Wed, Aug 17, 2022: 4:15 PM-4:30 PM
513B
Background/Question/MethodsThe spatial variability of soil enzymes is very important for assessing soil fertility conditions, enrichment of chemical fertilizer, and crop yield management. This research was conducted with the aim of spatial modeling and predicting soil urease and phosphatase enzymes in southwest Iran. For this aim, a number of 60 topsoil samples (0-30 cm) were collected and then transferred to the laboratory. After field and laboratory analysis, a set of freely and easily accessible environmental covariates including remote sensing indices and topographic attributes were extracted from the satellite images and digital elevation model. Then random forest (RF), as a popular machine learning model, was used to estimate the activities of soil urease and phosphatase using environmental covariates and soil properties. Variance inflation factor (VIF) and Pearson regression were applied to select the best environmental covariates and decrease the model complexity..
Results/ConclusionsThe RF model indicated that soil organic carbon (SOC) and Band 6 had the highest rank among other environmental covariate in the prediction of soil urease and phosphatase activities, respectively. The RF model showed similar performance in prediction of soil urease and phosphatase activities (R2=0.54), while in the case of error indices (NMAE and NRMSE) urease has a lower amount in comparison to phosphatase. Our findings could provide a reference for studying the relationships between soil extracellular enzyme activities and environmental covariates in areas with similar environmental conditions. Generally, modeling and spatial prediction of these two soil biological properties act as natural indicators. The low values of them indicate the presence of the high amount of chemical compounds such as ammonium and phosphate, which enter the food cycles, and Underground water sources can have nutritional and environmental health risks.Keywords: Random forest, soil-forming factors, soil enzymes activity, Spatial variability
Results/ConclusionsThe RF model indicated that soil organic carbon (SOC) and Band 6 had the highest rank among other environmental covariate in the prediction of soil urease and phosphatase activities, respectively. The RF model showed similar performance in prediction of soil urease and phosphatase activities (R2=0.54), while in the case of error indices (NMAE and NRMSE) urease has a lower amount in comparison to phosphatase. Our findings could provide a reference for studying the relationships between soil extracellular enzyme activities and environmental covariates in areas with similar environmental conditions. Generally, modeling and spatial prediction of these two soil biological properties act as natural indicators. The low values of them indicate the presence of the high amount of chemical compounds such as ammonium and phosphate, which enter the food cycles, and Underground water sources can have nutritional and environmental health risks.Keywords: Random forest, soil-forming factors, soil enzymes activity, Spatial variability