نوع مقاله : مقاله پژوهشی
نویسندگان
1 استادیار پژوهش، مرکز تحقیقات و آموزش کشاورزی و منابع طبیعی خراسان رضوی
2 عضو هیئت علمی مرکز ملی تحقیقات شوری، سازمان تحقیقات، آموزش و ترویج کشاورزی، یزد، ایران
3 عضو هیات علمی مرکز ملی تحقیقات شوری، سازمان تحقیقات، آموزش و ترویج کشاورزی، یزد، ایران.
4 دانش آموخته کارشناسی ارشد باغبانی، دانشگاه ایالتی کالیفرنیا، دیویس، یزد، ایران
5 مرکز ملی تحقیقات شوری، سازمان تحقیقات، آموزش و ترویج کشاورزی ، یزد، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
A total of 720 soil samples collected from 240 sampling locations within pistachio orchards in the provinces of Yazd and Khorasan Razavi were used for training and testing the Random Forest algorithm. The auxiliary variables used in this modeling included the median values of 32 surface reflectance variables and spectral indices derived from Sentinel-2 satellite images, covering the period from March 1 to October 1 of the sampling year. These variables were extracted for the sampling points using the Google Earth Engine platform. The Random Forest model was developed and optimized through coding in the R environment. The Leave-One-Out Cross-Validation (LOOCV) method was used to validate the model. After identifying and removing outliers, 191 points remained for retraining and retesting the Random Forest algorithm. The RMSE was 1.1 dS/m in the training set and 2.6 dS/m in the test set. The R² value was 93% in both datasets. This algorithm was then used to predict changes in soil salinity in the studied areas and years. Based on the results, in the Tanour Lahour drainage project, Class 3 of salinity is decreasing while Class 2 land area is increasing. In the Mortazieh farm, Class 3 salinity has decreased but shifted to Class 4. In the Rezaei farm, Class 4 land area is significantly increasing. Likewise, in the Dadyar farm in Khorasan Razavi Province, Class 4 salinity area has relatively increased. The results of this study demonstrated that the Random Forest algorithm is capable of successfully predicting soil salinity changes within the study areas and similar regions.
کلیدواژهها [English]
Given the limitations of traditional soil salinity assessment methods, alternative techniques such as remote sensing (RS) have been used to predict soil salinity and sodicity in unsampled areas. Vegetation indices (VIs), as indirect indicators of salinity, have been used to assess soil salinity through their negative effects on crop growth and plant stress. On the other hand, salinity indices, which serve as direct indicators of salinity, highlight the spectral reflectance of salt crusts on the soil surface, especially in the visible and near-infrared (NIR) regions of the electromagnetic spectrum. The scientific literature includes many successful estimations of soil properties using various soil sensing technologies. Despite significant advances in developing soil salinity prediction models, such models have not been widely developed for pistachio orchards, which typically lack full vegetation cover. Therefore, this study aimed to develop a Random Forest model to predict soil salinity using auxiliary variables derived from satellite imagery.
For this study, 669 soil samples collected from 223 sampling locations within four large pistachio-growing areas in Yazd and Khorasan Razavi provinces were used to train and test the Random Forest algorithm. At each sampling location, soil was collected down to a depth of 90 cm in 30 cm increments. The ground-truth data corresponded to the years 2018, 2022, and 2023. To train the Random Forest algorithm, spectral reflectance values and vegetation indices derived from Sentinel-2 satellite images were used. The surface reflectance values and spectral indices for the sampling points were extracted using the Google Earth Engine (GEE) platform. The Random Forest algorithm was trained, tested, and validated using code written in the R programming environment. The spectral indices used in this study were important variables commonly used as auxiliary inputs for soil salinity modeling.
According to the correlation matrix, although there was no strong correlation between the main variable (soil salinity) and individual auxiliary variables, some of the auxiliary variables showed positive or negative correlations with one another. Due to multicollinearity among the auxiliary variables, stepwise regression models could not be used to determine the relationship between soil salinity and these variables. The Random Forest model was trained using 500 decision trees. The RMSE of the model was 1.10 dS/m for the training set and 2.59 dS/m for the testing set. The R² value was 0.932 for both sets. The spectral indices used in this study are related to vegetation stress conditions and thus can also reflect plant performance under stress. The algorithm was used to predict changes in soil salinity across the study regions and years. The results showed that in the Tanour Lahour drainage project, the area of Class 3 salinity land is decreasing while Class 2 land is increasing. In Mortazieh farm, Class 3 salinity has declined but shifted toward Class 4. In Rezaei farm, Class 4 salinity has increased significantly. Similarly, in Dadyar farm in Khorasan Razavi province, Class 4 salinity area has relatively increased. These findings indicate that the Random Forest algorithm is capable of successfully predicting soil salinity changes in the study areas and similar regions.
The results of this study demonstrated that the Random Forest model, utilizing variables extracted from Sentinel-2 satellite imagery, has a high capability for predicting soil salinity changes in pistachio orchards in Yazd and Khorasan Razavi provinces. The spatial analysis of salinity changes also showed that in certain areas, such as the Tanour Lahour drainage project, salinity has decreased, indicating the positive effects of management practices. In contrast, in areas like Rezaei farm, the increasing trend of salinity is alarming. These results highlight the importance of continuous monitoring and the use of advanced tools such as the integration of remote sensing data and machine learning algorithms for effective soil salinity management. This approach can also be applied in other saline agricultural regions to prevent further soil degradation and yield reduction.
Conceptualization, Y. H. and F. D.; methodology, Y.H. and H.B.; software, Y.H.; validation, H. B., M. S. and M.M.; formal analysis, Y.H.; investigation, Y.H., F.D. and H.B.; resources, F.D. and M.M.; data curation, Y.H.; writing—original draft preparation, Y.H.; writing—review and editing, Y.H., F.D. and H.B.; visualization, Y.H.; supervision, Y.H.; project administration, Y.H.. All authors have read and agreed to the published version of the manuscript.
Not applicable.
The authors would like to thank the National Salinity Research Center for support of the present study.
The authors avoided data fabrication, falsification, plagiarism, and misconduct.
The author declares no conflict of interest.
)