Resolving temperature limitation on spring productivity in an 1 evergreen conifer forest using a model-data fusion framework 2 3

24 The flow of carbon through terrestrial ecosystems and the response to climate is a critical but highly uncertain 25 process in the global carbon cycle. However, with a rapidly expanding array of in situ and satellite data, there is an 26 opportunity to improve our mechanistic understanding of the carbon (C) cycle’s response to land use and climate 27 change. Uncertainty in temperature limitation on productivity poses a significant challenge to predicting the 28 response of ecosystem carbon fluxes to a changing climate. Here we diagnose and quantitatively resolve 29 environmental limitations on growing season onset of gross primary production (GPP) using nearly two decades of 30 meteorological and C flux data (2000-2018) at a subalpine evergreen forest in Colorado, USA. We implement the 31 CARDAMOM model-data fusion network to resolve the temperature sensitivity of spring GPP. To capture a GPP 32 temperature limitation—a critical component of integrated sensitivity of GPP to temperature—we introduced a cold 33 temperature scaling function in CARDAMOM to regulate photosynthetic productivity. We found that GPP was 34 gradually inhibited at temperature below 6.0 °C (± 2.6 °C) and completely inhibited below -7.1 °C (± 1.1 °C). The 35 addition of this scaling factor improved the model’s ability to replicate spring GPP at interannual and decadal time 36 scales (r = 0.88), relative to the nominal CARDAMOM configuration (r = 0.47), and improved spring GPP model 37 predictability outside of the data assimilation training period (r = 0.88) . While cold temperature limitation has an 38 important influence on spring GPP, it does not have a significant impact on integrated growing season GPP, 39 revealing that other environmental controls, such as precipitation, play a more important role in annual productivity. 40


43
Northern hemisphere evergreen forests contribute significantly to terrestrial carbon (C) storage and exchange 44 (Beer et al., 2010;Thurner et al., 2014). High-latitude and high-elevation evergreen forests show increasing gross 45 primary productivity (GPP) with increasing temperature driven in large part by earlier growing seasons (Myneni et  influence the timing and magnitude of GPP (Bowling et al., 2018). In particular, warmer springs can also lead to 50 earlier snowmelt, which can reduce spring C uptake through increased surface exposure to colder ablation-period air 51 temperatures (Winchell et al., 2016), and can reduce summer C uptake via drought (Hu et al., 2010). Many 52 subalpine forests in western North America are also highly water limited, with warming and earlier snow melt 53 creating accumulated water deficits, increased drought stress, and growing season C uptake losses (Wolf et al.,    validation against subsets of data. We also leverage a recent model intercomparison study (Parazoo et al., 2020), 99 driven by site level meteorological data at US-NR1, to provide a model benchmark assessment, and extract any 100 common environmental controls on modeled GPP. Finally, we examine whether using a decade of flux tower-101 derived GPP observations to train the model is sufficient to match and predict seasonal to annual patterns in GPP.

102
Given the complexity of carbon-water cycle interactions during the growing (summer) season in this highly water 103 limited ecosystem, and the relatively weak correlation between tower-derived spring and summer GPP (r = -0.31, p 104 = 0.20), we focus on spring GPP-temperature interactions, with the aim to resolve just one piece of the larger, 105 complex problem of understanding changes in C uptake in a subalpine evergreen ecosystem.   Turnipseed et al., 2004). Average annual precipitation is 800 mm, with a majority of precipitation falling in 115 the winter as snow (Greenland, 1989;Knowles et al., 2015), which creates a persistent winter snowpack from     (Williams et al., 1997), that are produced from a sensitivity analysis of GPP estimates from 145 5 the more comprehensive SPA land surface model scheme (Williams et al., 1996, Williams et al., 2001. ACM GPP 146 estimates are contingent on plant structural and biochemical variables (including LAI, foliar nitrogen and nitrogen-147 use efficiency) and meteorological forcing (total daily irradiance, maximum and minimum daily air temperature, day 148 length, atmospheric CO2 concentration). In DALEC2, water limitation on ACM is prescribed as a linear response to 149 soil water deficit (Bloom et al., 2020). For more details on the model-data fusion methodology and CARDAMOM

177
We also find that cold temperature has an important limitation on seasonal GPP at US-NR1. The seasonal 178 cycle of GPP shows peak productivity in early summer (around June) and falling to near-zero values by early winter 179 (November), continuing through late winter (February-March). Comparison of monthly GPP and minimum, 180 maximum, and mean monthly air temperature shows an initiation of photosynthesis at monthly maximum air 181 temperature above 0 °C (Fig. 3a) and monthly minimum air temperature above -5 °C (Fig. 3b). The strong 182 dependence of monthly GPP on temperature is consistent with previous findings that temperature is an important

