1. Introduction
Phenological indicators (PI)—markers of the timing of plant and animal behavioral cycles in response to seasonal changes—can be useful metrics to analyze the impact of climate change on ecosystem services, agricultural systems, and even human behavior. A review by Donnelly and Yu (2017) found that phenological changes received an increased emphasis in climate change research during 2007–16 relative to prior decades, owing to greater availability of phenological information. PI-based studies of climate change, while mostly focused on phenological models of plants, have examined disparate topics ranging from the behavior of migrating butterflies (Karlsson 2014) to the timing of peak attendance at U.S. national parks (Buckley and Foushee 2012). The current study focuses on three PI that are used to assess changes in the life cycles of plants during the winter and following spring. Chilling units (CU) are used to capture the effects of cooler wintertime temperatures on deciduous fruit trees and other plants as they undergo a period of dormancy that can impact their productivity the following spring. Next, we examine the influence of warming temperatures on the onset of spring, with the first appearance of leaves and blooms [extended spring indices (SI)], and, last, the potential for damaging hard freezes to occur after spring onset (referred to as a “false spring”). The effects of these PI are synergistic, as an earlier spring onset can lengthen the growing season, but also increase vulnerability to early frost for some plants.
The chilling requirement of fruit and nut trees must be considered relative to local winter conditions to yield a productive growing season, as an insufficient or surplus CU can negatively impact the health of the plants. Orchards may otherwise have the potential to remain productive for years or even decades, but future warming and the resulting milder winters could adversely affect productivity with negative economic effects for agricultural stakeholders. In an observational study over 1956–2003, Schwartz and Hanes (2010) found reduced chilling over parts of the western contiguous United States (CONUS), including fruit producing areas of California. Using statistical downscaling, Luedeling et al. (2009) found that insufficient chilling in future warmer conditions by the end of the century may no longer support agricultural production of some of California’s main fruit and nut tree crops in areas where they are currently being grown. Such results would have significant implications for agricultural systems if similar impacts were found to be widespread over the CONUS.
An earlier onset of springtime events (breaking hibernation, flowering, etc.) has been found by observation-based studies, as well as those projecting future changes (e.g., Schwartz et al. 2006; IPCC 2014; Monahan et al. 2016; Lipton et al. 2018; IPCC 2022). As summarized within the IPCC Fifth Assessment Report (AR5), springtime events were observed to occur approximately 1–3 days earlier per decade over the Northern Hemisphere, with the magnitude of the trend depending on the plant species being analyzed (IPCC 2014, section 4.3.2.1.1). Work summarized in the Sixth Assessment Report (AR6) emphasized the effects on human health and ecosystems as well as regional spring advancement trends (IPCC 2022, section 2.4.2). Observed spring advancement trends differ over various regions. Over the period 1955–2002, first leaf and first bloom dates advanced 1.2 and 1.0 days per decade, respectively, over the Northern Hemisphere (Schwartz et al. 2006). In a study examining spring onset across 276 U.S. national parks, 76% had experienced spring advancement in recent (previous 10–30) years relative to a historical baseline dating back to 1901, with over half having recent average spring onsets earlier than the 95th percentile (Monahan et al. 2016). A review article by Piao et al. (2019) examined mean spring advancement within the Northern Hemisphere and found the trend most pronounced in China (at 5.5 days per decade on average), while a lower rate of 0.9 days per decade was derived from observations over North America for 1982–2011. Menzel et al. (2006) examined PI over Europe for the period 1971–2000 and found a trend of 2.5 days per decade of spring advancement. Spring advancement and the rate at which it occurs have significant implications for ecosystems and human health effects. In an observation-based study over North America, Lian et al. (2020) found that increased evapotranspiration due to earlier spring onset and green-up drove deficits in soil moisture that persist into the summer. Earlier spring onset has also been linked to increases in the duration of the pollen season in North America, with an increase in asthma hospitalizations (IPCC 2022, section 7.2.3.2).
Statistical methods have been used to project future SI changes. Allstadt et al. (2015) analyzed future projections of springtime changes, specifically SI and the frequency of false springs, using statistical downscaling under representative concentration pathway (RCP) 8.5 (Riahi et al. 2011). Applying statistical downscaling to an ensemble of GCMs from phase 5 of the Coupled Model Intercomparison Project (CMIP5; Taylor et al. 2012), Allstadt et al. (2015) used daily model outputs and generally found earlier spring onset, with a mean spring advancement of 23 days across the CONUS by end of century (2071–2100) relative to the analyzed historical period (1950–2005). Changes in false springs varied across the CONUS but were more frequent in the Great Plains and Midwest. Meanwhile, end-of-century (2070–99) advancement in SI dates of ∼6 to 16 was found by Hayhoe et al. (2007) using statistical techniques to downscale CMIP3 GCMs over the northeastern CONUS. Parker and Abatzoglou (2016) used statistically downscaled daily minimum temperatures to examine projected changes to winter temperatures and the accompanying changes in cold hardiness zones delineated by the U.S. Department of Agriculture. They also compared results from dynamically downscaled simulations with statistically downscaled results. While both methods produced similar patterns of warming minimum temperatures, the dynamically downscaled results showed more heterogeneity and increased intraensemble variability through capturing snow–albedo feedbacks. In their PI-based study, Parker and Abatzoglou (2016) concluded that dynamical downscaling can represent local effects that are inadequately captured by statistical techniques.
The current study complements prior work where statistical downscaling was used to examine future PI changes by utilizing dynamical downscaling, a physics-based approach to obtain future projections with fine temporal and spatial resolution. Here, changes in indicators of plant dormancy (CU), spring onset (SI), and the frequency of false springs are assessed both in historical simulations and projections of 2025–2100 that are dynamically downscaled from multiple GCMs and under two greenhouse gas emission scenarios (i.e., RCPs; van Vuuren et al. 2011). The Weather Research and Forecasting (WRF) Model (Skamarock and Klemp 2008) is used to dynamically downscale the GCMs to 36-km grid spacing. The availability of simulated hourly temperature over a 76-yr future period is advantageous not only because of increased spatial resolution (relative to the GCMs), but also the availability of finer temporal resolution. All PI examined here are derived from hourly 2-m temperatures. The availability of WRF-simulated hourly data from dynamical downscaling allows these fields to be computed without extrapolation from consolidated daily temperature values like minima and maxima, as was done in prior studies.
Within dynamical downscaling and regional climate modeling applications, the evaluation and study of future extremes in temperature are often focused on heat extremes that occur in the summer, such as heat wave events or the number of days exceeding “hot” thresholds (e.g., Pan et al. 2011; Otte et al. 2012; Mallard and Spero 2019). However, extreme temperatures during transitional periods, such as spring and autumn periods, also have significant effects on ecological and agricultural interests. Here, the emphasis on wintertime plant dormancy and springtime conditions provides insight into important processes that could affect agricultural productivity and other related impacts. More detailed descriptions of the PI used in the current study are given in section 2 below, followed by results of observational comparisons over the historical simulated period and a discussion of projected PI changes over mid- to end-of-century within section 3. Section 4 offers a summary and conclusions.
2. Methods
a. Dynamically downscaled simulations
WRF (Skamarock and Klemp 2008) is used to downscale two GCMs from CMIP5 (Taylor et al. 2012) over a historical (1995–2005) and future period (2025–2100). The NOAA GFDL-CM3 is downscaled from its archived resolution of 2° × 2.5° (Donner et al. 2011) using WRF version 3.6. CESM (Gent et al. 2011) has an archived resolution of 0.9° × 1.25° and is downscaled using WRF, version 3.4.1. Two scenarios (RCP4.5 and RCP8.5) are utilized, where the resulting CESM-based simulations are referred to as “WRF-CESM4.5” and “WRF-CESM8.5,” respectively. GFDL is downscaled only under RCP8.5 and this run is referred to as “WRF-GFDL8.5.” RCP8.5 is a high emissions scenario that has greenhouse gases (GHGs) rising substantially throughout the twenty-first century with 8.5 W m−2 of radiative forcing by the year 2100 (Riahi et al. 2011). RCP4.5 is a milder scenario, where GHG emissions peak at midcentury and radiative forcing is at 4.5 W m−2 by 2100 (van Vuuren et al. 2011).
