This paper examines recent southern New York State climate changes as reflected in a detailed hourly climate record collected about 110 km due north of New York City since 1988, including comprehensive surface radiation data. Comparing 1988–2000 and 2001–14 means, the area has warmed, dominated by a 0.5–0.7-K summer warming. Daytime warming exceeds nighttime’s warming. Warming is not due to enhanced downward longwave flux but arises from increased incident solar fluxes accompanying declining aerosol loads. Local warming is shown to stem from a large-scale response to increased solar forcing, the key element of which is an accelerated summer hydrological cycle: increased precipitation, with smaller evaporation increases leading to large, significant soil moisture and runoff increases. Much of the accelerated summer hydrological cycle is shown to arise as a result of an anomalous low-level cyclonic motion centered on the mid-Atlantic U.S. coast, rendering the results regional rather than local. Analyzing the stability and CAPE budgets of mean and individual summer profiles over the studied site provides a diagnostic explanation of the observed warming and accelerated hydrometeorology due to enhanced solar fluxes. The study reveals a complex suite of (thermo)dynamic feedbacks to radiative forcing of which surface warming is but one element, reiterating and re-emphasizing that surface temperature trends may be embedded in far richer physics than greenhouse gas–induced radiative forcing alone.
Anthropogenic climate change dominates current climate science (IPCC 2013). While there is firm theoretical basis for expecting surface warming to accompany greenhouse gas (GHG) buildup, attributing observed climate changes to greenhouse warming (Wallace et al. 2013; Imbers et al. 2014) is still incomplete (e.g., Pall et al. 2011; Parmesan et al. 2011), for at least two reasons. First, observational attribution efforts suffer from a dearth of long-term surface radiation datasets (Wang et al. 2012). The key observational surface radiation networks have only one or two short-term (only several years long) sets of U.S. station data between 80°W and the Atlantic Ocean (Ohmura et al. 1989, 1998; Augustine et al. 2000, 2005; Wild et al. 2005). Second, local or regional warming may or may not be related to GHG buildup (e.g., Ramanathan and Feng 2009).
Because of the above ambiguities, key evidence for anthropogenic climate change—site-specific analyses explaining mechanistically observed trends and attributing them either directly to rising GHG concentrations or secondarily to regional dynamic responses to this rise—is still incomplete. Expanding the body of such analyses is the key contribution of this paper.
2. Data and methods
a. The main dataset
This paper is primarily based on an original instrumental dataset spanning 1988–present from the Cary Institute of Ecosystem Studies (CIES; http://www.caryinstitute.org), at 41°47′N, 73°44′W (hereafter “the CIES site”). The site is in the Hudson River valley, near Millbrook, New York, about 110 km north of New York City.
The instruments are located 128 m above sea level, amidst a periodically mowed flat open grassland embedded in a regional mosaic of mature forests, rural hamlets, and farmland characteristic of much of the rural northeastern United States. The CIES site and its broader area have changed remarkably little in recent decades. While U.S. Census Bureau total population estimates only span 2010–14, over this time, Millbrook’s population range was 22 people. Nearby Stanford village exhibited a population range of 19 people. Both records show no steady upward or downward trend and comprise under 1% of the respective populations.
Of the variables measured at the CIES comprehensive climate monitoring site, I use the ones shown in Table 1. Rather uniquely, the analyzed record includes radiometer-based data for most key radiative terms, and resolves both the seasonal and diurnal cycles. High-frequency (0.25 Hz) meteorological measurements are processed in real time, and the minimum, maximum, and mean of each 1-h period are recorded. The hourly means are analyzed here. Preliminary quality control is performed at the CIES site, and data are periodically compared with data from a sister station 13 km to the east (Kelly et al. 2009; Temimi et al. 2014). Additional technical information about the sensors, their technical history, and the quality control protocol can be found online (http://www.caryinstitute.org/science-program/research-projects/environmental-monitoring-program). The dataset contains very few brief temporal gaps. Because of these, and because several instrument changes and replacements occurred since 1988 (see the abovementioned URL), I examined closely potential discontinuities. For each gap or instrumental change event I compared both means and distributions of the before and after subpopulations. Using a standard Monte Carlo resampling protocol described in section 2c below to derive uncertainty ranges for the sample means and the relative frequency of the discrete distribution bins, I found no evidence for discontinuity.
Because a handful of observations fall far enough outside the range spanned by the bulk of the data to appear erroneous, I exclude hourly data failing to satisfy minimum ≤ mean ≤ maximum, or falling outside 1) physically plausible, expected ranges (e.g., [−30°, 45°C] for temperature or [5%, 100%] for relative humidity); 2) a reasonable range of temporally neighboring data points (eliminating a handful of unusually high-amplitude spikes that may represent faulty readings); and 3) ±4.5 standard deviations of the long-term mean. Criteria 2 and 3 mostly eliminated the same few likely erroneous points. While they exclude a handful of potentially real extreme events, they are too few to impact meaningfully the analyses and conclusions.
To get a deeper dynamical understanding of the local observations at the CIES site, I augment them spatially by using the National Centers for Environmental Prediction (NCEP)–U. S. Department of Energy (DOE) AMIP phase 2 (AMIP-II) reanalysis (R-2; Kanamitsu et al. 2002). While not purely observed, observed data strongly impact reanalysis states every few hours, and the observational array in the considered area is very dense. The reanalyzed data thus closely track observations, with potential loss of observational fidelity more than made up for in continuity, dynamical consistency, uniformity, and stringent quality control. R-2 variables used in this analysis are also shown in Table 1.
c. Climatological change estimates
The key change indicator I use here is the 2001–14 climatology minus that of 1988–2000, denoted by δ and referred to as anomaly or epochal or decadal differences. I compute those differences individually for each available variable at each hour of the day, in fifteen 25-day-wide day-of-year bins (centered on Julian days 12, 37, 62, …, 362). I also condense those into daily, seasonal, or annual mean climatology differences.
