基于机器学习算法的华中天然林土壤有机碳特征与关键影响因子

doi:10.11707/j.1001-7488.LYKX20230051

摘要/Abstract

摘要：

目的: 比较4种机器学习算法在模拟华中地区3种典型天然林土壤有机碳含量上的表现，筛选最优模型算法，明确影响该地区天然混交林土壤有机碳富集与空间分布的关键气候环境因子，为森林土壤有机碳分布格局研究提供技术参考。方法: 以华中地区3种典型天然林（常绿针叶混交林、落叶阔叶混交林和常绿阔叶混交林）为研究对象，引入4种机器学习算法（支持向量机、人工神经网络中的多层感知器、随机森林和分位数回归森林），模拟0~60 cm土层土壤有机碳含量，比较模型解释量及表现稳定性，筛选最优模型算法。结果: 4种机器学习算法均能成功模拟天然林0~60 cm土层土壤有机碳含量，多层感知器、随机森林、分位数回归森林模拟结果明显优于支持向量机，其中随机森林模型表现最稳定，决定系数最高达0.620。母质、土壤密度、孔隙度、地理位置、海拔、植被和水分亏损情况等共同影响华中地区天然林0~60 cm土层土壤有机碳含量，但显著影响表层（0~20 cm）、中层（20~40 cm）与深层（40~60 cm）土壤有机碳含量的因子并不一致且影响机制不同。在0~20 cm土层，显著影响因子最复杂，除土壤密度以外，土壤孔隙度、地形、植被和气候均产生显著影响（P<0.05）；在20~40 cm土层，土壤密度和地理位置依然有显著影响（P<0.05），各因子影响呈现复杂性和过渡性；在40~60 cm土层，成土母质是最重要的影响因子，其次为土壤密度和水分亏损指数，植被的影响下降（P<0.05）。从地理分布上看，0~20 cm表层土壤有机碳含量东南高、西北低，中层和深层土壤有机碳含量表现为西部较高、东部稍低；海拔较高的南部山区土壤有机碳含量更高。蒸发强烈或供给森林的水分不足会限制各层土壤有机碳富集。森林土壤有机碳含量随土层加深显著下降，常绿针叶混交林土壤有机碳含量在各土层均最高，落叶阔叶混交林居中，常绿阔叶混交林最低。结论: 华中地区天然林土壤有机碳含量分布呈现明显差异，常绿针叶混交林土壤有机碳贡献最大，成土母质和土壤物理属性对土壤有机碳富集与分布起决定性作用；适宜天然林生长的地理立地条件和气候环境，共同造就该地区天然林土壤有机碳富集。在营林和管理时可加大本地树种混交比重，提升森林土壤碳汇功能。

关键词: 森林土壤有机碳, 空间分布, 天然林, 机器学习, 华中地区

Abstract:

objective: Soil organic carbon contents were simulated for typical forests (evergreen coniferous, deciduous broadleaved and evergreen broadleaved forests) in central China. The optimal models were used to reveal the key factors influencing the accumulation and spatial distribution of soil organic carbon in mixed forests in central China, which would technically improve the understanding of spatial pattern of forest soil carbon. Method: Forest soil organic carbon content of 0–60 cm was modeled by four advanced machine learning algorithms, including support vector machine, multi-layer perceptron of artificial neural networks, random forests and quantile regression forests. Model selection was conducted by comparing their model explanation and performance stability. Result: Models for organic carbon content of forest soil were developed successfully using all the 4 algorithms for 0–60 cm soil depths. The results of multi-layer perceptron, random forests, and quantile regression forests were significantly better than support vector machine, among which random forests processed the most stable results along soil layers, with the highest R² at 0.620. Parent material, bulk density, soil porosity, topography, elevation, vegetation, and moisture deficit conditions jointly influenced the soil organic carbon content of 0–60 cm in the mentioned forests, while the significant factors differed among the topsoil (0–20 cm), middle (20–40 cm) and deep soil layers (40–60 cm) due to different mechanisms. Forest soil organic carbon content in the topsoil was comprehensively affected by soil porosity, geographic factors, vegetation, and climate, besides soil density as the most significant covariate (P<0.05). In the middle soil layer, soil properties and topography were still significant, while the influence of each factor on the soil organic carbon content showed complexity and transitional characteristics (P<0.05). In the deep soil layer, the parent material was the most important influencing factor, followed by soil properties and moisture insufficiency, while the influence of vegetation decreased (P<0.05). Geographically, the 0–20 cm surface soil organic carbon was higher in the southeast than in the northwest, while contents were observed higher in the west than in the east for the two deeper soil layers. Forest soil organic carbon contents were higher in mountainous forests with lower latitudes and higher elevations. Strong evaporation or insufficient moisture supply would limit the accumulation of forest soil organic carbon in all soil layers. Forest soil organic carbon content decreased significantly along soil layers. The highest soil organic carbon contents were found in evergreen coniferous forests, followed by deciduous broadleaved and evergreen broadleaved forests. Conclusion: The distribution of soil organic carbon in natural forests is characterized by spatial heterogeneity and differences among forest compositions in central China. Evergreen coniferous forests performed the largest contribution to soil organic carbon among these forests. Parent material and soil physical properties played a decisive role in the enrichment and distribution of soil organic carbon in the regional forests. Both suitable geographical and favorable topographical conditions contributed to the enrichment of soil organic carbon in forests. The proportion of mixed forests with local species could be promoted in forest management and silviculture to enhance forest soil carbon sinks.

Key words: forest soil organic carbon content, spatial variability, natural mixed forests, machine learning, central China

中图分类号:

S714

沈琛琛,肖文发,朱建华,曾立雄,陈吉臻,黄志霖. 基于机器学习算法的华中天然林土壤有机碳特征与关键影响因子[J]. 林业科学, 2024, 60(3): 65-77.

Chenchen Shen,Wenfa Xiao,Jianhua Zhu,Lixiong Zeng,Jizhen Chen,Zhilin Huang. Characterization of Soil Organic Carbon and Key Influencing Factors of Natural Forests in Central China Based on Machine Learning Algorithms[J]. Scientia Silvae Sinicae, 2024, 60(3): 65-77.

