Interactive comment on “ Quantifying energy use efficiency via maximum entropy production : A case study from longleaf pine ecosystems

“Our study provides foundational evidence of how MEP can be used to determine resiliency across ecosystems globally” I am not at all convinced and doubt this conclusion. The authors provide no discussion why a diagnosis based on entropy fluxes yields more or better insights than the diagnosis of energy, water, and carbon balances. I see this as a critical missing bit in this manuscript. Authors’ response: We will adjust the discussion and methodology accordingly, to show that entropy production indeed gives more insights about differences in resilience at the three longleaf pine sites, in contrast to solely using energy fluxes. In our revisions we will focus more on the entropy import and export, as well as the internal entropy production, to quantify how close these systems are to a thermodynamic steady state. This could not be accomplished using solely energy fluxes.


Introduction
Energy use efficiency in ecosystems comprises the efficient utilization and conversion of resources, such as solar radiation, nutrients and water (Amthor, 2010;Beer et al., 2009;Finzi et al., 2007;Thomas et al., 2016).Understanding the elements affecting energy use efficiencies is crucial, as anthropogenic and climate induced changes around the globe threaten to alter resources and the ability of ecosystems to utilize them (Haddeland et al., 2014;Porter et al., 2012;Reinmann and Hutyra, 25 2016;Thom et al., 2017).
Ecosystem energy use efficiency is often described based on the ability of the system to partition energy into sensible and latent heat fluxes (Huryna and Pokorny, 2016;Ripl, 1995;Wilson, 2002).However, energy fluxes differ substantially within and among ecosystems, due to differences in structure and function (Kang et al., 2015), which regulate their base rate of energy use (Brunsell et al., 2011).For example, ecosystems exhibit small-scale changes in environmental variables based on 30 their structural properties (Perry, 2002).This spatial heterogeneity can drive differences in microclimate which lead to Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-322Manuscript under review for journal Biogeosciences Discussion started: 8 August 2018 c Author(s) 2018.CC BY 4.0 License.
LE results in greater photosynthetic capacity, which enables maintenance of structural components (Marín et al., 2014).On the contrary, ecosystems that dissipate entropy more through sensible heat fluxes (H) can rapidly deplete the temperature gradient between the surface and canopy, which would eventually lead to lower entropy production (LeMone et al., 2007).
Here we build upon the techniques proposed by Stoy et al. (2014), Holdeway et al. (2010) and Brunsell et al. (2011) by calculating entropy ratios to compare and contrast three sites (mesic, intermediate and xeric) within a longleaf pine (Pinus 5 palustris Mill.) ecosystem.The three longleaf pine sites differed in ecosystem structure (i.e., basal area, EVI, etc.; Table 1) and biodiversity based on differences in soil water holding capacity, as well as different levels of anthropogenic legacy.The sites were exposed to a severe drought in the beginning of this study, which we used to quantify ecosystem resilience through the calculation of entropy efficiency ratios.We focused on entropy production from the absorption of radiation (eff rad ) and metabolic activity (eff M ), as well as overall ecosystem function (i.e., entropy production from LE, H, metabolism 10 and radiation; eff s ).
We hypothesize that (1) the xeric site will have a higher entropy flux from sensible heat (J H ) due to its lower vegetation coverage and lower basal area, and therefore higher partitioning into H rather than LE, which will lower its overall entropy efficiency compared to the mesic site, (2) the mesic site will be able to maintain higher entropy efficiencies (eff rad , eff M and eff s ) due to its structural complexity (i.e., species diversity and basal area), and (3) the intermediate site will have lower eff rad , 15 eff s and eff M compared to the mesic and xeric sites, as a result of its lower biodiversity and structural complexity, a consequence of its anthropogenic legacy.
We show that a combination of different entropy efficiency ratios revealed how structural and functional characteristics in this ecosystem contribute to energy use efficiencies at the three sites.Our results demonstrate that resilience to disturbance is a function of entropy efficiency, as low efficiencies at anthropogenically altered sites prolonged the recovery from drought 20 by approximately a year.In contrast, more biodiverse stands (in overstory and understory) were able to maintain high entropy/energy use efficiencies, which increased their resilience to and recovery from drought.

