Topoclimate effects on growing season length and montane conifer growth in complex terrain

Spatial variability in the topoclimate-driven linkage between forest phenology and tree growth in complex terrain is poorly understood, limiting our understanding of how ecosystems function as a whole. To characterize the influence of topoclimate on phenology and growth, we determined the start, end, and length of the growing season (GSstart, GSend, and GSL, respectively) using the correlation between transpiration and evaporative demand, measured with sapflow. We then compared these metrics with stem relative basal area increment (relative BAI) at seven sites among elevation and aspects in a Colorado montane forest. As elevation increased, we found shorter GSL (−50 d km−1) due to later GSstart (40 d km−1) and earlier GSend (−10 d km−1). North-facing sites had a 21 d shorter GSL than south-facing sites at similar elevations (i.e. equal to 200 m elevation difference on a given aspect). Growing season length was positively correlated with relative BAI, explaining 83% of the variance. This study shows that topography exerts strong environmental controls on GSL and thus forest growth. Given the climate-related dependencies of these controls, the results presented here have important implications for ecosystem responses to changes in climate and highlight the need for improved phenology representation in complex terrain.


Introduction
Improved understanding of spatially variable ecosystem functioning in topographically complex regions is essential given that complex terrain yields > 60% of total carbon uptake in the western United States, covers > 50% of the land surface globally, and is a source of uncertainty in global carbon cycling estimates (Rotach et al 2014, 2008, Schimel et al 2002. Much of this spatial variability is due to topographic-driven gradients in temperature and vegetation structure that result in a heterogeneous patchwork of topoclimates that can be decoupled from the mean regional climate (Daly et al 2010, Dobrowski 2011, Novick et al 2016. As a result, the timing of phenological events (e.g. onset of spring growth) can vary among topoclimates and influence tree growth and ecosystem productivity (Churkina et al 2005, Rossi et al 2013. However, studies that include direct measurements of spatial phenological variation in complex terrain are few despite the known impacts of phenology on regional and global climate, and forest growth and susceptibility to disturbance (Bonan 2008, Dale et al 2001, Kramer et al 2000, Peñuelas and Filella 2009.
The influence of topography on ecosystem and critical zone functioning has become the focus of renewed interest and several dendrochronological studies have identified an effect of topoclimate on tree growth and the location of tree line and ecotone boundaries (Körner 2012, Adams et al 2014. Tree growth rates are, in general, inversely proportional to elevation due to decreasing temperatures at higher elevations (Körner andPaulsen 2004, Bruening et al 2017). Slope and aspect effects on solar insolation can interact with elevation changes to affect site temperature and moisture regimes as well, resulting in high spatial variability in the factors that limit growth (e.g. energy, moisture) ( Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. on growing season length (GSL) has received much attention, and GSL is commonly used as a predictor of tree growth and forest productivity in complex terrain Körner 2014, Bruening et al 2017). However, GSL is typically determined as a function of air temperature alone (often averaged over periods > 1 d); an approach that likely neglects topographic and short-term effects on forest-and species-level physiological activity and phenological transitions. Given that GSL is often used to describe vegetation growth across topographic boundaries (Churkina et al 2005, Rossi et al 2013, Paulsen and Körner 2014, the impact of topography on GSL and forest growth represents a substantial knowledge-gap in the current understanding of forest ecosystem function. Measuring spatially explicit topo-climatic controls on forest productivity in complex terrain is difficult and new procedures to characterize spatially-explicit physiological activity are needed. One potential solution is to use spatially-distributed sapflow measurements of tree transpiration (Burgess et al 2001). Transpiration is an inevitable consequence of carbon uptake for photosynthesis (Collatz et al 1991, Farquhar and Sharkey 1982, Jarvis and Davies 1998 and may subsequently serve as an indicator of active canopy gas exchange. Because conifer gas exchange is well-coupled to atmospheric demand (Jarvis and McNaughton 1986), the relationship between vapor pressure deficit (VPD) and sapflow can be used as a metric for physiological activity while intrinsically accounting for the effects of temperature and moisture availability. Temporal dynamics of physiological activity can be used to identify phenological phase shifts such as the transition from dormant to active periods and vice-versa. These phenology metrics can then be compared with measurements of radial tree growth rates to clarify the relationship between GSL and forest growth in complex terrain.
