机器学习用于耕地土壤有机碳空间预测对比研究——以亚热带复杂地貌区为例

耕地土壤有机碳(SOC)是土壤质量的重要指标,也是生态系统健康的重要表征。当前机器学习(Machine Learning, ML)用于SOC数字制图日益热门,但不同算法在高空间分辨率SOC数字制图中的对比研究尚有欠缺。本研究以福建省东北部复杂地形地貌区为例,采用10m空间分辨率Sentinel-2影像数据,选取地形、气候、遥感植被变量为驱动因子,重点分析当前常用的机器学习算法——支持向量机(SupportVector Machine,SVM)、随机森林(RandomForest,RF)在SOC预测中的差异,并与传统普通克里格模型(Ordinary Kriging, OK)进行比较。结果表明:基于地形、遥感植被因子和气候因子构建的RF模型表现最佳(RMSE=2.004,r=0.897),其精度优于OK模型(RMSE=4.571, r=0.623),而SVM模型预测精度相对最低(RMSE=5.190, r=0.431);3种模型预测SOC空间分布趋势总体相似,表现为西高东低、北高南低,其中RF模型呈现的空间分异信息更加精细;最优模型反演得到耕地土壤有机碳平均含量为15.33 g·kg-1; RF模型和SVM模型变量重要性程度表明:高程和降水是影响复杂地貌区SOC空间分布的重要变量,而遥感植被因子重要性程度低于高程。     

基于多变量与RF算法的耕地土壤有机碳空间预测研究——以福建亚热带复杂地貌区为例

耕地土壤有机碳(Soil Organic Carbon,SOC)含量不仅是土壤质量的重要表征,还是农业温室气体的重要源库,而基于环境变量建立的随机森林算法(Random Forest,RF)是当前提高土壤有机碳空间预测精度的方法,但不同组合环境变量对RF模型预测精度的影响仍需深入研究。本文以福建闽东南复杂地貌区为例,以两种环境变量组合(遥感变量+气候因子和遥感变量+气候因子+土壤属性)为输入数据集,利用RF算法对耕地表层SOC含量进行模拟预测和精度对比,并与普通克里格(OrdinaryKriging,OK)插值模型进行比较。结果表明,基于全部环境变量构建的RF模型表现最佳,其模型拟合度和预测精度相较于未加入土壤属性的模型有显著提高(r提高7.95%,为0.95,RMSE下降45.13%),且对SOC空间分异信息的捕获更精确,OK模型总体预测精度最弱。利用最优模型反演得到的研究区耕地SOC含量为14.70±2.95 g·kg~(–1),东部沿海低于西部内陆。变量贡献率分析显示,除了与土壤碳紧密相关的水解性氮(N),遥感变量中数字高程模型(Digital Elevation Modecs,DEM)也是影响闽东南地区SOC预测精度的重要变量,因此,遥感变量、气候因子和土壤属性共同驱动的随机森林模型可作为闽东南复杂地貌区耕地有机碳含量空间预测的有效方法。     

冀西山前冲积扇区土壤机械组成模型制图对比研究

探索适合地形平缓的山前冲积扇地区土壤机械组成的空间预测方法。以河北省灵寿、行唐、曲阳县400 m高程以下区域为研究区,结合地形因子、土壤类型、归一化植被指数、地表温度等环境变量,选择基于对称对数比(SLR)转换的普通克里格法(SLR-OK)、回归克里格法(SLR-RK)、随机森林法(SLR-RF)3种方法,对训练集114个样点表层土壤机械组成的空间分布进行预测,并通过验证集50个样点比较了3种方法的预测精度。(1)从空间预测图来看,砂粒呈现出西北低、东南高的空间分布趋势;粉粒和黏粒与砂粒相反。与SLR-OK法相比,SLR-RK法和SLR-RF法能够更好地反映局部变异并减小平滑效应。(2)对于砂粒和粉粒,SLR-RF法对验证集含量预测的平均绝对误差(MAE)和均方根误差(RMSE)均低于其他两种方法,且决定系数最高,表明SLR-RF的预测精度最高;对于黏粒,SLR-OK法对验证集含量预测的MAE和RMSE均低于其他两种方法,且决定系数最高,表明SLR-OK法的预测精度最高。(3)线性回归预测模型的辅助变量包括高程、土壤类型和风力作用指数;随机森林法模型的辅助变量包括高程、土壤类型、归一化植被指数、地表温度、风力作用指数和流量累积,对于砂粒和粉粒,土壤类型和高程是重要的辅助变量,归一化植被指数、地表温度、风力作用指数和流量累积重要性相对较低。研究区砂粒和粉粒空间预测的最优方法为SLR-RF法,在山前冲积扇地区地形较平缓,砂粒和粉粒对地形变量敏感;黏粒空间预测的最优方法为SLR-OK法。     

