中国禹城土壤盐渍化的时空变异及其预测

This research used both geostatistics and GIS approach to compare temporal change of soil salt between 1980 and 2003, to analyze the spatial distribution of surface soil salt, to developed methods for predicting soil salinization potential based on recent improvements to the Dempster-Shafer theory, and to develop probability maps of potential salinization in Yucheng City, China. A semivariogram model of soil salt content was developed from the spherical model, and then employing kriging interpolation the spatial distribution of salt content in 2003 was obtained utilizing data from 100 soil sampling points. Potential salinization distribution was mapped using an approach that integrated soil data of the second general survey in 1980 in Yucheng City, which included groundwater salinity, groundwater depth, soil texture, soil organic matter content, and geomorphic maps. With the support of Dempster-Shafer theory and fuzzy set technique the factors that affected potential soil salinization were characterized and integrated;and then soil salinization was predicted. Finally a prognosis map of potential salinization distribution in the research area was obtained, with higher probability values indicating higher hazards to salinity processes. The distribution of the potential soil salinization probability was a successive surface.     

自动土壤图基于知识的分类

A machine-learning approach was developed for automated building of knowledge bases for soil resources mapping by using a classification tree to generate knowledge from training data. With this method, building a knowledge base for automated soil mapping was easier than using the conventional knowledge acquisition approach. The knowledge base built by classification tree was used by the knowledge classifier to perform the soil type classification of Longyou County, Zhejiang Province, China using Landsat TM bi-temporal images and GIS data. To evaluate the performance of the resultant knowledge bases, the classification results were compared to existing soil map based on a field survey. The accuracy assessment mad maalysis of the resultant soil maps suggested that the knowledge bases built by the machine-learning method was of good quality for mapping distribution model of soll classes over the study area.     

遥感和GPS支持的高山草甸土草地盖度精确量化研究

The principles of remotely estimating grassland cover density in an alpine meadow soil from space lie in the synchronous collection of in situ samples with the satellite pass and statistically linking these cover densities to their image properties according to their geographic coordinates. The principles and procedures for quantifying grassland cover density from satellite image data were presented with an example from Qinghai Lake, China demonstrating how quantification could be made more accurate through the integrated use of remote sensing and global positioning systems （GPS）. An empirical model was applied to an entire satellite image to convert pixel values into ground cover density. Satellite data based on 68 field samples was used to produce a map of ten cover densities. After calibration a strong linear regression relationship （r^2 = 0.745） between pixel values on the satellite image and in situ measured grassland cover density was established with an 89% accuracy level. However, to minimize positional uncertainty of field samples, integrated use of hyperspatial satellite data and GPS could be utilized. This integration could reduce disparity in ground and space sampling intervals, and improve future quantification accuracy even more.     

结合土地利用的城镇化流域土壤养分非函数空间插值

Integrating land use type and other geographic information within spatial interpolation has been proposed as a solution to improve the performance and accuracy of soil nutrient mapping at the regional scale.This study developed a non-algorithm approach,i.e.,applying inverse distance weighting (IDW) and ordinary kriging (OK),to individual land use types rather than to the whole watershed,to determine if this improved the performance in mapping soil total C (TC),total N (TN),and total P (TP) in a 200-km2 urbanizing watershed in Southeast China.Four land use types were identified by visual interpretation as forest land,agricultural land,green land,and urban land.One hundred and fifty soil samples (0 10 cm) were taken according to land use type and patch size.Results showed that the non-algorithm approach,interpolation based on individual land use types,substantially improved the performance of IDW and OK for mapping TC,TN,and TP in the watershed.Root mean square errors were reduced by 3.9％ for TC,10.7％ for TN,and 25.9％ for TP by the application of IDW,while the improvements by OK were slightly lower as 0.9％ for TC,7.7％ for TN,and 18.1％ for TP.Interpolations based on individual land use types visually improved depiction of spatial patterns for TC,TN,and TP in the watershed relative to interpolations by the whole watershed.Substantial improvements might be expected with denser sampling points.We suggest that this non-algorithm approach might provide an alternative to algorithm-based approaches to depict watershed-scale nutrient patterns.     