In this study, we used the correlation between transpiration and VPD (i.e. transpiration-atmosphere coupling) measured from seven spatially-distributed sapflow sites to examine forest physiological activity throughout the year and to describe variability in the start, end, and length of the growing season (GS start , GS end , and GSL, respectively). Sites were located across an 800 m elevation gradient and among slope aspects within a montane forest ecosystem. We addressed two main research questions: (1) does topoclimate variation or topographic location cause differences in growing season metrics or annual patterns of transpiration-atmosphere coupling?; (2) how does topographic variation in growing season metrics influence forest productivity as indicated by annual radial tree growth?

Methods
During 2013À2015, we collected data from seven sites (each with a 10 m radius) within two catchments of the Boulder Creek watershed in the Colorado Front Range, Colorado, USA (table 1). The sites have considerable variation in topography with an elevation range of 1884À2691 m. Betasso is a south-facing, lower montane catchment that is forested with Pinus ponderosa. Mean annual precipitation at Betasso is 500 mm, of which 20% falls as snow (Berryman et al 2015). The one Betasso site was selected to represent a lower elevation extreme and because it is collocated with a meteorological station that included air temperature (T air ) and relative humidity (RH) sensors (Vaisala HMP45AC, Vantaa, Finland) at 2 m above the ground, and soil temperature (T soil ; Decagon EC-5, Decagon Devices, USA) at 15 and 40 cm below the soil surface.
Gordon Gulch is an upper montane catchment with distinct north-and south-facing slopes. Southfacing slopes are more arid with a transient snowpack and low to moderate density P. ponderosa (Hinckley et al 2014). North-facing slopes are more mesic, with a shallow seasonal snowpack and densely spaced Pinus contorta (Hinckley et al 2014). Mean annual precipitation at Gordon Gulch is 510 mm of which 50% falls as snow (Berryman et al 2015). The six site locations within Gordon Gulch were selected to Table 1. Characteristics of the seven study sites including, mean annual air temperature (Mean T air ) and mean annual soil temperature (Mean T soil ). Error ranges represent one standard deviation. represent a range of elevations on south, west, and north facing slopes within the catchment. All Gordon Gulch sites were instrumented with T air and RH sensors at mid-canopy height (107 L, Campbell Scientific, Logan, UT). Four of the Gordon Gulch sites were instrumented with T soil sensors, installed at 10 and 30 cm below the soil surface (HydraProbe, Stevens Water Monitoring Systems Inc., Portland, OR) whereas, the other two (i.e. 2536 m and 2572 m) were instrumented with T soil (Decagon 5TE, Decagon Devices Inc., Pullman, WA) at 5 and 25 cm. Soil temperature measurements at all sites are reported as the mean of the two sensor depths. Three sites (1884 m, 2536 m and 2572 m) were also collocated with ultrasonic snow depth sensors (Judd Communications, Salt Lake City, UT). Meteorological data was measured at 10 m intervals, which were checked for missing and duplicate entries and then averaged to 30 min intervals for comparison to sapflow measurements. We measured sapflux velocity (V s ) at 30 min intervals (i.e. 48 measurements per day) using heatpulse sapflow sensors (Burgess et al 2001). Sensors were installed at 1.4 m aboveground in eight randomly selected trees per site to a depth of 2 and 5 cm inward from the cambium. Sapflux velocity is most commonly at its maximum near the sapwood-cambium boundary (Meinzer et al 2010) and the sapwood of P. contorta is commonly < 5 cm deep (Barnard et al 2013), therefore we used only V s measured at 2 cm depth for this study. The highest elevation southfacing site (2675 m) was mixed-species (P. ponderosa and P. contorta), so we instrumented n ¼ 4 of each species at that site.