Simulations are run over a 36-km CONUS domain with 34 vertical layers and a 50-hPa model top (Fig. 1). When downscaling the GFDL model, which has coarser resolution than the CESM, an intermediate 108-km parent domain is included in the WRF configuration [following the configuration used in Otte et al. (2012) and Bowden et al. (2012)]. All runs are initialized on 1 October to provide a 3-month spinup prior to the analyzed period, and the historical and future periods are continuous simulations. For all domains, spectral nudging (Miguez-Macho et al. 2004) is applied to temperature, geopotential height, and horizontal wind components only above the PBL following Otte et al. (2012), Bowden et al. (2012), and Spero et al. (2016). Nudging coefficients used on both the 36-km domain, as well as the outer 108-km domain that is employed for the GFDL-driven runs, are described in Otte et al. (2012, their Table 1).
The 36-km WRF domain, along with the nine NCEI regions.
Citation: Journal of Applied Meteorology and Climatology 62, 12; 10.1175/JAMC-D-23-0071.1
In these runs, the Kain–Fritsch convective parameterization scheme (Kain 2004) with radiative feedbacks following Herwehe et al. (2014) and the WRF single-moment 6-class microphysics scheme (Hong and Lim 2006) are employed. Processes in the PBL are simulated using the Yonsei University scheme (Hong et al. 2006). Land and surface processes are parameterized using the Revised MM5 Monin–Obukhov surface scheme (Jimenéz et al. 2012) and Noah land surface model (Chen and Dudhia 2001). Land-use information is based on the 24-category USGS dataset (Loveland et al. 2000). Within the CESM-driven runs, lake temperatures are from the Community Land Model following Spero et al. (2016), while the WRF-GFDL simulation used a coupled lake temperature model (Mallard et al. 2014). The nine U.S. climate regions utilized by the NCEI (Karl and Koss 1984) are used for regional analyses over the CONUS (Fig. 1).
Several aspects of the WRF Model configuration were vetted and analyzed by Otte et al. (2012), Bowden et al. (2012, 2013), Herwehe et al. (2014), Mallard et al. (2014), and Spero et al. (2016). The CESM-driven simulations utilized in this work were also employed by Nolte et al. (2018a) to drive model projections of future changes to air quality, with that study examining historical error in mean and maximum daily temperatures. Those CESM simulations also appeared in Fann et al. (2015, 2016) and informed the Air Quality chapter of the Fourth National Climate Assessment (NCA4; Nolte et al. 2018b). Nolte et al. (2021) expanded this analysis to include the GFDL-driven simulation. Overall, this ensemble was found to be adequately suited for air quality applications and future projections. Nolte et al. (2018a) found cool biases in most seasons and regions for the WRF-GFDL run, while WRF-CESM was shown to have more regionally variable temperature biases. Nolte et al. (2018a, 2021) conclude that temperature biases in these runs resemble those in the driving GCMs and are inherited from the global model being downscaled, which is an expected artifact of using spectral nudging.
b. Phenological parameters
Here, modeled phenological indicators are examined over the historical period 1995–2005 and compared with observation-based temperature and PI to assess the degree to which GCMs downscaled with WRF at this scale and with these settings capture agricultural and ecosystem cycles. This assessment uses decadal-averaged PI, as the GCM’s representation of the historical period is not analogous to an annual seasonal forecast.
CU are calculated from simulated hourly 2-m air temperatures using the Utah model (Richardson et al. 1974) during the period of 1 October to 1 May. This model is one of several methodologies that has been used to determine when deciduous fruit trees and other plants have completed a period of “rest” or dormancy needed before new growth occurs (e.g., Chandler 1942; Fishman et al. 1987; Luedeling and Brown 2011). The Utah model is selected here because it has been heavily used by prior work to study the chill requirements of a variety of plants ranging from pecans (Kuden et al. 2013) to grapevines (Camargo-Alvarez et al. 2020). As defined in the Utah model, CU values accumulate each hour based on how air temperatures fall into the ranges given in Table 1 of Richardson et al. (1974). The maximum value (1 CU) is given for air temperatures around 6.0°C, while 0 CU is awarded at temperatures cooler than 1.4°C. Meanwhile, negative values (as low as −1 CU) are given when hourly temperatures are warmer than 16°C. The Utah model, along with similar PI, is based on work by Erez and Lavee (1971), where it was observed that an air temperature of 6.0°C was most favorable for rest completion. The number of CU required to accumulate for rest completion varies among plant species and cultivated varieties (i.e., cultivars). Representative values leveraged from prior studies give a range of ∼800–1200 CU for some apricot varieties (Ruiz et al. 2007), ∼900–1400 for several varieties of apples (Parkes et al. 2020), ∼800–900 for peach trees varieties studied by Richardson et al. (1974), and ∼850 CU for Bing cherries (Richardson et al. 1986). Perennial plants adapt to cold winter temperatures by developing cold hardiness (acclimation) in the fall, maintaining this hardiness during the winter, and subsequently losing cold hardiness (deacclimation) prior to budbreak and green-up in the spring, with increases in the rate of deacclimation found as chilling accumulates (e.g., Okie and Blackburn 2011; Nanninga et al. 2017; Kovaleski et al. 2018; Kovaleski 2022). If insufficient chilling accumulated for these crops over the period preceding the growing season, then a poor harvest results with fruiting or flowing that occurs late or not at all (Kovaleski 2022). Meanwhile, a winter period with unusually increased chilling accumulation can cause plants to lose cold hardiness and break dormancy earlier in the year, leaving them vulnerable to frost or disease.
The date of spring onset is calculated using SI values (Schwartz et al. 2013). These PI predict the date (taken as the number of days since 1 January on an annual basis) of leaf out (LO) and first bloom (FB) in lilac and two honeysuckle clones (Schwartz 1993). Although SI are calculated based on these lilac and honeysuckle varieties, these plants effectively serve as proxies, as SI have been shown to capture spring onset in a wide variety of natural and agricultural plant species in both temperate and subtropical environments (Schwartz et al. 2013; Gerst et al. 2020). Gerst et al. (2020) compared SI with observed leaf and flowering dates from 19 types of trees and shrubs. They concluded that, although relationships varied by latitude and among different species, SI correlates to a wide range of species across the CONUS and can be used as a “yardstick” for evaluating future changes in spring onset. The day of the year when LO occurs is estimated for each of the three plant species (lilac and honeysuckle) using empirically derived equations based on the number of days since 1 January (MDS0), growing degree hours (GDHs), and the accumulated number of high energy synoptic events or “warm spells,” where a warm spell occurs when GDHs exceed a threshold value for a consecutive 3-day periods (Ault et al. 2015). GDHs are computed by taking the hourly temperature greater than a threshold, here taken as 0.6°C. LO “scores” are generated for each day based on these variables until the average score among all 3 plants exceeds a threshold value, and the date of that exceedance is taken as the LO date. The date of FB is constrained to occur after LO. FB scores are calculated based on the number of days since LO (MDSL) and the accumulated GDHs (AGDHs) since LO. As the season progresses, the daily increasing MDSL (which is scaled with a negative coefficient) reduces the score, while any positive-accumulating AGDH increases it. Again, the average score exceeding a key threshold signals the occurrence of the FB date for a given year. Additional information on the derivation of SI can be found in Ault et al. (2015). In this study, we calculate all SI using explicitly modeled hourly temperatures rather than artificially imposing a diurnal cycle between daily minimum and maximum, as has been done in other studies (e.g., Ault et al. 2015; Martinuzzi et al. 2016; Izquierdo-Verdiguier et al. 2018). Earlier SI dates, or spring advancement, are linked to several ecosystem impacts, including increases in summertime soil moisture deficits from earlier green up of vegetation, lengthening of the pollen season, and additional challenges with managing ecosystem impacts such as invasive species that can better adapt to changing conditions (e.g., Monahan et al. 2016; Lian et al. 2020; IPCC 2022). Here, projected SI changes are examined alongside the frequency of damaging hard freezes, as earlier spring onset and green up can lead to increased risk of damaging frost if a cold snap occurs.