匈牙利半湿润气候地区田间尺度下土壤侵蚀引起的土壤有机碳再分布

Soil organic carbon(SOC) has primary importance in terms of soil physics, soil fertility and even of climate change control. One hundred soil samples were taken from an intensively cultivated Cambisol to quantify SOC redistribution triggered by soil erosion under a subhumid climate, by the simultaneous application of diffuse reflectance(240–1 900 nm) and traditional physico-chemical methods.The representative sample points were collected from the solum along the slopes at the depth of 20–300 cm with a mean SOC content of 12 g kg~(-1). Hierarchical cluster analyses were performed based on the determined SOC results. The spatial pattern of the groups created were similar, and even though the classifications were not the same, diffuse reflectance had proven to be a suitable method for soil/sediment classification even within a given arable field. Both organic and inorganic carbon distributions were found to be a proper tool for estimations of past soil erosion processes. The SOC enrichment was found on two sedimentary spots with different geomorphological positions. Soil organic matter composition also differed between the two spots due to selective deposition of the delivered organic matter. The components with low-molecular-weight reached the bottom of the slope where they could leach into the profile, while the more polymerised organic matter compositions were delivered and deposited even before on a higher segment of the slope in an aggregated form. This spatial difference appeared below the uppermost tilled soil layer as well, referring the lower efficiency of conventional ploughing tillage in soil spatial homogenisation.     

依据传统土壤数据应用数字土壤制图法对香港山地地区土壤建立土纲信息

Based on legacy soil data from a soil survey conducted recently in the traditional manner in Hong Kong of China, a digital soil mapping method was applied to produce soil order information for mountain areas of Hong Kong. Two modeling methods (decision tree analysis and linear discriminant analysis) were used, and their applications were compared. Much more eflort was put on selecting soil covariates for modeling. First, analysis of variance (ANOVA) was used to test the variance of terrain attributes between soil orders. Then, a stepwise procedure was used to select soil covariates for linear discriminant analysis, and a backward removing procedure was developed to select soil covariates for tree modeling. At the same time, ANOVA results, as well as our knowledge and experience on soil mapping, were also taken into account for selecting soil covariates for tree modeling. Two linear discriminant models and four tree models were established finally, and their prediction performances were validated using a multiple jackknifing approach. Results showed that the discriminant model built on ANOVA results performed best, followed by the discriminant model built by stepwise, the tree model built by the backward removing procedure, the tree model built according to knowledge and experience on soil mapping, and the tree model built automatically. The results highlighted the importance of selecting soil covariates in modeling for soil mapping, and suggested the usefulness of methods used in this study for selecting soil covariates. The best discriminant model was finally selected to map soil orders for this area, and validation results showed that thus produced soil order map had a high accuracy.     

中国新疆绿洲地区多时空尺度上土壤盐分变化特征及其相关驱动因素研究

Located in the inland arid area of central Asia, salt-affected farmlands take up one third of the total irrigated land area in Xinjiang of Northwest China. Spatio-temporal variability of soil salinity and the underlying mechanism are fundamental problems challenging the sustainability of oasis agriculture in China. In this study, the data of total dissolved solids(TDS) measured for soil samples collected from 27 representative study areas in the oasis areas of Xinjiang were analyzed and the coefficient of variation(CV) and stratification ratio(SR) of TDS were used to describe the lateral and vertical soil salinity variations, respectively. Weekly, monthly,and annual changes in soil salinity were also summarized. Results showed that the top(0–20 cm) soil salinity was highly variable(CV 75%) for most studied areas. Lateral variation of soil salinity was significantly correlated with the sampling interval; as a result, a maximum sampling interval of 0.9 m was found for reducing evaluation uncertainty. The top 0–20 cm soil salt accounted for about25.2% of the total salt in the 0–100 cm soil profile. The stratification ratio values(the ratio of TDS at the 20–40 cm depth to that at the 0–20 cm depth) were mostly smaller than 1 and on average 0.92, illustrating that the top 0–20 cm soil contained slightly more salt and a considerable amount of salt still existed in subsurface and deep horizons. Irrigation reduced top soil salinity by 0.52 g kg-1, or14.6%, within the first week. On average, the relative range of soil salinity, calculated to indicate monthly changes in soil salinity, was58.2% from May to September. A 27-year experiment indicated that cultivation increased soil salinity by 44.4% at a rate of 0.14 g kg-1year-1. At small spatio-temporal scales, soil salinity variation was mainly affected by anthropogenic factors, such as irrigation and land use. However, natural factors, including groundwater, topography, and climate conditions, mainly influenced soil salinity variation at large spatio-temporal scales. This study displayed the highly variable nature of soil salinity in space and time. Those driving factors identified in this study could provide guidelines for developing sustainable agriculture in the oasis areas and combating salinization in arid regions of China.     

