基于机器学习模型的农田土壤容重模拟及其影响因素

doi:10.13287/j.1001-9332.202509.012

摘要/Abstract

摘要： 为明确典型农田土壤容重空间变异规律及关键驱动因素,本研究采集江苏省611个农田样地0～10、10～20和20～40 cm剖面土壤样品,结合实测的土壤容重数据与多种物理化学性质及气候环境变量,分析了土壤容重的空间分布规律及主要影响因素,并采用随机森林和神经网络模型对土壤容重进行模拟。结果表明: 江苏省农田土壤容重随土层深度增加而显著增大。0～10和20～40 cm土壤容重空间分布规律总体一致,10～20 cm土壤容重受秸秆还田影响空间分布更具随机性;土壤容重高值区域相对集中分布于江苏省南部、徐州及沿海区域。11种作物种植模式下,仅0～10 cm土壤容重存在显著差异。农田0～20 cm土壤容重空间异质性受耕作与秸秆施用影响,20～40 cm土壤容重主要由母质和水热条件控制。农田土壤容重与土壤有机碳、全氮含量呈显著负相关,与pH和C/N呈显著正相关,且土壤容重受土壤深度、区域、年均温度和纬度的共同影响。基于10个土壤理化性质、环境因素和气候变量的随机森林模型和7个相关变量的神经网络模型可分别模拟农田土壤容重62%和53%的变异。随机森林模型中的重要环境因素为纬度、经度和年均温度,其次是土壤C/N及有机碳含量;神经网络模型中以年均降水量和C/N重要性最高。农田土壤容重受土壤理化性质、地理位置、作物种植模式和气候因子的共同影响,随机森林模型在模拟0～40 cm深度土壤容重方面效果更好,可基于易获取的变量便捷模拟长三角区域农田土壤容重。

关键词: 土壤容重, 农田土壤, 空间分布, 机器学习

Abstract: To elucidate the spatial variations and key drivers of soil bulk density (BD) in typical farmland across Jiangsu Province, we collected soil samples from 611 sites at depths of 0-10 cm, 10-20 cm, and 20-40 cm. By combining measured BD data with a suite of physicochemical properties and climatic variables, we analyzed the spatial variations of BD and its main influencing factors, and then simulated BD using random forest (RF) and neural network (NN) models. Results showed that soil BD increased significantly with soil depth. The spatial patterns of BD in the 0-10 cm and 20-40 cm layers were largely consistent, whereas the 10-20 cm layer exhibited greater spatial randomness due to straw incorporation. Areas with high soil BD values were mainly concentrated in southern Jiangsu, Xuzhou, and coastal zones. Among the examined 11 crop planting systems, significant differences in BD were observed only in the 0-10 cm layer. The spatial variation of BD in the 0-20 cm soil layer was influenced primarily by tillage practices and straw incorporation, while BD in the 20-40 cm layer was mainly governed by parent material and hydrothermal conditions. BD was negatively correlated with soil organic carbon and total nitrogen, and positively correlated with pH and C/N. Soil depth, located region, mean annual temperature, and latitude jointly influence BD value. The RF model based on 10 soil physicochemical traits, environmental factors, and climatic variables explained 62% of BD variation, while the NN model based on 7 key variables explained 53% of BD variation. In the RF model, latitude, longitude, and mean annual temperature were the most important predictors, followed by C/N and soil organic carbon. In the NN model, mean annual precipitation and C/N ranked the highest among all variables. Our findings suggested that farmland BD is jointly controlled by soil physicochemical properties, geographic location, cropping systems and climate factors, and that the RF model offers higher simulation performance of BD for 0-40 cm soil layer, enabling convenient simulation of farmland BD in the Yangtze River Delta region using readily available variables.

Key words: soil bulk density, farmland soil, spatial distribution, machine learning

丁司丞, 方超, 卓玛拉姆, 冯兆忠. 基于机器学习模型的农田土壤容重模拟及其影响因素[J]. 应用生态学报, 2025, 36(9): 2827-2835.

DING Sicheng, FANG Chao, ZHUOMA Lamu, FENG Zhaozhong. Simulation of farmland soil bulk density and its influencing factors based on machine learning models[J]. Chinese Journal of Applied Ecology, 2025, 36(9): 2827-2835.