基于环境变量的中国土壤有机碳空间分布特征

研究中国土壤有机碳(Soil Organic Carbon,SOC)的空间分布特征对SOC储量估算以及农业生产管理具有重要意义。以全国第二次土壤普查2473个土壤典型剖面的表层(A层)SOC含量为研究对象,探寻地形、气候和植被等环境因素对SOC空间异质性分布的影响;以普通克里格法为对照,利用地理加权回归、地理加权回归克里格、多元线性回归和回归克里格模型建立SOC空间预测模型;并分别绘制了中国SOC的空间分布预测图。结果表明:(1)SOC含量与年均降水量、年均温、归一化植被指数、高程以及地形粗糙指数呈极显著相关关系;(2)平均绝对估计误差、均方根误差、平均相对误差和皮尔逊相关系数等模型验证指标表明地理加权回归的预测精度优于其他模型,可以更好地绘制SOC在大尺度上的空间分布特征;(3)较高SOC含量主要分布在研究区东北部、西南部以及东南部,而西北部SOC含量普遍偏低。本文以期从大尺度上探讨土壤属性与环境变量之间的相关关系,为全国土壤属性的空间制图提供一定的解决方案和思路。     

基于环境相关法和地统计学的土壤属性空间分布预测

土壤属性是土壤质量的重要决定因素,并强烈影响土地利用和生态过程。正确理解并充分考虑土壤空间变异,对于在景观尺度上建立生态、环境过程模型是必不可少的。在黄土高原横山县采集了254个样点,应用数字地形与遥感影像分析技术,获取相关地形因子与遥感指数,分析土壤属性（土壤容重、有机质和全磷）与环境因子相互关系,并利用环境变量进行空间预测。结果表明,土壤容重、有机质与地形因子和遥感指数之间存在较好相关性,而全磷与地形因子相关性不大;多元线性逐步回归模型对于土壤容重和有机质拟合较好,而对于全磷,预测结果较差;回归－克里格预测有效地减小了残差,消除了平滑效应,与实测值较为接近。     

黄土高原县域土壤养分空间变异特征及预测——以陕西省横山县为例

研究土壤属性空间变异及其分布特征与环境因子的关系,对于了解生态系统的过程具有重要意义。以横山县为例,采集了254个耕层(0~20 cm)土样,利用数字地形与遥感影像分析技术,提取相关地形与遥感指数,分析不同土地利用、地形条件下土壤养分空间变异及分布特征,并结合回归分析与地统计学进行空间分布预测。结果表明,不同土地利用类型其养分含量差异显著,水田和川地的有机质和全氮含量明显高于其他土地利用类型,而全磷含量以梯田最高。不同坡度分析表明,"0~3°"坡度等级有机质和全氮含量显著高于其他坡度等级;不同坡向土壤养分含量差异均不显著,但存在一个明显的趋势,即阴坡有机质和全氮含量整体上要较阳坡高。土壤有机质与高程H呈现负相关,与坡向的余弦值cosα、复合地形指数CTI、修正后的土壤调节植被指数MSAVI及湿度指数WI呈正相关。土壤全氮与相关环境因子的关系基本与有机质的一致。而土壤全磷与修正后的土壤调节植被指数MSAVI及湿度指数WI正相关。土壤有机质和全氮用相关环境变量的多元线性逐步回归模型拟合预测较好,而对于全磷,预测结果较差;回归—克里格预测有效地减小了残差,消除了平滑效应,与实测值较为接近,预测精度高于多元线性逐步回归预测。     