田间尺度下测定土壤有效K、P的采样方法

Field nutrient distribution maps obtained from the sutdy on soil variations within fields are the basis of precision agriculture,The quality of these maps for maagement depends on the accuracy of the predicted values,which depends on the initial sampling.to Produce reliable predictions efficently the minimal sampling size and combination should be decided firstly,which could avoid the misspent funds for field sampling work.A 7.9 hectare silage field close to the Agricultural Research Institute at Hillaborough,Northern Irealnd,was selected for the study,Soil samples were collected from the field at 25m intervals in a rectangular grid to provide a database of selected soil propoerties.Different data combinations were subsequently abstracted from this database for comparison purposes,and ordinary krging used to produce interpolated soil maps.These prediced data groups were compared using least significant difference(LSD) test method.The results showed that the 62 sampling sizes of triangle arrangement for soil available K were sufficient to reach the required accuracy.The triangular sample combination proved to be superior to a rectangular one of similar sample size.     

土地混合使用制度下土壤硝态氮分布的地理空间制图研究

Mapping the spatial distribution of soil nitrate-nitrogen (NO₃-N) is important to guide nitrogen application as well as to assess environmental risk of NO₃-N leaching into the groundwater. We employed univariate and hybrid geostatistical methods to map the spatial distribution of soil NO₃-N across a landscape in northeast Florida. Soil samples were collected from four depth increments (0-30, 30-60, 60-120 and 120-180 cm) from 147 sampling locations identified using a stratified random and nested sampling design based on soil, land use and elevation strata. Soil NO₃-N distributions in the top two layers were spatially autocorrelated and mapped using lognormal kriging. Environmental correlation models for NO₃-N prediction were derived using linear and non-linear regression methods, and employed to develop NO₃-N trend maps. Land use and its related variables derived from satellite imagery were identified as important variables to predict NO₃-N using environmental correlation models. While lognormal kriging produced smoothly varying maps, trend maps derived from environmental correlation models generated spatially heterogeneous maps. Trend maps were combined with ordinary kriging predictions of trend model residuals to develop regression kriging prediction maps, which gave the best NO₃-N predictions. As land use and remotely sensed data are readily available and have much finer spatial resolution compared to field sampled soils, our findings suggested the effcacy of environmental correlation models based on land use and remotely sensed data for landscape scale mapping of soil NO₃-N. The methodologies implemented are transferable for mapping of soil NO₃-N in other landscapes.     

土壤制图中多等级代表性采样与分层随机采样的对比研究

采样设计是土壤地理研究中备受关注的重要问题。本文以区域尺度土壤属性制图为例,将多等级代表性采样与经典采样中的分层随机采样进行对比研究。以安徽宣城研究区的表层砂粒含量为目标要素,采集数量均为59个的两套样点,设计不同数量(46、58和59)的样点分组,采用两种制图方法进行制图并利用独立验证点进行评价。结果表明:1)无论是采用多元线性回归方法还是基于环境相似度的制图方法,在同等样点数量下,利用代表性样点所得土壤图精度均高于利用随机样点所得精度,并且利用少量代表性样点(46个)所得土壤图精度也高于利用多量随机样点(59个)所得精度;2)随着代表性较低样点的增加,土壤制图精度基本有一个提高的趋势,而采用随机样点所得土壤图的精度波动较大。因此,可认为多等级代表性采样方法是一种可用于区域尺度土壤调查的有效采样方法,且比分层随机采样高效、稳定。     

