Supplement of Addressing biases in Arctic – boreal carbon cycling in the Community Land Model Version 5

Abstract. The Arctic–boreal zone (ABZ) is experiencing amplified warming, actively
changing biogeochemical cycling of vegetation and soils. The
land-to-atmosphere fluxes of CO2 in the ABZ have the potential to
increase in magnitude and feedback to the climate causing additional large-scale warming. The ability to model and predict this vulnerability is critical
to preparation for a warming world, but Earth system models have biases that
may hinder understanding of the rapidly changing ABZ carbon fluxes. Here we
investigate circumpolar carbon cycling represented by the Community Land Model
5 (CLM5.0) with a focus on seasonal gross primary productivity (GPP) in plant
functional types (PFTs). We benchmark model results using data from satellite
remote sensing products and eddy covariance towers. We find consistent biases
in CLM5.0 relative to observational constraints: (1) the onset of deciduous
plant productivity to be late; (2) the offset of productivity to lag and
remain abnormally high for all PFTs in fall; (3) a high bias of grass, shrub,
and needleleaf evergreen tree productivity; and (4) an underestimation of
productivity of deciduous trees. Based on these biases, we focus on model
development of alternate phenology, photosynthesis schemes, and carbon
allocation parameters at eddy covariance tower sites. Although our
improvements are focused on productivity, our final model recommendation
results in other component CO2 fluxes, e.g., net ecosystem exchange
(NEE) and terrestrial ecosystem respiration (TER), that are more consistent
with observations. Results suggest that algorithms developed for lower
latitudes and more temperate environments can be inaccurate when extrapolated
to the ABZ, and that many land surface models may not accurately represent
carbon cycling and its recent rapid changes in high-latitude ecosystems,
especially when analyzed by individual PFTs.


. CLM5.0 Output During Spin-Up of Model Development. We see an initial spike in GPP at the start of the simulation. Within 20 years, the variability is due to climate forcing from cycling the years 1901 to 1920.
We also find that there is a bug in the computation of the 10-day leaf temperature, where the day time temperature was double counted in Equation 1, and we corrected it to have both T daytime and T nighttime included.
We find that correcting this 10-day leaf temperature calculation, generally increases productivity in CLM5.0 PFTs in the Arctic, but by less than 0.5 gC/m 2 . T leaf,10day is used to calculate the maximum daily change in J max and V cmax by calculating the enzyme turnover rate at a particular temperature. Colder temperatures allow J max and V cmax to change less each time step, and as we noted previously, the default winter predictions are high for the Arctic. Thus, without our modification of average 10 J max and V cmax in winter, this bug fix would increase the high productivity bias.

S4 Improvements to LUNA
The LUNA equations and procedure are described below: J max,opt (V cmax,opt ) is predicted by LUNA as the optimal J max (V cmax,opt ) for the plant, which as the name states is optimal and does not account for limitations on enzyme resources. The maximum change constraint (mxcon) limits the amount of change for J max (and V cmax ) based on the resources available to that plant, which can change every time step. This scheme allows J max and V cmax and thus photosynthesis to be co-limited by resources. We find the change constraint to be a reasonable 20 one to place on J max and V cmax , as it allows for the climatic history on the grid cell to influence the future prediction of leaf photosynthetic traits. However, this scheme for J max and V cmax is only active during the growing season. In winter when LAI=0 and plants are dormant, J max and V cmax are not predicted by LUNA and instead are given a default global place holder value. Thus, at the start of the growing season (or first day of spring), J max,t and V cmax,t are directly calculated from the last day of winter: We find that this global default winter value strongly influences the prediction of J max and V cmax throughout the entire growing season (Supplement Fig. S4). In all of the ABZ PFTs, raising these default values increases mean growing season GPP, whereas decreasing them lowers GPP (Supplement Fig. S4). Furthermore, the constant winter values in Equation 3 30 represent a high bias globally in V cmax (?), contributing additional bias. Due to the sensitivity of this choice and in an effort to leverage the physiological history of a given location, we choose to save the average predictions of J max and V cmax from the previous growing season for all PFTs (J max,prevyr and V cmax,prevyr ). We use these pft specific values to initialize J max and V cmax , such that on the first day of the growing season, J max,pf t,last day of winter = J max,pf t,prevyr

35
V cmax,pf t,last day of winter = V cmax,pf t,prevyr Shrubs at ru-skp tower Leuning Dyn S/L R/L Fluxnet Figure S9. GPP from Intermediate Model Development Steps at Flux Towers for carbon allocation parameters. Dynamic Stem Leaf generally improves the GPP simulation (orange). Sensitivity tests are done using a static value (green) for stem-leaf allocation, but static values here were not clearly supported by observations and did not improve the simulation of shrubs. Observationally based values for root-leaf allocation (R/L= red,green) generally increase productivity at these points. Comparisons are done against previous model development step incorporating the Leuning scheme (blue). The GPP measured (black) at specific flux towers is also included for comparison.