Soil physicochemical properties and terrain information predict soil enzymes activity in phytophysiognomies of the Quadrilatero Ferrifero region in Brazil

Soil enzymes act in biogeochemical cycles of elements and are indicators of soil quality since they rapidly reflect changes of the environmental conditions. Moreover, enzymes are related to soil physicochemical properties, but their spatial distribution has been rarely evaluated. The hypothesis of this work is that soil properties related to fertility and texture (F), total contents of chemical elements obtained by portable X-ray fluorescence (pX) spectrometry and terrain attributes (T) can be used as predictor variables to soil enzyme activity, along with phytophysiognomy and season information. The objective of this work was to predict soil enzymes activity and assess its spatial variability in the most common phytophysiognomies of the Quadrilatero Ferrifero mineral province, in Brazil. Soil samples were collected in four phytophysiognomies during both dry and humid seasons. Activity of beta-glucosidase, acid phosphatase, alkaline phosphatase, urease, and hydrolysis of fluorescein diacetate (FDA) was determined. Phytophysiognomy, season, F, T, and pX, were used together or separately to predict the enzymes activity through conditional random forest algorithm and the accuracy was assessed via leave-one-out cross validation. The generated models were accurate, with coefficient of determination (R-2) varying from 0.63 (FDA by pX) to 0.82 (beta-glucosidase by F). F variables were more important for the predictions, while pX variables were more important for predicting acid phosphatase and urease. The accurate models using T variables allowed the generation of maps showing the enzymes variability along the phytophysiognomies. This approach can accelerate the determination of soil enzymes activity across the landscape.

Automated summary

This study predicted soil enzyme activity using various predictor variables and successfully generated accurate models. The results indicate that F variables are more important for predicting enzyme activity, while pX variables also play a role in predicting enzyme activity. Models using T variables allowed for the generation of maps showing spatial variability of enzyme activity.