We calculated transpiration-atmosphere coupling as the coefficient of determination (R 2 ) between daily linear regressions of VPD and V s , co-measured every 30 min (48 measurements per day) for the three years of this study (figure 1). We used a probability distribution function of values of transpirationatmosphere coupling to characterize the threshold between periods of dormancy and non-dormancy. The probability density distribution was bimodal with two peak densities corresponding to the two main phenological phases: the mean of forest dormant periods (0.175) and the mean of the forest growing season (0.58) (figure 2). Given a transition-period minima of 0.375, we chose one-half the distance from the dormant period peak to the transition-period Environ. Res. Lett. 12 (2017) 064003 minima to represent the threshold between the two phenological phases (0.275). The GS start was then quantified as the first day of a period in spring that the transpiration-atmosphere coupling 10 day moving average was greater than the threshold for more than 10 d. The converse was used to quantify GS end in fall.
The GSL was then calculated as GS end -GS start . To characterize differences in forest productivity among the sites, we collected increment cores at 1.4 m from three dominant trees per site. Following sanding, ring widths were measured using WinDENDRO software (Regent Instrumentation, Quebec, Canada). To examine differences in stand level growth variation, we calculated relative basal area increment (relative BAI). Relative BAI normalizes individual growth rings by basal area allowing for comparisons among sites (Bigler et al 2004). Relative BAI was calculated as where r t is stem radius at breast height (cm) at year ¼ t, r t−1 is stem radius at breast height in year ¼ t−1, and BA is the stem basal area.

Results
The largest differences in topoclimate were between the Betasso and Gordon Gulch catchments whereas the differences among sites at Gordon Gulch were not significant except for T soil (table 1). Annual T air at Betasso was on average 4.4°C warmer than the mean of T air at all of the Gordon Gulch sites (p < 0.01). Mean annual T air among sites at Gordon Gulch did not differ significantly by elevation or aspect (p ¼ 0.51 and p ¼ 0.13, respectively). Annual T soil was highest at the 1884 m site (10.8°C) and significantly different than the mean of the six Gordon Gulch sites (5.9°C, p < 0.01). Among all sites at Gordon Gulch, T soil was significantly different (p < 0.01), generally decreasing as elevation increased. These differences were especially distinct between north-and south-facing sites (4.5 and 6.7°C, respectively; p < 0.01). North-facing sites at Gordon Gulch accumulated a shallow (30-40 cm) seasonal snowpack. Betasso and the south-facing sites at Gordon Gulch had transient snowpack development, typically with shallow, semipersistent, heterogeneous snow cover (10-20 cm). Transpiration-atmosphere coupling followed a general pattern with defined periods of physiological activity in late-spring through fall, dormancy in winter and early spring, and a few periods of physiological activity that were out of phase with the overall annual cycle (figure 1). Outside of the growing season, the lowest elevation south-facing site (1884 m) had the greatest percentage of days within the year with transpiration-atmosphere coupling > 0.275 (29%) and the highest elevation north-facing site had the fewest (16%). Across all sites, transpirationatmosphere coupling was significantly correlated with T soil (p < 0.01, 3rd degree polynomial adjusted R 2 ¼ 0.65) and T air (p < 0.01, 3rd degree polynomial adjusted R 2 ¼ 0.56; figure 3).