Site description
This study was conducted at the Joseph W. Jones Ecological Research Center in southwestern Georgia, USA (31.2201 o N,  25   84.4792 o W) from January 2009 to December 2016.The three sites are maintained by frequent low intensity fire on a twoyear return interval and were last burned in 2015 (Starr et al., 2016).The climate is humid subtropical with a mean annual precipitation of 1310 mm (Kirkman et al., 2001).Mean temperature extremes range from 3 °C to 16 °C in winter and 22 °C to 33 °C in summer (NCDC, 2011).
The three sites differ based on soil moisture availability, as a result of differences in soil drainage.The mesic site lies on 30 somewhat poorly drained sandy loam over sandy clay loam and clay textured soils (Goebel et al., 1997;2001).Soils at the intermediate site are well drained and have a depth to the argillic horizon of ~165 cm (Goebel et al., 1997) on well-drained deep sandy soils with no argillic horizon (Goebel et al., 1997).All sites are situated within 10 km of each other with average elevations of 165, 155, and 160 m for the mesic, intermediate, and xeric sites, respectively.95-year-old longleaf pine trees (Pinus palustris Mill.) dominate the overstory of all sites, and overall basal area (B A ) and diameter at breast height (DBH) varied by site (Table 1).The xeric site has the highest proportion of oak trees in the overstory with 22 %, versus 8 % and 7.7 % at the mesic and intermediate sites, respectively.The understory at the mesic and 5 xeric sites is largely covered with perennial C 4 grass species, such as wiregrass (Aristida beyrichiana [Trin.]),whereas woody species dominate at the intermediate site.The composition and abundance of other overstory and understory species varies by site (Kirkman et al., 2001;2016).Soil perturbation at the intermediate site affected species richness, so that wiregrass is almost absent, resulting in an understory dominated by woody species.
We acquired monthly EVI for 2009 through 2016 for all three sites from the online data pool at lpdaac.usgs.govvia the 10 NASA Land Processes Distributed Active Archive Center (LP DAAC) and the USGS Earth Resources Observation and Science Center (EROS), using MODIS Aqua and Terra data products (MYD13Q1 and MOD13Q1;DAAC, 2008) to quantify changes in ecosystem structure from disturbance.We also acquired Palmer Drought Severity Indices (PDSI) for Southwest Georgia from the National Oceanic and Atmospheric Administration data archive for 2009 to 2016 to identify the months of drought disturbance (Dai et al., 2004).
Net ecosystem exchange of CO 2 measurements.Net ecosystem exchange (NEE) was measured continuously at 10 Hz at all three sites from January 2009 to December 2016 using open-path eddy covariance (EC) techniques (Whelan et al., 2013).
The data was stored on CR-5000 dataloggers (Campbell Scientific, Logan, UT).CO 2 and water vapor concentration were measured with an open path infrared gas analyzer (IRGA, LI-7500, LI-COR Inc., Lincoln, NE) and wind velocity and sonic temperature were measured with a three-dimensional sonic anemometer (CSAT3, Campbell Scientific, Logan, UT).These 20 sensors were installed ~4 m above mean canopy height at each site (34.5, 37.5, and 34.9 m for the mesic, intermediate and xeric sites, respectively), ~0.2 m apart to minimize flow distortion between the two instruments and vertically aligned to match the sampling volume of both instruments.

Sensible and latent heat flux measurements
Net energy fluxes of LE and H were estimated in W m -2 using temperature and wind velocity measurements from the sonic 25 anemometer, as well as water vapor density measurements from the IRGA: where  is the latent heat of vaporization (J kg -1 ),  & is the density of air (kg m -3 ),  , is the specific heat of air (kJ kg -1 K -1 ), ′ is the instantaneous deviation of vertical wind speed (w, m s -1 ) from the mean, and ′ and  .′ are the instantaneous 30 deviations of water vapor concentration (kg kg -1 ) and sonic temperature (Kaimal and Gaynor, 1991)  For entropy calculations, H and LE were corrected using the Bowen method following Twine et al. (2000), where the fluxes were adjusted using residual energy of net radiation (R n ) when subtracting ground heat flux (G), and the estimated Bowen ratio from H and LE ( = H/LE), under the assumption that  was correctly measured by the EC system.The corrected 5 values of LE and H were then:
Soil temperature (T soil ), volumetric water content of the soil (SWC) and G were measured in one location near the base of each tower at each site every 15 seconds and averaged every 30 minutes on an independently powered CR10X datalogger.
T soil was measured at depths of 4 and 8 cm with insulated thermocouples (Type-T, Omega Engineering, INC., Stamford, CT), and G was measured at a depth of 10 cm with soil heat flux plates (HFP01, Hukesflux, Delft, The Netherlands).SWC 20 was measured within the top 20 cm of the soil surface using a water content reflectrometer probe (CS616, Campbell Scientific, Logan, UT).

Data processing
Raw EC data were processed using EdiRe (v.1.4.3.1184;Clement, 1999), which carried out a 2-d coordinate rotation of the horizontal wind velocities to obtain turbulence statistics perpendicular to the local streamline.Fluxes were calculated for 25 half-hour intervals and then corrected for mass transfer resulting from changes in density not accounted for by the IRGA.
Barometric pressure data were used to correct fluxes to standard atmospheric pressure.Flux data screening was applied to eliminate 30-min fluxes of NEE, H and LE, resulting from systematic errors as described in Whelan et al. (2013) and Starr et al. (2016).Such errors encompassed (amongst other things): rain, poor coupling of the canopy and the atmosphere (defined by the friction velocity, ustar), and excessive variation from half-hourly means.
Gross ecosystem exchange (GEE) and ecosystem respiration (R eco ) were estimated from eddy covariance measurements of net ecosystem exchange of CO 2 (NEE; µmol m -2 s -1 ) at a time resolution of half an hour, from which GEE and R eco can be estimated as follows: Missing half hourly data were gap-filled using separate functions for day and night as described in Whelan et al. (2013) and 5 Starr et al. (2016).

