Warmer spring temperatures in temperate deciduous forests advance the timing of tree growth but have little effect on annual woody productivity

Our finding that interannual variation in woody growth is more strongly linked to conditions during the peak growing season than to growing season length aligns with parallel findings for NEE 13,14 . However, there is also a disconnect with findings that NEE is at least modestly greater in years with warm springs 2 or long growing seasons 4,13,14 . Warming advances spring phenology and may advance or delay autumn senescence depending on timing of warming and water availability 12,34,35 , with delays more common across eastern North America, 2–4 implying that temperatures are lengthening the period from peak stem growth to the cessation of CO 2 uptake by the ecosystem. We show that the extra C fixation in years with warm springs does not substantially augment woody growth, but it remains unclear how it is allocated within the ecosystem. There are two main possibilities, which hold contrasting implications for the response of forest C balance to rising spring temperatures.

Growth milestones for both canopy foliage phenology and stem growth occurred 6-10 days 106 earlier, on average, at SCBI than at Harvard Forest (Fig. 2, Extended Data Table 2) Table 2). Peak growing season 111 length was similar across sites, with being, on average, only two days longer at SCBI for 112 ring-porous species and less than one day longer for diffuse-porous species (Extended Data 113 Table 2). 114 show the dates at which stem growth milestones were achieved, on average, for sampled populations of 120 ring-porous and diffuse-porous trees at SCBI (2011-2020) and Harvard Forest (1998)(1999)(2000)(2001)(2002)(2003). Mean 121 temperature was calculated for each wood-type/site combination over the respective critical T max 122 window, then turned into a ratio and assigned a color on a gradient where the coldest year in the sample 123 is blue and the warmest is red.

125
Both MODIS-derived canopy foliage phenology and dendrometer band measurements of stem 126 growth phenology generally shifted backwards as spring temperatures increased (Fig. 2 Figure 6).
To test whether the negative effect of summer temperatures might offset an enhancement of 171 growth by warmer spring temperatures, we tested for the joint effects of April and June-July 172 on RWI. Results were qualitatively similar to the univariate correlations (Fig. 3), with 173 significant (at p = 0.05) positive correlations to April for only 4% of chronologies and 174 significant negative correlations with June-July for 77% of chronologies, supporting that 175 summer temperatures were the more important driver of annual stem growth (Extended Data 176 Table 3).  forests of eastern North America advance the phenology of tree stem growth but have little 190 effect on annual woody productivity (Figs. 1-3). The observed phenological advance in the start 191 of stem growth under warmer springs parallels phenological advances observed for canopy 192 foliage ( Fig. 2a-b) 2,4,5 and NEE 2,4 . However, inconsistent with the concept that an earlier start to 193 growth would increase annual woody productivity, we demonstrate that warmer springs 194 hasten the cessation of stem expansion and thereby have negligible effect on total annual 195 growth for most species and locations (Fig. 3). Our observations suggest that the cessation of 196 rapid stem expansion, which occurs mid-summer near the time of peak canopy greenness 197 (Extended Data Figure 2) 4 , is likely driven by cues other than photosynthate limitation, such as 198 daylength or sink limitation, which also play an important role in autumn leaf senescence 17,23,31 . 199 Our tree-ring analysis (Fig. 3) demonstrates that the primary effect of warming temperatures on 200 annual tree growth is not an augmentation through an earlier start to growth, but rather a 201 reduction associated to drought stress during the peak growing season 26 . Warm springs may 202 also amplify summer drought stress in some times and places, effectively canceling out any 203 positive effects of an extended growing period 3,32 ; however, spring temperatures and summer 204 Standardized Precipitation Evapotranspiration Index 33 were uncorrelated within our 205 dendrometer band analysis, implying that the effects of warm spring temperatures on growth 206 phenology elucidated here ( Fig. 1) were not attributable to summer drought. 207 Our finding that interannual variation in woody growth is more strongly linked to conditions 208 during the peak growing season than to growing season length aligns with parallel findings for 209 NEE 13,14 . However, there is also a disconnect with findings that NEE is at least modestly greater 210 in years with warm springs 2 or long growing seasons 4,13,14 . Warming advances spring phenology 211 and may advance or delay autumn senescence depending on timing of warming and water 212 availability 12,34,35 , with delays more common across eastern North America, 2-4 implying that 213 warming temperatures are lengthening the period from peak stem growth to the cessation of 214 CO2 uptake by the ecosystem. We show that the extra C fixation in years with warm springs 215 does not substantially augment woody growth, but it remains unclear how it is allocated within 216 the ecosystem. There are two main possibilities, which hold contrasting implications for the 217 response of forest C balance to rising spring temperatures. 