نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه علوم خاک، دانشکده کشاورزی، دانشگاه بوعلیسینا، همدان، ایران
2 گروه خاکشناسی، دانشکده کشاورزی، دانشگاه بوعلیسینا، همدان، ایران
3 گروه علوم و مهندسی خاک، دانشکده کشاورزی، دانشگاه کردستان، سنندج، ایران.
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Soil water retention characteristics, such as field capacity (FC) and permanent wilting point (PWP), are critical for efficient water management in agriculture. However, direct measurement of these parameters at the field scale is not always practical. This study aimed to estimate FC and PWP using visible-near infrared (Vis-NIR) spectral data combined with soil physicochemical properties through random forest (RF) and multiple linear regression (MLR) models. A total of 130 soil samples were collected from five provinces in Iran, and their spectral and soil properties were measured. The dataset was divided into training (90 samples) and testing (40 samples) sets, and 11 pedotransfer functions (PTFs) were developed in three steps. To improve model performance, in addition to the no-preprocessing (NP) method, multiplicative scatter correction (MSC), first and second derivatives with Savitzky–Golay filtering (FD-SG, FD-SG2), and standard normal variate (SNV) were applied to the spectral data prior to model development. The results showed that the RF model (RMSE = 0.050) outperformed MLR (RMSE= 0.057) during the training stage. However, in the testing stage, no statistically significant difference was observed between the two methods in estimating FC. In contrast, for PWP estimation, RF generally yielded better results than MLR across most functions; however, these differences were not statistically significant, except for PTF2 (AIC = −264.3), where a significant difference was observed. During the training stage, PTF11 exhibited the best performance for FC estimation (AIC= −540.2), while PTF7 showed the highest performance for PWP estimation (AIC = −612.4). PTF3, incorporating sand, clay, and organic matter as input variables, was identified as the most accurate estimator of FC (AIC = −553.3). Similarly, PTF6, using sand, clay, organic matter, and total porosity, was identified as the most effective estimator of PWP (AIC = −616.2). Principal component analysis identified key wavelengths at 409 nm for FC and 1414, 1912, and 2150 nm for PWP. Overall, soil-property-based PTFs outperformed spectral-only models, but combining spectral and soil data with machine learning improved prediction accuracy.
کلیدواژهها [English]
Soil moisture is one of the key factors influencing plant growth and development and is recognized as a fundamental driving force for the sustainable development of many terrestrial ecosystems. The importance of this factor has become even more pronounced under current conditions, where global water challenges and climate change have cast a shadow over the sustainability of natural resources. Changes in soil moisture can have significant effects on vegetation cover and on the physical and chemical properties of soils. The soil water retention curve (SWRC) describes the relationship between soil moisture content and matric suction. Among the most important points on the SWRC are the soil moisture contents at field capacity (FC) and at the permanent wilting point (PWP). The soil's water-holding capacity reflects the availability of soil water and the distribution of soil pore sizes, which are of great importance for studying soil water storage, conservation, movement, and supply. However, due to the difficulty, time-consuming nature, and high cost of direct measurement of these properties, the development of indirect and rapid methods for estimating these parameters—particularly through the use of pedotransfer functions (PTFs) and modern visible–near infrared (Vis–NIR) spectroscopy—has become both a scientific and practical necessity. Therefore, the objectives of this study were as follows: 1- Estimation of soil hydraulic parameters (FC and PWP) using a combination of soil physico-chemical properties and spectral reflectance in the visible-near infrared (Vis-NIR) region, and development and validation of RF and MLR models to enhance the accuracy and generalization of predictions across soils with diverse textures. 2- Evaluation of the effects of spectral preprocessing methods (MSC, FD-SG1, FD-SG2, and SNV) on improving the accuracy of models estimating FC and PWP parameters. 3- Identification of key wavelengths and soil variables influencing the variability of FC and PWP through Principal Component Analysis (PCA) and selected pedotransfer functions.
In this study, a total of 130 disturbed and undisturbed soil samples were collected from five provinces of Iran to estimate the FC and PWP parameters. The physical, chemical, and hydraulic properties of the soils were measured, and their spectral reflectance was recorded using a spectroradiometer. The soil spectral curves in the visible to near-infrared range (350-2500 nm) were measured under standardized spectroscopic conditions in a darkroom environment. To improve the estimation accuracy, several spectral preprocessing techniques were applied to the reflectance data, including Multiplicative Scatter Correction (MSC), first derivative Savitzky–Golay filtering (FD-SG1), second derivative Savitzky–Golay filtering (FD-SG2), and Standard Normal Variate (SNV). Principal Component Analysis (PCA) was employed to reduce the dimensionality of the spectral data and to extract the most relevant features. For the development of pedotransfer functions (PTFs), eleven models based on different algorithms were constructed in three stages using various combinations of input variables to predict FC and PWP parameters. The datasets were used to develop regression- and Random Forest-based PTFs using STATISTICA 14 software. The entire dataset, consisting of both input and output variables, was divided into two subsets: 90 samples were randomly selected for model training and 40 samples for testing. This process was repeated ten times, and a new model was executed for each iteration. The mean results of the ten runs were reported as the final outcomes. Finally, the PTFs were developed using Multiple Linear Regression (MLR) and Random Forest (RF) methods to estimate the FC and PWP parameters.
The results indicated that the Random Forest (RF) model outperformed the Multiple Linear Regression (MLR) model during the training phase. Among the soil and spectral datasets, the pedotransfer functions developed in the second stage (PTF3), which included sand, clay, and organic matter as input variables, performed as the best estimators of FC, showing a 3.5% improvement compared to the baseline model. Moreover, PTF6, which used sand, clay, organic matter, and total porosity as input variables, was identified as the most effective function for estimating PWP, with a 5% improvement over the baseline model. Comparative analysis among the PTFs in the third stage revealed that PTF11 (SNV) for FC and PTF7 (NP) for PWP exhibited the best performance during the training phase, significantly reducing the Akaike Information Criterion (AIC) values. Furthermore, Principal Component Analysis (PCA) identified key wavelengths at 1414, 1912, and 2150 nm for PWP, and at 409 nm for FC.
Overall, integrating spectral and soil data with machine learning algorithms, especially random forest, offers a more accurate and cost-effective approach compared to existing methods for estimating FC and PWP parameters and can significantly contribute to improving soil management.
The official and fluent translation of your text into English is as follows This study was conducted with the financial and moral support of the Vice-Chancellery for Research of Bu-Ali Sina University, Hamedan. Financial support for this research was provided by Bu-Ali Sina University, Faculty of Agriculture, through a student thesis grant for the first author and research grants for the other authors.
The authors participated in all stages of research design and execution, statistical data analysis, analysis and interpretation of information and results, report preparation, manuscript drafting, results review and control, correction, revision, and finalization of the manuscript, and their roles in order of contribution are as follows: Chiman Mahdizadeh, Hossein Bayat and Masoud Davari.”
During the preparation of this work the author(s) used ChatGPT in order to assist with translation and language refinement. After using this tool, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.
The data from the present study are available from the authors upon reasonable request.
The authors would like to express their sincere appreciation to the Bu-Ali Sina University, Hamedan, for its financial and moral support of the present study.
The authors are grateful to the anonymous reviewers for their constructive scientific and structural comments.
This study did not involve human participants or animals, and no sensitive data were used. The authors followed standard research ethics.
The authors declare no conflict of interest.
)