[1] LIU Yifei, NI Junjun, SADEGHI Hamed,. Optimized TabPFN-gradient boosting multimodel ensemble strategy for predicting soil saturated hydraulic conductivity [J]. Journal of Southeast University (English Edition), 2026, 42 (3): 318-328. [doi:10.3969/j.issn.1003-7985.2026.03.005]

Copy

Optimized TabPFN-gradient boosting multimodel ensemble strategy for predicting soil saturated hydraulic conductivity()

预测土饱和导水率的优化TabPFN-梯度提升多模型集成策略

Journal of Southeast University (English Edition)[ISSN:1003-7985/CN:32-1325/N]

Volumn:

42

Issue:

2026 3

Page:

318-328

Research Field:

Civil Engineering

Publishing date:

2026-09-06

Info

Title:

Optimized TabPFN-gradient boosting multimodel ensemble strategy for predicting soil saturated hydraulic conductivity

预测土饱和导水率的优化TabPFN-梯度提升多模型集成策略

Author(s):

LIU Yifei¹, NI Junjun¹, SADEGHI Hamed²

1.School of Transportation, Southeast University, Nanjing 211189, China
2.Department of Civil Engineering, Sharif University of Technology, Tehran 145888 9694, Iran

刘翼飞¹, 倪钧钧¹, SADEGHI Hamed²

1.东南大学交通学院, 南京 211189
2.Department of Civil Engineering, Sharif University of Technology, Tehran 145888 9694, Iran

Keywords:

soil saturated hydraulic conductivity;  machine learning;  tabular prior-data fitted network (TabPFN);  gradient boosting;  ensemble strategy

土壤饱和导水率;  机器学习;  TabPFN;  梯度提升;  集成策略

PACS:

TU42

DOI:

10.3969/j.issn.1003-7985.2026.03.005

Abstract:

Soil saturated hydraulic conductivity is a critical hydraulic parameter in geotechnical engineering and hydrology. Traditional experimental measurement methods are time-consuming and labor-intensive, making it difficult to meet the rapid application requirements of engineering practice. This study constructs prediction models based on easily measurable basic soil properties, using particle composition percentages and bulk density as input features to predict saturated hydraulic conductivity. The study proposes an optimized tabular prior-data fitted network (TabPFN) mo-del that integrates an expert mechanism with a two-stage optimization strategy and compares its performance with traditional gradient boosting models and random forest (RF). A multimodel stratified weighted ensemble strategy is constructed to synthesize the advantages of all models. The results show that the expert mechanism and the two-stage optimization strategy significantly improve TabPFN performance. The optimized TabPFN mo-del achieved a maximum prediction accuracy of R² = 0.956 8 and root mean square error (RMSE) of 0.001 1 cm/s, representing a 16.78% improvement in R² over the preoptimization mo-del and outperforming all gradient boosting and RF models. The stratified weighted ensemble strategy effectively integrated the prediction advantages of each sub-model across different soil textures, achieving the highest prediction accuracy (R² = 0.961 7, RMSE = 0.001 0 cm/s) and effectively reducing prediction errors and heteroscedasticity. The proposed optimized ensemble mo-del provides a new intelligent method for the rapid estimation of saturated hydraulic conductivity in engineering practice.

土的饱和导水率是岩土工程和水文学中的关键水力参数,传统试验测定方法耗时费力,难以满足工程实践中快速应用的需求。本文基于易于测得的土壤基本性质构建预测模型,以颗粒组成质量分数和容重作为输入特征,对饱和导水率进行预测。提出了一种优化的表格先验数据拟合网络(TabPFN)模型,该模型融合了专家机制和两阶段优化策略,并将其性能与传统梯度提升模型和随机森林模型进行了比较。构建了多模型分层加权集成策略,以综合各类模型的优势。结果表明,专家机制和两阶段优化策略显著提升了TabPFN的性能。优化后的TabPFN模型最高预测精度R²为0.956 8,RMSE为0.001 1 cm/s;与优化前模型相比,R²提高了16.78%,且优于所有梯度提升模型和随机森林模型。分层加权集成策略能够有效整合各子模型在不同土壤质地条件下的预测优势,取得了最高的预测精度(R²为0.961 7,RMSE为0.001 0 cm/s),并有效降低了预测误差和异方差性。本文提出的优化集成模型为工程实践中快速估算饱和导水率提供了一种新的智能化方法。

References:

[1]GARCÍA-GUTIÉRREZ C, PACHEPSKY Y, MARTÍN M Á. Technical note: Saturated hydraulic conductivity and textural heterogeneity of soils[J]. Hydrology and Earth System Sciences, 2018, 22(7): 3923-3932.
[2]CHAPUIS R P. Predicting the saturated hydraulic conductivity of soils: A review[J]. Bulletin of Engineering Geology and the Environment, 2012, 71(3): 401-434.
[3]DEXTER A R, CZYŻ E A, GAŢE O P. Soil structure and the saturated hydraulic conductivity of subsoils[J]. Soil and Tillage Research, 2004, 79(2): 185-189.
[4]QI L, LI X, PAN Y, et al. Impacts of forest fire on hydro-sediment using the SWAT mo-del: A case study in the Zhangweinan Canal, China[J]. Acta Geophysica, 2025, 73(5): 4761-4780.
[5]WANG X H, MA C, WANG Y Q, et al. Effect of root architecture on rainfall threshold for slope stability: Variabilities in saturated hydraulic conductivity and strength of root-soil composite[J]. Landslides, 2020, 17(8): 1965-1977.
[6]DOS SANTOS R C V, VARGAS M M, TIMM L C, et al. Examining the implications of spatial variability of saturated soil hydraulic conductivity on direct surface runoff hydrographs[J]. Catena, 2021, 207: 105693.
[7]ZHAO Y L, WANG Y Q, ZHANG X C. Spatial and temporal variation in soil bulk density and saturated hydraulic conductivity and its influencing factors along a 500 km transect[J]. Catena, 2021, 207: 105592.
[8]LIU J Y, WANG X, REN X W. Hydraulic conductivity and particle size of soils: Modeling and experiment[J]. Acta Geotechnica, 2025, 20(7): 3395-3409.
[9]MCDOWELL G, DE BONO J. Relating hydraulic conductivity to particle size using DEM[J]. International Journal of Geomechanics, 2021, 21: 06020034.
[10]PHAM K, WON J. Enhancing the tree-boosting-based pedotransfer function for saturated hydraulic conductivity using data preprocessing and predictor importance using game theory[J]. Geoderma, 2022, 420: 115864.
[11]ZHANG Y G, SCHAAP M G. Estimation of saturated hydraulic conductivity with pedotransfer functions: A review[J]. Journal of Hydrology, 2019, 575: 1011-1030.
[12]REHMAN M, HAFEEZ M B, KRAWCZUK M. A comprehensive review: Applications of the Kozeny-Carman mo-del in engineering with permeability dynamics[J]. Archives of Computational Methods in Engineering, 2024, 31(7): 3843-3855.
[13]DIAZ-GONZALEZ F A, VUELVAS J, CORREA C A, et al. Machine learning and remote sensing techniques applied to estimate soil indicators—Review[J]. Ecological Indicators, 2022, 135: 108517.
[14]LIU Y F, NI J J, ZHANG F, et al. Bayesian-informed DNN and ensemble learning for predicting soil water characteristic curves from easily measurable parameters[J]. Transportation Geotechnics, 2026, 56: 101791.
[15]DONG S, RUI X F, ZUO C. Research progress on intelligent detection and analysis algorithms for pavement damage[J]. Journal of Southeast University (Natural Science Edition), 2025, 55(4): 1131-1153. (in Chinese)
[16]SONG Y, WANG Y. Multi-frame radar HRRP target recognition using MFA-Net[J]. Journal of Southeast University (English Edition), 2025, 41(3): 384-391.
[17]SINGH V K, PANDA K C, SAGAR A, et al. Novel genetic algorithm (GA) based hybrid machine learning-pedotransfer function (ML-PTF) for prediction of spatial pattern of saturated hydraulic conductivity[J]. Engineering Applications of Computational Fluid Mechanics, 2022, 16(1): 1082-1099.
[18]ALBALASMEH A, MOHAWESH O, GHARAIBEH M, et al. Artificial neural network optimization to predict saturated hydraulic conductivity in arid and semi-arid regions[J]. Catena, 2022, 217: 106459.
[19]RUNDEL D, KOBIALKA J, VON CRAILSHEIM C, et al. Interpretable machine learning for TabPFN[C]//Explainable Artificial Intelligence. Cham: Springer, 2024: 465-476.
[20]DONG X B, YU Z W, CAO W M, et al. A survey on ensemble learning[J]. Frontiers of Computer Science, 2020, 14(2): 241-258.
[21]GUPTA S, HENGL T, LEHMANN P, et al. SoilKsatDB: Global database of soil saturated hydraulic conductivity measurements for geoscience applications[J]. Earth System Science Data, 2021, 13(4): 1593-1612.
[22]NATEKIN A, KNOLL A. Gradient boosting machines, a tutorial[J]. Frontiers in Neurorobotics, 2013, 7: 21.
[23]HU S Y, XIE Z N, YANG Y. Interference wind pressure prediction of high-rise buildings with square section based on machine learning[J]. Journal of Southeast University (Natural Science Edition), 2024, 54(6): 1425-1437. (in Chinese)
[24]CHEN R, CHEN H, FAN Z H, et al. Experimental study on the unsaturated shear strength of stabilized aeolian sand considering geopolymer content[J]. Journal of Southeast University (Natural Science Edition), 2026, 56(2): 302-311. (in Chinese)
[25]MOZAFFARI H, PAKJOO M, NEMATOLLAHI M A, et al. Predicting soil hydraulic conductivity: A review of artificial neural networks applications[M]//Artificial Intelligence Applications for a Sustainable Environment. Cham: Springer, 2025: 441-462.

Memo

Memo:

Received: 2025-11-17; Revised: 2026-05-12.
Biographies: LIU Yifei (1997—), male, Ph.D.candidate; NI Junjun (corresponding author), male, doctor, professor, nijunjun@seu.edu.cn.
Foundation items: The National Natural Science Foundation of China (No.U2340227, 52308342), Fundamental Research Funds for the Central Universities (No.RF1028623071, 2242024k30066), Graduate Research and Practice Innovation Program of Jiangsu Province (No.SJCX25_0128).
Citation: LIU Yifei, NI Junjun, SADEGHI Hamed. Optimized TabPFN-gradient boosting multimodel ensemble strategy for predicting soil saturated hydraulic conductivity[J]. Journal of Southeast University (English Edition), 2026, 42(3): 318-328. DOI: 10. 3969/j. issn. 1003-7985. 2026. 03. 005.

Common functions