Both elevation and aspect had significant impacts on growing season metrics across all sites (figure 4). At north-facing sites, GS start occurred an average of 4-6 d later, and GS end 12-14 d earlier than south-facing sites at similar elevations. The west-facing site at 2589 m had an average GS start three days later and a GS end one day earlier than the south-facing site at 2572 m. The variation in GS start and GS end resulted in GSLs that were 21 d longer on average at south-facing sites than north-facing sites at Gordon Gulch, with the westfacing site having an eight day longer GSL than the closest elevation north-facing site (2536 m). The GSL was up to 50 d longer at the lowest versus highest T air °C T soil °C Transpiration-atmosphere coupling (unitless) Figure 3. Transpiration-atmosphere coupling is correlated with air temperature (T air ) and mean soil temperature (T soil ) across all seven sites. Each point represents a single day and lines represent a least squares fit of a third degree polynomial.
Environ. Res. Lett. 12 (2017) 064003 elevation south-facing sites. Also at 2675 m, P. contorta had a GSL that was ten days longer on average than the co-occurring P. ponderosa with GS start occurring four days earlier and GS end occurring six days later in P. contorta.
Growing season metrics generally varied linearly with elevation, but did not vary significantly (p > 0.05) with mean site T air or T soil . Across all sites, the GS start occurred later, and the GSLs were shorter with increasing elevation (40 d km À1 and À50 d km À1 , respectively, p < 0.01 for both, figure 4). We acknowledge that the lowest and highest points being south-and north-facing, respectively, may influence these regression slopes but we found no significant interactions among elevation, aspect, or species (p > 0.20 for all). The GS end was not significantly correlated with elevation across all sites (p ¼ 0.11), but there was an interaction between aspect and elevation (p ¼ 0.01). There was a significant linear trend of earlier GS end with increasing elevation for south-facing sites (À10 d km À1 , p < 0.01) but not for north-or northand west-facing sites (p > 0.2 for all comparisons).
Growing season length was the best predictor of relative BAI, explaining > 83% of the variance among Relative BAI (unitless) Growing season length (days) Growing season length (days) Figure 5. Relationship between growing season length and basal area growth increment in the seven study sites (a). Relationship was not significant among only P. ponderosa sites (b), but was in the P. contorta sites (c). Error ranges represent one standard deviation.
Environ. Res. Lett. 12 (2017) 064003 sites, species, and aspects (figure 5). In general, northfacing sites with only P. contorta had lower relative BAI than sites with only P. ponderosa. However, the mixed species site at 2675 m suggests that growth differences may not be due to only north-versus south-facing topoclimates given that individuals of P. contorta had a higher relative BAI and longer GSL than the co-located P. ponderosa. When all sites were subsampled by species, GSL was not a significant predictor of relative BAI in P. ponderosa, but was for P. contorta, explaining 96% of the variance in relative BAI. We also tested mean site T air and T soil ; GS start ; GS end ; and the total number of days with transpiration-atmosphere coupling > 0.275 as predictors of relative BAI, but none produced significant relationships (p > 0.05 for all).

Discussion
This study is, to our knowledge, the first to explore how complex terrain can influence variation in direct measurements of montane forest GSL. In this context, we found that GSL varied greatly over small spatial scales but was not clearly linked with differences in site T air or T soil , indicating that complex topographyenvironment interactions may ultimately drive phenology and subsequently tree growth (Adams et al 2014, Salzer et al 2014. Within Gordon Gulch, sites separated by as little as 0.3 km of linear distance (i.e. north-versus south-facing sites) had GSLs that varied by up to 21 d. The large differences in GSL between aspects were primarily due to earlier GS end at north-versus southfacing sites which is notable given that previous studies have shown GS end to be determined by temperature or day length (Kramer et al 2000, Rossi et al 2007. Because plant-perceived photoperiod is responsive to direct, diffuse, and reflected light (Holmes and Smith 1977), it was not likely to vary among slope aspects and was thus doubtful to have influenced the timing of GS end in this study. It is also improbable that species differences in sensitivity to day length drove this disparity given that P. contorta had a greater GSL than P. ponderosa at the mixed-species site, contrary to the aspect-driven trend. In addition, the T air difference between the south-facing sites at 2478 and 2675 m was similar to the T air difference between north-and south-facing aspects at similar elevations (i.e. about 0.9°C). However, the difference in GS end was disproportionately greater between the sites on opposing aspects (up to 12 d) versus those at the different elevations (2 d).