Hourly means use 3-h windows—e.g., the 0300–0400 LT bin is based on 0200–0500 LT observations—with relative weights of 3 and 2 for the central and fringe hours. The total number of a given variable’s hourly realizations is upward of 2 × 105, with over 90% of individual hourly/day-of-year bins incorporating more than 103 hourly means.
Following Eshel et al. (2000) and Eshel (2011), I derive significance estimates at specified p values from 250-member Monte Carlo (MC) ensembles, a nonparametric approach that does not require that the variables are from a Gaussian distribution. In each MC realization, the climatologies are recalculated using only 70% of the observations available for each of the two climatological periods, with the discarded 30% randomly chosen individually for each period, bin and variable. Of the sorted 250 MC members of each experiment, for example, elements 248 and 2 (the closest integers corresponding to a two-tailed p value ≤ 0.01 criterion) give the top and bottom of the distribution’s central 99%. Thus all significance levels in this paper are two-tailed. A climatology difference is considered significant when it falls inside the range the two bounds span and that range excludes zero.
Because of the possibility of spurious trends in R-2 data due to shifting remote sensing algorithms or analytic methods, I have repeated all calculations for mean differences reported here using two other differenced epochs: 1989–99 versus 2002–13 and 1988–2002 versus 2003–14. In both test cases, all results were fully reproduced within 4%–6% of the results reported here.
3. Observed temperature changes
Figure 1 presents the temperature differences δT. Figure 1a shows that the later period is p < 0.01 significantly warmer than the earlier one throughout nearly all considered hours, except during the year’s 4–5 coldest weeks. Consequently, in the remainder of the paper I focus on summer, defined expansively based on Fig. 1a as Julian days 75–349.
Figure 1b presents mean daytime and nighttime (2000–0400 LT) δT values over summer (red) and the full year (green). While (not shown) winter changes are small and insignificant, summer warming is p < 0.01 significant in the 3-hourly ranges considered. Throughout, warming is most pronounced during warmer parts of the day, with daytime and nighttime warming differences 0.1 < p < 0.05 significant. Annual (summer) mean warming is 0.43 (0.62) K. These trends—0.32 and 0.46 K decade−1, respectively, given the 13 years separating the centers of the two differenced epochs—are consistent with warming nearby R-2 grid points exhibit (not shown) and also with the long-term station at Albany International Airport, some 100 km to the northwest, with values of 0.40 and 0.57 K decade−1 for the full year and summer, respectively. These comparisons place the current analysis in a broader context.
4. Observed radiation changes
Given the dominance of the idea that “warming is due to increased atmospheric GHGs” in current climate discourse, I begin by testing its applicability to observed warming at the CIES site (Fig. 1) by comparing observed radiative changes at the CIES site to those expected as a result of the 47-ppmv CO2 buildup from 1998 to 2014.
Observed radiative changes are epochal incident radiation differences between net () and longwave radiation (L↓; notation is standard but defined formally in appendix A). Figure 2 reveals O(1–10) W m−2 epochal differences, with annual means of +1.6 and +1.5 W m−2 for total and longwave radiation (rightmost bars in Figs. 1c,d; p < 0.01 and p < 0.61, respectively). This δR↓net corresponds to δtR↓net ≈ 1.2 W m−2 decade−1, significantly smaller than the δtR↓net ≈ 8.2 W m−2 decade−1 reported by Augustine and Dutton (2013) for the surface radiation budget observing network (SURFRAD; with only the Penn State location east of 88°W). However, while these authors do not provide station specific trends, their Fig. 3 suggests a much smaller δtR↓net at the Penn State location.
For expected radiative changes, I use the Fu–Liou radiative transfer model (FL2; Fu and Liou 1992, 1993; Fu et al. 1999), which permits custom estimates with specification of not only GHG mixing ratios, but also temperature, water vapor, and cloud cover profiles. While the model can also consider aerosols (which I use later), here I highlight the role of anthropogenic CO2 buildup by excluding aerosols.
Using R-2 temperature, water vapor, and cloud cover climatologies over the CIES site, the model maps 47 ppmv of CO2 onto δR↓net ≈ 0.14 W m−2. Adding water vapor forcing by replacing qclim with qclim + δq raises the estimate slightly to 0.17 W m−2. These 0.1–0.2 W m−2 additions to R↓net are 10–100 times smaller than the actual changes observed at the CIES site (Fig. 2) summarized above. Using the nonlocal approximation of Pierrehumbert [2014; his Eq. (1)] yields δR↓net ≈ 0.67 W m−2, still only less than to almost half the observed R↓net. Observed warming at the CIES site (Fig. 1) is thus not primarily attributable to GHG warming.
Attributing warming at the CIES site to GHGs also contradicts several other observations at the CIES site. First, surface GHG warming is expected to maximize at night, when IR cooling to space dominates (Wild et al. 2007). Yet daytime (0900–1700 LT) additions dominate radiative changes at the CIES site, and both the annual and summer positive mean δR↓net values are exclusively due to daytime increases overwhelming nighttime (2000–0400 LT) decreases (Figs. 2a,c), with summer and annual nighttime R↓net of −1 and −2 W m−2, respectively.
Second, δT evolution at the CIES site is very different from the reasonably expected response—calculated in appendix B—to the observed epochal differences of either radiation alone (δR↓net) or of radiation plus turbulent surface fluxes [δ(R↓net + QL + QS) (Boer 1993; Trenberth 1999)]. Figure 3 presents the actual and expected δT values at the CIES site (using δR↓net from the CIES site data, and R-2 turbulent fluxes δQL+S). Figure 3 shows that surface fluxes can produce large thermal responses, and that some of the CIES site δT seasonal evolution may be partly forced by those anomalous surface fluxes. Yet the gross mismatch between observed and expected δT in Fig. 3 (which is conservative because the assumed depth of 4 m for the thermally active layer is an upper bound) indicates that there is clearly more to warming at the CIES site than surface fluxes.