基于地形与遥感辅助信息的小流域尺度高分辨率有机碳空间分布预测研究

土壤有机碳(soil organic carbon,SOC)既是衡量土壤质量的重要指标，也是影响全球碳氮循环的关键因素之一。作为数字土壤制图(digital soil mapping,DSM)研究中起主要作用的环境变量，地形元素在SOC预测制图中也是无可替代的。应用机器学习模型，通过引入不同超参数设置下获得的高分辨率(5 m)Geomorphons(GM)地形分类图作为丘陵地形特征信息的补充，结合数字高程模型(digital elevation model,DEM)衍生变量和光学、合成孔径雷达(synthetic aperture radar,SAR)遥感数据对句容市黄梅镇北部小流域尺度(1︰25 000)丘陵地貌区地表层SOC含量进行预测制图，并评估不同GM变量在SOC含量预测中的表现。基于74个土壤样本和不同环境变量组合，分别采用袋装决策回归树(bagged classification and regression tree,bagged CART)、随机森林(random forest,RF)和立体派(cubist)三种方法构建SOC含量预测模型，并通过四个精度验证指标，采用十...     

利用地形和遥感数据预测土壤养分空间分布

在GIS支持下,选择地形因子和遥感植被指数,建立土壤养分空间分布预测模型,应用回归克里格（Kriging）方法,预测吉林省农安县土壤养分（有机质和全氮）的空间分布。结果表明,11个环境因子中,相对高程、坡度、地形起伏度、坡度变率、归一化植被指数（NDVI）与土壤有机质和全氮含量均具有显著的相关性。地面粗糙度和地形湿度指数与有机质具有显著相关性,而与全氮的相关性不显著。相对高程、坡度、地面粗糙度、河流动能指数以及NDVI在土壤养分的多元回归预测模型中贡献较大,是预测土壤养分空间分布的最优因子。有机质和全氮在研究区的空间分布格局呈现由东南向西北逐渐减少的趋势,这种分布格局受地形和植被的综合作用,同时与土壤类型密不可分。精度检验结果表明,回归克里格方法能够提高土壤养分空间分布预测精度,是一种有效的空间分布插值方法。     

广义线性地统计模型在典型亚热带丘陵区数字土壤制图中的应用

数字土壤制图(DSM)在当前的应用越来越广泛,其方法主要包括环境相关模型、空间自相关模型,以及这些模型的混合模型。理论上,混合模型相对单一模型具有明显的优势。广义线性地统计模型(GLGM)也是一种混合模型,相对于最常用的混合模型——回归克里格(RK),又具有能加入随机效应来解决土壤变异的非平稳性等优势。然而,GLGM因计算繁琐等缺点,在国内外应用较少。本文以广西南宁高峰林场内一小面积(3.03km2)丘陵为研究区,以14个地形因子为预测变量,使用GLMM及其与普通克里格(OK)相结合的模型(GLGM),对土壤有机碳(SOC)、pH、黏粒和阳离子交换量(CEC)的空间分布进行预测,并与常用的多元线性回归(MLR)、地理加权回归(GWR)、回归森林(RF)、普通克里格(OK)、RK和广义可加模型(GAM)进行比较。结果表明：GLMM在预测黏粒上准确度较高;GLMM和GLGM在预测CEC上准确度中等;GLMM和GLGM在预测SOC和pH上准确度较低。综合线性回归模型的调整决定系数、块金效应和全局莫兰指数,本文认为,当土壤属性与环境变量具有较低的线性回归调整决定系数(即小于5%)、土壤属性具有较弱的空间自相关性(即块金效应大于71%)和较强的局部空间变异(即全局莫兰指数小于0.09)时,GLMM和GLGM具有较高的适用性,例如本文中的黏粒。反之,GLMM和GLGM的适用性不好,例如本文中的SOC和pH。鉴于土壤空间变异的高度异质性和多尺度性,本文认为GLMM和GLGM具有较好的应用前景。但是,在未来的研究中还需要进一步研究如何提高GLMM和GLGM的模拟效率。     