图/表 10

表1

表2

表3

表4

表5

表6

表7

图1

图2

图3

参考文献 0

	丁建军, 章　盛, 孙　超, 等. 基于PLSR的土壤有机碳预测模型中建模组与验证组最优数量关系. 江汉大学学报(自然科学版), 2018, 46 (5): 404- 408.
	Ding J J, Zhang S, Sun C, et al. Optimal quantity relationship between modeling group and prediction group in soil organic carbon prediction model based on PLSR. Journal of Jianghan University (Natural Science Edition), 2018, 46 (5): 404- 408.
	冯　广, 姚　兰, 艾训儒, 等. 鄂西南亚热带常绿落叶阔叶混交林土壤空间异质性. 生态学报, 2016, 36 (23): 7635- 7643.
	Feng G, Yao L, Ai X R, et al. Analysis of soil spatial variability in a subtropical evergreen and deciduous broad-leaved mixed forest in southwest Hubei Province. Acta Ecologica Sinica, 2016, 36 (23): 7635- 7643.
	郭李娜, 樊贵盛. 基于网格搜索和交叉验证支持向量机的地表土壤容重预测. 土壤通报, 2018, 49 (3): 512- 518.
	Guo L N, Fan G S. Support vector machines for surface soil density prediction based on grid search and cross validation. Chinese Journal of Soil Science, 2018, 49 (3): 512- 518.
	郭璐璐, 李安迪, 商宏莉, 等. 川西贡嘎山不同森林生态系统土壤有机碳垂直分布与组成特征. 中国农业气象, 2018, 39 (10): 636- 643. doi: 10.3969/j.issn.1000-6362.2018.10.002
	Guo L L, Li A D, Sang H L, et al. Total and labile organic carbon in soils of three subalpine forest types in Gongga Mountain, Western Sichuan. Chinese Journal of Agrometeorology, 2018, 39 (10): 636- 643. doi: 10.3969/j.issn.1000-6362.2018.10.002
	简尊吉, 倪妍妍, 徐　瑾, 等. 2022. 马尾松人工林土壤碳氮磷生态化学计量学特征的纬度变化. 林业科学研究, 35(2): 1–8.
	Jian Z J, Ni Y Y, Xu J, et al. 2022. Latitudinal variations of soil C-N-P stoichiometry in Pinus massoniana (Lamb.) plantations. Forest Research, 35(2): 1–8. ［in Chinese］
	雷相东. 机器学习算法在森林生长收获预估中的应用. 北京林业大学学报, 2019, 41 (12): 23- 36. doi: 10.12171/j.1000-1522.20190356
	Lei X D. Applications of machine learning algorithms in forest growth and yield prediction. Journal of Beijing Forestry University, 2019, 41 (12): 23- 36. doi: 10.12171/j.1000-1522.20190356
	李双成, 吴绍洪, 戴尔阜. 生态系统响应气候变化脆弱性的人工神经网络模型评价. 生态学报, 2005, 25 (3): 621- 626. doi: 10.3321/j.issn:1000-0933.2005.03.032
	Li S C, Wu S H, Dai E B. Assessing the fragility of ecosystem using artificial neural network model. Acta Ecologica Sinica, 2005, 25 (3): 621- 626. doi: 10.3321/j.issn:1000-0933.2005.03.032
	李家湘, 游健荣, 徐永福, 等. 湖南植被研究: 植被类型、组成和分布格局. 中国科学:生命科学, 2021, 50 (3): 275- 88.
	Li J X, You J R, Xu Y F, et al. Vegetation research in Hunan Province: vegetation types, composition, and distribution pattern. Scientia Sinica(Vitae), 2021, 50 (3): 275- 88.
	梁萌杰, 陈龙池, 汪思龙. 湖南省杉木人工林生态系统碳储量分配格局及固碳潜力. 生态学杂志, 2016, 35 (4): 896- 902.
	Liang M J, Chen L C, Wang S L. The allocation pattern of carbon storage and carbon sequestration potential in Chinese fir plantation ecosystems in Hunan Province. Chinese Journal of Ecology, 2016, 35 (4): 896- 902.
	刘佳斌, 郜允兵, 李永涛, 等. 基于高斯混合模型的土壤环境质量分区研究. 农业环境科学学报, 2021, 40 (8): 1746- 1757. doi: 10.11654/jaes.2020-1503
	Liu J B, Gao Y B, Li Y T, et al. Soil environmental quality zoning based on the Gaussian mixture model. Journal of Agro-Environment Science, 2021, 40 (8): 1746- 1757. doi: 10.11654/jaes.2020-1503
	刘捷豹, 陈光水, 郭剑芬, 等. 森林土壤酶对环境变化的响应研究进展. 生态学报, 2017, 37 (1): 110- 117.
	Liu J B, Chen G S, Guo J F, et al. Advances in research on the responses of forest soil enzymes to environmental change. Acta Ecologica Sinica, 2017, 37 (1): 110- 117.
	刘　璐, 赵常明, 徐文婷, 等. 神农架常绿落叶阔叶混交林凋落物养分特征. 生态学报, 2019, 39 (20): 7611- 7620.
	Liu L, Zhao C M, Xu W T, et al. Litter nutrient characteristics of mixed evergreen and deciduous broadleaved forests in Shennongjia, China. Acta Ecologica Sinica, 2019, 39 (20): 7611- 7620.
	聂浩亮, 薄慧娟, 张润哲, 等. 北京海坨山典型林分土壤有机碳含量及有机碳密度垂直分布特征. 林业科学研究, 2020, 33 (6): 155- 162.
	Nie H L, Bo H J, Zhang R Z, et al. Vertical distribution characteristics of soil organic carbon content and organic carbon density of typical forest stands at Haituo Mountain, Beijing. Forest Research, 2020, 33 (6): 155- 162.