Entropy production calculations
LE, H, and G fluxes were converted from W m -2 to kJ m -2 and summed to monthly estimates.Net ecosystem exchange of CO 2 was converted to kJ m -2 , using the assumption that one micromole of CO 2 stores approximately 0.506 J (Nikolov et al., 1995), and then summed to monthly metabolic energy values.10 For entropy production and fluxes of shortwave (R s ) and longwave radiation (R l ) we followed established approaches of Brunsell et al. (2011), Holdaway et al. (2010), and Stoy et al. (2014).The entropy flux produced through emission of R s at the surface of the sun (J Rs , kJ m -2 K -1 ) was calculated as: where sun surface temperature (T sun ) was assumed to be ~5780 K, with R s,in defined as incident R s .For entropy flux of R l 15 (J Rl , kJ m -2 K -1 ), we calculated: where R l,in is incoming R l , R l,out is outgoing R l , and surface temperature (T srf ; K) was calculated from upwelling R l : with emissivity of the surface calculated as e srf = 0.99-0.16α(Juang et al., 2007), the surface area A was 1 m 2 , and the Stefan-20 Boltzmann constant k B = 5.67 × 10 -8 W m -2 K -4 .The shortwave albedo (α) was calculated as the monthly average of noontime R s,out , defined as outgoing R s , divided by R s,in .The sky temperature, T sky (K), was calculated from R L,in using the Stefan-Boltzmann equation: where the emissivity of the atmosphere (e atm ) was assumed to be 0.85 for R l , following Campbell and Norman (1998).25 All other ecosystem entropy fluxes J LE , J H , J G , and J NEE (kJ m -2 K -1 ) were calculated by dividing the energy fluxes by temperature as: where x = LE, H, G and NEE e , and y = T air (for J LE , J H and J NEE ; K) or T soil (for J G ).We used the energy of NEE (NEE e ) directly measured with the EC towers, to calculate change in entropy of the metabolic system.
We also calculated entropy produced from evaporation associated with mixing of saturated air from the canopy with the fraction of air in the atmosphere that has RH below 100 % (JLE mix ), following Holdaway et al. (2010): where the evapotranspiration rate, ET = LE/,  is the latent heat of vaporization of water, and  d is the gas constant of water vapor (0.461 kJ kg -1 K -1 for moist air).
The sum of entropy of ecosystem fluxes (J, kJ m -2 K -1 ) was then calculated by adding all entropy fluxes between the surface and atmosphere: The conversion of low entropy R s and R l to high entropy heat at the surface through absorption of R s and R l , respectively, was calculated as: where T srf is the radiometric surface temperature (Eq.3.2.1)and σ RS and σ Rl are in kJ m -2 K -1 .15 The overall entropy production (σ, kJ m -2 K -1 ) was then calculated as the sum of the entropy productions of R s and R l : (3.8) Finally, we estimated overall change in entropy production (S) over time (t) in kJ m -2 K -1 of the ecosystem by adding entropy flux and entropy production: Note that this approach does not account for entropy production due to frictional dissipation of entropy from rainfall or subsurface water flow, as these would be of even smaller magnitude than entropy production from metabolic activity of the ecosystem (Brunsell et al., 2011).

MEP calculations
We estimated MEP of the radiation budget (MEP rad ), metabolic entropy budget (MEP M ), and overall entropy (MEP s ) in kJ m - 25 2 K -1 , to compare their effectiveness in quantifying ecosystem function.We assume that under ideal conditions, an ecosystem maximizes its entropy production when it converts all incoming R s and R l into work (Stoy et al., 2014).Note that MEP Rl is often negative or 0 because an efficient ecosystem would convert less energy into sensible heat, such that T srf would approach T air .

MEP of the radiation budget (MEP
The entropy efficiency ratio of radiation is then calculated using σ of the radiation from Eq. 3.8 as follows: 5 Maximum entropy of metabolism, as the metabolic energy use efficiency, was calculated by estimating maximum possible entropy production assuming that the ecosystem would be able to use all incoming energy from GEE (E in ) as a function of T air (K) and therefore maximize its entropy dissipation: The entropy efficiency ratio of metabolism was then calculated from metabolic entropy flux of actual available energy J NEE , estimated as NEE e /T air (in K), divided by MEP M : Maximum entropy of all energy fluxes (MEP s ) was assumed as the maximum efficiency of an ecosystem, if all incoming energy from R s and R l was converted to work, which would only depend on T srf : 15 We then estimated overall site entropy efficiency as:

Statistical analyses
We tested for significant differences in environmental and structural variables among the three sites prior to the entropy 20 analysis.We estimated simple general linear mixed models (GLMM) to look at differences among sites and of SWC, vapor pressure deficit (VPD), EVI, albedo, T srf , T air and T sky , as well as R s,in , R s,out , R l,in and R l,out via the R function nmle.All response variables were monthly means except for rainfall and the radiation components, which were monthly sums.We used a generalized linear model to analyze rainfall sums, utilizing a quasi-Poisson distribution with a log function to link linear predictors to the mean of the response variable via the function glm in R. We included a random effect for year and 25 month, to account for repeated measurements, as well as an AR(1) structure to account for possible temporal autocorrelation among measurements.Independent variables were month, year and site, as well as their interactions.
Subsequently, we estimated GLMMs of monthly entropy fluxes (J LE , J H , J G , and J NEE ), entropy production (σ), and entropy efficiency ratios (eff rad , eff s , and eff M ) to quantify differences by environmental and structural variables and by site.As in the models of environmental variables, we included random effects and an AR(1) autoregressive correlation structure to account Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-322Manuscript under review for journal Biogeosciences Discussion started: 8 August 2018 c Author(s) 2018.CC BY 4.0 License.
for repeated monthly measurements.All models initially included independent variables for site, year and month, mean monthly T air (except for J G , which included T soil ), SWC, rainfall, VPD and EVI, as well as their interactions with site.We also included the interactions of site with year, to determine changes in the energy efficiency pre, during and post drought at the three sites.Independent variables and their interactions were deemed significant when p<0.05.We used a Tukey adjustment to test for significant differences among sites.We also calculated cumulative sums of the entropy efficiency 5 ratios (i.e., eff rad , eff M , eff s ) to quantify the total difference among the sites over the course of this study.Analyses were performed via the R packages nlme, lsmeans, and car (Fox and Weisberg, 2011;Lenth, 2016;Pinheiro et al., 2014).

Differences in environmental, radiative and temperature variables among sites
All three sites experienced a severe drought from mid-2010 through mid-2012 (Fig. 1).Rainfall decreased (by ~50 mm per 10 month) and VPD increased during the drought on average by ~ 0.4.SWC was significantly lower at the xeric (<20 %) compared to mesic and intermediate sites (~20 %) for all years of this study, but decreased more at the mesic and intermediate site during the drought.EVI was on average 0.02 lower at the xeric site compared to the other two sites, which was significant for all years compared to the mesic, except for 2011, 2013 and 2016; and compared to the intermediate site except for 2013.During the drought EVI decreased at all sites, which was only significant at the mesic site (Fig. 1).15 The albedo was significantly lower at the xeric site (~0.125), but did not differ between the mesic and intermediate sites We did not find any significant differences in R s,in by site.However, we did see significant differences in R s,out , R l,in and R l,out , with less R l,out at the mesic site compared to the intermediate site, and significantly more R l,in at the xeric site, compared to the mesic site (Fig. 2).Monthly R s,out was higher at the xeric site from 2009 through 2011, compared to the other two sites 30 (Fig. 2).From 2012 on, the mesic site had significantly lower R s,out compared to the other two sites.

Understory wiregrass and woody abundance at the sites
Wiregrass was virtually absent at the intermediate site for all years of this study (Fig. 3), whereas woody species were significantly more abundant at that site compared to the mesic and xeric sites.The mesic and xeric sites both had significantly higher proportions of wiregrass in the understory (~25 % versus 5 % at the intermediate site), which slightly decreased during 2011 (Fig. 3).During 2011, woody biomass increased to ~75 g m -2 at the xeric site, but not at the mesic 5 site.In 2012, woody biomass decreased to ~40 g m -2 at the xeric and intermediate sites and remained low during the following years at the xeric site, but increased at the intermediate site (>100 g m -2 ).

Changes in the entropy fluxes of J H , J LE , and J NEE and J G and entropy production
The xeric site had significantly higher monthly J LE , ranging from ~650 to 750 kJ m -2 K -1 , versus the intermediate site with ~480-600 kJ m -2 K -1 (Fig. 4) for all years except 2014.J LE at the xeric site was also higher than the mesic site in 2011-2013 10 and 2015, ranging from 550 to 650 kJ m -2 K -1 per month.The mesic site had ~100-150 kJ m -2 K -1 higher J LE compared to the intermediate in 2010, 2012 and 2016.J LE significantly increased with greater T air , SWC (data not shown) and EVI (Fig. 4), independent of site.J LE was significantly higher at the mesic site, compared to the intermediate site when VPD was above 0.4 (Fig. 4).The xeric site had significantly greater J LE compared to the mesic site when VPD was below 1 kPa, and significantly greater J LE compared to the intermediate site when VPD was below 1.5 kPa.15 J H was significantly higher at the mesic site in 2011 and 2015 (~400-500 kJ m -2 K -1 ) compared to the intermediate (~ 500-600 kJ m -2 K -1 ; Fig. 4) and xeric sites (~550-700 kJ m -2 K -1 ).In 2012 the xeric site had significantly lower J H compared to the other sites (by ~ 100 kJ m -2 K -1 ) and in 2016 the mesic site had significantly lower J H compared the intermediate site (by ~100 kJ m -2 K -1 ).J H decreased with increasing SWC (data not shown) and EVI (Fig. 4) and was significantly higher with increasing T air , independent of site.VPD significantly increased J H at all three sites.J H was significantly lower at that xeric 20 site compared to the mesic and intermediate sites when VPD was below 0.4 kPa (Fig. 4).
J NEE was always negative at the mesic site throughout this study, indicating more energy storage which decreased entropy production locally (Fig. 4).The mesic site had significantly lower J NEE compared to the intermediate site for all years, except 2011, and 2015.The mesic site had significantly more negative J NEE compared to the xeric site in 2009 and 2012, whereas the xeric site had significantly more negative J NEE compared to the intermediate site for the years 2010, and 2012-2014.The 25 mesic site had significantly more negative J NEE compared to the other two sites when EVI was below 0.35 (Fig. 4).In addition, the xeric site had less positive J NEE (by ~ 1 kJ m -2 K -1 ) compared to the intermediate site when EVI was <0.44.For T air between 17-23 °C, the mesic site had significantly more negative J NEE compared to the xeric site.When T air was above 17 °C, J NEE at the intermediate site was significantly more positive compared to the other sites (by > 2 kJ m -2 K -1 ; Fig. 4).
J NEE became more negative (from -1 to -4 kJ m -2 K -1 ) with increasing VPD, independent of site, and was not affected by 30 SWC or rainfall (data not shown).
Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-322Manuscript under review for journal Biogeosciences Discussion started: 8 August 2018 c Author(s) 2018.CC BY 4.0 License.J G was significantly higher at the xeric site, compared to the intermediate site in 2012.In 2016, J G at the intermediate site (-35 kJ m -2 K -1 ) was significantly more negative compared to the mesic and xeric sites (0-10 kJ m -2 K -1 ; Fig. 5).In contrast, J G was significantly more positive at the xeric site, compared to other sites when SWC was >22 %.J G increased from negative to positive with an increase in T soil , which was not significantly different by site (Fig. 5).J G was not affected by VPD, rainfall or EVI. 5 In 2010, 2012 and 2013, σ was significantly higher at the mesic site compared to the xeric site (by > 50 kJ m -2 K -1; Fig. 5).
In 2009, 2011 and 2012 σ was significantly higher at the intermediate site (by > 50 kJ m -2 K -1 ) compared to the xeric, but not the mesic site (Fig 5).Entropy production was unaffected by SWC, T air , and EVI, but significantly increased with higher VPD (from <1400 to 1800 kJ m -2 K -1 ; Fig. 5).

