Prediction of soil salinity using tree regression and artificial neural network in Ghorveh soils, Kurdistan Province

Soil salinity is one of the major problems in arid and semi-arid area. In this condition, soluble salts accumulate in the soil surface and reduce yield and soil fertility. Soils survey and mapping can help to improve these soils. The investigation of variability of soil salinity using traditional methods is expensive and time consuming. Therefore, one of the ways to solve this challenge is using digital soil mapping that soil characteristics were mapped using auxiliary data. The aim of this research is using tree regression (TR) and artificial neural network (ANN) models and auxiliary data to prepare soil salinity map.
Using Hypercube soil sampling method, 100 soil samples in depths 0-30 cm of Ghorveh soils, Kurdistan Province (covers 30000 ha) were taken and soil electrical conductivity was measured. Auxiliary data in this study were terrain attributes and Landsat 8 ETM data. Terrain parameters (include 15 parameters) and salinity index (SI) and normalized difference vegetative index (NDVI) were computed and extracted using SAGA and ArcGIS software, respectively. To make a relationship between soil salinity and auxiliary data, TR and ANN models were applied and were validated using cross validation method. Finally, soil salinity map were made using better model.
To predict soil salinity, auxiliary variables include salinity index, wetness index, index of valley bottom flatness, NDVI index, Band 3, and Band 7 were the most important. The results of the study showed that ANN model (0.70, 0.036 and 0.190, respectively for determination of coefficient, mean error, and root mean square root) has more accuracy compared to TR model to predict soil salinity. Soil salinity content ranged between 0.23 to 6.93 dSm1 and the highest content of soil salinity located in central regions (lowland and bare land). In these central regions, auxiliary data include salinity index, index of valley bottom flatness, wetness index, band 7 and band 3 had the highest values and NDVI index had the lowest values.
Salinity index is the most important auxiliary data to predict soil salinity of the study area. Strong link between soil data and auxiliary data can impact on the accuracy of the model. In general, the results showed that pedometrics techniques in a wide range can be used for digital mapping of soil properties. It is suggested ANN model and auxiliary data such as terrain attributes and satellite images were applied to prepare map of soil properties in future studies.