应用土壤-景观定量模型预测土壤属性空间分布及制图

以土壤-景观定量模型为基础的土壤制图方法在世界范围内得到了广泛研究。本文在皖南宣城的丘陵地带内选择研究区,从该区的数字高程模型(DEM)中获取景观信息:地形因子,定量地分析了土壤属性与地形因子之间的相关关系,并建立基于该关系的线性土壤-景观定量模型,最后应用该模型来预测土壤属性在空间上的分布并制图。结果表明:土体厚度和表层有机质含量与地形因子之间有着显著相关性;建立的线性回归模型分别能解释土体厚度、表层有机质含量空间变异的32.2%和35.3%;依据该模型预测的土体厚度和表层有机质含量具有较高的准确度,并能制图表达土壤属性在空间上的自然连续性。     

Potential of wild species for genetic enhancement of some semi-arid food crops

Discovery and incorporation of genes from wild species provide means to sustain crop improvement, particularly when levels of resistance in the cultigens are low and virulent strains of pests and pathogens overcome the host plant resistance. The extent of utilization and the potential of the wild genepool for genetic enhancement were reviewed in five important food crops viz. sorghum, pearl millet, chickpea, pigeonpea and groundnut grown in the semi-arid tropics. Introgression from compatible wild germplasm in the primary gene pool resulted in transfer of new cytoplasmic male sterility systems in pearl millet and pigeonpea, development of high protein, cleistogamous flower and dwarf pigeonpea lines and foliar disease resistant groundnut cultivars. Utilization of wild species in secondary and tertiary gene pools has been generally limited due to sterility, restricted recombination or cross incompatibility. Nevertheless, these species are extremely important as they contain high levels of resistance to several important biotic and abiotic stresses. Several of them, like those belonging to the Parasorghum section and the rhizomatous Arachis species are sources of multiple resistances and hold great promise to sustain crop productivity.     

“绿色革命”以来大气CO2浓度升高对C3作物营养品质的影响

自20世纪60年代"绿色革命"以来,育种技术和农耕技术的发展促进了农作物产量的大幅提升,然而作物的营养品质出现下降趋势。在相似的遗传背景下,大气CO_2浓度升高会使单位体积农作物产品的营养元素含量下降,因此"绿色革命"至今,农作物产品的营养元素下降可能受大气CO_2浓度升高影响。通过植物生长箱模拟"绿色革命"初期和目前的大气CO_2浓度水平(310μmol/mol和400μmol/mol),针对主要C_3作物水稻、小麦和大豆,研究"绿色革命"以来大气CO_2浓度升高对其籽粒的C、N、Fe、Zn元素含量的影响,结果表明:CO_2浓度升高对3种作物籽粒的C元素含量几乎没有影响,变化幅度在±1.5%之间;籽粒的N、Fe、Zn元素含量普遍呈现下降趋势,但均未达到显著水平。     

Faculty positions:Center for Agricultural Resources Research,Chinese Academy of Sciences

正The Center for Agricultural Resources Research(CARR),the Institute of Genetics and Developmental Biology(IGDB),Chinese Academy of Sciences,invites applicants for several research group leader positions.CARR is one of the research organizations in Chinese Academy of Sciences(CAS).We seek nominations and applications     

Collection and preliminary evaluation of native turfgrass accessions in Italy

Despite a raising interest on turfs in Italy, all theavailable varieties of this kind in the Country are of foreign origin, and areoften poorly adapted to the prevailing climatic conditions. This prompted tobegin a collection activity of indigenous turfgrass species, with the ultimategoal of identifying promising materials for future breeding based on localgenetic resources. The collection was carried out in three areas of Italy, viz.the northern Po Plain, the coastal region of Liguria, and the island of Sardiniathat are characterised, respectively, by subcontinental, warm temperate, andtypical Mediterranean climate. Altogether, 141 sites were visited, yielding 226accessions belonging to eight species of potential interest for turfs:Poa pratensis, Poa trivialis,Festuca rubra, Festuca arundinacea,Lolium perenne, Agrostis stolonifera,Agrostis tenuis, and Cynodon dactylon,this last being a warm-season grass. Poa pratensis andCynodon dactylon were mostly collected in northern Italyand Sardinia, respectively, whereas Festuca arundinacea andLolium perenne were rather ubiquitous. The collection sitesranged from 0 to 1040 m asl, but sites over 750 m wereonly visited in the inner part of Sardinia. All the accessions, collected aswhole plants, were transplanted at Lodi, northern Italy, where they are beingevaluated. Their preliminary evaluation for traits of importance for turf use,such as sward colour and overall quality, highlighted the great variation andthe occurrence of interesting accessions in all species. Other characters wererecorded, bearing specific importance in individual species, and in all casespromising accessions were identified. The germplasm of Festucarubra, Festuca arundinacea, and Loliumperenne proved highly infected by endophytic (symbiotic) fungi of thegenus Neotyphodium.     