Entropy efficiency ratio models 10
Values of eff rad were significantly higher at the intermediate site, compared to the mesic and xeric sites in 2009, followed by no significant difference between the mesic and intermediate sites in 2010 and 2011 (Fig. 6).eff rad was significantly higher at the mesic site (~0.9),compared to the intermediate (~0.89) site from 2012 through 2016, whereas eff rad at the xeric site was significantly lower compared to the mesic and immediate sites throughout this study.There was no significant change in eff rad at the mesic and intermediate sites in from 2009 through 2011, but eff rad at the xeric site significantly increased starting 15 2010 throughout the remainder of this study.eff rad at the mesic site significantly increased in 2012 and remained high (>0.9)for the remainder of this study.There was no significant change in eff rad at the intermediate site in this study, with the exception of 2012, which was significantly lower compared to 2009 and 2015.Values of eff rad were significantly greater when EVI increased, independent of site (Fig. 6).Values of eff rad only significantly increased with increasing SWC at the mesic site and exhibited no significant change at the other two sites (Fig. 6).The mesic site had significantly higher, whereas 20 the xeric site had significantly lower values of eff rad compared to the other two sites for all levels of SWC and VPD.Higher VPD significantly decreased values of eff rad only at the xeric site (Fig. 7).Values of eff rad significantly increased with higher T air only at the mesic and xeric sites (Fig. 7).The mesic site had higher values of eff rad compared to the intermediate site when T air was above 10 °C.The xeric site had significantly lower values of eff rad compared to the other two sites for all levels of T air (Fig. 7).Rainfall significantly decreased values of eff rad by > 0.005 regardless of site (data not shown).The 25 cumulative sum of eff rad at the intermediate site exceeded 100 % of that at the mesic site up until the beginning of 2012 followed by consistent decreases below 99 % of eff rad at the mesic site.During the drought the xeric site experienced a steep increase in eff rad relative to the mesic site, ranging from 96.5-98 %, which remained at 98 % for the remainder of the study.
Values of eff M were significantly higher at the mesic site compared to the intermediate for all years in this study except for 2015, and compared to the xeric for all years except for 2014 (Fig. 6).Values of eff M were not significantly different between 30 had significantly higher eff M compared to the other sites when EVI was below 0.4 (Fig. 6).The xeric site had significantly higher values of eff M compared to the mesic site when EVI increased above 0.5.Values of eff M were significantly higher at the mesic site compared to the other sites, when SWC was between 19 and 23 % (Fig. 7).When SWC was above 23 %, eff M at the xeric site was significantly lower compared to the other sites (Fig. 7).Greater VPD significantly increased eff M by ~0.1, whereas T air significantly decreased eff M by more than 0.1 independent of site (data not shown).Rainfall had no effect 5 on E M .The cumulative sum of eff M at the intermediate and xeric sites ranged between 90-95 % and 71-94 % of eff M at the mesic site, respectively.During the drought the intermediate site experienced a decrease in eff M , whereas the xeric site experienced a constant increase in eff M relative to the mesic site.
Values of eff s were significantly higher at the mesic site, compared to the intermediate site for all years except 2013 and 2014 (by 0.02-0.04;Fig. 6).eff s was significantly higher at the xeric site (by ~0.02) compared to the intermediate site for all 10 years, except 2009all 10 years, except , 2010all 10 years, except and 2014 (Fig. 6 (Fig. 6).While the mesic site had significantly higher eff s compared to the intermediate for all levels of EVI (by ~ 0.02-0.025), the xeric was significantly higher (~0.45) than the intermediate site (~0.425) when EVI was above 0.22, and significantly lower than the mesic when EVI was below 0.25 (Fig. 6).eff s significantly increased with higher SWC, VPD and T air (by ~0.05-0.1),independent of site (data not shown).Rain had no effect on eff s .The cumulative sum of eff s at the intermediate site was ~97 % of that at the mesic site during the drought, which then decreased 15 to 96 % for the remainder of this study.Similar to the relative cumulative sum of eff rad and eff M , the xeric site experienced an increase in the cumulative sum of eff s relative to the mesic site, ranging from 90 to 97 %.

