the incubation experiments, to evaluate the transferability. The results show that the incubation parameters are transferable to the field site, with rRMSE < 10% for the modelled SOC and rRMSE between 10% and 15% for the SON dynamics. This approach can help to analyse long-term effects of unexplored and unusual organic fertilizers under field conditions, whereat the model is used to upscale the C dynamics from incubation experiments, considering environmental conditions.     

A comparison of two methods for the isolation of free and occluded particulate organic matter

Various methods exist for the isolation of particulate organic matter (POM), one of the soil‐organic‐matter (SOM) fractions reacting most sensitive on land‐use or soil‐management changes. A combination of density separation and ultrasonic treatment allows to isolate two types of POM: (1) free POM and (2) POM occluded in soil aggregates. POM fractions are closely linked to their biochemical function for the formation and stabilization of aggregates, therefore methods using different aggregate sizes may result in different POM fractions isolated. We evaluated two physical fractionation procedures to reveal whether they yield different POM fractions with respect to amount and composition, using grassland and arable soils with sandy‐loam to sandy–clay‐loam texture and thus low macroaggregate stability. Method I used air‐dried aggregates of <2.0 mm size and a low‐energy sonication for aggregate disruption, method II used field‐moist aggregates <6.3 mm and a high‐energy–sonication procedure for aggregate disruption. POM fractions were analyzed by elemental analysis (C, N) and CPMAS ¹³C‐NMR spectroscopy. With both methods, about similar proportions of the SOM are isolated as free or occluded POM, respectively. The free‐ and occluded‐POM fractions obtained with method I are also rather similar in C and N concentration and composition as shown by ¹³C‐NMR spectroscopy. Method II isolates a free‐ and occluded‐POM fraction with significantly different C and N concentrations. NMR spectra revealed significant differences in the chemical composition of both fractions from method II, with the occluded POM having lower amounts of O‐alkyl C and higher amounts of aryl C and alkyl C than the free POM. Due to the use of larger, field‐moist aggregates with minimized sample pretreatment, two distinctly different POM fractions are isolated with method II, likely to be more closely linked to their biochemical function for the formation and stabilization of aggregates. High‐energy sonication as in method II also disrupts small microaggregates <63 µm and releases fine intraaggregate POM. This fraction seems to be a significant component of occluded POM, that allows a differentiation between free and occluded POM in sandy soils with significant microaggregation. It can be concluded, that microaggregation in arable soils with sandy texture is responsible for the storage of a more degraded occluded POM, that conversely supports the stabilization of fine microaggregates.