To better understand observed warming at the CIES site—in particular the relative roles of radiation, surface fluxes, and atmospheric column processes—requires a deeper understanding of radiative and thermal balances at the CIES site. That is our next objective.
a. A closer look at surface radiation
1) Calculated L↓
The key to GHG warming is increased downward surface longwave flux L↓. Consistently, L↓ at the CIES site—calculated as in appendix A as it is not explicitly measured at the CIES site—has indeed increased (Fig. 2d) by 5.5 and 1.5 W m−2 when averaged over the full summer and year, respectively (with the latter only significant). Yet surface warming at the CIES site has induced L↑ increases significant at p ≪ 0.1. [This is in contrast to the essentially unchanged network-wide L↑ values that Augustine and Dutton (2013; their Fig. 5d and Table 2) report over 1996–2011.] Summer and year δL↑ values are +3.2 and +2.2 W m−2, skewed toward daytime; the respective night (day) values are 2.7 (3.9) and 1.8 (2.9) W m−2, consistent with the stronger daytime warming (Fig. 1) reported earlier.
Thus net incident longwave radiation L↓ − L↑ at the CIES site has actually modestly decreased, inconsistent with attributing the CIES site warming to GHG buildup. Figures 2b and 2d cast further doubt on this attribution, showing that—as in the δR↓net case discussed earlier—L↓ increases are likewise dominated by 10–20 W m−2 summer daytime (0900–1700 LT) increases.
If not GHGs, what causes most of the positive surface longwave forcing at the CIES site? Apart from air pollution, which I address later, such changes may accompany changing atmospheric temperatures, water vapor, or cloud cover [e.g., Schmetz 1989; Räisänen 1996; see also section 8.6.3 of IPCC (2007)].
To resolve δL↓ values due to warming, water vapor δe, and cloud changes δc, the combined R-2 δL↓ does not help. Rather, an explicit L↓ = L↓(T, e, c) formalism is needed. I devise such a formalism by linearizing the Sedlar and Hock (2009, hereinafter SH9) atmospheric emissivity parameterization (see appendix C), mapping observed δT, δe, and δc onto expected δL↓ values. Because decadal differences are 1%–10% of the respective climatologies, I assume linearity, with the sum of individual contributions yielding the full longwave response, δL↓ ≈ δL↓(δT) + δL↓(δe) + δL↓(δc). I also calculated individual δL↓ using the Fu–Liou radiative transfer model introduced earlier, with resultant δL↓ values within ±1–2 W m−2 of the reported SH9 δL↓ values and of like sign.
Using observed CIES hourly δT and δq and 6-hourly R-2 total cloud cover as input into SH9 (see appendix C) yields the expected summer surface δL↓ Fig. 4 presents. The sums of these expected values (rightmost red bars in Fig. 4) can be readily compared with the actual δL↓ observed at the CIES site (Figs. 2b,d and the gray bar in Fig. 4). While nighttime (daytime) total expected δL↓ values overestimate (underestimate) observed ones, the full-day calculated summer δL↓ tracks observations reasonably well (3.9 and 5.5 W m−2, respectively) and thus permit the desired partitioning into δL↓(δT), δL↓(δe), and δL↓(δc).
Summer full-day cloud cover differences are trivial, inducing no expected surface longwave forcing. Conversely, δe ≈ 51.3 Pa and δT ≈ 0.62 K, inducing δL↓(δe) ≈ +1.2 W m−2 and δL↓(δT) ≈ +2.7 W m−2. (Note that this value of δe ≈ 51 Pa is less than half the 118 Pa needed to maintain relative humidity given the 0.62-K warming. Something is drying the surface layer, as will be discussed later.) While it is possible that δe and δT are shallow, thus representing L↓ poorly, the summer R-2 δe(p) and δT(p) profiles—with δ > 0 over 1000–300 mb (1 mb = 100 Pa), δT actually increasing from the surface to 700 mb and maintaining a fixed amplitude to 300 mb, and δe decaying upward slowly, with δe500mb ≈ 0.25δe1000mb—show that this is not so. Thus the observed summer downward longwave radiation increase at the CIES site is 70% due warming of the troposphere, and 30% due to its moistening.
2) Solar Radiation S
Solar radiation at the CIES site exhibits large, coherent changes (Fig. 5). The three left bars show that direct beam, global, and net downward solar radiation [S↓u, S↓g, and S↓g(1 − as), respectively; appendix A] have all increased significantly in summer and annually. The annual and summer mean δS↓net ≈ 4 W m−2 at the CIES site are key to reversing the δ(L↓ − L↑) < 0 into the summer and annual δR↓net ≈ 4.4 and 1.6 W m−2.
The one downward solar flux that has actually p ≪ 0.1 significantly decreased is the scattered (diffuse) flux, δS↓d ≈ −2.5 and −1.4 W m−2 in summer and annually (Fig. 5, rightmost bars). Thus the summer and annual δS↓net > 0 arise because δS↓u > |δS↓d|.
While it is possible that cloud changes have caused the above observed longwave and shortwave radiative changes, this proves not to be the case. To yield the above observed large S↓u (incident direct solar) increases or S↓d (scattered incident solar) declines require total and/or low cloud cover to decrease, yet both have minutely increased: R-2 total and low cloud epochal differences over the CIES site are +0.78% and +0.61% of the sky, respectively, diametrically opposite to the above expectations. When forcing the Fu–Liou radiative transfer model with otherwise climatological summer conditions characteristic of the CIES site, these cloud cover increases reduce surface noon incident global solar flux from 877.2 to 874.3 W m−2 (i.e., −3 W m−2 instead of the observed +4 W m−2) and increase the calculated S↓d from 244.4 to 248.9 W m−2 (i.e., +4.5 W m−2 instead of the observed −2.5 W m−2).
A plausible explanation for this is that the main atmospheric scatterer over the CIES site in recent decades has been pollution, not clouds, and that the increased incident solar fluxes observed over the CIES site are the local manifestation of the established reversal of “global dimming” into “global brightening” (where global refers to global solar radiation, not to global scale; e.g., Liepert 2002; Kim and Ramanathan 2008; Wild 2012; Long et al. 2009). In addition to this body of literature, this notion is consistent with several geographically pertinent observations.