	欧强新, 雷相东, 沈琛琛, 等. 基于随机森林算法的落叶松-云冷杉混交林单木胸径生长预测. 北京林业大学学报, 2019, 41 (9): 9- 19.
	Ou Q X, Lei X D, Shen C C, et al. Individual tree DBH growth prediction of larch-spruce-fir mixed forests based on random forest algorithm. Journal of Beijing Forestry University, 2019, 41 (9): 9- 19.
	庞健峰, 丁孝忠, 韩坤英, 等. 1: 100万中华人民共和国数字地质图空间数据库. 中国地质, 2017, 44 (S1): 8- 18,125−138.
	Pang J F, Ding X Z, Han K Y, et al. The national 1: 1 000 000 geological map spatial database. Geology in China, 2017, 44 (S1): 8- 18,125−138.
	王春燕, 何念鹏, 吕瑜良. 中国东部森林土壤有机碳组分的纬度格局及其影响因子. 生态学报, 2016, 36 (11): 3176- 3188.
	Wang C Y, He N P, Lü Y L. Latitudinal patterns and factors affecting different soil organic carbon fractions in the eastern forests of China. Acta Ecologica Sinica, 2016, 36 (11): 3176- 3188.
	王鹏程, 邢乐杰, 肖文发, 等. 三峡库区森林生态系统有机碳密度及碳储量. 生态学报, 2009, 29 (1): 97- 107. doi: 10.3321/j.issn:1000-0933.2009.01.013
	Wang P C, Xing L J, Xiao W F, et al. Organic carbon density and storage of forest ecosystems in Three Gorges Reservoir Area. Acta Ecologica Sinica, 2009, 29 (1): 97- 107. doi: 10.3321/j.issn:1000-0933.2009.01.013
	徐　佳, 刘　峰, 吴华勇, 等. 基于人工神经网络和随机森林学习模型从土壤属性推测关键成土环境要素的研究. 土壤通报, 2021, 52 (2): 269- 278.
	Xu J, Liu F, Wu H Y, et al. Predicting of key environmental factors from soil properties based on artificial neural network and random forest learning model. Chinese Journal of Soil Science, 2021, 52 (2): 269- 278.
	许昌建, 刘迎春, 左丽君, 等. 基于仿真大光斑激光雷达和多层感知器的森林地上生物量估算模型构建. 林业资源管理, 2021, (1): 50- 60.
	Xu C J, Liu Y C, Zuo L J, et al. Estimation on forest above-ground biomass based on simulated large-footprint LiDAR and multi-layer perceptron. Forest Resources Management, 2021, (1): 50- 60.
	张宏帅, 朱高龙, 吴家煜, 等. 基于BP神经网络与Kriging结合的土壤有机质空间分布模拟—以福建省华安县为例. 亚热带农业研究, 2021, 17 (1): 40- 47.
	Zhang H S, Zhu G L, Wu J Y, et al. Simulation of spatial distribution simulation of soil organic matter based on BP neural network and Kriging interpolation—taking Hua'an County, Fujian Province as an example. Subtropical Agriculture Research, 2021, 17 (1): 40- 47.
	张　强, 黄生志, 陈晓宏. 基于支持向量机的土壤湿度模拟及预测研究. 土壤学报, 2013, 50 (1): 59- 67. doi: 10.11766/trxb201112210497
	Zhang Q, Huang S Z, Chen X H. Simulation and prediction of soil moisture based on support vector machine technique. Acta Pedologica Sinica, 2013, 50 (1): 59- 67. doi: 10.11766/trxb201112210497
	赵彦锋, 李怡欣, 马盼盼, 等. 近30年河南省耕地土壤有机碳的三维变化与关键因素研究. 土壤学报, 2023, 1 (9): 1- 13. doi: 10.11766/trxb202201290041
	Zhao Y F, Li Y X, Ma P P, et al. Three dimensional changes of soil organic carbon content and its’ key factors in cultivated soils of Henan Province during the recent 30 years. Acta Pedologica Sinica, 2023, 1 (9): 1- 13. doi: 10.11766/trxb202201290041
	周　鑫, 姜　航, 孙金兵, 等. 地形因子和物理保护对张广才岭次生林土壤有机碳密度的影响. 北京林业大学学报, 2016, 38 (4): 94- 106.
	Zhou X, Jiang H, Sun J B, et al. Soil organic carbon density as affected by topography and physical protection factors in the secondary forest area of Zhangguangcai Mountains, Northeast China. Journal of Beijing Forestry University, 2016, 38 (4): 94- 106.
	周艳翔, 吕茂奎, 谢锦升, 等. 深层土壤有机碳的来源、特征与稳定性. 亚热带资源与环境学报, 2013, 8 (1): 48- 55. doi: 10.3969/j.issn.1673-7105.2013.01.009
	Zhou Y X, Lü M K, Xie J S, et al. Sources, characteristics and stability of organic carbon in deep soil. Journal of Subtropical Resources and Environment, 2013, 8 (1): 48- 55. doi: 10.3969/j.issn.1673-7105.2013.01.009
	Feng Y H, Schmid B, Loreau M, et al. Multispecies forest plantations outyield monocultures across a broad range of conditions. Science, 2022, 376 (6595): 865- 868. doi: 10.1126/science.abm6363
	Garg P K, Garg R D, Shukla G, et al. 2020. Digital mapping of soil landscape parameters.Studies in Big Data. Springer Singapore, Singapore.
