A Comparison of Spatial Interpolation Approaches for Mapping Habitat and Peatland Ecohydrological Metrics

Abstract

Accurate spatial interpolation of peat thickness and surface elevation are essential for estimating carbon storage and habitat mapping for species at risk. The objectives of this research were to (a) assess three interpolation methods’ (Universal Kriging, Inverse Distance Weighting (IDW) and Random Forest (RF) machine learning) ability to predict peat thickness and surface elevation, (b) compare method performance when creating derivative surfaces for habitat mapping and carbon metrics, and (c) make recommendations on spatial interpolation workflows when assessing peatland ecohydrological metrics and habitat mapping. Methods were applied in three peatlands in eastern Georgian Bay, Ontario, Canada. Kriging and IDW produced similar spatial patterns for habitat mapping and carbon metrics; although Kriging achieved slightly higher predictive accuracy, all three methods produced ecologically negligible differences. Surfaces generated using RF had comparable accuracy to IDW and Kriging but lower R² values, likely reflecting limitations in predictor variables and the absence of explicit spatial autocorrelation modeling. However, the RF method holds potential for extrapolating peat thickness to unsampled areas at the landscape scale. For small (≤ 1 ha) basin-confined peatlands with representative field data, IDW offers a practical alternative to Kriging; although Kriging remains preferable when computational resources and variogram modelling expertise are available. We suggest that RF methods require further refinement to enable reliable interpolation of ecohydrological variables within small, basin-confined peatlands. Our findings demonstrate that choice of interpolation method can influence habitat estimates but provide similar carbon estimates, highlighting the importance of robust spatial modelling for informing peatland conservation in a changing climate.