First, summer daily mean concentrations of particulate matter with diameter of less than 2.5 μm (PM2.5) at the Poughkeepsie–Newburgh–Middletown air quality site—core-based statistical area (CBSA) station 39100, averaging point measurements from an approximate 50-mile radius with the CIES site near its center—has been p ≪ 0.1 declining by 5.6 ± 0.08 μg m−3 decade−1 since 1999. In addition, Fig. 6 presents sensitivity of FL2 surface incident solar fluxes to near-surface aerosol (assumed sulfate droplets) optical depth (AOD) τ. The shown range of 0.05 ≤ τ ≤ 0.75 spans most days [e.g., Ramanathan et al. (2001, their Fig. 2) or Ginoux et al. (2006, their Fig. 7)] except for extreme “air quality action days” [e.g., Dickerson et al. (1997) report τ ≈ 1–2 for the historic 1995 pollution event there]. Figure 6a herein shows that ∂τS↓u < 0 (open squares) and ∂τS↓d > 0 (open diamonds), as expected, but |∂τS↓d| < |∂τS↓u| so that overall ∂τS↓u+d > 0 (solid circles). That is, lower aerosol loads (leftward in Fig. 6a) enhance net downward solar flux by increasing unscattered flux more strongly than the scattered flux is decreasing, as observed at the CIES site. In Fig. 6b, I use these calculations to estimate the τ ranges characterizing 1988–2000 (green) and 2001–14 (yellow): any given S↓u value (a horizontal line) intersects the S↓u(τ) curve, and the τ value at the intersection point (a vertical line) determines the AOD to which the specific S↓u value corresponds. Figure 6b shows that the central 99% of the CIES site epochal S↓u distributions (Fig. 6b’s vertical axis) correspond to nonoverlapping calculated τ distributions, emphasizing the impact of improving air quality and explaining δS↓u,g,net > 0 at the CIES site.
b. Summary of radiation results
Net downward surface radiation R↓net has increased in summer and annually.
Nighttime R↓net has decreased throughout the year.
While L↓ has increased, especially in summer, L↓net = L↓ − L↑ has actually declined.
Most L↓ and R↓net gains are realized in times of maximum insolation.
Nighttime L↓ has increased minimally, in summer only.
Summer L↓ increase is mostly due to warming and moistening.
Incident direct (scattered) solar fluxes have increased (decreased) throughout the year.
Direct and net solar flux increases are mostly due to improving air quality.
5. Spatial relevance
Figure 7 addresses statistically the spatial generality of the key radiative changes reported above. It shows maps of temporal correlation over January 1988–June 2014 between various radiation time series observed at the CIES site and corresponding R-2 flux time series throughout the shown geographical domain. The maps show that all radiative fluxes measured at the CIES site also are not unique, occurring instead over a wide swath of eastern North America between North Carolina and Nova Scotia. The following section builds on and further illuminates this point.
6. Mechanistic interpretation
Because the characteristic response time scales of most elements of the land–atmosphere system to radiative anomalies span hours to weeks, over the decadal time scale addressed here the system is quasi-equilibrated. Consequently, the following diagnostics distinguish various physical consistencies that characterize the post-2000 land–atmosphere system from those that characterized the pre-2000 system, assuming a statistical equilibrium throughout. As discussed in the remainder of this paper, accelerated surface hydrology, air column hydrometeorology, and land–atmosphere fluxes are the key elements of the overall physical response of the northeastern United States to imposed changes.
Of the plethora of epochal differences highlighted above, declining aerosol loads, in all likelihood due to improved air quality measures (e.g., Kim and Ramanathan 2008; Leibensperger et al. 2012a,b; Mickley et al. 2012), come closest to purely external forcing. They thus make a logical starting point for the following mechanistic discussion of summer changes.
Recall that summer air quality improves significantly, as quantified by the aforementioned p < 10−8 significance, with a decline of 5.6 ± 0.08 μg m−3 decade−1 over 1999–2014 of daily mean PM2.5 concentrations at the Poughkeepsie–Newburgh–Middletown air quality site. Consistently, the summer solar flux response is large, coherent, and significant: approximately +7.2 and −2.5 W m−2 changes in daily mean direct and scattered downward solar radiation fluxes (leftmost and rightmost bars in Fig. 5d), with a net change of δS↓net ≈ +4 W m−2 (Fig. 5a). Substituting the CIES site summer δS↓net ≈ δR↓net ≈ 4 W m−2 (Fig. 2c) in the numerator of Eq. (B1) and integrating over the seven summer months yields expected warming of 12–16 K by late October, well in excess of the observed (Fig. 1). As Fig. 3 has already suggested, secondary internal responses (notably accelerated hydrology) must negate much of the warming the added incoming radiation would have caused by itself, keeping it K.
To begin our inquiry into these moderating processes, Fig. 8a shows regional near-surface temperature and flow epochal differences. The δT(x, y) pattern resembles observed aerosol removal patterns [e.g., Figs. 2, 4, and 5 in Leibensperger et al. (2012a)] and corresponding patterns of calculated resultant radiative forcing [e.g., top panel of Fig. 7 in Leibensperger et al. (2012a) or Fig. 1c in Mickley et al. (2012)]. These patterns depict a large-scale climate response, into which the data from the CIES site afford a highly resolved window.
Figure 8a places the CIES site at the northern flank of a large-scale thermal anomaly ringed by dynamically consistent anomalous cyclonic flows. On the eastern flank of this cyclone are strong anomalous southerlies, and to its north are weakened westerlies. Figure 8b shows that these flows produce near-surface convergence of moist static energy s over the Atlantic Ocean centered at 38°N, 68°W and divergence over Pennsylvania centered at 40°N, 79°W. While the CIES site is in neither of these nodes, it is within the weakened westerlies to the north of the cyclone; averaged over 1000–925 mb, δ(u, υ) ≈ (−0.52, −0.22) m s−1 over the CIES site, with the resultant northeasterly anomaly of around 0.6 m s−1 placing the CIES site downstream of the cool, moist Gulf of Maine. Because the response is strongly baroclinic, at 850–700 mb, δ(u, υ) ≈ (−0.45, 0.33) m s−1, yielding a southeasterly anomaly of 0.6 m s−1 (not shown) that places the CIES site downstream of the subtropical Atlantic.
a. Summer energy anomalies
Because of the sharp, height dependent thermodynamic gradients they traverse, the wind anomalies strongly impact stability over the CIES site. To quantify this, I distinguish dry and saturated advective moist static energy s tendencies. For simplicity, I hold the latent heat L = 2.42 × 106 J kg−1 and specific heat at constant pressure cp = 1005 J kg−1 K−1 fixed (rounded to the nearest integer, this cp holds for −50° ≤ T ≤ 40°C, over which L varies by ±2%). With these approximations,
where the d and s subscripts denote the dry and moist contributions to overall moist static energy; and denote the lateral flow and gradient on pressure (p) surfaces, with vertical counterparts ω and ∂p; and θυ denotes the virtual potential temperature.
