عنوان مقاله [English]
Background and objectives: Soil quality is one of the most important soil properties which investigation of it's changes is essential to soil management and degradation. Quantifying soil quality using soil quality index to improve understanding of soil ecosystems is have been wieldy used. The soil quality index is calculated by measuring some soil characteristics which measuring these properties is expensive and time consuming. Therefore, one of the solutions is the use of digital soil mapping technique that can digitally predict soil properties using auxiliary data and data mining models. The purpose of this research is using a random forest model and auxiliary data for mapping the soil quality index.
Materials and Methods: Based on the geomorphology map, 17 soil profiles and 105 auger samples were taken from a depth of 0-20 cm in the Ghorveh area of Kurdistan Province ( covers 6500 ha) and soil texture, organic carbon, cation exchange capacity, electrical conductivity, pH, carbonate calcium equivalent, total nitrogen, available phosphorus, microbial respiration rate, sodium adsorption ratio (SAR), bulk density, and gravel percentage were measured and calculated then the soil additive weighted index was calculated. Environmental variables in this research were map geomorphology, terrain attributes and data of ETM+ image. Geomorphology map was prepared based on zinc method. Terrain attributes (including 10 parameters), soil adjust vegetative index index (SAVI), and normalized difference vegetative index (NDVI), and brightness index (BI) were computed and extracted using SAGA and Arc GIS software, respectively. To make a relationship between soil quality index and auxiliary data, random forest (RF) model were applied and using cross validation method and statistic criteria including coefficient of determination (R2), mean error (ME) and root mean square error (RMSE) was validated.
Results and Discussion: According to the communality (share of each soil indicator), bulk density, sand, cation exchange capacity and clay had the highest weight (≥ 0.1) and gravel and SAR had the lowest weight (≤0.05) among the soil quality properties. To predict soil quality index, auxiliary variables including slope, SAVI index, wetness index, MrVBF index, LS factor, elevation, NDVI index and geomorphology map were the most important. The results of this study showed that the random forest model with 0.65, 0.042 and 0.062 for determination of coefficient (R2), mean error (ME), and root mean square root (RMSE) had a fairly suitable accuracy for prediction of soil quality index. The soil quality index was ranged between 0.3-0.65 and its mean values in geomorphologic units with low gradient and low soil depth (Mo131, Mo141 and Hi231) were the lowest and in geomorphologic units with low slope and high soil depth (Pi111, Pi311, Pi322, Pi211 and Pi312) were the highest which these differences were statistically significant.
Conclusion: In this research, a randomized forest model was used to study the spatial variation of soil quality index in Ghorveh area of Kurdistan province. The geomorphologic conditions of the study area have affected many soil characteristics and subsequently the soil quality index in the region. The soil quality index content was the lowest in highlands of north, northwest and northeast with high slope and low soil depth. The slope was the most important auxiliary variables to predict soil quality index in the region. Based on the results of statistical indices, random forest model also had relatively accurate estimation of the soil quality index. Therefore, it is suggested to map soil properties podometric techniques (such as randomized forest) and auxiliary data such as geomorphologic map, terrain attributes, and satellite images were applied.