玉米秸秆覆盖与深翻两种还田方式对黑土有机碳固持的影响

秸秆还田是实现东北黑土肥力提升与保障区域生态环境安全的有效措施。明确玉米秸秆覆盖与深翻还田下土壤有机碳(SOC, Soil Organic Carbon)的变化及其在团聚体中的固持特征,对于揭示秸秆还田后黑土有机碳的稳定机制与固碳潜力具有重要意义。该研究基于黑土区中部6 a定位试验,选择常规种植(CK)、秸秆覆盖还田(SM, Stovers Mulching)和秸秆深翻还田(SI, Stovers Incorporation)3个处理,对0～10、10～20、20～30及30～40 cm土层SOC含量、容重、水稳性团聚体分布及团聚体中有机碳(OC, Organic Carbon)含量进行了分析与测定,并对各处理年均碳投入量、SOC储量与土壤固碳速率等进行了估算。与CK相比,SM处理显著增加了0～10 cm土层SOC含量,增幅为22.4%,但对10～40cm土层SOC含量无显著影响;SI处理显著增加了0～40cm土层SOC含量,增幅为18.1%～41.5%,以20～30cm的增幅最突出。与SM处理相比,SI处理0～10 cm土层SOC储量显著低于前者,而20～30 cm土层SOC储量反之。6 a间,SM处理耕层(0～20 cm)与亚耕层(20～40 cm)土壤固碳速率分别为1.34和0.77 Mg/(hm2·a),SI处理为0.85和1.74 Mg/(hm2·a)。秸秆不同还田方式显著改变了0～40 cm土层团聚体分布及其中OC含量。与CK相比,SM显著增加了耕层大团聚体(0.25 mm)比例与平均质量直径(MWD, Mean Weight Diameter),SI显著提高了0～40 cm土层团聚体MWD,且对10～40 cm土层团聚结构的改善作用优于SM;SM处理显著增加了0～10 cm土层2和0.053 mm粒级团聚体OC含量,SI处理不仅增加了0～10 cm土层2 mm粒级团聚体OC含量,也显著提高了10～40 cm土层各粒级团聚体OC含量。在黑土区,秸秆覆盖还田对SOC的提升主要集中于表层,秸秆深翻还田促进了0～40cm土层SOC积累与土壤团聚结构的改善。     

复杂景观环境下土壤厚度分布规则提取与制图

复杂景观环境下,土壤—环境关系知识的获取是预测性土壤制图的基础。为了探究复杂景观下土壤厚度分布与环境条件的关系,该文以黑河上游祁连山区典型小流域为研究区,应用模糊c均值聚类(fuzzy C-means cluster,FCM)和决策树(decision Tree,DT)方法,建立了一套获取土壤厚度分布与环境间关系知识的方法。利用2种方法结合获得流域内土壤厚度各分布等级的环境要素关键阈值与土壤-环境关系知识集,将所得环境阈值和知识集进行预测性制图,并通过野外独立样点对制图结果进行精度评价。结果表明:土壤厚度图的总体精度为74.2%,Kappa系数为0.659。该研究将2种方法结合获得了土壤厚度分布对应的土壤环境关键阈值和土壤-环境关系知识集,为复杂景观环境下土壤厚度的预测性制图提供了一种有效的解决方案。     

Estimation of soil organic carbon stock and its spatial distribution in the Republic of Ireland

Data scarcity often prevents the estimate of regional (or national) scale soil organic carbon (SOC) stock and its spatial distribution. This study attempts to overcome the data limitations by combining two existing Irish soil databases [SoilC and national soil database (NSD)] at the national scale, to create an improved estimate of the national SOC stock. Representative regression models between the near‐surface SOC concentration and those of deeper depths, and between SOC concentration and bulk density (BD) were developed based on the SoilC database. These regression models were then applied to the NSD derived SOC concentration map, resulting in an improved SOC stock and spatial distribution map for the top 10 cm, 30 cm and 50 cm depths. Western Ireland, particularly coastal areas, was found to have higher SOC densities than eastern Ireland, corresponding to the spatial distribution of peatland. We estimated the national SOC stock at 383 ± 38 Tg for the near‐surface of 0–10 cm depth; 1016 ± 118 Tg for 0–30 cm depth; and 1474 ± 181 Tg for 0–50 cm depth.     

A multiple regression approach to assess the spatial distribution of Soil Organic Carbon (SOC) at the regional scale (Flanders,Belgium)

Estimates of the amount of Soil Organic Carbon (SOC) at the regional scale are important to better understand the role of the SOC reservoir in global climate and environmental issues. This study presents a method for estimating the total SOC stock using data from Flanders (Belgium). More than 6900 SOC measurements from the national soil survey (database ‘Aardewerk’) are combined with a digital land use map and a digital soil map of Flanders. The spatial distribution of the SOC stock is studied in its relation to factors such as soil texture, soil moisture (drainage class) and land use. The resulting map with a resolution of 15 m consists of different classes forming a combination of these environmental factors. The results show that the lowest SOC amount (kg m^? 2) is stored under cropland whereas the highest amount is found under grassland. Regarding the effect of soil properties, a significant correlation between SOC stock and depth of the ground water table is observed. Sandy loam soils stock the lowest SOC amount (kg m^? 2), whereas clay soils retain the highest SOC amount. First, the mean SOC amounts of the land use–soil type classes are calculated and assigned to the corresponding cells in order to obtain a total SOC stock with its spatial distribution for Flanders. Then, a multiple regression model is applied to predict the SOC value of a particular land use–soil type class on the map. This model is based on the observed relationships between SOC and land use–soil type characteristics, using the entire dataset. The first approach does not allow to obtain a (reliable) SOC value for all land use–soil type classes due to a lack of samples in some classes. A major advantage of the regression model approach is the attribution of class specific SOC values to each land use–soil type class, regardless of the number of observations in the classes. Consequently, by applying the model approach instead of the mean approach, the area for which a reliable SOC estimate could be obtained increased by 8.1% (from 9420 km² to 10179 km²) and the total predicted SOC stock increased by 10.1% (from 88.7 ± 5.6 Mt C to 97.6 ± 1.1 Mt C).     

