| Abstract: | Reforestation of agricultural lands is an important means of restoring land and sequestering carbon (C). At large scales, the labour and costs of direct measurement of ecosystem responses can be prohibitive, making the development of models valuable. Here, we develop a new sampling scenario‐based modelling approach coupled with Bayesian model averaging to build predictive models for absolute values in mixed‐species woody plantings and differences from their adjacent pasture, for litter stocks, soil C stocks and soil C:N ratios. Modelling scenarios of increasing data availability and effort were tested. These included variables that could be derived without a site visit (e.g. location, climate and management) that were sampled in the adjacent pasture (e.g. soil C and nutrients) or were sampled in the environmental planting (e.g. vegetation, litter properties, soil C and nutrients). The predictive power of models varied considerably among C variables (litter stocks, soil C stocks and soil C:N ratios in tree plantings and their differences to their adjacent pastures) and the model scenarios used. The use of a sampling scenario‐based approach to building predictive models shows promise for monitoring changes in tree plantings, following reforestation. The approach could also be readily adapted to other contexts where sampling effort for predictor variables in models is a major potential limitation to model utilization. This study demonstrates the benefit of exploring scenarios of data availability during modelling and will be especially valuable where the sampling effort differs greatly among variables. Copyright © 2016 John Wiley & Sons, Ltd. |