The frequency of false spring is assessed following Peterson and Abatzoglou (2014) and Allstadt et al. (2015). We define false spring in two ways: 1) as the occurrence of a hard freeze (Tmin < −2.2°C) 7 or more days after LO (“early” false spring), and 2) the occurrence of a hard freeze any time after FB (“late” false spring). Both metrics of false spring are only considered prior to 1 July within the given calendar year.
c. Observational data and computation of parameters
Observation-based SI values are available from 1981 to present as a 4-km gridded product from the National Phenological Network (NPN) (USA National Phenology Network 2019). The NPN product is constructed by applying phenology models to historical daily minimum and maximum 2-m air temperature data from the 4-km PRISM dataset (Daly et al. 2008). The software needed to run this phenological model was acquired from NPN and modified to use downscaled hourly 2-m temperatures as input to derive SI values. Therefore, while the NPN and simulation-based SI dates are similarly derived, the simulated SI uses explicitly modeled hourly diurnal temperatures, while the observation-based NPN SI includes a superimposed diurnal cycle bounded by the PRISM maximum and minimum. As PRISM daily minimum and maximum temperatures are used to calculate the observation-based SI used here, these temperature data are also used to compare with simulated 2-m temperatures. Here, the NPN SI and PRISM data are interpolated to the WRF domain for comparison with the simulated fields.
To assess the skill of simulated CU, hourly 2-m temperatures were taken from CFSR, a global reanalysis product available at 0.31° resolution (Saha et al. 2010). After CFSR temperatures were interpolated to the 36-km WRF grid, CU were derived using the Utah model described in section 2b.
3. Results
Here, mid- to end-of-century future projections for each PI will be discussed, along with comparisons with observational data over the historical period. The discussion of PI follows their seasonal progression, with CU results presented first followed by springtime PI.
a. Chilling units
1) Model error
Bias in accumulated CU is shown for each month over the period in which CU are accumulated for both the CESM- and GFDL-driven downscaled simulations for the historical period 1995–2005, where bias is taken relative to observational CFSR-based CU (Fig. 2). Using bias to highlight systematic error is advantageous here, as the accumulation of CU over time is critical to understanding the impacts of weather on the growing seasons. Both runs produce biases of ±200 CU for most months and regions. However, negative biases exceeding this magnitude occur in the WRF-CESM run for most of the CU season in the Northeast and Southeast regions. An underestimation of CU is consistent with the warm biases in the eastern CONUS during similar seasonal periods reported by Nolte et al. (2018a; their Fig. 2). WRF-GFDL has large positive biases in the South and Southeast, especially in April and by the end of the CU season. Again, this overestimation of CU is consistent with the cold bias in WRF-GFDL over the historical period reported by Nolte et al. (2021, their supplemental Fig. 1).
Average bias in accumulated CU for the historical period (1995–2005) from the (left) WRF-CESM and (right) WRF-GFDL simulations. This regionally averaged bias in CU is taken against CFSR-derived CU values at the end of each month of the season over which CU is accumulated, as well as at the end of the season. The y axis is consistent between both plots.
Citation: Journal of Applied Meteorology and Climatology 62, 12; 10.1175/JAMC-D-23-0071.1
Nolte et al. (2018a, 2021) demonstrate that the biases in these regional climate simulations emulate those of the downscaled GCMs, as the WRF Model is constrained toward the GCM’s solution via spectral nudging. Therefore, model error assessed here reflects the accuracy of both the GCM and regional model. Also, given the novelty of applying downscaling with a focus on PI (and this subset of PI specifically), the effects of contextualizing model error using a decadal time slice are not well known. The timing and frequency of phenomena that enhance or suppress CU (as well as the other PI studied here) affects the model error found when comparing these runs with observational data, because the timing of individual weather systems in both the GCM and downscaled models do not mirror that of observed systems. In several regions, CU biases in Fig. 2 change by hundreds of units from month to month (e.g., March to April transitions in the South and Upper Midwest in WRF-CESM and WRF-GFDL), implying that even CU totals that have been accumulating for a period of months can be altered by submonthly cold or warm periods. Therefore, trends in future projections, taken relative to the simulated historical period, will be emphasized both here and in subsequent sections, rather than comparing future projections with observed PI values.
2) Future projections of CU
Figure 3 shows average end-of-century (2090s) accumulated CU changes progressing through the season for all 3 simulations. Early in the season (November–January), all simulations show generally decreased CU throughout the CONUS, with the largest decreases at lower latitudes in the Southeast, South, Southwest, and portions of the West. Both runs driven with RCP8.5 show reductions of greater than 1000 CU in these areas, while the WRF-CESM4.5 simulation projects a less severe reduction of over 500 CU. As the CU season progresses into March and beyond, this decrease is amplified in the lower latitudes from the West to the Southeast NCEI regions. By the end of the CU season, regionally averaged decreases drop below −500 CU in the South for WRF-CESM4.5 and below −1000 CU in the South and Southeast in both the WRF-CESM8.5 and WRF-GFDL8.5 runs. Meanwhile, CU increases in northern portions of the CONUS and complex terrain of the western United States, with WRF-GFDL8.5 showing regionally averaged increases exceeding 500 CU in the Northwest, Northern Rockies and Plains, and Upper Midwest. This spatial pattern of CU change is robust over the projected period. Figure S2 in the online supplemental material shows midcentury (2050s) projections, which are lower in magnitude relative to end-of-century but similarly have decreased CU at lower latitudes and slight increases in northern parts of the CONUS. Overall, these results suggest that projected changes in the climate under both emissions scenarios by midcentury and beyond could inhibit rest completion for some fruiting and flowering plants in portions of the southern United States, especially for cultivars that have higher CU requirements. Meanwhile, in portions of the northern CONUS, cultivars with lower CU requirements may break dormancy too early under future conditions. Note that rest completion and subsequent fruiting or flowering is influenced by several factors such as photoperiod (Laube et al. 2014), water availability (Shellie et al. 2018), and the amount of deacclimation (or loss of cold hardiness) needed by a specific plant or species before budbreak can occur (Kovaleski 2022). Work by Kovaleski (2022) indicates that for a given loss of accumulated chilling, decreases in the rate of deacclimation can vary in magnitude, as the relationship between these metrics is nonlinear.
Average changes (2090–2100 minus historical values) in accumulated CU at the end of (top) November, (top middle) January, (middle) March, and (bottom middle) the season (ending 1 May) for (left) WRF-CESM4.5, (center) WRF-CESM8.5, and (right) WRF-GFDL8.5. (bottom) Spatially averaged changes (relative to the corresponding historical simulation) in accumulated CU plotted for each month, as well as the season, for each region. Average accumulated CU are provided in Fig. S1 in the online supplemental material for the historical period, from WRF-CESM and WRF-GFDL, for the same months.