195
In the baseline version of CARDAMOM, seasonal GPP in DALEC2 is limited primarily by incoming 196 shortwave radiation. This light-focused limitation works well for deciduous forests where spring temperature and 197 sunlight are correlated, as well as high latitude regions where sunlight is limited. However, for reasons discussed 198 above, this method fails in evergreen forests such as Niwot Ridge whose green canopies are exposed to high sunlight 199 and below-freezing temperature in spring. As temperature increases, evergreen stems slowly thaw, which enables 200 the trees to access available soil moisture and slowly reactivate their carbon and water exchange processes (Mayr et

212
To represent the integrated impact of the cold weather processes, here we implement a cold temperature 213 scaling factor (g) in DALEC2. This scaling factor is developed by analyzing the relationship between monthly 214 minimum & maximum air temperature with tower-derived monthly GPP, where Tmin(t) is the observed minimum air temperature at Niwot Ridge at time t, GPP(t) is the nominal ACM-based 220 DALEC2 GPP estimate (see section 2.3) and GPPcold is the corresponding cold temperature GPP estimate. Equation

221
(2) may represent the integrated effect of all cold weather biophysical limitations, including processes such as the 222 impact of cold weather on plant hydraulics, and changes to carotenoid-chlorophyll ratios. We also theorize that our 223 temperature scaling factor partially captures soil moisture disruptions due to changing soil temperature. The

300
The cold experiments exhibit an improved fit to the observed IAV in spring productivity (Fig. 5), relative to 301 CARD, (r = 0.47, std = 0.03 g C m -2 day -1 for CARD; r = 0.88, std = 0.27 g C m -2 day -1 for CARDcold). CARDcold 302 also has slightly reduced RMSE (-0.01 g C m -2 day -1 ) and larger MBE (0.13 g C m -2 day -1 ). Similar to the seasonal  Table S2). Although the cold temperature limitation 311 improves IAV slightly, it is still small compared to observed variability (mean annual std = 0.14 g C m -2 day -1 ).

312
Correlations to tower-derived GPP at the annual scale are small for both CARD and CARDcold (r = 0.19 and r = 313 0.22, Fig. S7a-b). Overall, the cold temperature limitation substantially improves agreement between the model and

322
The standard deviation in tower-derived mean spring GPP (March-May) is approximately 0.25 g C m -2 day -323 1 . The addition of the cold temperature limitation improves the model's ability to match the IAV of mean spring 324 GPP ( Fig. 6a-b). An examination of all modeled scenarios for CARD and CARDcold (i.e. all 4000 DALEC2

378
With the inclusion of the cold temperature limitation on GPP and its application in CARDAMOM, we 379 provide a data-constrained estimate of the climate sensitivity of the Niwot Ridge forest to spring temperature.

380
Posterior estimates indicate that GPP is gradually inhibited below 6.0 °C ± 2.6 °C (Tg) and completely inhibited year (e.g., Xu et al., 2016). However, these spring gains in GPP have been shown to not offset the losses of carbon 395 due to summer droughts (e.g., Buermann et al., 2013;Knowles et al., 2018). It is also unclear how the long-term 396 stress of increased temperature could affect forest productivity directly.

397
This study focuses on the relationship between temperature and GPP and its usefulness on model versus rainfall affects productivity, or how resulting changes to winter snowpack could alter productivity long-term.

409
Since annual average GPP appears to be more dependent on winter precipitation/snowpack (Pearson's linear r =   (Table S1). CLM4.5 also shows the smallest

449
There is also large variability in the modeled seasonal cycle (Fig. 7b) and mean annual GPP (Fig. S9). For 450 mean annual GPP estimates, Pearson's r values are reduced for all models (Table S2). Once again, ORCHIDEE-451 exp2 and ORCHIDEE-exp3 stand out with some of the higher correlations (r = 0.60 and r = 0.64) and p-values 452 below 5% significance level. Furthermore, ORCHIDEE-exp3 (temperature stress with SIF data assimilation) has 453 the lowest RMSE and MBE of the model set. SiB3-exp2 (fixed LAI) has a standard deviation closest to 454 "observations" (0.14 gC m -2 day -1 ), and the smallest RMSE and MBE of the TBM models.

497
These differences could be due to variations in other variables (e.g., soil temperature, irradiance, etc.) and/or 498 physiological differences in the vegetation species. Identifying how photosynthesis temperature thresholds vary 499 across space and ecosystem type would be beneficial in improving model performance in simulating productivity.

500
Our model intercomparison study also provide insights on how we may improve our ability to model seasonal GPP.

501
For example, in Fig. 7b

525
Joint assimilation of these datasets, coupled with observed meteorological forcing, has potential to introduce more