218 One possibility is that extra photosynthate in years with warm springs may be allocated to 219 woody growth without affecting diameter growth in the current year. It is theoretically possible 220 that extra C is allocated to cell wall thickening, a process that lags behind stem expansion 36 , or 221 to a thicker layer of higher-density latewood, resulting in formation of more C-dense wood in 222 years with warm springs. However, existing evidence indicates that warm springs have a 223 neutral or negative effect on latewood width 37-39 , which is more strongly controlled by summer 224 drought stress 37,38 , suggesting that a positive effect of warm springs on the total C content of 225 annual rings is unlikely. Extra C could also be saved within trees as non-structural 226 carbohydrates and used towards growth the following year 40,41 , potentially including an earlier 227 start to growth 31 . Extension of our tree-ring analysis revealed weak correlation between April 228 and growth the following year (sig. pos. correlations for 5/142 RP and 3/66 DP species-site combinations, Fig. Extended Data Figure 7), although predominantly positive (non-significant) 230 correlations in RP species suggests that this dynamic may weakly influence their annual 231 growth. Thus, warm springs are unlikely to provide substantial, sustained C sinks under 232 warming spring temperatures. 233 A second possibility is that any additional C fixed during years with warm springs may be 234 allocated to plant functions other than stem growth, including respiration, reproduction, and 235 production of foliage, roots 24 , or root exudates. Much of this C would have a relatively short 236 residence time within the ecosystem, and C loss through fall or winter respiration may offset 237 gains from an earlier spring 3,42 . However, C allocated to nonstructural carbohydrates or 238 relatively short-lived plant tissues would typically remain in the ecosystem beyond the end of 239 the year 40 , such that the long-term effect of warm springs on the forest C balance would not be 240 captured in analyses of interannual variation 2,13,14 . Studies within or including the temperate 241 deciduous biome that examined long-term trends in growing season length and ecosystem C 242 uptake 2,4,10,11 -as opposed to their interannual variation -showed increasing trends in both 243 variables, suggesting that the C not allocated to woody productivity within the current year has 244 a multi-year residence time within the ecosystem. However, given our finding that warm 245 springs do not significantly enhance woody productivity, this C is likely to have a relatively 246 short residence time within the ecosystem. 247 Thus, a distinction between interannual variation and directional change may be critical when 248 considering how directional climate change is likely to affect tree growth and ecosystem C 249 dynamics. As discussed above, temporal lags between C uptake and release imply that the full 250 effects of warm spring temperatures on forest woody productivity and C cycling are unlikely to 251 be apparent in analyses of interannual variation (including this analysis) 43 . Moreover, 252 acclimation of trees to warming temperatures 44 and, on longer time scales, species adaptations 253 and shifts in community composition 45 are likely to alter the phenology of forest C cycling. If we 254 look across spatial gradients where the latter have had time to play out, we see that warmer 255 spring temperatures are associated with earlier leaf-out 6 and longer growing seasons, which in 256 turn are are correlated with greater tree growth 46 , woody productivity 47 , and NEE 48 . Thus, 257 warming spring temperatures are expected to increase the biophysical potential for annual tree 258 growth, but that potential is not being realized on an interannual time frame. 259 As climate change accelerates and spring temperatures become increasingly warmer, growing 260 seasons will start earlier; however, barring rapid acclimation of forests to the warming 261 conditions, an earlier onset of growth in the spring is unlikely to provide the sustained increase 262 in CO2 sequestration and ensuant negative climate change feedback that is anticipated in most 263 climate forecasting models 2,3,17,18 . Rather, the dominant effect of rising temperatures on forest 264 woody productivity will be a negative effect of high summer temperatures, which constitutes a 265 positive feedback to climate change. 266

Dendrometer band analysis 268
Dendrometer band measurements were collected at SCBI 49  The raw dendrometer band data were manually inspected before analysis. We screened the 303 data for three classes of errors. First, when a measurement was drastically different from 304 previous and following measurements, it was assumed to be a human error and the datapoint 305 was removed. Second, when measurements remained essentially unchanged for several 306 readings, followed by a sudden jump then return to a normal growth pattern, this was assumed 307 to be a case where the band was stuck on the tree bark and then released. In these cases, the full 308 annual record for the tree was removed. Third, data points that deviated substantially from 309 normal growth patterns, but for unknown causes, were removed. If a majority of the data points 310 fell into this class within a tree-year, the entire year was removed from the analysis.
