Abstract
Soil quality, encompassing biotic and abiotic properties, is a key contributor to forest health. Wood harvesting can impact soil quality by altering nutrient availability, pH, moisture, and the soil microbiome. However, it is unclear whether wood harvesting affects abiotic and biotic soil properties similarly across different sites. Identifying robust indicators of wood harvesting can help develop improved forest recovery monitoring and management strategies. Machine learning (ML) models are beneficial for monitoring and identifying indicators of forest recovery, but presently no model has been developed using soil quality to assess forest perturbation and recovery following wood harvesting. To address these gaps, this study developed a Random Forest (RF) model using abiotic soil properties and microbial taxa to predict the level of organic matter (OM) removal in BC forests and identify which soil properties and taxa are the most important predictors. We used soil samples collected from 6 BC Long-Term Soil Productivity (LTSP) program sites, with data collected for 16S ribosomal RNA sequences, as well as the abiotic properties. These samples were taken from sites subjected to different severities of OM removal that correspond to common wood harvesting methods, along with unharvested reference sites. Genera associated with OM removal were identified and used with abiotic soil properties to develop RF models that predicted OM removal with 78–89% accuracy. Abiotic factors, specifically carbon content and pH, were more important for predicting OM removal than microbial genera. Amongst microbial genera, Actinophytocola, Inquilinus, and Puia were most important for predicting OM removal.