Calibrating the soil organic carbon model Yasso20 with multiple datasets

Toni Viskari,Anna Repo,Julius Vira,Jari Liski,Janne Pusa,Istem Fer

doi:10.5194/gmd-15-1735-2022

Abstract

Abstract. Soil organic carbon (SOC) models are important tools for assessing global SOC distributions and how carbon stocks are affected by climate change. Their performances, however, are affected by data and methods used to calibrate them. Here we study how a new version of the Yasso SOC model, here named Yasso20, performs if calibrated individually or with multiple datasets and how the chosen calibration method affects the parameter estimation. We also compare Yasso20 to the previous version of the Yasso model. We found that when calibrated with multiple datasets, the model showed a better global performance compared to a single-dataset calibration. Furthermore, our results show that more advanced calibration algorithms should be used for SOC models due to multiple local maxima in the likelihood space. The comparison showed that the resulting model performed better with the validation data than the previous version of Yasso.

Highlights

Soils are the second-largest global carbon pool, and even small changes in this pool impact the global carbon cycle (Peng et al, 2008)
This is indicative of what would happen if a simple single-chain calibration was done with soil organic carbon (SOC) models
Soil organic carbon (SOC) models should be constrained by data from multiple different ecosystems reflecting the various dynamics affecting the SOC decomposition process

Summary

Introduction

Soils are the second-largest global carbon pool, and even small changes in this pool impact the global carbon cycle (Peng et al, 2008). More complicated SOC models addressing these arguments have been developed, for example Millennial (Abramoff et al, 2018), and modules including additional drivers affecting the C pools have been included in existing SOC models, such as nitrogen (Zaehle and Friend, 2010) and phosphorus (Davies et al, 2016; Goll et al, 2017) cycles. Their implementation is hindered, though, by the fact that detailed data are needed to constrain the model parameterization, but individual measurement campaign datasets are often limited in size and lacking in nuance of the SOC state (Wutzler and Reichstein, 2007; Palosuo et al, 2012). The chosen calibration methodology is affected by the same issues based on its approach of fitting the data

Methods

Results

Discussion

Conclusion