We fit a five-parameter logistic growth model 28 to dendrometer band data from each tree-year 312 to define phenological dates and growth rates (Fig. 1). In particular, we model the observed 313 diameter at breast height (DBH) on a given day of the year (DOY; i.e., julian days) as: 314 Here, and are lower and upper asymptotes of the model, corresponding to DBH at the 316 beginning and end of the year, respectively.
is the day of year where the inflection point 317 in growth rate occurs, shapes the slope of the curve at the inflection point, and is a tuning 318 parameter controlling the slope of the curve toward the upper asymptote. The DOY on which 319 maximum growth occurs, (Fig. 1), occurs on when = 1. The model was fit in R 320 v4.0 using the functions developed in the Rdendrom package 28 . These functions take the time-321 series of manual dendrometer band measurements and return maximum-likelihood optimized 322 values of the above five parameters that best predict DBH for each day of year. We then 323 modeled DBH using these optimal parameter values in our logistic growth model and extracted 324 the intra-annual growth variables of interest (Fig. 1). 325 After fitting the growth model, we removed tree-years with poor fits. Models were judged to be 326 poorly fit if certain modeled growth characteristics fell outside of the logical range. Modeled fits 327 for tree-years were removed under five conditions: (1) single day growth rates were ≥ 2 328 standard deviations away from the mean for each wood-type (SCBI = 2, Harvard Forest = 34); 329 (2) was ≥ 2 standard deviations away from the mean for its xylem architecture group, climpact (see www.climpact-sci.org) 54 was used to plot temperatures for visual inspection and to 351 identify readings that were >3 standard deviations away from yearly means, which were labeled as outliers and removed from the dataset. Gaps in the SCBI meteorological tower data 353 were subsequently filled using temperature readings obtained from a National Center for 354 Environmental Information (NCEI) weather station located in Front Royal, Virginia 355 (https://www.ncdc.noaa.gov/cdo-web/datasets/GHCND/stations/GHCND:USC00443229/detail). 356 Daily temperature records for Harvard Forest, which had already been gap-filled based on 357 other local records, were obtained from the Harvard Forest weather station 55,56 . For each site, we 358 used records of daily maximum ( ) and minimum temperatures ( ). 359 The critical temperature window (CTW, Fig. 1 To ensure that patterns were robust under an alternative definition of CTW, and to parallel the 373 monthly time windows used in our tree-ring analysis (detailed below; Fig. 3, Extended Data 374 Figure 6-7), we also ran analyses where we fixed the CTW to be the month of April. This was 375 consistent with the periods identified by climwin for ring-and diffuse-porous species groups at 376 both sites, all of which included all or part of April (Extended Data The tree-ring records from our focal sites were complemented with a much larger collection 407 spanning 106 deciduous and mixed forest sites in Eastern North America 26,65 . Again, records 408 were limited to broadleaf deciduous species with clearly defined xylem porosity (i.e., excluding 409 semi-ring porous). 410 For each species-site combination, we converted tree-ring records into the dimensionless RWI to 411 emphasize interannual variability associated with climate. 66 A 2/3rds n spline was applied to 412 each core using ARSTAN to produce standardized ring-width series; n is the number of years in each series 66,67 . An adaptive power transformation, a process that also stabilises the variance 414 over time 68 , was used to minimize the influence of outliers in all series. Low series replication, 415 often in the earliest portions of a chronology collection, can also inflate the variance of tree-ring 416 records 69 . The 1/3rds spline method was chosen when replication in the inner portion of each 417 chronology (ca. inner 30-50 yr of each record depending on full chronology length) was less 418 than three trees. When replication was greater than n = 3 trees, we used the average correlation 419 between raw ring-width series (rbar) method. The robust biweight mean chronology (RWI) for 420 each species-site combination was calculated from the ring-width indices following variance 421 stabilisation 67 . We defined chronology start year (Extended Data were assessed using 'dplR' 71 and 'bootRes' 72 in R v 4.0 (R Core Team, 2020), which correlated 427 functions and bootstrapped confidence intervals for these relationships 73 . We analyzed these 428 correlations for January through September of the current year (presented in Fig. 3, Extended 429 Data Figure 6). To test for potential lag effects of spring temperatures on growth the following 430 year, we also ran a version of the analysis extending back to include climate of every month of 431 the previous year (Extended Data Figure 7). Correlations and significance levels for months 432 April-August are given in SI Table 1.  433 We used a multivariate model to test for joint effects of April and summer on RWI. We 434 began by testing univariate correlations of over three summer windows: June, June-July, 435 and May-August. Having determined that, among these, June-July explained the most 436 variation, we then analyzed the joint effects of April and June-July on RWI for each 437 chronology independently using the base lm() function in R. Slopes and p-values for each 438 chronology are given in SI Table 1. 439