Leaf position and genotype differences for mineral element concentrations in sorghum grown on tropical acid soil 1

Mineral element deficiencies and toxicities are common problems associated with sorghum [Sorghum bicolor (L.) Moench] production on acid soils. To better understand some of the mineral element problems and the analysis of plant tissue of sorghum plants grown on acid soils, four sorghum genotypes were grown on an acid Oxisol at Carimagua, Colombia limed with dolomite at 2 and 6 Mg ha^‐1.

Samples for mineral element analyses were obtained from leaves at different positions on the four genotypes. Concentrations of P and Mg were highest in the flag leaf (Leaf No. 1) and decreased as the position on the plant declined from the top of the plant for plants grown at 2 Mg lime ha^‐1. Similar decreases in P, Mg, K, and Zn concentrations occurred in plants grown with 6 Mg lime ha^‐1. Concentrations of Ca, S, Si, Mn, Fe, Cu, and Al increased as leaf position declined from the flag leaf for plants grown at 2 and 6 Mg lime ha^‐1. The higher lime supply enhanced Ca and reduced Mn and Fe concentrations in leaves. Differences in mineral element concentrations for the four genotypes used were fairly extensive. The elements to show the greatest range among genotypes were Al and Si and the elements to show the least range among genotypes were P, K, and S. Care should be used in collecting leaf samples for plant analysis and genotypic differences for accumulation of mineral elements should be considered in interpretation of results.     

Short-term variation in the concentration of selected ions within individual tropical rainstorms

Rainwater was collected at the campus of the University of Brunei Darussalam in Bandar Seri Begawan, Brunei Darussalam, using a funnel-in-bottle sampler. Polypropylene bottles were changed at intervals during rainstorm events. The pH and conductivity were determined immediately after collection on aliquots of the sample. Samples were refrigerated at 5°C for subsequent chemical analysis. Analyses for Na, Mg, Ca, Zn and Fe were carried out by means of inductively coupled plasma atomic emission spectroscopy (ICP-AES); Cu and Mn were analysed by graphite furnace atomic absorption spectroscopy (GFAAS); K was analysed using flame atomic emission spectroscopy (FAES); and Cl^–, NO₃ ^– and SO₄ ^2– were analysed by ion chromatography (IC). Concentration versus time profiles are reported for three rainstorm events. All ions exhibited a decrease in concentration during the rainstorm. The first sample contained the highest concentration of ions, consistent with a first-flush effect. The contribution of the initial stages of the shower to the total quantity of ion deposited during the entire rainstorm is quite overwhelming; in many cases 20 to 30% of the mass was deposited in less than 5% of rainstorm duration. On the other hand, the pH and conductivity variation during rainstorms did not exhibit a consistent pattern.     

Leaf litter selection by detritivore and geophagous earthworms

Summary Litterbag experiments with 10 different kinds of leaf litter showed that detritivore (Lumbricus species) and geophagous (Aporrectodea species) earthworms prefer certain litter types over others, since different numbers of worms were found below the litter after 50–52 days of exposure in a pasture. The detritivores preferred Fraxinus, Tilia, and predecomposed Ulmus and Fagus litter to Fagus litter and paper, while geophages preferred Tilia litter to Alnus and Ulmus litter, so that the two groups of earthworms showed different preferences. The detritivores seemed to be more selective than the geophages. The palatability of the litter was examined in relation to the C: N ratio, the lignin concentration and the initial and final polyphenol concentration. The numbers of detritivores were significantly correlated with the C: N ratio and the final polyphenol concentration, so that selection of litter seems to be related to palatability. The numbers of geophages were not significantly correlated with any of the parameters for palatability. The disappearance of litter after 50–52 days appeared to be due to detritivore activity, since the numbers found below the litter were positively and significantly correlated with the litter disappearance. There was no significant correlation with geophage activity. This indicates that detritivores use litter as food, and therefore influence the composition of the litter layer.     