I analyze the lower troposphere, distinguishing two sublayers: surface–888 mb (top of the R-2 925-mb level) and 888–550 mb (top of the R-2 600-mb level). Reported values are vertical averages for lateral terms (because of motions on isobaric x–y planes) and the top and bottom interface flux differences for vertical terms. To recast energy convergence as equivalent surface fluxes and vice versa, I use summer climatologies over the CIES site of geopotential height, h550,880 ≈ (5021, 1216) m, and vertical mean (from the surface to the subscript p) density ρa550,880 ≈ (0.94, 1.14) kg m−3.
Figure 9 presents the summer mean epochal differences in moist static energy tendencies advectively induced by lateral and vertical flow anomalies. Lateral sensible heat convergence anomalies (purple in Fig. 9a) stabilize the troposphere for p ≥ 550 mb, essentially warming the upper layer while cooling the lower. Saturation reduces but does not qualitatively change this, with the cpθυ + Lq sum (red in Fig. 9a) still an energy source aloft and a near-surface energy sink.
Vertical contributions (Fig. 9b) are shaped by later period suppression of climatological summer mean subsidence; δω < 0 for 180 mb, with δω(700 mb) ≈ −16 mb day−1 most negative, and mb day−1. Because ∂zθυ > 0 (all time-mean profiles are stable), the weakened subsidence means reduced warming (purple in Fig. 9b are both negative). Vertical Lq contributions operate differently in the two layers. In the lower layer, ∂zq < 0 (drying with height) dominates, and the weakened subsidence means reduced drying or moistening, with Lq addition nearly canceling out the strong cooling (Fig. 9b, lower bars). In the upper layer, (while drying with height continues, its magnitude drops markedly), and ∂zδω < 0 (decreased subsidence drops with height) dominates, weakly drying the upper layer and thus increasing its anomalous energy sink -related to cpθυ (in the upper bars of Fig. 9b, the red is a bit more negative than the purple).
Figure 9c presents the sums of lateral and vertical advective terms, the ∂ts induced by the differences between the two analyzed periods in terms of interactions of three-dimensional flow and gradients. With no condensation (Fig. 9c, purple), the energy sink is smaller aloft than near the surface, and the layer of p ≥ 550 mb is stabilized at an approximate rate1 of 420 − 229 = 191 J kg−1 day−1. I discuss stability under saturation after properly introducing the impact of surface flux changes below.
b. Summer hydrological responses
Consistent with the expectation that accelerated hydrometeorology keeps warming at the CIES site well below levels dictated by anomalous radiation, or even radiation plus surface latent and sensible flux anomalies, summer precipitation at the CIES site has likely been increasing. Mean summer epochal differences of total and convective R-2 precipitation rates are δPt ≈ 0.86 kg m−2 day−1 and δPc ≈ 0.73 kg m−2 day−1, both significant at p < 0.01. (With ρw = 103 kg m−3, these P values are numerically the same in millimeters per day.) The measured precipitation change at the CIES site—δPt ≈ 0.4 kg m−2 day−1, with the range [−0.15, 0.95] kg m−2 day−1 significant at p ≤ 0.05 and δPt > 0 kg m−2 day−1 significant at —tentatively corroborates this.
Summer R-2 mean upward latent surface heat flux epochal difference is δQL ≈ 12.2 W m−2, corresponding to δE = δQL/L ≈ 0.43 kg m−2 day−1 added evaporation. Thus, despite the added precipitation’s predominantly convective nature (suggesting a high runoff fraction), approximately 50% re-evaporates locally (δE ≈ 0.43 kg m−2 day−1 ≈ δPt/2). That is, as expected, hydrometeorology and surface hydrology have accelerated in tandem, with enhanced precipitation—predominantly convective—minus evaporation (P − E) a centerpiece of this acceleration. The R-2 summer mean P − E from 1988 to 2000 at the CIES site was −0.18, reversing to +0.27 kg m−2 day−1 in 2001–13 [a δ(P − E) significant at p < 0.05]. (For completeness, the winter net water balance at the CIES site is essentially unchanged, with δP > 0 kg m−2 day−1 only significant.)
In keeping with the thrust of section 5, these trends are not unique to the CIES site. Averaged over 40°–44°N, 70°–80°W, for example, R-2 April–October δ(P − E) ≈ +0.21 kg m−2 day−1 (but this difference is only 0.1 < p < 0.2 significant). The excess P − E is balanced mostly by added runoff, with a small additional storage. Averaged over the above area, the April–October mean R-2 volumetric soil moisture fractions, 0.28 and 0.30 in the early and late epoch, respectively, differ significantly at p < 0.01. Valid over a 0.1–2-m soil depth range, the 0.02 addition amounts to 38 kg m−2 of additional resident water.
To quantify runoff trends, I use the only long-term streamflow record in the area, that of Wappinger Creek (USGS hydrologic unit number 02020008), whose catchment, fortuitously, includes the broad area surrounding the CIES site. The gauging station is at 41°39′N, 73°52′W, roughly 25 km southwest of the CIES site, in the town of Wappinger Falls, New York, upstream of which the catchment area is 470 km2. The creek’s climatological flow peaks in March at 14 m3 s−1, declining to an August–September minimum of 3 m3 s−1. Considering the catchment surface area, Wappinger Creek annual and winter flow rates have been rising insignificantly [with best estimates and ranges significant at p ≤ 0.05 of 0.83 (−0.15, 1.87) and 0.35 (−1.21, 2.20) kg m−2 day−1 decade−1]. Yet at 1.33 (0.29, 2.47) kg m−2 day−1 decade−1, summer flow increase since 1988 is p ≪ 0.01 significant.