	Ghannam R B, Techtmann S M. Machine learning applications in microbial ecology, human microbiome studies, and environmental monitoring. Computational and Structural Biotechnology Journal, 2021, 19, 1092- 1107. doi: 10.1016/j.csbj.2021.01.028
	Gomes L C, Faria R M, de Souza E, et al. Modelling and mapping soil organic carbon stocks in Brazil. Geoderma, 2019, 340, 337- 350. doi: 10.1016/j.geoderma.2019.01.007
	Hengl T, Mendes de Jesus J, Heuvelink G B, et al. SoilGrids250m: global gridded soil information based on machine learning. PLoS One, 2017, 12 (2): e0169748. doi: 10.1371/journal.pone.0169748
	Jenny H. 1941. Factors of soil formation. a system of quantitative pedology. McGraw-Hill, New York.
	Korouzhdeh T, Eskandari-Naddaf H, Kazemi R. Hybrid artificial neural network with biogeography-based optimization to assess the role of cement fineness on ecological footprint and mechanical properties of cement mortar expose to freezing/thawing. Construction and Building Materials, 2021, 304, 124589. doi: 10.1016/j.conbuildmat.2021.124589
	Leuschner C, Wulf M, Bäuchler P, et al. Soil C and nutrient stores under Scots pine afforestations compared to ancient beech forests in the German Pleistocene: the role of tree species and forest history. Forest Ecology and Management, 2013, 310, 405- 415. doi: 10.1016/j.foreco.2013.08.043
	Li M M, Meador T, Sauheitl L, et al. Substrate quality effects on stabilized soil carbon reverse with depth. Geoderma, 2022, 406, 115511. doi: 10.1016/j.geoderma.2021.115511
	Liu F, Wu H Y, Zhao Y G, et al. Mapping high resolution national soil information grids of China. Science Bulletin, 2022, 67 (3): 328- 340. doi: 10.1016/j.scib.2021.10.013
	Liu L N, Lei Y L. An accurate ecological footprint analysis and prediction for Beijing based on SVM model. Ecological Informatics, 2018, 44, 33- 42. doi: 10.1016/j.ecoinf.2018.01.003
	Lu N, Tian H Q, Fu B J, et al. Biophysical and economic constraints on China’s natural climate solutions. Nature Climate Change, 2022, 12 (9): 847- 853. doi: 10.1038/s41558-022-01432-3
	Ma M, Chang R Y. Temperature drive the altitudinal change in soil carbon and nitrogen of montane forests: implication for global warming. CATENA, 2019, 182, 104126. doi: 10.1016/j.catena.2019.104126
	Ma Y X, Minasny B, McBratney A, et al. 2021. Predicting soil properties in 3D: should depth be a covariate? Geoderma, 383: 114794.
	Malone B P, McBratney A B, Minasny B, et al. Mapping continuous depth functions of soil carbon storage and available water capacity. Geoderma, 2009, 154 (1/2): 138- 152.
	Millard K, Richardson M. On the importance of training data sample selection in random forest image classification: a case study in peatland ecosystem mapping. Remote Sensing, 2015, 7 (7): 8489- 8515. doi: 10.3390/rs70708489
	Mishra U, Riley W J. Alaskan soil carbon stocks: spatial variability and dependence on environmental factors. Biogeosciences, 2012, 9 (9): 3637- 3645. doi: 10.5194/bg-9-3637-2012
	Pedregosa F, Varoquaux G, Gramfort A, et al. Scikit-learn: machine learning in python. Journal of Machine Learning Research, 2011, 12, 2825- 2830.
	Poggio L, De Sousa L M, Batjes N H, et al. SoilGrids 2.0: producing soil information for the globe with quantified spatial uncertainty. Soil, 2021, 7 (1): 217- 240. doi: 10.5194/soil-7-217-2021
	Shangguan W, Dai Y j, Liu B y, et al. A China data set of soil properties for land surface modeling. Journal of Advances in Modeling Earth Systems, 2013, 5 (2): 212- 224. doi: 10.1002/jame.20026
	Shearman T M, Varner J M, Hood S M, et al. 2019. Modelling post-fire tree mortality: can random forest improve discrimination of imbalanced data? Ecological Modelling, 414: 108855.
	Sonali S, Reddy C S, Pasha S V, et al. Modeling the spatial dynamics of deforestation and fragmentation using Multi-Layer Perceptron neural network and landscape fragmentation tool. Ecological Engineering, 2017, 99, 543- 551. doi: 10.1016/j.ecoleng.2016.11.047