The 5S rRNA gene diversity in the genus <Emphasis Type="Italic">Secale</Emphasis> and determination of its closest haplomes

We analyzed 127 rDNA sequences (5S DNA units) obtained from 23 seed accession samples from more or less 10 taxa in wild and cultivated rye, genus Secale L. The sequences fell into two known groups, here assigned to two unit classes, viz. long R1 and short R1 (designations to reflect on R haplome of rye). The different taxa could not be fully differentiated based on the 5S DNA units. We searched for 5S DNA sequences from known unit classes most closely similar to the long R1 and the short R1. One set with the long R1 unit class contained sequences of the long P1 unit class from Agropyron (P haplome) and from Kengyilia (StYP haplome), long J1 from Thinopyrum (J haplome), whereas the set with the short R1 included the long S1 from Pseudoroegneria (St haplome) and Kengyilia (StYP haplome), the short J1 from Thinopyrum (J haplome) and the short V1 from Dasypyrum (V haplome). Each of the two sets was analyzed separately by maximum likelihood (ML) phylogenetic analysis from which we were able to infer that the 5S DNA units of Secale differentiated in a non-clock fashion and followed the HKY substitution model in the gene tree with the long R1 unit class and the HKY + G in the gene tree with the short R1 unit class. A complementary Bayesian analysis yielded identical tree topologies to the ML ones for each of the two sequence sets. In the tree with the long R1 units the long P1 and long J1 unit classes were closest to the long R1 unit class, whereas in the tree with the short R1 units the long S1 and short J1 unit classes were closest to the short R1 unit class, indicating possibly a close relationship between the St, J and R haplomes.     

Nematode succession and microfauna-microorganism interactions during root residue decomposition

The quality of plant material affects the vigor of the decomposition process and composition of the decomposer biota. Root residues from hairy vetch (Vicia villosa Roth), rye (Secale cereale L.) and vetch+rye, packed in litterbags were placed in pots of soil at 15 C and the content of the bags was analyzed after 2, 4, 8 and 12 weeks. Bacterial biomass did not differ between residues with contrasting composition. Among bacterivores groups of nematodes that require high bacterial production dominated in fast decomposing resources whereas flagellates with smaller requirements prevail in slower decomposing resources. Biomass of bacterial feeding nematodes correlated positively with early phase (0-2 wk) decomposition that increased in the order: rye< vetch+rye<vetch. Bacterial biomass therefore seems to be under top-down (predation) control during early decomposition. In contrast, the fungal biomass differed between resources with highest values for rye. Moreover, this increase in fungal biomass occurred later during succession and was correlated with decomposition activity for rye in that period. Fungal biomass therefore seems to be under bottom-up (resource) control. The composition of the nematode assemblages (composed of 25 taxa) showed a clear relationship to initial plant resource quality as well as decomposition phase. Early successional microbivorous nematodes vary according to resource quality with demanding bacterivores+predators (Neodiplogasteridae) dominating in vetch and less demanding bacterivores (Rhabditidae) and fungivores (Aphelenchus) being equally common in vetch and rye. Later in the succession (2-4 wk) bacterivorous Cephalobidae and fungivorous Aphelenchoides prevailed similarly on the different root materials whereas bacterivorous protozoa and the amoebal fraction thereof dominated in rye. At week 12 no species dominated the nematode assemblages that were similar between the resources. The differences between nematode assemblages among plant resources at 2 week were similar to the results of a field study sampled after 6 weeks with the same soil and plant resources. This lends support to the relevance of the successional patterns observed in this incubation study.     

Cycling of extracellular DNA in the soil environment

Upon entering the soil environment, extracellular DNA is subjected to dynamic biological, physical, and chemical factors that determine its fate. This review concerns the fate of both recombinant and non-recombinant sources of DNA. A schematic of DNA cycling coupled with genetic transformation is presented to understand its behavior in soil. Extracellular DNA may persist through cation bridging onto soil minerals and humic substances, be enzymatically degraded and restricted by DNases of microbial origin, and/or enter the microbial DNA cycle through natural transformation of competent bacteria. Lateral gene transfer may disseminate DNA through the microbial community. An understanding of DNA cycling is fundamental to elucidating the fate of extracellular DNA in the soil environment.