Returning to atmospheric stability, added evaporation can impact stability in three ways. First, water vapor enrichment reduces air density via the virtual effect, equivalent to eroding stability by surface warming. Second, evaporation is a surface heat sink, stabilizing by redirecting excess energy from raising temperatures toward phase change instead. Third, if recondensation occurs, and high enough above the surface, it stabilizes by liberating aloft latent heat energy consumed at the surface. As before, I adhere to the sign convention of added buoyancy being positive.
Neglecting again T dependence on cp and L, the virtual effect impact on the lower layer is
where ≈ 286.9 K, h880 ≈ 1216 m, and ρa880 ≈ 1.14 kg m−3 are again the climatological summer R-2 air temperature, height and density of the surface–880-mb layer. Added upward vapor flux thus destabilizes at a rate equivalent to 53 J kg−1 day−1 of warming. The column over the CIES site is further destabilized by the effect of anomalous surface fluxes, (δR↓net + δQS+L)/(h880ρa880) ≈ 138 J kg−1 day−1 of added surface warming. Modified surface fluxes thus destabilize the column by 138 + 53 = 191 J kg−1 day−1. However, recall that with no saturation, anomalous advective cpθυ flux divergences stabilizes the column with p ≤ 500 mb by 420 − 229 = 191 J kg−1 day−1. Since these are equal and opposite, absent condensation, overall lower troposphere stability is unaltered.
Saturation (Fig. 9c, red) changes matters. Aloft, moisture anomalously diverges, so a putative saturated upper layer experiences a 52 J kg−1 day−1 anomalous (δ) equivalent cooling. In the surface layer, moisture anomalously converges, so saturation (tapping into the potential energy source Fig. 9c represents) is equivalent to low-level warming of ≤277 J kg−1 day−1. Since any combination of cooling aloft and/or surface warming erodes the lower troposphere (p > 550 mb) static stability, these potential destabilization sources are additive, for a total destabilization rate of ≤329 J kg−1 day−1. This presents a potential nonlinearity. If some lofting mechanism (e.g., enhanced mechanical stirring) is vigorous enough to overcome ambient convective inhibition An, the more condensation occurs, and the more likely subsequent condensation becomes. This scenario—in which most of the added surface radiative energy is directed not toward warming, but toward irreversible work against otherwise stable stratification and toward accelerating the hydrological cycle—in fact dominates recent climate change over the CIES site, where precipitation has been indeed rising, as reported earlier, mostly convectively.
Figure 10 further explains the added prevalence of moist processes at the CIES site, revealing the thermodynamic roots of the accelerated hydrological cycle there. The key calculation of Fig. 10 follows Eshel and Farrell (2001), who in turn followed Emanuel (1994). Surface parcels, characterized by temperature and humidity observed at the CIES site, are lifted dry adiabatically to their lifting condensation level pLCL. From there, they rise pseudoadiabatically to 100 mb in increments of dp = 1 mb. Because Γs and the Tυ in the next level up are mutually dependent, both are calculated simultaneously iteratively at each level starting from the calculated LCL Tυ. The negative and positive areas are then calculated by vertical summation between pLCL and the level of neutral buoyancy (i.e., over pLNB ≤ p ≤ pLCL) of
where Rd is the dry air gas constant, i is the level superscript, the hat and its absence respectively denote the lifted parcel and the environment, summer θυ(p) is calculated from R-2 temperature and relative humidity profiles, and An,p ≥ 0. Figures 10a–d analyze climatological profiles, while Fig. 10e summarizes analyses of individual R-2 summer 6-hourly profiles (11 027 in 1988–2000 and 11 266 in 2001–13).
The climatological profiles (Figs. 10a–d; the mean vertical structures that impact, but are mostly shaped by, moist ascent), show a slightly increased likelihood for moist convection in the latter epoch. First, convective inhibition is declining; ≈ 263 J kg−1 while ≈ 273 J kg−1. Second, the LCL is slightly lower, from 957 mb to ≈ 956 mb. Third, while the earlier period Ap = 0, the post-2000 period has an Ap ≈ 4.2 J kg−1 midtropospheric residual time mean moist static instability. Thus in recent years, condensation has been slightly easier to reach (δAn ≈ −10 J kg−1 and δpLCL ≈ 1 mb), and has offered a slightly larger mean potential energetic gain (δAp ≈ 4 J kg−1) when moist convection occurs. While minuscule relative to energetics of typical midlatitude moist convection, these changes are nonetheless revealing because they exist in the mean states that are in equilibrium with convection climatologies. In particular, the recent value of Ap > 0 indicates that, on average, added destabilization slightly exceeded neutralization by added convection.
The observed added precipitation indicates that saturation has become more frequent, and—by appeal to Fig. 9c—that with this added saturation, advective destabilization has also intensified. That is, the potential nonlinearity introduced earlier—condensation intensifying instability, promoting additional condensation—is in fact present at the CIES site post-2000. But what helped overcome the resident mean nonzero An?
The first possibility is added mechanical stirring, but the actual ≈ −0.035 m s−1 significant at p > 0.05 is opposite in sign to what is needed for added mechanical stirring overpowering An > 0.
Instead, the question is best answered by analyzing moist static stability of individual profiles, not climatologies, summarized in Fig. 10e. The main part of the panel reveals that moist static stability of individual profiles over the CIES site falls into three categories: strongly stable (An ≥ 400 J kg−1), potentially explosively unstable (Ap ≥ 4500 J kg−1), and an intermediate regime (An < 400 J kg−1 and Ap < 4500 J kg−1).
