Plain Language Summary
In the Arctic, winter warming events (WWEs) are episodes of exceptionally warm weather that last from hours to a few days and often occur in combination with rainfall. The combined effect of multiple processes triggered by WWEs (involving, for example, changes in snow depth and snow properties, and heat exchanges between air, rain and meltwater, and soils) results in profound changes in ground temperatures and water fluxes, which many ecosystem processes depend on (vegetation dynamics, organic matter decomposition by microbes, etc). However, large-scale ecosystem models, which are used to study how arctic ecosystems will change in the future, often overlook the effects of WWEs because they oversimplify some complex physical processes and operate at longer temporal scales than WWEs. WWEs will likely become more frequent and intense in the future as the climate continues to warm. This study used a widely used ecosystem model, LPJ-GUESS, to investigate how the ecosystems will respond to more frequent and intense WWEs. The ecosystem responses that we observed were notable but in the opposite direction as observed in field measurements. We identified the processes that are lacking in the model and are causing this mismatch which, if implemented in the model, would significantly improve the predictions of future ecosystem changes in response to climate change.
1 Introduction
The Arctic is warming three times faster than the global average, with the strongest warming occuring in autumn and winter (AMAP 2021). The occurrence and impacts of winter warming events (WWEs), i.e. short-lasting extraordinarily warm spells, often accompanied by rainfall (rain on snow; ROS), are increasing rapidly and expanding geographically (e.g. Bartsch et al., 2010; Vikhamar-Schuler et al., 2016; Pan et al., 2018). Climate models predict a further increase in the coming decades. Despite their short duration, these extreme events could cause societal and environmental impacts that can override the impacts of long-term climatic trends (e.g., Phoenix & Bjerke, 2016).
In the winter, the impacts of WWEs are currently considered a research priority for better understanding future ecosystem dynamics in the subarctic (Pascual et al., 2020). WWEs can affect ground temperatures (GT) in multiple ways, mostly through: (1) direct heat transfer from the air, (2) latent heat release from refreezing melt and rainwater, and (3) changing the snowpack properties, such as depth and density, which influence the energy exchanges between the atmosphere, snow, and soil. These altered winter processes can further influence ground albedo and groundwater content (GWC), which has impacts lasting to the growing seasons (Pascual & Johansson, 2022). These WWE-associated environmental changes can further alter microbial activity, greenhouse gas (GHG) emissions (e.g., Natali et al., 2019), and permafrost and vegetation dynamics (Bruhwiler et al., 2021), ultimately altering the arctic carbon budget.
These interlinked processes and feedbacks related to WWEs are difficult to disentangle in observational data and thus challenging to implement in models. Moreover, large-scale ecosystem models which run with monthly climate data do not explicitly account for the impacts of such stochastic climate extremes (e.g. Tang et al., 2015).
In this study, we investigate the potential effects of predicted WWE scenarios on winter and growing-season physical and biogeochemical variables using the latest version of a widely used dynamic ecosystem model, LPJ-GUESS. We aim to evaluate the model’s performance and identify model gaps in representing ecosystem responses to future WWEs.
2 Materials and Methods
2.1 Study sites
The Torneträsk area, in northern Sweden, is a topographically heterogeneous area that aligns along a strong west-east oceanic-continental climatic gradient, with precipitation and winter temperature decreasing eastwards due to the increasing continentality and the rain shadow effect caused by the Scandes Mountains. The area has experienced rapid climate warming (Callaghan et al., 2010) and a substantial increase in the frequency and intensity of WWEs (Pascual & Johansson, 2022).
Vegetation in the area varies following its climatic and altitudinal gradients. Birch forests occur below c. 600 and 800 m.a.s.l (Van Bogaert, 2010). Tundra species dominate above the tree-line, while in the lowlands, birch forests alternate with peat plateaus underlain by permafrost, and non-permafrost fens.
In this modelling study, we selected four sites representing these dominant ecosystem types in the Torneträsk area, including (1) a birch forest (~370 m.a.s.l) located <10 km east of the Abisko Station (ANS) (Heliasz et al., 2012); (2) a tundra site (~410 m.a.s.l), located <1 km to the southeast of the ANS (Michelsen et al., 2012 and references therein); (3) a peat plateau (~380 m.a.s.l) known as Storflaket, located c. 6 km east of the ANS (Johansson et al., 2013), and (4) a fen (~515 m.a.s.l) located c. 25 km west of the ANS, near the Katterjokk Station (SMHI). The dominant vegetation species at these sites are found in Appendix A.
