نوع مقاله : پژوهشی
نویسنده
بلوار شهید قدوسی مرکز تحقیقات کشاورزی و منابع طبیعی استان تهران
چکیده
کلیدواژهها
عنوان مقاله [English]
نویسنده [English]
Soil pH is an important soil characteristic that is a measure of soil acidity or alkalinity and profoundly affects nutrient availability and microbial activity, directly affecting plant growth and health. Different crops grow within specific pH ranges, and maintaining an optimal pH level ensures that essential nutrients are readily available to plants. The present study aimed to digitally map soil pH using environmental covariates, Landsat 8 satellite images, and predictive models, and to introduce the best models, in the Badr watershed in southern Qorveh County. To conduct this research, in the first stage, a geomorphological map was drawn using a geological map and based on the Zink geopedology method in a geographic information system environment. In the second stage, the location of 125 study maps was determined based on the Latin supercube technique, and the pH in saturated mud was measured by a pH meter. Auxiliary variables included derivatives of digital elevation model, remote sensing indices received from Landsat 8 satellite and geopedology map, and selection of appropriate auxiliary variables was done using principal component analysis (PCA) method. In the third stage, modeling was performed, digital maps of soil classes and properties were prepared and models were evaluated. Important auxiliary variables in predicting soil pH value in order of importance are: geomorphology, watershed network base level, carbonate index, slope direction, relative slope position, slope length factor (LS factor) and curvature index. pH prediction was done by K-nearest neighbor (KNN), decision tree analysis (DTA), artificial neural network (ANN), random forest (R.F.) and multiple linear regression (MLR) models. Among the models used to predict pH, using the 10-fold cross validation method, the multiple linear regression (MLR) model had the highest prediction accuracy with a coefficient of determination of 0.698 and a root mean square error of 0.190. Based on the 5-fold cross validation methods, the accuracy and precision of the nearest neighbor model performed best in predicting soil pH. In general, based on the 10-fold cross validation and 5-fold cross validation methods, it can be stated that the combined methods have a greater ability to predict pH. This means that combining the prediction results of other models can produce maps with higher accuracy.
کلیدواژهها [English]
)