Citation: Journal of Applied Meteorology and Climatology 62, 12; 10.1175/JAMC-D-23-0071.1
As discussed in section 2b and defined in Richardson et al. (1974), the Utah model awards CU values of −1, −0.5, 0, 0.5, or 1 that are accumulated each hour based on whether temperatures fall within certain ranges, with optimal temperatures between 1.4° and 12.5°C resulting in the maximum CU value. In Fig. 4, historical and end-of-century monthly mean temperatures are expressed as CU values to examine changes in the progression of CU throughout the season and as way of illustrating whether the resulting mean temperatures favor accumulation of positive CU. The simulations under RCP8.5 demonstrate the most dramatic end-of-century changes. Notably, during the first and last months (October and April, respectively) of the CU season, these projections show a decrease in the spatial coverage of areas where temperatures result in positive CU values. This would shorten the period where CU are likely to accumulate in the CONUS, both at the beginning and end of the season, in the end-of-century projections under RCP8.5. Meanwhile, during the “interior” months (November–March), there is a northward shift in positive CU values in the future, resulting in increased CU in northern portions of the CONUS (Figs. 3 and 4). This pattern of changes in temperatures and CU is robust across both CESM- and GFDL-driven downscaling, despite differences in historical bias in CU.
The mean monthly temperature (expressed in units of CU, according to the Utah model) from October to April for WRF-CESM (in the first two columns) and WRF-GFDL (in the second two columns) over the (left),(right center) historical and (left center),(right) projected end-of-century (2090–2100) periods under RCP8.5.
Citation: Journal of Applied Meteorology and Climatology 62, 12; 10.1175/JAMC-D-23-0071.1
Exploring changes in CU also highlights the local effects of projected changes in 2-m temperature. For most of the CU season (November–March), each simulation shows a robust northward shift of positive CU and of temperatures in ranges that increase CU accumulations (Figs. 3 and 4). Such a latitudinal shift is not evident when viewing projected changes to end-of-century mean temperature (Fig. S3 in the online supplemental material). Although areas of the CONUS may experience the same magnitude of warming in 2-m temperatures, the effect on the frequency with which temperatures meet or exceed key values for PI differs, resulting in differential local impacts to ecosystem services and agriculture.
b. Extended spring indices
1) Model error
Average simulated LO and FB dates from both historical runs are positively biased relative to those in the NPN dataset (which is based on PRISM temperature data) (Table 1). While both runs have SI that are biased high, indicating that simulated spring onset occurs later than observed, FB dates are simulated with less skill, with higher biases and mean absolute error (MAE, shown in Table S1 in the online supplemental material), in the CONUS and all NCEI regions when compared with LO dates. As FB is constrained to follow the date of LO (see section 2b), positive biases in LO dates would tend to adversely affect the skill of the following FB dates. Simulated spring onset is generally more accurate in the eastern CONUS for both runs, with the Upper Midwest, Ohio Valley, Northeast, Southeast, and South having lower bias than the West, Northwest, and Southwest. The latter regions show large biases in areas of complex terrain and near some parts of the coast (Fig. S4 in the online supplemental material). As expected, these biases would affect the timing of spring differently across the regions. Over the historical period, the Southeast and South have the earliest simulated spring onset among the NCEI regions with averaged LO at ∼63–80 days and FB ranging between ∼110 and 133 days; meanwhile, CONUS-wide SI range between ∼103–117 for LO and ∼149–165 for FB (Table S2 in the online supplemental material).
Biases averaged over the historical period for simulated LO and FB (given as the number of days since 1 Jan), as well as average 2-m temperature (relative to PRISM data, with temperatures taken each year between 1 Jan and the date of observed LO at each grid point) for each of the NCEI regions and the CONUS.
The temperature biases (relative to PRISM at each grid cell from 1 January until observed LO is reached at that cell) are negative for the CONUS and for all regions for WRF-GFDL (Table 1). An underprediction of 2-m air temperatures produces positive SI biases because the cooler temperatures reduce GDH accumulation; therefore, longer periods of slowly accumulating GDH are needed before reaching thresholds critical for SI. However, several regions within WRF-CESM exhibit a positive temperature bias over this period, reaching approximately 2°C in the Southeast and Northeast, and these regions produce LO biases that are relatively small (14–22 days) in comparison with other regional biases but positive. Comparison between the NPN observational data (which are based on PRISM minimum and maximum daily temperature data) and the simulated SI (which are derived from hourly temperatures) is complicated by the fact that the former is derived using a prescribed diurnal temperature profile while the latter is not. However, it can be concluded that WRF-CESM better captures historical SI over this period, when compared with WRF-GFDL.
2) Future projections of SI
Projected trends and changes in SI, taken relative to the historical period, are examined in the proceeding discussion using SI values derived from hourly 2-m temperatures. Changes in regionally and decadally averaged LO and FB dates are shown throughout the simulated period (Fig. 5) and at the end of the century (Table 2). Spring advancement (where LO and FB are projected to occur earlier in the year relative to the historical period) is evident in all three simulations over each region and the CONUS. CONUS-average spring onset at the end-of-century is approximately 16–21 days earlier for WRF-CESM8.5 and 27–33 days earlier for WRF-GFDL8.5, with changes in FB dates being larger than those for LO dates (Table 2). Under WRF-CESM4.5, less drastic changes are projected to the onset of spring, with CONUS-average changes of approximately 9–11 days by end-of-century. The trend for spring advancement is robust across each NCEI region and both SI metrics, with earlier spring onset projected by all three future projections. The largest changes in end-of-century LO dates generally occur in the westernmost regions (Northwest, West, and Southwest), while all three scenarios show the smallest changes in the Southeast. However, there is less consensus among the three future projections for regional changes in FB dates, as WRF-GFDL favors larger reductions in FB dates in the Northwest and Northeast, while both CESM-driven runs show the largest changes in FB in the South, Southwest, and West.
Regional-averaged (inset colors, as in Fig. 1) and CONUS-averaged (black) spring advancement is shown for (top) LO and (bottom) FB indices, expressed as the number of days earlier in the year that the onset of spring is occurring on average for a given decade relative to the historical 1995–2005 period, for the (left) WRF-CESM4.5, (center) WRF-CESM8.5, and (right) WRF-GFDL8.5 runs. The inset box shows the trend in spring advancement (in days per decade) for each NCEI region and the CONUS (CN). The R2 values for each of the linear fits are shown in Table S3 in the online supplemental material. Note that the y axis is consistent across all three runs for each metric.
Citation: Journal of Applied Meteorology and Climatology 62, 12; 10.1175/JAMC-D-23-0071.1
End-of-century changes (days), taken as the 2090–2100 mean LO date minus the 1995–2005 mean and averaged over each of the NCEI regions and the CONUS. FB changes are also shown similarly.
