Data-mining analysis of the global distribution of soil carbon in observational databases and Earth system models

Shoji Hashimoto,Aleksi Lehtonen,Boris Ťupek,Kazuki Nanko

doi:10.5194/gmd-10-1321-2017

Abstract

Abstract. Future climate change will dramatically change the carbon balance in the soil, and this change will affect the terrestrial carbon stock and the climate itself. Earth system models (ESMs) are used to understand the current climate and to project future climate conditions, but the soil organic carbon (SOC) stock simulated by ESMs and those of observational databases are not well correlated when the two are compared at fine grid scales. However, the specific key processes and factors, as well as the relationships among these factors that govern the SOC stock, remain unclear; the inclusion of such missing information would improve the agreement between modeled and observational data. In this study, we sought to identify the influential factors that govern global SOC distribution in observational databases, as well as those simulated by ESMs. We used a data-mining (machine-learning) (boosted regression trees – BRT) scheme to identify the factors affecting the SOC stock. We applied BRT scheme to three observational databases and 15 ESM outputs from the fifth phase of the Coupled Model Intercomparison Project (CMIP5) and examined the effects of 13 variables/factors categorized into five groups (climate, soil property, topography, vegetation, and land-use history). Globally, the contributions of mean annual temperature, clay content, carbon-to-nitrogen (CN) ratio, wetland ratio, and land cover were high in observational databases, whereas the contributions of the mean annual temperature, land cover, and net primary productivity (NPP) were predominant in the SOC distribution in ESMs. A comparison of the influential factors at a global scale revealed that the most distinct differences between the SOCs from the observational databases and ESMs were the low clay content and CN ratio contributions, and the high NPP contribution in the ESMs. The results of this study will aid in identifying the causes of the current mismatches between observational SOC databases and ESM outputs and improve the modeling of terrestrial carbon dynamics in ESMs. This study also reveals how a data-mining algorithm can be used to assess model outputs.

Highlights

Soil is the largest organic carbon stock in terrestrial ecosystems (Batjes, 1996; IPCC, 2013; Köchy et al, 2015)
Observational estimations of soil organic carbon (SOC) are still under development and have significant uncertainty, the consistency between observational SOC database results and Earth system models (ESMs) outputs will enhance our confidence in predicting SOC dynamics under climate change
The same data-mining Boosted regression trees (BRT) algorithm was applied to observational databases of SOC stocks and ESM outputs

Summary

Introduction

Soil is the largest organic carbon stock in terrestrial ecosystems (Batjes, 1996; IPCC, 2013; Köchy et al, 2015). The soil organic carbon (SOC) stock represents a balance between carbon inputs to soil and carbon losses from soil via decomposition and dissolved organic carbon, and this influx and efflux of soil carbon is controlled directly and indirectly by environmental conditions (Carvalhais et al, 2014; Schimel et al, 1994). Earth system models (ESMs) were developed to understand the current climate and provide future climate projections, and these models incorporate the terrestrial carbon cycle, including SOC (Arora et al, 2013; Friedlingstein et al, 2014). In ecosystem carbon cycle models of ESMs, SOC is calculated as the balance between carbon inputs via dead organic matter and carbon emissions via organic matter decomposition, with both processes influenced by temperature and water conditions.

Methods

Results

Discussion

Conclusion