Konferans bildirisi Açık Erişim

Digital mapping of soil copper contamination: comparison of gaussian process regression and random forest models

2021    Keshavarzi, Ali; Ertunç, Güneş; Kaya, Fuat

Copper (Cu) is a trace element that is important for humans, plants, animals, and micro-organism health. It can be presented in many proteins and enzymes. Soils with low bioavailability of Cu can lead to crop yield losses and symptoms of deficiencies in livestock, especially in intensive farming systems. Digital Soil Mapping (DSM) can be characterized as the development of spatial soil information systems through numerical models inferring the spatial and temporal variations of soil types and soil properties from soil observation and knowledge, as well as related environmental variables. The current study was designed to map and foresee the Cu content by two different data mining techniques: (i) Gaussian Process Regression (GP), and (ii) Random Forest (RF). Collected data (620 observations) from the Tellus project in Northern Ireland were used to develop the models. Some soil characteristics (i.e., pH and Phosphorus), DEM-based topographical attributes, and remotely sensed data (Landsat 8 Satellite Imagery) such as Topographic Wetness Index (TWI), Valley Depth, Band 1, Band 5, and Band 9 were entered as input parameters for prediction and mapping of soil Cu. To evaluate the performance of two various techniques used in this study, statistical indexes including the correlation coefficient (CC) and root mean square error (RMSE) were assessed through 10-fold cross-validation mode. The results of the two models suggested that the RF model with higher CC (0.664) and lower RMSE (11.678 mg/kg) is the best one for the appraisal of soil Cu. This study revealed that machine learning models can be successfully implemented to the rapid prediction and mapping of soil heavy metals using soil properties, topographical attributes, and remotely sensed data. Digital maps may be used to prioritize remediation steps and can be applied to other areas with similar environmental conditions and sources of pollution.