Discussion
Here we describe differences in energy use efficiencies of sites with varying structural complexities (i.e., understory composition, basal area, DBH) using entropy fluxes and production and entropy efficiency ratios.We show that more 20 structurally complex sites -namely the xeric and mesic sites -have higher metabolic and overall energy use efficiencies compared to a site with anthropogenic modifications (i.e., intermediate site).Although the radiation entropy efficiency (eff rad ) indicated that both the mesic and intermediate sites were equally energy efficient, eff s and eff M revealed lower ecological function at the intermediate site.From these results we conclude that a combination of entropy efficiency ratios can give more valuable insights about the energy use efficiency and resiliency of ecosystems, as opposed to focusing only on 25 individual parts of the entropy budget.
We hypothesized that the xeric site would have higher J H , due to its sparser canopy and sandy soils and therefore lower volumetric heat capacity, in addition to the effect of lower soil water availability.However, in contrast to our first hypothesis, the mesic and intermediate sites and not the xeric site experienced a more severe increase in J H when EVI decreased during the disturbance of drought (Fig. 1).Furthermore, the effect of SWC and rainfall on J H was not significantly 30 different by site, suggesting that plant transpiration, and thus the cooling of leaf surfaces, was the driving force for changes in J H and not differences in water availability.Furthermore, even though plant abundance was lower at the xeric site, its Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-322Manuscript under review for journal Biogeosciences Discussion started: 8 August 2018 c Author(s) 2018.CC BY 4.0 License.
species composition was better adapted to drought conditions, which allowed for the preservation of J LE .In contrast, the mesic site was affected by the interaction of physical (i.e., albedo; Fig. 1) and biological forces, as albedo increased with decreasing plant biomass (lower EVI), which was not observed at the xeric site (Fig. 1).As a consequence, J LE decreased during the drought, which resulted in lower overall entropy efficiency (Fig. 6).eff s decreased due to lower leaf area at the mesic site, which resulted in more radiation reaching the ground, thereby increasing T srf (Figs. 1 and 2) and evaporation of 5 water from the soil.This depleted soil moisture storage (Fig. 1) and therefore decreased plant hydraulics and with it J LE (Kim and Wang, 2012;Lauri et al., 2014), which caused a shift in the energy balance to higher rates of J H . Additionally, the xeric site experienced more J LE compared to the intermediate site, even though its basal area was almost half that of the intermediate (Table 1).This result is consistent with the notion that the understory plays a crucial role in the structural and functional complexity of more open forest ecosystems (Aoki, 2012;Lin, 2015).For example, C 4 grasses and oak species at 10 the xeric site were better adapted to drought (Osborne and Sack, 2012;Roman et al., 2015), which enabled high quality entropy production and the maintenance of its structural integrity (i.e., lower decreases in EVI; Fig. 1).We found further support for this trend in the steep increase in J NEE , eff M , and eff s with greater EVI at the xeric site, suggesting that greater plant biomass and leaf area increased ecosystem function (Peng et al., 2017;Zhu et al., 2016) which allowed for higher quality entropy production.For instance, high overstory oak abundance at the xeric site (~20 %) in addition to the C 4 15 understory resulted in more transpiration during spring and summer, as well as during drought (i.e., anisohydric response; Roman et al., 2015), compared to stands containing just pine trees (Klein et al., 2013;Renninger et al., 2015;Stoy et al., 2006).This led to similar entropy efficiencies compared to the mesic site, as evidenced by both sites having comparable eff s , even though B A was substantially lower (Table 1).
Our results demonstrate the role of active biological functions in controlling energy and entropy fluxes in ecosystems, in 20 addition to the underlying reflective capacities of the land surface (i.e., albedo; Stoy et al., 2014).For example, higher energy use efficiency during drought and low SWC (eff M , Fig. 7) at the xeric site suggests an evolutionary adaptation of the plant species to low soil water availability and high VPD (Basu et al., 2016;Brodribb et al., 2014).In contrast, plant species at the mesic were adapted to higher soil water conditions, such that eff s decreased, as the quality of entropy dissipation through other fluxes leads to lower J LE (Kuricheva et al., 2017).This is because J LE couples both mass and heat dynamics (Brunsell et 25 al., 2011), whereas J H is only a function of the thermal gradient (Kleidon, 2010;LeMone et al., 2007).Nevertheless, J H increased at all sites during the first year of drought and then remained higher compared to pre-drought conditions, indicating a shift of ecological function (Ban-Weiss et al., 2011) towards lower quality energy degradation at all sites (Kuricheva et al., 2017).However, the rapid increase in J LE and decrease in albedo in 2012 at the mesic and xeric sites indicated ecosystem recovery.This provides evidence of rapid adaptation of the plant canopy to drought, which allowed for 30 greater eff rad .Nevertheless, the xeric site was less efficient in using available radiation energy, indicated by high R l,out (Brunsell et al., 2011), as well as a higher albedo compared to the mesic site.Structural limitations due to lower productivity (i.e., lower EVI), a consequence of the sandy soil characteristics at the site, impeded the efficient absorption of available radiation, therefore lowering eff rad (Norris et al., 2011).One reason for this lower entropy efficiency at the site was the large Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-322Manuscript under review for journal Biogeosciences Discussion started: 8 August 2018 c Author(s) 2018.CC BY 4.0 License.
proportion of deciduous oak trees (Table 1), which typically shed their leaves during the winter and therefore decrease the absorptive capacity (Baldocchi et al., 2004).However, high eff s relative to the intermediate site indicated that the xeric site was using the remaining energy available to the site more efficiently.These results demonstrate that sites with higher biodiversity are better adapted to changes in resource availability by way of how they alter their reflective properties, such that energy partitioning is optimized (i.e., high absorption of solar radiation) and high entropy production can be maintained 5 (Gunawardena et al., 2017;Otto et al., 2014;Taha et al., 1988).
We found support for our second hypothesis, that entropy efficiencies (cumulative sums of eff rad , eff M and eff s ) are higher at more biodiverse sites (i.e., mesic site).Accordingly, the mesic site efficiently used available energy from incoming solar radiation (Fig. 2) through lower reflection of R s and by emitting less longwave radiation (Lin, 2015).This suggests higher ecological function of the plant species, especially longleaf pine in the overstory and wiregrass in the understory.Wiregrass 10 is a C 4 species that can maintain photosynthetic rates under high temperatures (Osborne and Sack, 2012;Ward et al., 1999), which allows for greater energy conversion and high-quality entropy dissipation (Brunsell et al., 2011).We confirmed these results at the xeric and mesic sites as indicated by J NEE below 0 kJ m -2 K -1 when T air increased.In contrast, the metabolism at the intermediate site was a source for entropy production when T air was above 20 °C demonstrating an inefficiency in maintaining optimal function when environmental pressure was imposed on the system.This also demonstrates 15 unsustainable entropy dissipation, as more energy was used to sustain metabolic function than was available.In contrast, greater metabolic efficiency at the mesic and xeric sites allowed for more rapid increases in the structural complexity at the sites as indicated by a steep decrease in albedo (Brunsell et al., 2011;Holdaway et al., 2010).The lower ability to adapt to changes in resource availability at the intermediate site could induce its degradation, if environmental fluctuations become more frequent and severe or air temperatures continue to rise with climate change (Mori, 2011;Siteur et al., 2016).This 20 could further exacerbate instabilities for nearby sites, as changes in the reflective properties of degraded sites can alter microclimate and weather patterns across whole ecosystems (Norris et al., 2011).
In contrast to the mesic and xeric sites, metabolic activity at the intermediate site was largely dependent on EVI (i.e., leaf area and biomass), demonstrating lower biological control of individual plant species (i.e., stomatal control; Urban et al.

