Determining TTOP model parameter importance and overall performance across northern Canada

Abstract

Modelling current permafrost distribution and response to a changing climate depends on understanding which factors most strongly control ground temperatures. The Temperature at the Top of Permafrost (TTOP) model provides an analytical framework for estimating permafrost presence and thermal state, yet its sensitivity to key parameters remains poorly quantified across diverse northern environments. This study evaluates the relative influence of TTOP model parameters using ground and air temperature data from 330 sites across northern Canada. A leave-one-out cross-validation approach to determine model sensitivity was combined with random forest analysis to rank variable importance. Results show that TTOP performance is dominated by freezing-season conditions – particularly the freezing nfactor and freezing degree days – while thaw-season parameters exert less control. Sensitivity varies by region, with thawing parameters becoming more influential where the duration of the freezing and thawing seasons is similar. Machine learning results also highlighted the importance of thermal offset and mean surface temperatures which are strongly influenced by substrate properties. While the model generally reproduces observed ground temperatures well (RMSE of 0.2 °C), parameters derived from landcover classes were not transferable between sites, underscoring the importance of locally calibrated inputs. Overall, this study is the first empirically-based Canada-wide assessment of how different climatic and environmental factors affect the accuracy of permafrost temperature modelling and provides practical guidance for improving parameterization in regional and global permafrost models.