The absence of clear day length or T air influences on GS end suggests that other environmental factors may drive phenological transitions e.g. enhanced stem and canopy surface temperature due to radiative warming (Smith et al 1984). Previous studies have shown that conifer leaf and stem temperatures can exceed T air by up to 4 and 15°C, respectively, on clear days with high insolation (Pomeroy et al 2009), but this effect would be substantially limited on northfacing slopes during fall and winter when solar angles are low (Wang et al 2002). While a comprehensive analysis of the mechanisms that drive dormancy onset was beyond the scope of this study, it is probable that limited amplification of stem and canopy temperature on north-facing sites due to lower insolation may have inhibited thermal recovery (i.e. leaf and stem thaw) following strong freeze events. This lowered thermal recovery could have resulted in the earlier dormancy onset we observed at the north-versus south-facing sites. Future studies would benefit from co-measuring air, stem, and canopy temperatures, and radiation and moisture regimes on different aspects in conjunction with transpiration-atmosphere coupling to clarify the impact of different components of the whole-plant energy balance on phenological shifts and plant growth.
Our analysis confirms that radial-stem growth was correlated with GSL, but not with GS start or GS end , or with T air or T soil . It was particularly surprising that relative BAI was not linked with GS start or T air variation. Earlier studies identified the spring onsettiming of radial cell growth, often determined by T air , as a key determinant of annual radial growth (Lupi et al 2010, Oberhuber et al 2014, Rossi et al 2013, Royce and Barbour 2001 and that this period can shift overtime in relation to mean night time temperature (Sullivan et al 2015). Other studies have identified a photoperiodic control on radial growth rates that lead to peak growth in early July and ending in August (Cuny et al 2012, Oberhuber et al 2014, Rossi et al 2006, Seo et al 2011. Hence, an earlier GS start would expand the period of growth unilaterally. In contrast, Rossi et al (2006) identified the onset of summer drought as responsible for declines in radial growth and Adams et al (2014) found a significant relationship between interannual precipitation variability and ring growth in Gordon Gulch, both suggesting that moisture availability may control radial growth rates. It is more likely that the GSL influence on growth varies with topographic gradients in temperature in our study domain. For example, Villalba et al (1994) found moisture-versus energy-limitations on growth to vary with slope aspect, Trujillo et al (2012) found moisture limitations on photosynthetic capacity to be elevation dependent, and Maxwell and Condon (2016) identify ground water depth as a key determinant of transpiration (and thus photosynthesis).
In this study, GSL was a significant predictor of relative BAI across all sites, but it was more predictive for north and west-facing sites (P. contorta) than at south-facing sites with P. ponderosa. The significant relationship between GSL and relative BAI at northfacing sites suggests an energy limitation to radial tree growth whereas the lack of relationship at south-facing sites implies that any beneficial effect of increasing Environ. Res. Lett. 12 (2017) 064003 GSL may be constrained by other resource limitations (e.g. water, nutrients), such as that reported in previous dendrochronological studies (e.g. Tran et al 2017). Given our findings and variation in results from past studies, it would be beneficial to investigate radial growth rates relative to total carbon uptake at finer temporal scales in conjunction with measurements of transpiration-atmosphere coupling to better determine the effects of growing season metrics on forest growth.