	Sparks A. Nasapower: a NASA POWER global meteorology, surface solar energy and climatematology data client for R. Journal of Open Source Software, 2018, 3 (30): 1035.
	Tang X L, Xia M, Pérez-Cruzado C, et al. Spatial distribution of soil organic carbon stock in Moso bamboo forests in subtropical China. Scientific Reports, 2017, 7, 42640. doi: 10.1038/srep42640
	Vaysse K, Lagacherie P. Using quantile regression forest to estimate uncertainty of digital soil mapping products. Geoderma, 2017, 291, 55- 64. doi: 10.1016/j.geoderma.2016.12.017
	Wang C, Xue L, Dong Y, et al. Soil organic carbon fractions, C-cycling hydrolytic enzymes, and microbial carbon metabolism in Chinese fir plantations. Science of the Total Environment, 2021, 758, 143695. doi: 10.1016/j.scitotenv.2020.143695
	Wang T L, Wang G Y, Innes J L, et al. ClimateAP: an application for dynamic local downscaling of historical and future climate data in Asia Pacific. Frontiers of Agricultural Science and Engineering, 2017, 4 (4): 448- 458. doi: 10.15302/J-FASE-2017172
	Whalen E D, Grandy A S, Sokol N W, et al. Clarifying the evidence for microbial‐and plant‐derived soil organic matter, and the path toward a more quantitative understanding. Global Change Biology, 2022, 28 (24): 7167- 7185. doi: 10.1111/gcb.16413
	Xing A J, Du E Z, Shen H H, et al. Nonlinear responses of ecosystem carbon fluxes to nitrogen deposition in an old‐growth boreal forest. Ecology Letters, 2022, 25 (1): 77- 88. doi: 10.1111/ele.13906
	Xiong C, Singh B K, He J Z, et al. Plant developmental stage drives the differentiation in ecological role of the maize microbiome. Microbiome, 2021a, 9 (1): 171. doi: 10.1186/s40168-021-01118-6
	Xiong X, Liu J, Zhou G, et al. Reduced turnover rate of topsoil organic carbon in old-growth forests: a case study in subtropical China. Forest Ecosystems, 2021b, 8, 58. doi: 10.1186/s40663-021-00337-5
	Yang C, Zhang X, Ni H, et al. Soil carbon and associated bacterial community shifts driven by fine root traits along a chronosequence of moso bamboo (Phyllostachys edulis) plantations in subtropical China. Science of the Total Environment, 2021, 752, 142333. doi: 10.1016/j.scitotenv.2020.142333
	Yu W J, Deng Q H, Kang H Z. Long-term continuity of mixed-species broadleaves could reach a synergy between timber production and soil carbon sequestration in subtropical China. Forest Ecology and Management, 2019, 440, 31- 39. doi: 10.1016/j.foreco.2019.03.004
	Zhang Q H, Cao BG, Jiang L. 2002. Development and construction of the 1: 500000 geological map database. Geol. China 2, 208–212.
	Zhang M, Shi W, Ren Y, et al. Proportional allocation with soil depth improved mapping soil organic carbon stocks. Soil and Tillage Research, 2022a, 224, 105519. doi: 10.1016/j.still.2022.105519
	Zhang Q F, Cheng L, Feng J G, et al. Nitrogen addition stimulates priming effect in a subtropical forest soil. Soil Biology and Biochemistry, 2021, 160, 108339. doi: 10.1016/j.soilbio.2021.108339
	Zhang Z X, Hao M, Li Y Q, et al. Effects of vegetation and terrain changes on spatial heterogeneity of soil C–N–P in the coastal zone protected forests at northern China. Journal of Environmental Management, 2022b, 317, 115472. doi: 10.1016/j.jenvman.2022.115472
	Zhou G Y, Xu S, Ciais P, et al. Climate and litter C/N ratio constrain soil organic carbon accumulation. National Science Review, 2019, 6 (4): 746- 757. doi: 10.1093/nsr/nwz045
	Zhou Y, She J Y, Zhu X M. Dynamic analysis of biodiversity, carbon storage and environmental factors of coniferous forest in Loudi City, Hunan Province. International Journal of Low-Carbon Technologies, 2022, 17, 831- 840. doi: 10.1093/ijlct/ctac037