基于时间稳定性和降维因子分析的土壤水分监测优化

以南京市高淳区青山茶场中相邻的茶园和竹林坡地为研究区,对研究区土壤水分进行长期定点监测。基于土壤水分时间稳定性,结合因子分析选取典型样点组合,采用多元线性回归模型构建各监测点土壤水与典型样点间的数量关系。通过典型样点预测各监测点土壤水分,并检验预测结果,以期通过少数样点的监测来反映研究区的整体概况,优化研究区土壤水分监测。结果表明:在茶园仅监测7个点时,验证期RMSE小于1.5 cm3/cm3;竹林仅监测5个样点时,验证期RMSE小于1.7 cm3/cm3,模型能很好地预测研究区各样点土壤水分,为优化土壤水监测、减少野外工作量提供了理论依据。不同土地利用方式、不同深度处土壤水分分布特征存在显著差异;竹林土壤水分具有较强的时间稳定性,土壤水分的空间自相关性较茶园强;30 cm深处比10 cm深处土壤水分具有更稳定的空间分布结构。     

基于各向异性的区域土壤有机碳三维模拟与空间特征分析

为探索更加科学的土壤属性三维空间模拟方法,以各项同性三维普通克里格法为对比方法,采用均方根误差（root mean squared errors,RMSE）和标准化克里格方差（mean squared deviation ratio,MSDR）以及空间模拟方差图等,评价比较了各项同性和顾及各项异性的三维模拟方法的模拟效果。结果显示：三种方法模拟的土壤有机碳三维空间分布格局基本一致。随着土壤深度的不断增加,土壤有机碳含量较高的斑块逐渐减少,垂直方向上总体呈现出土体上部高下部低的格局。顾及各向异性能在一定程度上克服普通克里格法常出现的牛眼和趋中效应等缺陷问题。顾及各向异性基于Markov 的同位置协同格里格法模拟效果最佳。该法的 RMSE 值最小（1.6215）,相比于各项同性三维普通克里格法 RMSE提高将近50%,特异值覆盖比率最大（76.15%）,模拟精度最高,能够更好地突出波动性,体现特异值;该方法的 MSDR最接近1（1.4409）,且模拟的土壤有机碳质量分数总体方差均值最小（2.08）。研究成果将为区域土壤属性三维空间有效模拟提供方法参考。     

冲积平原区土壤碳密度估算及其空间分布

冲积平原区通常具有复杂的剖面质地层次排列,为了准确估算冲积平原区土壤碳密度的空间分布特征,该文在华北冲积平原区的河北曲周县选取了121个土壤剖面,测定了各土层有机碳含量,构建了基于负指数函数的土壤有机碳垂向分布模型,结合地统计学方法绘制了该县土壤碳密度的空间分布图。结果表明,土壤有机碳含量随深度增加呈逐渐递减的趋势,各土层有机碳含量均属于中等变异程度。0~20和20~40 cm土壤有机碳空间连续性较好,它们的空间相关距离分别为14和3 km,而下层(40 cm)土壤有机碳均表现为纯块金效应结构。土壤有机碳垂向分布模型可以很好地描述剖面土壤有机碳含量的变化特征,且预测与实测的土壤有机碳含量的均方根误差仅为0.70 kg/m3,决定系数达到了0.95。曲周县土壤有机碳密度的空间分布总体表现为西北高东南低的趋势。其空间分布主要受土壤类型和质地的影响,其中潮土和盐化潮土的碳密度明显高于褐土化潮土,质地较细的土壤(轻壤、中壤和粘土)碳密度明显高于质地较粗的土壤(砂土和砂壤)。该研究为冲积平原区土壤碳密度的估算提供了一种新的方法。     