Here, trends in spring advancement over the simulated future period (which spans 2025–2100) are examined and compared with the range of spring advancement derived from observation-based studies, which generally give 1–3 days of spring advancement per decade over the Northern Hemisphere (IPCC 2014). Regional- and decadal-averaged changes in spring onset are shown in Fig. 5 for the 2030s through the 2090s, where the slopes of linear trend lines fitted to the data express the rate of change in spring onset, given as days of spring advancement per decade. R2 values are used to assess the fraction of the variance in LO and FB changes that can be explained by a linear model (Table S3 in the online supplemental material). R2 values can range from 0 to 1 where higher values indicate greater adherence to a linear model. Generally, WRF-GFDL8.5 and WRF-CESM8.5 projections have R2 values closer to 1 (0.94–1 over the CONUS), indicating changes in those SI adhere closely to a linear trend as compared with somewhat lower R2 values for WRF-CESM4.5 (0.75–0.91). Averaged over the CONUS, spring advancement trends of around 2 days per decade are projected by WRF-CESM8.5, with WRF-GDFL8.5 projecting a larger trend of 3.5–4 days per decade, depending on which SI is considered. Positive spring advancement is robust across all regions but varies notably with regional-averaged trends of 1.5–2.9 days per decade for WRF-CESM8.5 and an accelerated spring advancement of 2.6–4.6 days per decade projected by WRF-GFDL8.5 across the regions. The larger trends in spring advancement for WRF-GFDL8.5 are consistent with it producing the most warming among the ensemble, taken relative to the historical simulation driven by the same GCM (Fig. S3 in the online supplemental material). Nolte et al. (2021) qualified that WRF-GFDL8.5 produced the greatest warming trends among the ensemble, but absolute temperatures remained slightly cooler than WRF-CESM8.5, due to the cool bias over the historical period for WRF-GFDL. Meanwhile, changes taken from the more moderately forced WRF-CESM4.5 simulation show spring advancement of only 0.7–0.8 days per decade over the CONUS, with regional trends varying between 0.3 and 1.2 days per decade.
Overall, results from downscaling CESM tend to produce similar rates of spring advancement as the 1–3 days per decade suggested by historical trends, with an acceleration when driven with the higher-emission scenario. WRF-GFDL8.5 amplifies trends in spring advancement relative to those found from observational records, with CONUS-averaged trends exceeding 3 days per decade for both SI while regional-averaged results show spring advancement exceeding 3 days per decade for all regions for FB and 7 of 9 regions for LO. It should be noted that historical trends generalized here are derived from a variety of plant species and metrics (IPCC 2014, 2022), and some sensitivity is expected to the metric used to quantify spring onset. Notably, all runs here show greater trends when considering FB versus LO, which highlights this expected sensitivity. The earlier spring onset found here is generally consistent with the statistical downscaling study conducted by Allstadt et al. (2015) that examined projected changes in SI across the CONUS under RCP8.5.
c. False springs
1) Model error
Over the historical period, both WRF-CESM and WRF-GFDL underpredict the frequency of early false springs (hard freezes following LO; Table S4 in the online supplemental material). The probability of false spring is expressed here using the simulated number of false springs found to occur over a given time period normalized by the total number of simulated years examined; therefore, it represents the average risk of false spring over a given period. Over the CONUS, the average bias in the risk of early false springs for both simulations is ∼−0.6, with the risk of early false springs being underestimated most severely in the western CONUS. This result can be anticipated, given that this metric is derived based on the presence of freezing temperatures following LO. LO dates are overpredicted and simulated to occur later in the year (e.g., when temperatures can be anticipated to be warmer) relative to dates taken from the observation-based NPN dataset. However, both mean bias (Table S4) and mean absolute error (not shown) generally show improvement in model error for late false springs (hard freezes following FB), as compared with early false springs. Biases in the risk of late false springs are more varied across the regions and CONUS-averaged biases are less than 0.1 in magnitude. Allstadt et al. (2015, their appendix C) found early false springs to be more frequent than late false springs over their analyzed historical period. Here, this trend is also present when comparing early and late false springs in WRF-GFDL; however, the trend is absent from the historical WRF-CESM simulation due to underpredicting the frequency of early false springs (Table S4).
2) Future projections of false spring risk
While long-term warming in near-surface temperatures suggests a reduction in freezing events, the consistently projected trend of spring advancement and earlier LO and FB dates would leave crops vulnerable to freezing events for a longer period during the early spring. When changes are averaged over the entire CONUS, both types of false spring risk are projected to decrease by the end of the century by both RCP8.5 projections, with mixed results from WRF-CESM4.5 (Table 3; Fig. 6). Relative to SI, the projected changes to the probability of false springs are found to be not only more regionally dependent but sensitive to decadal variability.
End-of-century-averaged (2090–2100) changes in the probability of false springs proceeding LO and FB, taken relative to the simulated historical period. Positive values, indicating increased risk, are italicized for emphasis.
As in Fig. 5, but for the probability of (top) early false springs (those after LO) and (bottom) late false springs (those after FB). Trends were not computed.
Citation: Journal of Applied Meteorology and Climatology 62, 12; 10.1175/JAMC-D-23-0071.1
By the end of the century, decreased risk of false springs is generally most pronounced in the eastern United States, with the largest reductions in early false springs in the Southeast (between −0.13 and −0.22) in all three simulations (Table 3). Similarly, Allstadt et al. (2015) also projected a reduced risk of early false springs over the southern CONUS by end-of-century. Here, late false spring decreases are also most pronounced in regions of the eastern CONUS, with the largest decreases found in the Ohio Valley for WRF-GFDL8.5 (−0.18) and the Upper Midwest for WRF-CESM8.5 and WRF-CESM4.5 (from −0.41 to −0.30). Meanwhile, increases in early false springs are projected for the Northern Rockies and Plains and Southwest for all three projections, and an increased risk of late false springs is projected by WRF-GFDL8.5 for the Southwest, West, and Northwest.
An increased risk of false spring occurrence is expected where spring advancement trends are largest, given that this expansion of the growing season would leave plants vulnerable to hard freezes for a longer period. As discussed in the previous section, spring advancement by the end of the century (Table 2) is projected to be larger in the western CONUS, which aligns with the areas where false spring risk is projected to increase rather than decrease. For example, WRF-GFDL8.5 projected the most pronounced changes in FB dates for the Northwest, West, and Southwest; alongside the Northeast (Table 2). These westernmost regions also show increased risk of late false springs (i.e., hard freezes occurring after FB) (Table 3). Conversely, the earlier onset of LO is less pronounced in the Southeast, which exhibits the most robust decreases at the end-of-century for early false spring risk (i.e., hard freezes that follow LO).
The change in false springs relative to the simulated historical period is shown throughout the projected period in decadal-averaged risk in Fig. 6. The Upper Midwest shows a somewhat consistent trend of decreasing risk of late false springs. However, generally the trend in changes to false spring risk is more variable from decade to decade than SI trends (Fig. 5). Notably, the probability of early false springs in the Southeast has the most pronounced decrease relative to the historical period during the 2050s under WRF-CESM4.5 and WRF-GFDL8.5 and during the 2080s under WRF-CESM8.5. The midcentury peaks in false springs increase the risk in some regions (such as in the 2050s and 2060s for the early false springs within the Ohio Valley, Southwest, and South regions projected by WRF-CESM8.5; or increasing risk of late false springs in the 2040s and 2050s projected by WRF-GFDL8.5 for the Northern Rockies and Plains, Northeast, and Southwest). These false springs could adversely affect agricultural yields, and the nonlinear trends in the changes would make false springs difficult to anticipate and prepare for.
4. Summary and conclusions
Here, dynamically downscaled global climate simulations are assessed over historical and future periods, focusing on PI that quantify the dormancy period of overwintering plants [chilling units (CU), which must accumulate to adequate values for plants to complete dormancy and produce new growth], the onset of spring [using the extended spring indices (SI)], and the prevalence of proceeding hard freezes (“false springs”).