The mid-winter periods of physiological activity that we report confirms previous studies that have identified the ability of evergreen conifer forests to rapidly upregulate physiological activity in response to warm synoptic climate patterns (Adams et al 2002, Sevanto et al 2006, Verhoeven et al 1999. These periods of mid-winter photosynthesis have also been shown to contribute significantly to annual sums of carbon uptake in Mediterranean montane forests (Kelly and Goulden 2016). In this study however, adding mid-winter periods of physiological activity to the GSL did not improve the prediction of radial tree growth, suggesting that winter-time carbon uptake is not partitioned to stem radial growth in montane ecosystems. Overall biomass accumulation is not well correlated with ecosystem productivity (Litton et al 2007) and winter photosynthate production is often stored for use in spring (Schaberg 2000), potentially explaining the disparity in results between this study and Kelly and Goulden (2016). Nevertheless, we urge future studies to consider mid-winter physiological activity as it has the potential to influence annual sums of carbon, water, and energy exchange.
The 791 m elevation difference between the lowest and highest south-facing sites resulted in a mean annual T air difference of 4.4°C, which is within the range of forecasted increase in T air in the Rocky Mountains by the year 2100 (Baldwin et al 2003). Comparisons between these two sites may provide a proxy for predicting forest responses to future climates. A temperature increase of 4.4°C at our highest south-facing site could extend the GSL by 50 d or more and potentially shift elevational growth dependencies on moisture availability (Trujillo et al 2012). While our results indicate a positive relationship between GSL and radial growth, as climate warming is amplified in mountainous regions and the hydrologic cycle intensifies (Huntington 2006, Rangwala and Miller 2012, Trenberth 1999, the potential strain of a longer growing season on plant available water may be augmented by shifting precipitation patterns, potentially increasing forest disturbance, mortality, and fire severity and influencing the hydrologic cycle (Allen et al 2010, McDowell et al 2008, Westerling 2016, Penn et al 2016, Pribulick et al 2016. The spatial variability in growing season metrics we report has implications for modeling watershed hydrology and vegetation-atmosphere exchange. Hydrologic and land-surface models often use static dates for phenology or, in the case of dynamic phenology sub-models, apply singular GS start and GS end to the entire watershed or model-pixel (Oleson et al 2010, Tague and Band 2004, Wigmosta et al 1994. Given that we show a 21 d variability in GSL within a relatively small watershed, future efforts need to concentrate on improving spatial resolution of topography-vegetation interactions in complex terrain (Bierkens et al 2015). We also show that up to 29% of the days outside of the growing season have measurable responses of transpiration to atmospheric demand which will likely be neglected in standard hydrologic models. Moreover, accurate characterization of GS start is essential in systems where the majority of annual water inputs, become available in spring. A recent study (Winchell et al 2016) found that in subalpine forests, seasonal snowpack > 1 m can completely melt in as little as 17 d and up to 30% of total annual ecosystem carbon uptake can occur during this period. Due to the spatial and temporal interplay between GS start and changes in moisture availability, any error in the characterization of GS start during this brief period in the annual water cycle may lead to substantial errors in the timing of the biological component of evapotranspiration.

Conclusion
Using a novel approach to analyzing sapflow data, we found that relative elevation differences among sites and slope aspect have greater influence on montane forest phenology and tree growth than site specific environmental variables. Growing season length was linearly correlated with site elevation, resulting in GSL decreasing by 50 d per km of elevation gain. This difference in GSL was primarily due to higherelevation and north-facing sites having earlier GS end . There was also a large difference in GSL between north-and south-facing sites (up to 21 d) at similar elevation and the number of days outside of the growing season with physiological activity increased from 18% at the coldest and highest elevation site to 29% at the lowest and warmest site, with implications for calculating annual sums of water, carbon, and energy fluxes. The influence of spatial variability in growing season metrics may be especially important in estimates of carbon flux given the strong relationship between GSL and radial tree growth. While the true driver of growth differences among sites is more likely to be a suite of environmental variables, in this study domain a generalized topographic representation of elevation and slope aspect provided the greatest predictive power of GSL and tree growth. These findings underscore the need for high spatial resolution representations of ecosystem processes in complex mountainous terrain. Environ. Res. Lett. 12 (2017)