Pedotransfer函数
导水率
土壤科学
人工神经网络
保水性
土壤质地
参数统计
堆积密度
环境科学
土壤水分
保水曲线
可用水量
含水量
支持向量机
曲线拟合
土工试验
集成学习
均方误差
纹理(宇宙学)
功能(生物学)
试验数据
土壤分类
数学
集合预报
土壤级配
电导率
土壤结构
单调函数
作者
Amir Haghverdi,Mahdi Mohammadnezhad
出处
期刊:Geoderma
[Elsevier BV]
日期:2026-11-01
卷期号:475: 118129-118129
标识
DOI:10.1016/j.geoderma.2026.118129
摘要
Accurate estimation of soil hydraulic properties is essential for precision agriculture, vadose-zone modeling, and site-specific irrigation management, but direct measurement remains time-consuming and expensive. This study introduces NeuroCurve ( https://neurocurve-ptf.streamlit.app/ ), a web-based artificial neural network (ANN) ensemble tool for predicting soil water retention curves (WRCs) and hydraulic conductivity curves (HCCs) from basic soil properties. NeuroCurve was developed using the HYPROP-WP4C dataset of Hohenbrink et al. (2023) and the HYPROP dataset of Schindler and Müller (2017) , integrating measured observations with targeted gap filling to improve representation across the wet-to-dry range of both curves. NeuroCurve was further evaluated using an independent California dataset and compared with five existing pedotransfer function (PTF) tools: KURVE, NeuroFX, SPINN, Legacy Rosetta, and Rosetta3. For the fixed test subset (20 % of the combined development dataset), NeuroCurve with soil texture and bulk density (BD) as inputs achieved root mean square error (RMSE) values of 0.052 cm 3 cm −3 for WRC and 0.502 for logK. For the California soils, NeuroCurve with soil texture and BD as inputs was competitive with other PTF tools for WRC prediction (RMSE = 0.045 cm 3 cm −3 , R 2 = 0.76) and outperformed all other evaluated PTF tools for HCC prediction (logK RMSE = 0.531, R 2 = 0.69). Overall, NeuroCurve advances previous HYPROP-based pseudo-continuous PTFs (PC-PTFs) by using ANN ensembles to simultaneously predict WRCs and HCCs. The results suggest that the relatively simple PC-PTF approach used in NeuroCurve can provide accurate estimates of both hydraulic curves and that increased model complexity does not necessarily translate into improved predictive performance. However, broader independent comparisons across different soil types and regions are needed to determine the relative strengths and limitations of PTF approaches for predicting both water retention and hydraulic conductivity curves.
科研通智能强力驱动
Strongly Powered by AbleSci AI