By the end of the century and across the CONUS, CU are projected to decrease early during the “season” (1 October–1 May) over which they are accumulated (Fig. 3). Later in the CU season, larger decreases occur in the southern CONUS, while increases are projected in parts of the northern and western CONUS. These regionally dependent projected changes generally exceed ±500 CU and are occasionally as large as −1000 CU. Changes of such magnitudes would significantly affect some plant cultivars, given that several crops require an excess of 1000 CU to accumulate during dormancy (as discussed in section 2b). Increased accumulation of CU would cause plants to break their dormancy too early and become vulnerable to disease or freezing temperatures, potentially causing large losses due to damaged or lost crops (NOAA and USDA 2008; Ault et al. 2013). Conversely, reduced CU could fall below threshold values and inhibit fruiting or flowering, especially in the South and Southeast, which have the largest projected decreases in CU. These projected changes could necessitate the use of cultivars that are better adapted to winters with less chill, as well as the use of agricultural practices to mitigate low CU accumulation. For example, Parker and Abatzoglou (2019) studied changes in CU available in Georgia peach orchards and discussed the option of chemical applications or overhead irrigation to induce dormancy through enhanced evaporative cooling.
The simulations and metrics (SI) examined here projected earlier spring onset (Table 2, Fig. 5). Prior studies (e.g., IPCC 2014; Allstadt et al. 2015; Lipton et al. 2018; IPCC 2022) also project spring advancement. Here, spring onset is projected to advance by an average of 0.75 days per decade with WRF-CESM4.5 and by 1.9–4.0 days per decade within the runs driven by RCP8.5 throughout the projected future period and across the CONUS (Fig. 5). Despite the robust spring advancement across the CONUS, changes in false spring risk are more regionally and temporally variable than projected changes in SI. While long-term and widespread warming trends suggest a reduction in the frequency of hard freezes, an earlier onset of spring would leave plants vulnerable for prolonged periods to cold snaps as temperatures fluctuate throughout the transitional seasons. Here, the risk of early false springs decreases most in the Southeast, while the risk of late false spring is most decreased in the Ohio Valley or Upper Midwest regions (Fig. 6; Table 3). Regions in the western half of the CONUS tend to show increased risk of false spring. Adaptation to increases in the risk of false spring may be more difficult than adapting to SI changes, as false spring projections are highly variable among the regions and lack the more linear trend of projected change found for SI.
This work places emphasis on the importance of accurately simulating 2-m temperatures at subdaily scales, as biases in hourly 2-m temperature (whether derived from statistical functions or explicitly simulated) can impact the calculation of all PI examined here. The subdaily data that were simulated from dynamical downscaling enabled the calculation of the PI shown in the present work without relying on statistical functions to prescribe diurnal temperature patterns for each location. Another benefit of dynamically downscaled data is the self-consistent evolution of all atmospheric fields in response to a changing climate and unconstrained by observed data ranges in the historical record. A dynamically downscaled model can simulate changes to the ranges of extreme events without relying on the stationarity of historical relationships assumed in a statistically downscaled model. While the use of dynamical downscaling can augment our understanding of climate change impacts, it comes at a higher computational cost. Due to this computational expense, application of dynamical downscaling has and will continue in the foreseeable future to be more limited relative to statistical downscaling methods. However, it is important to consider climate change impacts using dynamically downscaled simulations for various reasons including the ability to simulate climate impacts at subdaily scales as well as changes that are more unpredictable, such as feedbacks in the climate system. Among other improvements, increasing the grid resolution to finer than that used here could reduce error for 2-m temperatures by better resolving land–water boundaries, spatial changes in land use, and topography, which could subsequently improve representation of historical PI. In future work, the use of finer resolution could alleviate model error over complex terrain, particularly refining these projections for agricultural operations concentrated in valleys where cold air outbreaks would affect the PI considered here. Additionally, increasing the ensemble size for dynamically downscaled data, as well as lengthening the period used for climatological representation (e.g., from decadal to 30 years) could increase the robustness of the findings shown here.
Using PI in a dynamical downscaling application to examine projected temperature changes is a novel aspect of this study. Using CU to express changes to monthly mean temperatures (as shown in Fig. 4) can communicate potential changes that are impactful to crops and horticulture. Temperature thresholds that are critical to capturing CU and SI can be compared with changes projected from regional climate simulations to better inform adaptation and resilience strategies for agriculture and ecosystems services management.
Acknowledgments.
The authors thank Dr. Mark Schwartz and two anonymous reviewers for their constructive feedback on this paper; Drs. Anna Jalowska and Shannon Koplitz also provided technical reviews of this article prior to its submission for publication. The views expressed in this article are those of the authors and do not necessarily represent the views or policies of the U.S. EPA.
Data availability statement.
The WRF Model is available through NCAR, which is sponsored and funded by the National Science Foundation. The versions of WRF that were used in this study can be downloaded from the internet (https://www2.mmm.ucar.edu/wrf/users/download/get_source.html). Extended Spring Indices data were obtained from the USA National Phenology Network (https://www.usanpn.org/geoserver-request-builder). The algorithm used to compute simulated SI values was also obtained from the USA NPN and is available at their GitHub repository (https://github.com/usa-npn/gridded_models). PRISM data were obtained from the PRISM Climate Group (http://prism.oregonstate.edu). Data used to generate the figures shown in this article are available online (https://data.gov/). The raw model outputs are available upon request from the corresponding author.
REFERENCES
Allstadt, A. J., S. J. Vavrus, P. J. Heglund, A. M. Pidgeon, W. E. Thogmartin, and V. C. Radeloff, 2015: Spring plant phenology and false springs in the conterminous US during the 21st century. Environ. Res. Lett., 10, 104008, https://doi.org/10.1088/1748-9326/10/10/104008.
Ault, T. R., G. M. Henebry, K. M. de Beurs, M. D. Schwartz, J. L. Betancourt, and D. Moore, 2013: The false spring of 2012, earliest in North American record. Eos, Trans. Amer. Geophys. Union, 94, 181–182, https://doi.org/10.1002/2013EO200001.
Ault, T. R., R. Zurita-Milla, and M. D. Schwartz, 2015: A Matlab© toolbox for calculating spring indices from daily meteorological data. Comput. Geosci., 83, 46–53, https://doi.org/10.1016/j.cageo.2015.06.015.
Bowden, J. H., T. L. Otte, C. G. Nolte, and M. J. Otte, 2012: Examining interior grid nudging techniques using two-way nesting in the WRF Model for regional climate modeling. J. Climate, 25, 2805–2823, https://doi.org/10.1175/JCLI-D-11-00167.1.
Bowden, J. H., C. G. Nolte, and T. L. Otte, 2013: Simulating the impact of the large-scale circulation on the 2-m temperature and precipitation climatology. Climate Dyn., 40, 1903–1920, https://doi.org/10.1007/s00382-012-1440-y.
Buckley, L. B., and M. S. Foushee, 2012: Footprints of climate change in US national park visitation. Int. J. Biometeor., 56, 1173–1177, https://doi.org/10.1007/s00484-011-0508-4.
Camargo-Alvarez, H., M. Salazar-Gutiérrezac, M. Keller, and G. Hoogenboomae, 2020: Modeling the effect of temperature on bud dormancy of grapevines. Agric. For. Meteor., 280, 107782, https://doi.org/10.1016/j.agrformet.2019.107782.
Chandler, W. H., 1942: Deciduous Orchards. Lea & Febiger, 438 pp.
Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569–585, https://doi.org/10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.
Daly, C., M. Halbleib, J. I. Smith, W. P. Gibson, M. K. Doggett, G. H. Taylor, J. Curtis, and P. Pasteris, 2008: Physiographically sensitive mapping of climatological temperature and precipitation across the conterminous United States. Int. J. Climatol., 28, 2031–2064, https://doi.org/10.1002/joc.1688.
Donnelly, A., and R. Yu, 2017: The rise of phenology with climate change: An evaluation of IJB publications. Int. J. Biometeor., 61, 29–50, https://doi.org/10.1007/s00484-017-1371-8.