森林类型 Forest types	土壤样点数量(占比) Number of soil sampling plots [percentage (%)]	土层 Soil layers/cm	土壤密度 Bulk density/ (g·cm^?3)	土壤有机碳含量 Soil organic carbon content/ (g·kg^?1)
常绿阔叶混交林 Evergreen broadleaved	128（47.9）	0~20	1.28±0.218	22.32±13.405
		20~40	1.36±0.205	11.46±8.798
		40~60	1.35±0.173	9.57±6.640
落叶阔叶混交林 Deciduous broadleaved	101（37.9）	0~20	1.18±0.229	27.82±17.777
		20~40	1.29±0.224	14.56±11.652
		40~60	1.30±0.215	12.01±9.396
常绿针叶混交林 Evergreen coniferous	38（14.2）	0~20	1.03±0.238	36.41±24.038
		20~40	1.15±0.247	21.45±14.069
		40~60	1.21±0.229	18.75±20.789

变量类型 Variable types	基本变量 Variables	变量简写与常用表达 Abbreviations and common expressions	单位 Units
地理地形因素 Geographic and topographical factors	经度Longitude	Longitude	°E
	纬度Latitude	Latitude	°N
	海拔 Elevation	Elevation	m
	坡度 Slope	Slope	°
	坡向 Aspect	Aspect	°
	景观位置指数Landscape position index	LPI
	高差 Relief	Relief
	海拔高差比率 Elevation relief ratio	ERR
成土因素 Soil forming and lithology factors	母质 Parent material	PM
成土因素 Soil forming and lithology factors	母质中的黏土矿物质比 Clay mineral ratio in parent material	CMR	%
土壤属性 Soil properties	土壤类型 Soil subgroup	Subgroup
	土壤质地 Soil texture	Sand, Clay, Silt	%
	土壤侵蚀程度 Degree of soil erosion	SEr
	土壤密度Bulk density	BD	g·cm^?3
	土壤孔隙度Soil porosity	Spo	%
	表层土壤温度 Surface soil temperature	SST	℃
	表层土壤湿度 Surface soil wetness	SSW	%
	深层土壤湿度 Root zone soil wetness	RZSW	%
植被因素 Vegetation factors	森林类型 Forest type	Forest
	树种结构 Species structure	SpS
	样地平均树高 Mean tree height	H	m
	叶面积指数 Leaf area index	LAI
	均一化植被指数的月均值与年均值 Mean normalized vegetation index	NDVI, NDVI_month
	植被净初级生产力多年均值 Net primary productivity of vegetation	NPP	kg·m^?2a^-1
气候因素 Climatic factors	太阳辐射 Solar radiation	SR	kWh·m^?2
	蒸发量 Evaporation	Evaporation	mm
	地表温度多年均值 Land surface temperature	LST	°C
	月度、生长季、年度气温多年均值 Mean temperature in months, growing seasons and annual mean temperature	MAT, Month_T, GST	°C
	月度、生长季、年度降水多年均值 Monthly, growing season and annual precipitation	MAP, Month_P, GSP	mm
	1—12月月均水分亏损Monthly hargreaves climatic moisture deficit	CMDMonth	mm