The full sample percentages these categories account for in the two epochs differ highly significantly. Using a χ2 independence test with df = 2 on the 1988–2000 period’s 25:59:16 turning into 24:56:20 over 2001–14 reveals p < 10−14 differences in the two periods’ proportions. That is, while very stable profiles became slightly less likely, a significant frequency shift from the potentially explosively moist convectively unstable category to the intermediate regime has occurred.
In addition, the insets of Fig. 10e examine, individually for each of the three regimes, epochal differences in (inset from left to right) po − pLFC (for profiles with Ap > 0), Ap, and An. The highly stable profiles (red) have only changed significantly in their height (the leftmost red bar), with δ(po − pLFC) ≈ 26 mb significant at p < 0.01. For the potentially highly unstable profiles (green bars), both areas rose significantly (p < 0.01), but starkly different in magnitude: δAp ≈ 191 J kg−1 versus δAn ≈ 4 J kg−1. Finally, the only change significant at p < 0.05 in the intermediate category is δAp ≈ 58 J kg−1 (center blue bar).
These changes clarify the thermodynamic roots of the added precipitation at the CIES site. 1) Highly stable profiles have not changed markedly in either frequency (reported above) or energetics. 2) Potentially highly unstable profiles, while less frequent, had an additional nearly 200 J kg−1 of potential gain while requiring only an additional 4 J kg−1 to begin realizing it. On average, therefore, high Ap profiles became more likely to yield additional precipitation, as observed. 3) The intermediate regime became significantly more prevalent, and—with essentially unchanged convective inhibition—offers ≤58 J kg−1 more energy, thus likely to produce additional precipitation.
Overall, therefore, moist convection has become more frequent and energetically “lucrative.” The fact that the added moist convection originates more from moderate Ap profiles and less in explosively unstable ones may help explain why such a large proportion of the added available surface water re-evaporates locally rather than running off; a moderate Ap profile is less likely to generate the kind of torrential, abrupt rainfall that is likely to mostly run off. The above CAPE-related observations also help explain Fig. 9c: With condensation more likely, the bars in Fig. 9c that better characterize the actual atmosphere over the CIES site are not the dry (purple) ones, but some combination of these and the saturated (red) bars. This destabilizes the column and promotes moist convection, making subsequent additional condensation and instability likelier still.
It may be illuminating to conclude by commenting briefly on the expected degree of saturation in the two layers. Because in the upper layer water vapor anomalously diverges, it can only contribute less condensation, not more. The observed added precipitation must therefore originate from the lower layer. Yet even if the full mass of additional converging water, = 277ρa880h880/L ≈ 0.16 kg m−2 day−1 (where 277 J kg−1 day−1 is the value of m in the lower bar of Fig. 9c), condenses and precipitates pseudoadiabatically, the resultant additional liquid water flux is smaller than the observed 0.4–0.8 kg m−2 day−1 added precipitation. The low end of the range can be fully resolved by invoking recondensation of the additional local evaporation introduced earlier, further emphasizing acceleration of local hydrology.
7. Discussion and summary
Southern New York has been warming. Summer warming of 0.5–0.7 K is most pronounced and highly significant, especially during the day.
While longwave and full spectrum net surface downward radiation have both increased in summer, warming over the CIES site does not fit well with direct attribution to increased atmospheric opacity to upwelling thermal IR radiation. First, robust warming coincides with high insolation, not nighttime, when the impact of diminished IR cooling is expected to dominate. In fact, annual nighttime net downward surface radiation has decreased significantly. Second, when the time-dependent climatological added net radiation is used to force a simple model of continental thermal evolution, model prediction and observed surface temperatures diverge wildly. Third, a longwave radiation model based only on observed cloud cover, humidity, and temperature changes reproduces reasonably well the summer net downward longwave anomalies, requiring no GHG buildup. Warming at the CIES site is thus not directly attributable to GHG warming.
A likelier explanation of the warming involves solar forcing. Summer and annual mean incident solar radiation at the CIES site have increased significantly, conforming to the established “global brightening.” Warming at the CIES site is thus an element of regional-scale dynamic and thermodynamic adjustments to enhanced incident solar radiation.
In turn, the solar increases are most likely due to aerosol load reduction by improved emission controls. Because these measures are national in scope, with similar solar flux increases observed throughout much of the United States, radiometer-based net downward longwave, shortwave, and full spectrum surface radiation at the CIES site are temporally well correlated with their R-2 counterparts over much of northeastern North America.
The key dynamical response to the enhanced solar warming at the CIES site is a near-surface anomalous cyclone centered over the mid-Atlantic coast. At the CIES site and throughout New England, this cyclone yields near-surface northeasterly anomalies and reduced subsidence. Resultant advective sensible heat divergence and surface flux anomalies mutually cancel, maintaining the lower atmosphere climatological stability over the CIES site. Conversely, considering moist energetics, height-dependent advective moisture convergence destabilizes the lower troposphere.
Most of the added surface radiative energy is thus directed not toward warming, but toward accelerating the hydrological cycle. Summer evaporation at the CIES site has increased, but precipitation has increased more, leading to increased P − E, soil moisture, and runoff.
Surface warming at the CIES site is best viewed as the key change in surface thermodynamics necessary for local closure under enhanced solar radiation of the vapor and liquid water and energy budgets. The magnitude of the added evaporation is narrowly bounded. Significantly reduced evaporation, below the difference between the added atmospheric flux convergence and precipitation, yields a water deficit. The upper bound stems from the dominance of stabilization by the evaporative surface heat sink over destabilization by the virtual effect. Because of this dominance, if evaporation becomes too high, the resultant surface heat sink stabilizes the atmosphere to a mean stability that is too high to produce the observed precipitation. Sandwiched between these two bounds is the actual evaporation (for a given value of radiative forcing and its corresponding added precipitation).
Perhaps the most general lesson this paper delivers is well known yet likely worth re-emphasizing: temperature trends often conceal richer dynamics and thermodynamics than the trend statistics reveal, pointing to a complex web of land–vegetation–atmosphere feedbacks (in some settings potentially involving ocean or cryosphere as well) that connect the disparate climate system responses to shortwave and longwave radiative forcing.