Donner, L. J., and Coauthors, 2011: The dynamical core, physical parameterizations, and basic simulation characteristics of the atmospheric component AM3 of the GFDL global coupled model CM3. J. Climate, 24, 3484–3519, https://doi.org/10.1175/2011JCLI3955.1.
Erez, A., and S. Lavee, 1971: The effect of climatic conditions on dormancy development of peach buds. I. Temperature. J. Amer. Soc. Hortic. Sci., 96, 711–714, https://doi.org/10.21273/JASHS.96.6.711.
Fann, N., C. G. Nolte, P. Dolwick, T. L. Spero, A. Curry Brown, S. Phillips, and S. Anenberg, 2015: The geographic distribution and economic value of climate change-related ozone health impacts in the United States in 2030. J. Air Waste Manage. Assoc., 65, 570–580, https://doi.org/10.1080/10962247.2014.996270.
Fann, N., and Coauthors, 2016: Air quality impacts. The Impacts of Climate Change on Human Health in the United States: A Scientific Assessment, U.S. Global Change Research Program, 69–98, https://doi.org/10.7930/J0GQ6VP6.
Fishman, S., A. Erez, and G. A. Couvillon, 1987: The temperature dependence of dormancy breaking in plants: Computer simulation of processes studied under controlled temperatures. J. Theor. Biol., 126, 309–321, https://doi.org/10.1016/S0022-5193(87)80237-0.
Gent, P. R., and Coauthors, 2011: The Community Climate System Model version 4. J. Climate, 24, 4973–4991, https://doi.org/10.1175/2011JCLI4083.1.
Gerst, K. L., T. M. Crimmins, E. E. Posthumus, A. H. Rosemartin, and M. D. Schwartz, 2020: How well do the spring indices predict phenological activity across plant species? Int. J. Biometeor., 64, 889–901, https://doi.org/10.1007/s00484-020-01879-z.
Hayhoe, K., and Coauthors, 2007: Past and future changes in climate and hydrological indicators in the US Northeast. Climate Dyn., 28, 381–407, https://doi.org/10.1007/s00382-006-0187-8.
Herwehe, J. A., K. Alapaty, T. L. Spero, and C. G. Nolte, 2014: Increasing the credibility of regional climate simulations by introducing subgrid-scale cloud-radiation interactions. J. Geophys. Res. Atmos., 119, 5317–5330, https://doi.org/10.1002/2014JD021504.
Hong, S.-Y., and J.-O. Lim, 2006: The WRF Single-Moment 6-class microphysics scheme (WSM6). J. Korean Meteor. Soc., 42, 129–151.
Hong, S.-Y., Y. Noh, and J. Dudhia, 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 2318–2341, https://doi.org/10.1175/MWR3199.1.
IPCC, 2014: Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. C. B. Field et al., Eds., Cambridge University Press, 1132 pp.
IPCC, 2022: Climate Change 2022: Impacts, Adaptation and Vulnerability. H.-O. Pörtner et al., Eds., Cambridge University Press, 3056 pp., https://doi.org/10.1017/9781009325844.
Izquierdo-Verdiguier, E., R. Zurita-Milla, T. R. Ault, and M. D. Schwartz, 2018: Development and analysis of spring plant phenology products: 36 years of 1-km grids over the conterminous US. Agric. For. Meteor., 262, 34–41, https://doi.org/10.1016/j.agrformet.2018.06.028.
Jimenéz, P. A., J. Dudhia, J. F. González-Rouco, J. Navarro, J. P. Montávez, and E. Garciá-Bustamante, 2012: A revised scheme for the WRF surface layer formulation. Mon. Wea. Rev., 140, 898–918, https://doi.org/10.1175/MWR-D-11-00056.1.
Kain, J. S., 2004: The Kain–Fritsch convective parameterization: An update. J. Climate Appl. Meteor., 43, 170–181, https://doi.org/10.1175/1520-0450(2004)043<0170:TKCPAU>2.0.CO;2.
Karl, T. R., and W. J. Koss, 1984: Regional and national monthly, seasonal, and annual temperature weighted by area, 1895–1983. Historical Climatology Series 4-3, National Climatic Data Center, 44 pp., https://repository.library.noaa.gov/view/noaa/10238/noaa_10238_DS1.pdf.
Karlsson, B., 2014: Extended season for northern butterflies. Int. J. Biometeor., 58, 691–701, https://doi.org/10.1007/s00484-013-0649-8.
Kovaleski, A. P., 2022: Woody species do not differ in dormancy progression: Differences in time to budbreak due to forcing and cold hardiness. Proc. Natl. Acad. Sci. USA, 119, e2112250119, https://doi.org/10.1073/pnas.2112250119.
Kovaleski, A. P., B. I. Reisch, and J. P. Lando, 2018: Deacclimation kinetics as a quantitative phenotype for delineating the dormancy transition and thermal efficiency for budbreak in Vitis species. AoB Plants, 10, ply066, https://doi.org/10.1093/aobpla/ply066.
Kuden, A. B., Ö. Tuzcu, S. Bayazit, B. Yildirim, and B. Imrak, 2013: Studies on the chilling requirements of pecan nut (Carya illionensis Koch) cultivars. Afr. J. Agric. Res., 8, 3159–3165, https://doi.org/10.5897/AJAR12.1983.
Laube, J., T. H. Sparks, N. Estrella, J. Höfler, D. P. Ankerst, and A. Menzel, 2014: Chilling outweighs photoperiod in preventing precocious spring development. Global Change Biol., 20, 170–182, https://doi.org/10.1111/gcb.12360.
Lian, X., and Coauthors, 2020: Summer soil drying exacerbated by earlier spring greening of northern vegetation. Sci. Adv., 6, eaax0255, https://doi.org/10.1126/sciadv.aax0255.
Lipton, D., and Coauthors, 2018: Ecosystems, ecosystem services, and biodiversity. Impacts, Risks, and Adaptation in the United States: Fourth National Climate Assessment, Vol. II, D. R. Reidmiller et al., Eds., U.S. Global Change Research Program, 268–321, https://doi.org/10.7930/NCA4.2018.
Loveland, T. R., B. C. Reed, J. F. Brown, D. O. Ohlen, Z. Zhu, L. Yang, and J. W. Merchant, 2000: Development of a global land cover characteristics database and IGBP DISCover from 1-km AVHRR data. Int. J. Remote Sens., 21, 1303–1330, https://doi.org/10.1080/014311600210191.
Luedeling, E., and P. H. Brown, 2011: A global analysis of the comparability of winter chill model for fruit and nut trees. Int. J. Biometeor., 55, 411–421, https://doi.org/10.1007/s00484-010-0352-y.
Luedeling, E., M. Zhang, and E. H. Girvetz, 2009: Climatic changes lead to declining winter chill for fruit and nut trees in California during 1950–2099. PLOS ONE, 4, e6166, https://doi.org/10.1371/journal.pone.0006166.
Mallard, M. S., and T. L. Spero, 2019: Effects of mosaic land use on dynamically downscaled WRF simulations of the contiguous United States. J. Geophys. Res. Atmos., 124, 9117–9140, https://doi.org/10.1029/2018JD029755.
Mallard, M. S., C. G. Nolte, O. R. Bullock, T. L. Spero, and J. Gula, 2014: Using a coupled lake model with WRF for dynamical downscaling. J. Geophys. Res. Atmos., 119, 7193–7208, https://doi.org/10.1002/2014JD021785.
Martinuzzi, S., A. J. Allstadt, B. L. Bateman, P. J. Heglund, A. M. Pidgeon, W. E. Thogmartin, S. J. Vavrus, and V. C. Radeloff, 2016: Future frequencies of extreme weather events in the National Wildlife Refuges of the conterminous U.S. Biol. Conserv., 201, 327–335, https://doi.org/10.1016/j.biocon.2016.07.007.