2.2 Model description and simulation setup
2.2.1 Model description
The Lund-Potsdam-Jena General Ecosystem Simulator (LPJ-GUESS) is a process-based dynamic ecosystem model widely used on regional and global scale studies (Smith et al., 2001; 2014). The model simulates vegetation dynamics (including vegetation establishment, mortality and competition, etc.), water, carbon and nitrogen cycles, and soil biogeochemistry. This study used the latest version of LPJ-GUESS (version 4.1, Smith et al., 2014), with the recently-developed dynamic, intermediate complexity snow scheme enabling the simulation of climate-snow-soil interaction. The model can simulate up to five snow layers, their physical and thermal properties, and their development throughout the cold season. Based on the individual snow layer properties (e.g., temperature, density, thermal conductivity), freeze-thaw processes in snow layers, and heat transport through the snowpack between the atmosphere and soil can be simulated, and ROS events can be accounted for (Pongracz et al., 2021). LPJ-GUESS includes detailed representations of permafrost and wetland processes, including peatland hydrology, peatland-specific PFTs, and CH4 emissions (see Wania et al., 2009a, 2009b, 2010).
2.2.2 Model setup
Daily climate data provided by the ANS (ANS 2020) and Katterjokk Station (SMHI), including air temperature, air temperature daily range, and precipitation, together with shortwave radiation (1913-1984, Sheffield et al., 2006; 1984-2018, ANS) and annual CO2concentrations obtained from the Global Monitoring Laboratory (https://gml.noaa.gov/ccgg/trends/), were used to drive the model from 1913 to 2018. Soil property data was extracted from the WISE5min, V1.2 Soil Property Database (Batjes 2012). More details about the setup and input data are found in Appendix B.
Representative plant functional types (PFTs) were selected for each site (Table A1). We enabled high-latitude and wetland-specific plant functional types (PFTs) in the simulations to better capture site-specific vegetation conditions (see Wania et al., 2009 for more details). The PFT parameters at each site followed previous studies (e.g., Tang et al., 2015; Gustafson et al., 2021) (Table A2).
2.2.3 Model calibration and evaluation
Sobol sensitivity analyses (SA) were conducted to explore the influence of different parameters and parameter interactions on the estimated seasonal snow density, snow depth, snow temperature, and GT at the study sites (except for the tundra site due to lack of observational data for calibration and evaluation) (Appendix D). A sampling of eight relevant parameters, using ranges based on literature values, and a certain percentage of changes from the original values (Table D1), was conducted. Among the most influential parameters, we selected the parameter values that minimized the absolute differences between the measured and the modeled seasonal snow depth and GT at each site (2006-2012 at the peat plateau, 2001-2010 elsewhere) (Figures D4-8). The model was subsequently evaluated with independent observational data (2011-2018) when possible (Appendix E).
2.2.4 Model simulations with future WWEs
We generated manipulation experiments with different levels of WWE frequencies and intensities imposed on the observation-based climate inputs (HISTORICAL dataset) to assess the responses of different ecosystem processes to these extreme events. The applied frequencies and intensities were based on different climate scenarios in the Coupled Model Intercomparison Project Phase 6 (CMIP6) (Eyring et al., 2016). We selected six climate scenarios from two general circulation models (GCMs) with different climate sensitivities, i.e., CanESM5 and GFDL-ESM4, and three shared socioeconomic pathways representing three levels of varying GHG projections, i.e., SSP119, SSP270, and SSP585. For each scenario (n=6), daily meteorological data (1950-2100) for the gridcell near the Torneträsk area was extracted, and then bias-corrected at a daily scale. Detailed descriptions of the CMIP6 scenarios and the bias-correction method are found in Appendix C.
The bias-corrected GCM’s outputs were used to create the manipulation experiments (Table 1). In addition to the HISTORICAL runs (S0), we designed three additional experiments in which the future monthly anomalies in the frequency and intensity of melt days (S1), ROS (S2), and both (S3) in the GCM’s outputs (Table C2) were added to the HISTORICAL climate inputs, maintaining the long-term climate means as unchanged as possible. These anomalies were calculated based on four indices modified from Vikhamar et al. (2016) (Table C1).
The effects of altered WWEs and those from the altered long-term climate trends were compared by using an additional experiment (S4) in which the future winter monthly anomalies (Table C3) were directly added to the historical daily air temperature and precipitation.
Table 1 . Description of the HISTORICAL and MANIPULATION runs.