参考文献

[1] 王思敏, 张红丽, 张恒硕, 等. 晋西黄土区典型小流域不同土层土壤容重分布特征及其影响因素. 生态学杂志, 2024, 43(3): 609-615
[2] Martín MÁ, Reyes M, Taguas FJ. Estimating soil bulk density with information metrics of soil texture. Geoderma, 2017, 287: 66-70
[3] 李孟霞, 文国松, 李永忠. 作物对土壤压实胁迫响应研究进展. 山东农业科学, 2019, 51(1): 154-160
[4] Panagos P, De Rosa D, Liakos L, et al. Soil bulk density assessment in Europe. Agriculture, Ecosystems & Environment, 2024, 364: 108907
[5] 韩光中, 王德彩, 谢贤健. 中国主要土壤类型的土壤容重传递函数研究. 土壤学报, 2016, 53(1): 93-102
[6] 刘亚男, 郗敏, 张希丽, 等. 中国湿地碳储量分布特征及其影响因素. 应用生态学报, 2019, 30(7): 2481-2489
[7] Lawrence PG, Roper W, Morris TF, et al. Guiding soil sampling strategies using classical and spatial statistics: A review. Agronomy Journal, 2020, 112: 493-510
[8] Lawrence GB, Fernandez IJ, Hazlett PW, et al. Methods of soil resampling to monitor changes in the chemical concentrations of forest soils. Journal of Visualized Experiments, 2016, 25: 54815
[9] Al-Shammary AAG, Kouzani AZ, Kaynak A, et al. Soil bulk density estimation methods: A review. Pedosphere, 2018, 28: 581-596
[10] Xu L, He NP, Yu GR. Methods of evaluating soil bulk density: Impact on estimating large scale soil organic carbon storage. Catena, 2016, 144: 94-101
[11] 何林倩, 刘倩, 王德彩, 等. 基于数字土壤制图技术的土壤有机碳储量估算. 应用生态学报, 2021, 32(2): 591-600
[12] Schillaci C, Perego A, Valkama E, et al. New pedotransfer approaches to predict soil bulk density using WoSIS soil data and environmental covariates in Mediterranean agro-ecosystems. Science of the Total Environment, 2021, 780: 146609
[13] Sequeira CH, Wills SA, Seybold CA, et al. Predicting soil bulk density for incomplete databases. Geoderma, 2014, 213: 64-73
[14] Chen SC, Chen ZX, Zhang XL, et al. European topsoil bulk density and organic carbon stock database (0-20 cm) using machine-learning-based pedotransfer functions. Earth System Science Data, 2024, 16: 2367-2383
[15] Jalabert SSM, Martin MP, Renaud JP, et al. Estimating forest soil bulk density using boosted regression modelling. Soil Use and Management, 2010, 26: 516-528
[16] Yi XS, Li GS, Yin YY. Pedotransfer functions for estimating soil bulk density: A case study in the three-river headwater region of Qinghai Province, China. Pedosphere, 2016, 26: 362-373
[17] Assouline S, Or D. Conceptual and parametric representation of soil hydraulic properties: A review. Vadose Zone Journal, 2013, 12: vzj2013-07
[18] 卢宏亮, 赵明松, 刘斌寅, 等. 基于随机森林模型的安徽省土壤属性空间分布预测. 土壤, 2019, 51(3): 602-608
[19] Robinson DA, Thomas A, Reinsch S, et al. Analytical modelling of soil porosity and bulk density across the soil organic matter and land-use continuum. Scientific Reports, 2022, 12: 7085
[20] 柴华, 何念鹏. 中国土壤容重特征及其对区域碳贮量估算的意义. 生态学报, 2016, 36(13): 3903-3910
[21] Sun XL, Minasny B, Wang HL, et al. Spatiotemporal modelling of soil organic matter changes in Jiangsu, China between 1980 and 2006 using INLA-SPDE. Geoderma, 2021, 384: 114808
[22] Zheng GH, Jiao CX, Xie XL, et al. Pedotransfer functions for predicting bulk density of coastal soils in East China. Pedosphere, 2023, 33: 849-856
[23] 江苏省统计局. 江苏统计年鉴. 南京: 中国统计出版社, 2004—2023