2016
), but a strong influence of the amount of biomass, and therefore higher photosynthetic capacity, present at the site 25 (Brunsell et al., 2011;Fig. 4 and 6).For example, even though the increase in EVI in 2012 was greater at the intermediate site (Fig. 1) this did not correspond to higher J LE or a decrease in albedo.To illustrate, the intermediate site was unable to maintain negative metabolic entropy production, when biomass was below a certain threshold (EVI <0.35).This resulted in lower ecological function and lower quality entropy production, as energy was dissipated more through J H , rather than J LE (Brunsell et al., 2011).This reduced entropy efficiencies (eff rad , eff M and eff s ) at the site when contrasted to the mesic and 30 xeric sites (Fig. 6 and 7), which affected its recovery from drought.In contrast, the mesic site was able to maintain a negative value when EVI was ³ 0.2, indicating greater adaptive capacity of individual plant species, which "rescued" ecosystem function (Elmqvist et al., 2003).Our third hypothesis was therefore partially supported, as the intermediate site had lower eff M and eff s .Although eff rad , as an indicator for ecosystem radiation efficiency, showed that the intermediate site was more Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-322Manuscript under review for journal Biogeosciences Discussion started: 8 August 2018 c Author(s) 2018.CC BY 4.0 License.efficient in using available energy and producing entropy compared to the xeric site; it did not acclimate eff rad to changes in SWC (Fig. 7), which resulted in slower adaptation to drought.The result of lower eff M and eff s at the intermediate site is intriguing as the mesic and intermediate sites were structurally similar, based on similar B A , mean DBH and overstory tree compositions (Table 1).The inefficiency was a consequence of anthropogenic modification, which homogenized the ecosystem, leading to a decrease in biodiversity (Table .1; Fig. 3).This was a results of soil perturbations from 5 anthropogenic management that took place prior to its establishment (>95 years ago).This suggests that lower plant biodiversity at the site decreased ecological function, which reduced entropy efficiency and therefore resilience to disturbance (Meysman and Bruers, 2010).This was especially apparent when plant biomass was low (low EVI = low J NEE and lower overall J LE ), such that plant functional types present at the site could not rescue ecosystem function during disturbance (Elmqvist et al., 2003)1. 10 We conclude that the quality of entropy fluxes, in relation to structural and environmental variables gives valuable insights into the functional complexity of ecosystems and their ability to adapt to drought.Furthermore, a combination of different entropy efficiency ratios revealed how structural and/or functional characteristics affect energy efficiencies in ecosystems.
Our results demonstrate that ecosystem resilience was mediated by ecosystem entropy efficiency, which was a function of plant biodiversity and ecosystem structure, as well as anthropogenic legacy.We show that low ecological function at 15 anthropogenically altered sites has the potential to decrease ecosystem function, and therefore inhibit optimal energy use efficiency (Meysman and Bruers, 2010), especially with disturbances events.Changes in climate and natural and human induced disturbances are becoming more frequent and severe (IPCC, 2014), demanding more predictive power about how changes in ecosystem structure and function will alter resilience to disturbances.Future policy, conservation or restoration applications depend on reliable theories such as the theory of MEP presented here, to maintain ecosystem services and 20 ecological function (Haddeland et al., 2014;Porter et al., 2012;Reinmann and Hutyra, 2016;Thom et al., 2017).This is especially critical for anthropogenically modified systems, as their land use history can affect changes in energy use efficiency and thus alter their ability to recover from disturbances (Bürgi et al., 2016;Foster et al., 2003).The application of MEP theory could improve our understanding of the interaction of structure, function and legacy on energy use efficiency across a variety of global ecosystems.
from their respective Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-322Manuscript under review for journal Biogeosciences Discussion started: 8 August 2018 c Author(s) 2018.CC BY 4.0 License.means.The overbars in Eqs.1.1 and 1.2 signify the time-averaged covariance.Missing H and LE were gap-filled using simple linear models on a monthly basis regressed with net radiation.