模型Models	模型优点Model advantages	重要参数代码 Code of important hyperparameters	解释Descriptions
支持向量机 Support vector machines	处理数据能力强 Good performance for both low and high dimensional data 兼具稳定性和灵活性 High stability and flexibility with large variety of kernel functions	‘kernel’	数据类型和相对转换算法 The type of data and relative transformation
		‘C’	控制数据正态分布的惩罚系数 This penalty parameter controls the regularization of the data
		‘gamma’	决定边界的平滑度 It decides the smoothness of the decision boundary
多层感知器 Multi-layer perceptron	从复杂数据和模糊数据中准确提取信息 The ability to extract patterns from complex or imprecise data	‘hidden_layer_size’	设定隐藏层 Hidden layer sizes
		‘activation’	对应隐藏层函数 Sigmoid function for the relative hidden layer
		‘solver’	权重优化器方法 Weight optimizer method
		‘alpha’	数据正则化强度 Strength of the regularization
		‘max_iter’	最大迭代次数 The maximum number of iterations
随机森林 Random forests	模拟准确性和效率高 High accuracy and efficiency 避免过渡拟合 Prevention of overfitting 适宜大数据量 Being fit for large database	‘n_estimators’	森林决策树子树估算器数量 The number of estimators of subtrees
分位数回归森林Quantile regression forests	优点同随机森林 The same advantages like RF 可以进行不确定性评估 Allowing uncertainty evaluation	‘max_depth’	拆分节点的最大回归深度 The maximum depth of regression to split a node
分位数回归森林Quantile regression forests		‘max_feature’	适用要素的最大特征值 The maximum values of used features

土层 Soil layers/ cm	主要模型参数 Primary modeling parameters		模型评价统计 Statistics of modeling goodness
土层 Soil layers/ cm	数据转换算法 Data transformation	惩罚系数 Penalty	EVS	R²	RMSE	MAE
0~20	线性Linear	1	0.522	0.517	15.309	8.899
20~40	多项式Poly	1	0.485	0.466	8.373	6.088
40~60	多项式Poly	1	0.451	0.449	7.823	5.129

土层 Soil layers/cm	主要模型参数 Primary modeling hyperparameters		模型评价统计 Statistics of modeling goodness
土层 Soil layers/cm	模型优化器 Weight optimizer	数据正则化强度 Strength of the regularization	EVS	R²	RMSE	MAE
0~20	准牛顿算法lbfgs	0.1	0.530	0.496	15.975	9.961
20~40	准牛顿算法lbfgs	0.01	0.531	0.524	7.459	6.206
40~60	准牛顿算法lbfgs	0.1	0.503	0.469	7.020	5.943