نوع مقاله : مقاله پژوهشی
نویسندگان
1 استادیار پژوهش، بخش تحقیقات خاک و آب، مرکز تحقیقات و آموزش کشاورزی و منابع طبیعی استان کرمانشاه، سازمان تحقیقات، آموزش و ترویج
2 بخش تحقیقات حفاظت خاک و آبخیزداری، مرکز تحقیقات کشاورزی و منابع طبیعی کرمانشاه
3 دانشیار موسسه تحقیقات آب و خاک
چکیده
کلیدواژه‌ها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
The digital mapping of sand, silt and clay is used in the design of irrigation systems, land evluation and soil erosion models. In contrast, the quantification of the spatial uncertainty of the mentioned soil properties provides users with a criterion for the reliability of these maps. For this purpose, in this research, two sources of uncertainty, including the uncertainty of variation of available data and the uncertainty of the model, were investigated using the standard deviation criterion. This research performed in an area of 57000 hectares in Ravansar plain of Kermanshah province. 120 observation points and derivatives of DEM and vegetation and parent materials indices obtained from Landsat 8 images were used as model inputs. To estimate the uncertainty caused by the variations in the available data, the random forest model was run 50 times based on the random separation of the data into two categories: training (25%) and validation (75%). Quantile regression forest model was used to predict soil properties and estimate model uncertainty. The results of the validation of the predictions of the random forest model showed that the value of R2 is 0.47, 0.38, and 0.27 for sand, silt, and clay, respectively and the RMSE value was determined for sand (8.61), silt (8.25), and clay (8.1). These results state that the predictor variables explain the sand changes well, but the ability of the random forest model to predict sand is less than the silt and clay variables. Also, the results showed that the uncertainty in variations of available data is about 1/4 the total uncertainty and uncertainty in the model is about 3/4 of the total uncertainty. Since the uncertainty of the model takes a larger share of the total uncertainty, in order to reduce the total uncertainty must be carefully chosen machine learning models.
کلیدواژه‌ها [English]
)