Menzel, A., and Coauthors, 2006: European phenological response to climate change matches the warming pattern. Global Change Biol., 12, 1969–1976, https://doi.org/10.1111/j.1365-2486.2006.01193.x.
Miguez-Macho, G., G. L. Stenchikov, and A. Robock, 2004: Spectral nudging to eliminate the effects of domain position and geometry in regional climate model simulations. J. Geophys. Res., 109, D13104, https://doi.org/10.1029/2003JD004495.
Monahan, W. B., A. Rosemartin, K. L. Gerst, N. A. Fisichelli, T. Ault, M. D. Schwartz, J. E. Gross, and J. F. Weltzin, 2016: Climate change is advancing spring onset across the U.S. national park system. Ecosphere, 7, e01465, https://doi.org/10.1002/ecs2.1465.
Nanninga, C., C. R. Buyarski, A. M. Pretorius, and R. A. Montgomery, 2017: Increased exposure to chilling advances the time to budburst in North American tree species. Tree Physiol., 37, 1727–1738, https://doi.org/10.1093/treephys/tpx136.
NOAA and USDA, 2008: The Easter freeze of April 2007: A climatological perspective and assessment of impacts and services. NOAA/USDA Tech. Rep. 2008-1, 56 pp., http://www1.ncdc.noaa.gov/pub/data/techrpts/tr200801/tech-report-200801.pdf.
Nolte, C. G., T. L. Spero, J. H. Bowden, M. S. Mallard, and P. D. Dolwick, 2018a: The potential effects of climate change on air quality across the conterminous US at 2030 under three representative concentration pathways. Atmos. Chem. Phys., 18, 15 471–15 489, https://doi.org/10.5194/acp-18-15471-2018.
Nolte, C. G., and Coauthors, 2018b: Air quality. Impacts, Risks, and Adaptation in the United States: Fourth National Climate Assessment, Vol. II, D. R. Reidmiller et al., Eds., U.S. Global Change Research Program, 512–538, https://doi.org/10.7930/NCA4.2018.CH13.
Nolte, C. G., T. L. Spero, J. H. Bowden, M. C. Sarofim, J. Martinich, and M. S. Mallard, 2021: Regional temperature-ozone relationships across the U.S. under multiple climate and emissions scenarios. J. Air Waste Manage. Assoc., 71, 1251–1264, https://doi.org/10.1080/10962247.2021.1970048.
Okie, W. R., and B. Blackburn, 2011: Increasing chilling reduces heat requirement for floral budbreak in peach. HortScience, 46, 245–252, https://doi.org/10.21273/HORTSCI.46.2.245.
Otte, T. L., C. G. Nolte, M. J. Otte, and J. H. Bowden, 2012: Does nudging squelch the extremes in regional climate modeling? J. Climate, 25, 7046–7066, https://doi.org/10.1175/JCLI-D-12-00048.1.
Pan, L. L., S. H. Chen, D. Cayan, M. Y. Lin, Q. Hart, M. H. Zhang, Y. Liu, and J. Wang, 2011: Influences of climate change on California and Nevada regions revealed by a high-resolution dynamical downscaling study. Climate Dyn., 37, 2005–2020, https://doi.org/10.1007/s00382-010-0961-5.
Parker, L. E., and J. T. Abatzoglou, 2016: Projected changes in cold hardiness zones and suitable overwinter ranges of perennial crops over the United States. Environ. Res. Lett., 11, 03400, https://doi.org/10.1088/1748-9326/11/3/034001.
Parker, L. E., and J. T. Abatzoglou, 2019: Warming winters reduce chill accumulation for peach production in the southeastern United States. Climate, 7, 94, https://doi.org/10.3390/cli7080094.
Parkes, H., R. Darbyshire, and N. White, 2020: Chilling requirements of apple cultivars grown in mild Australian winter conditions. Sci. Hortic., 260, 108858, https://doi.org/10.1016/j.scienta.2019.108858.
Peterson, A. G., and J. T. Abatzoglou, 2014: Observed changes in false springs over the contiguous United States. Geophys. Res. Lett., 41, 2156–2162, https://doi.org/10.1002/2014GL059266.
Piao, S., and Coauthors, 2019: Plant phenology and global climate change: Current progresses and challenges. Global Change Biol., 25, 1922–1940, https://doi.org/10.1111/gcb.14619.
Riahi, K., and Coauthors, 2011: RCP8.5–A scenario of comparatively high greenhouse gas emissions. Climatic Change, 109, 33–57, https://doi.org/10.1007/s10584-011-0149-y.
Richardson, E. A., S. D. Seely, and D. R. Walker, 1974: A model for the estimation of completion of rest for ‘Redhaven’ and ‘Elberta’ peach trees. HortScience, 9, 331–332, https://doi.org/10.21273/HORTSCI.9.4.331.
Richardson, E. A., J. L. Anderson, and R. H. Campbell, 1986: The OMNIDATA BIOPHENOMETER (Ta45-P): A chill unit and growing degree hour accumulator. Acta Hortic., 184, 95–100, https://doi.org/10.17660/ActaHortic.1986.184.10.
Ruiz, D., J. A. Campoy, and J. Egea, 2007: Chilling and heat requirements of apricot cultivars for flowering. Environ. Exp. Bot., 61, 254–263, https://doi.org/10.1016/j.envexpbot.2007.06.008.
Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System Reanalysis. Bull. Amer. Meteor. Soc., 91, 1015–1058, https://doi.org/10.1175/2010BAMS3001.1.
Schwartz, M. D., 1993: Assessing the onset of spring: A climatological perspective. Phys. Geogr., 14, 536–550, https://doi.org/10.1080/02723646.1993.10642496.
Schwartz, M. D., and J. M. Hanes, 2010: Continental-scale phenology: Warming and chilling. Int. J. Climatol., 30, 1595–1598, https://doi.org/10.1002/joc.2014.
Schwartz, M. D., R. Ahas, and A. Aasa, 2006: Onset of spring starting earlier across the Northern Hemisphere. Global Change Biol., 12, 343–351, https://doi.org/10.1111/j.1365-2486.2005.01097.x.
Schwartz, M. D., T. R. Ault, and J. L. Betancourt, 2013: Spring onset variations and trends in the continental United States: Past and regional assessment using temperature‐based indices. Int. J. Climatol., 33, 2917–2922, https://doi.org/10.1002/joc.3625.
Shellie, K., A. P. Kovaleski, and J. P. Londo, 2018: Water deficit severity during berry development alters timing of dormancy transitions in wine grape cultivar. Sci. Hortic., 232, 226–230, https://doi.org/10.1016/j.scienta.2018.01.014.
Skamarock, W. C., and J. B. Klemp, 2008: A time-split nonhydrostatic atmospheric model for weather research and forecasting applications. J. Comput. Phys., 227, 3465–3485, https://doi.org/10.1016/j.jcp.2007.01.037.
Spero, T. L., C. G. Nolte, J. H. Bowden, M. S. Mallard, and J. A. Herwehe, 2016: The impact of incongruous lake temperatures on regional climate extremes downscaled from the CMIP5 archive using the WRF Model. J. Climate, 29, 839–853, https://doi.org/10.1175/JCLI-D-15-0233.1.
Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485–498, https://doi.org/10.1175/BAMS-D-11-00094.1.
USA National Phenology Network, 2019: Gridded historical maps (1981–near real-time). USA-NPN, accessed 18 September 2019, https://doi.org/10.5066/F7SN0723.
van Vuuren, D. P., and Coauthors, 2011: The representative concentration pathways: An overview. Climatic Change, 109, 5–31, https://doi.org/10.1007/s10584-011-0148-z.