藏东南色季拉山西坡不同植被土壤有机碳垂直分布特征及其影响因素

  【目的】  藏东南地区高山生态系统有巨大的土壤碳汇潜力，研究其不同生态系统下土壤有机碳 (SOC) 储存的变化特征及其影响因子，有助于深入了解青藏高原土壤碳循环及区域碳源汇平衡。  【方法】  本研究在西藏色季拉山西坡海拔3000～4600 m开展密集土壤采样，研究不同海拔高度下不同植被类型SOC的储存特征，并分析其关键影响因子。  【结果】  表层0—5 cm的SOC含量随海拔升高而增加，4个植被带SOC含量平均值表现为高寒草甸 (8.31% ± 0.77%) > 暗针叶林 (7.20% ± 0.90%) > 高寒灌丛草甸 (6.74% ± 0.80%) > 针阔混交林 (3.88% ± 0.46%)。在剖面5—10、10—15、15—20、20—30、30—40、40—60 cm各层SOC含量随海拔升高呈先增加后降低趋势，SOC含量在4种植被带的平均值表现为暗针叶林 > 高寒灌丛草甸 > 高寒草甸 > 针阔混交林。SOC含量随剖面深度增加而显著下降，高寒草甸和高寒灌丛草甸SOC垂直分布特征为表层聚集型，而针阔混交林和暗针叶林SOC垂直分布特征为普通递减型。剖面0—20、20—40、40—60 cm的SOC储量随海拔升高呈先增加后降低的特征。在表层0—20 cm高寒草甸SOC储量最高 (C 95.66 ± 4.81 t/hm2)；在剖面20—40和40—60 cm暗针叶林SOC储量最高，且其在整个0—60 cm剖面的SOC总储量在所有植被类型中最高 (C 199.14 ± 11.10 t/hm2)；针阔混交林SOC储量在剖面各层均为最低，且其在整个剖面的SOC总储量 (C 111.45 ± 10.30 t/hm2) 显著低于其他植被类型。剖面各层SOC储量与年平均温度、凋落物碳氮比呈显著负相关，而与海拔高度、年平均降水量和土壤含水量呈显著正相关。逐步回归显示土壤含水量是影响剖面各层以及整个剖面SOC储存的关键因子。随机森林模型对SOC储存的解释度为50.32%～65.82%，土壤含水量对表层土体SOC预测的相对贡献最高，年平均温度、年平均降水量和凋落物质量对各层SOC预测均有显著贡献，而植被类型对SOC预测的相对贡献随剖面加深而逐步增加。  【结论】  色季拉山西坡不同海拔高度下SOC的储存特征随不同植被类型和剖面深度而发生显著变化，环境因子(如土壤水分) 对表层土体SOC储存有关键影响，植被类型对深层土体SOC储量变化的预测有重要贡献。     

秸秆带状覆盖下培土对马铃薯产量与水分利用效率的影响

气候变化效应评估、土壤固碳潜力和肥力管理等,迫切需要详尽的土壤有机质（soil organic matter, SOM）空间分布信息。该文以江苏省第二次土壤普查的1 519个典型土壤剖面的表层（0～20 cm）SOM含量为例,选择1 217个样本为建模集,302个为验证集,选取年均温度、年均降雨、物理性黏粒和土壤pH值等因子进行SOM的地理加权回归（geographically weighted regression, GWR）建模。从建模集中分别随机抽取100%（1 217个）、80%（973个）、60%（730个）、40%（486个）,20%（243个）的样点,对比不同样点数量下GWR和传统全局回归模型的精度差异,并选择最优模型进行SOM空间预测制图。结果表明：1）江苏省SOM含量在不同空间尺度上存在极显著的空间自相关性。不同样点数量的建模集的全局自相关性和局部空间自相关聚类图结果相似。全局Moran''s I值介于0.25～0.61（P<0.001）。SOM含量空间分布以空间聚集特征为主,"高-高"聚集区主要分布在苏中和苏南地区,"低-低"聚集区主要分布在苏北地区。2）GWR建模结果均优于传统的传统全局回归建模,其残差在不同的空间尺度上均不存在空间自相关性。不同建模集的GWR的R2adj较全局建模均提高0.15～0.20,其AIC和RSS均比全局模型有大幅降低,为56.08～360.19和17.40～76.67。不同建模样本数量的GWR模型对SOM的解释能力差异较小。3）建模样点数量（除建模样本n=243）对GWR预测制图结果的精度影响不大,RMSE介于5.56～5.75 g/kg之间,MAE介于3.87～4.05 g/kg之间,R2介于0.52～0.48之间,均优于全部建模样点的普通克里格插值验证结果。该研究可为样点数较少的省级尺度地区SOM空间建模与制图提供借鉴。