[24] John K, Abraham Isong I, Michael Kebonye N, et al. Using machine learning algorithms to estimate soil organic carbon variability with environmental variables and soil nutrient indicators in an alluvial soil. Land, 2020, 9: 487
[25] Chen Y, Ma LX, Yu DS, et al. Comparison of feature selection methods for mapping soil organic matter in subtropical restored forests. Ecological Indicators, 2022, 135: 108545
[26] Garson DG. Interpreting neural network connection weights. AI Expert, 1991, 6: 46-51
[27] 于冬雪, 贾小旭, 黄来明, 等. 黄土区不同土层土壤容重空间变异与模拟. 土壤学报, 2019, 56(1): 55-64
[28] 陈伟志, 张亚, 李静婷, 等. 川滇干热河谷区土壤容重空间变异特征: 以滇中楚雄地区为例. 中国地质调查, 2024, 11(2): 62-71
[29] 朱荣昱, 赵蒙杰, 姚云凤, 等. 秸秆还田方式与播种深度对夏直播花生土壤物理性状与出苗特性的影响. 作物学报, 2024, 50(8): 2106-2121
[30] Hu WF, Li Q, Wang WQ, et al. Straw mulching decreased the contribution of Fe-bound organic carbon to soil organic carbon in a banana orchard. Applied Soil Ecology, 2024, 194: 105177
[31] Liu H, Zak D, Rezanezhad F, et al. Soil degradation determines release of nitrous oxide and dissolved organic carbon from peatlands. Environmental Research Letters, 2019, 14: 094009
[32] Amorim HCS, Hurtarte LCC, Souza IF, et al. C:N ratios of bulk soils and particle-size fractions: Global trends and major drivers. Geoderma, 2022, 425: 116026
[33] 范倩玉, 李晋, 刘振华, 等. 不同轮作模式对潮土土壤物理性状的影响. 山西农业科学, 2020, 48(8): 1267-1270
[34] Chen JZ, Wu ZL, Zhao TM, et al. Rotation crop root performance and its effect on soil hydraulic properties in a clayey Utisol. Soil and Tillage Research, 2021, 213: 105136
[35] 卢立娜, 赵雨兴, 胡莉芳, 等. 沙棘(Hippophae rhamnoides)种植对鄂尔多斯砒砂岩地区土壤容重、孔隙度与贮水能力的影响. 中国沙漠, 2015, 35(5): 1171-1176
[36] Katuwal S, Knadel M, Norgaard T, et al. Predicting the dry bulk density of soils across Denmark: Comparison of single-parameter, multi-parameter, and vis-nir based models. Geoderma, 2020, 361: 114080
[37] Ao YL, Li HQ, Zhu LP, et al. The linear random forest algorithm and its advantages in machine learning assisted logging regression modeling. Journal of Petroleum Science and Engineering, 2019, 174: 776-789
[38] Ruehlmann J, Körschens M. Soil particle density as affected by soil texture and soil organic matter: 2. Predicting the effect of the mineral composition of particle-size fractions. Geoderma, 2020, 375: 114543
[39] 袁玉琦, 陈瀚阅, 张黎明, 等. 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. 土壤学报, 2021, 58(4): 887-899
[40] Francesca Cotrufo M, Lavallee JM, Zhang Y, et al. In-N-Out: A hierarchical framework to understand and predict soil carbon storage and nitrogen recycling. Global Change Biology, 2021, 27: 4465
[41] Tifafi M, Guenet B, Hatté C. Large differences in global and regional total soil carbon stock estimates based on SoilGrids, HWSD, and NCSCD: Intercomparison and evaluation based on field data from USA, England, Wales, and France. Global Biogeochemical Cycles, 2018, 32: 42-56
[42] 周洋, 赵小敏, 郭熙. 基于多源辅助变量和随机森林模型的表层土壤全氮分布预测. 土壤学报, 2022, 59(2): 451-460
[43] Liu F, Yang F, Zhang GL, et al. Predicting soil depth in a large and complex area using machine learning and environmental correlations. Journal of Integrative Agriculture, 2022, 21: 2422-2434
[44] 魏宇宸, 赵美芳, 朱昌达, 等. 基于景观及微地形特征的丘陵区土壤属性预测. 应用生态学报, 2022, 33(2): 467-476