I thank editor Rosana Nieto-Ferreira, and three exceedingly helpful anonymous reviewers for their fair and deeply informed reviews from which this paper benefited measurably. I also thank Dr. Loretta Mickley of Harvard, for her diligent, invaluable critical reading of an earlier draft of this paper. Loretta’s detailed comments have improved this work discernibly. I also thank the current and immediate past Presidents of the Cary Institute—Drs. Ginsberg and Schlesinger—for their support, and Ms. Victoria Kelly at Cary for her generous help with data-related queries.
Deriving Individual Surface Radiative Terms from Observations at the CIES Site
The approximate surface radiative budget is
where as denotes surface shortwave reflectivity, ε is surface emissivity, Ts is surface temperature, and I neglect surface reflectivity of thermal IR because of subsequent absorption by the overlaying atmosphere and assimilation into the measured L↓. Consistent with laboratory measurements of relevant materials (Nerry et al. 1990) that show ε ≈ 0.95–0.96 for thermal IR from rocky surfaces, and with NDVI-based calculated emissivity ε ≈ 0.97–0.98 for mixed soil and vegetation scenes not unlike the area surrounding the CIES site [see Table 1 of Valor and Caselles (1996)], I hold surface emissivity fixed at ε = 0.97.
The CIES site dataset contains downward radiation measurements of full spectrum net R↓net, diffuse shortwave S↓d, and global (diffuse plus unscattered, S↓g = S↓d + S↓u) shortwave radiation, S↓g being essentially S↓. On the other hand, the dataset contains no information about as, Ts, or L↓. I meet this challenge as follows.
First, following Mann and Schmidt (2003), who found that in Northern Hemisphere summer ground temperatures closely track near-surface air temperatures, I assume Ts ≈ T (the brightness temperature for the surface upward longwave flux is the air temperature measured approximately 2 m above the ground T2m, not the unknown solid surface temperature). While of practical necessity and imperfect, this assumption is also supported by observations and basic physical reasoning. Observationally, the R-2 full summer record Ts − T2m interpolated to the CIES site location has a mean and standard deviation of −0.16 and 0.61 K, respectively, with [−0.83, 0.73] K spanning the central 95% of the distribution. This Ts ≈ T2m is also expected given basic physical considerations. First, in summer, when the surface is forced by 20–30 MJ m−2 day−1, buoyancy alone is likely to mix the atmosphere between the surface and 2 m enough to equalize Ts and T2m. Second, in the characteristic hot, humid, and frequently rainy northeastern summer, liquid water is abundant in the uppermost soil (the distribution of days between rain events dominantly peaks at 5–6 days, precisely consistent with the synoptic time scale.), so that much of any putative surface energy surplus is expended on latent heat or enhanced evapotranspiration, not on maintenance of sharp thermal gradients. With these assumptions, R↓net + εσT4 is known.
Absent observations at the CIES site, I estimate surface shortwave reflectivity using the dataset of Muller et al. (2011, 2012), comprising independent albedo estimates for direct and scattered solar radiation. Some may consider surface albedo unimportant because of subsequent absorption and backscattering of upwelling solar radiation by overlaying atmosphere (Donohoe and Battisti 2011; Wendisch and Yang 2012). However, the observed S↓g, which includes the scattered contribution, is measured after the infinity of reflection, absorption, and backscattering events have already made their contributions, so allowing for surface albedo is necessary.
In the Muller et al. (2011, 2012) dataset, albedo values for S↓u and S↓d are reported as abs and aws, “black sky” and “white sky” albedo (Fig. A1). With the direct downward solar flux inferred from the S↓g = S↓d + S↓u at the CIES site, Eq. (A1) becomes
A final difficulty is that the time span of the CIES and Muller et al. (2011, 2012) datasets (1988–2014 and 1998–2011, respectively) overlap only partially, and that the albedo data have much coarser time resolution. Because the albedo data exhibit no trend, I use their 1998–2011 climatologies, summarized for the vicinity of the CIES site in Fig. 7. Using the shown abs and aws climatologies interpolated to the hourly resolution of the data at the CIES site (but with albedo’s diurnal cycle still absent) removes the final indeterminacy.
A Simple Model of Surface Temperature Response to Net Radiative Anomalies
The expected Earth surface temperature response to anomalous radiation is ∂tδT = δR↓net(t)/(ρhcp), where t, ρ ≈ 2400 kg m−3, cp ≈ 810 J kg−1 K−1, and h are time, characteristic density, heat capacity, and thickness of the thermally responsive soil/rock layer, respectively.B1 For seasonal responses, h ≤ 4 m is a reasonable upper bound. Integrating with a time increment Δt for the expected daily mean temperature perturbation considering seasonally varying anomalous net radiation epochal differences obtained from the CIES site data yields
Using this starting from the observed δTJan.1 ≈ 0.5 K yields expected temperatures in Fig. 3.
Parameterized Estimation of Individual Impacts on L↓
To estimate the individual impacts of temperature, humidity, and cloud cover anomalies on downward surface longwave radiation, δL↓(δT), δL↓(δe), and δL↓(δc), respectively, I follow SH9. Using their best-performing parameterization combination— Konzelmann et al.’s (1994) clear-sky emissivity in terms of near-surface vapor pressure e and temperature T, and König-Langlo and Augstein’s (1994) effective atmospheric emissivity εe in terms of c—yields
where σ is the Stefan–Boltzmann constant. Imperfectly, the empirical coefficients are optimized for a location well north of the CIES site; I found no nearer published fits.
Since δ values of all three participating variables are far smaller than the respective climatologies (denoted with overbars), I treat the δ values as small perturbations about the climatologies. The estimated first-order downward longwave responses are thus
yielding the predicted δL↓ values presented in Fig. 4.
Because differencing the two layers implies removal of the vertical mean, it is analogous to obtaining baroclinic flows by barotropic mode removal.
Assuming a 1:3 ratio of soil to rock mixture, and soil (rock) density and heat capacity of 1600 (2700) kg m−3 and 850 (800) J kg−1 K−1, respectively.