(
~0.108) for 2009 and 2010.The albedo increased at the mesic and intermediate sites, but decreased at the xeric site during the beginning of the drought to 0.115 and 0.125, respectively (Fig. 1).Values for albedo were not significantly different between the intermediate and xeric sites following 2010, but the albedo at the mesic site was significantly lower compared to the other sites following 2011 (Fig. 1).20 T srf at the xeric increased above that of the mesic site in 2009 and 2015.In 2016 the intermediate site had significantly higher T srf compared to the mesic site.T air was significantly lower at the mesic compared to the intermediate site for the years 2009, and 2014 -2016.In addition, the xeric had higher T air compared to the mesic site in 2012, 2013 and 2015.T air at the xeric site decreased below that of the intermediate during 2014 and 2015.For the years 2012, 2013, and 2015 the mesic site had significantly lower T air compared to the xeric (Fig. 1).T sky at the intermediate site was significantly lower compared to the 25 mesic in 2009 through 2014.The xeric site had significantly higher T sky compared to the mesic during 2012 -2016 and compared to the intermediate site for all years of this study (Fig. 1).
the intermediate and xeric sites.Values of eff M did not significantly change throughout this study at the mesic and intermediate sites.At the xeric site eff M was significantly higher in 2013, 2014 and 2016, compared to 2009.Values of eff M significantly increased with increasing EVI at the intermediate and xeric sites, but not the mesic site (Fig. 6).The mesic site Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-322Manuscript under review for journal Biogeosciences Discussion started: 8 August 2018 c Author(s) 2018.CC BY 4.0 License.

Figure 2 :
Figure 2: Incoming and outgoing longwave and shortwave radiation at the mesic, intermediate and xeric sites for the years 2009-2016, with monthly sums of (a) incoming shortwave radiation (R s,in ), (b) outgoing shortwave radiation (R s,out ), (c) incoming longwave radiation (R l,in ), and (d) outgoing shortwave radiation (R l,out ).

Figure 3 :
Figure 3: (a) Wiregrass and (b) woody understory biomass from 2009 through 2015 at the mesic, intermediate and xeric sites.Note that the 5

Figure 3 :
Figure 3: (a) Wiregrass and (b) woody understory biomass from 2009 through 2015 at the mesic, intermediate and xeric sites.Note that the sampling protocol changed to a 2-year measurements cycle in 2013, such that measurements were not made in 2014 and 2016.

Figure 4 :
Figure 4: Annual changes of entropy fluxes of (a-c) latent energy (J LE ), (d-f) sensible heat (J H ), (g-i) metabolic entropy production (J NEE ),and their changes in interaction with site and (a, d, g) year, (b, e, h) Enhanced Vegetation Index (EVI) and (c, f) vapor pressure deficit (VPD) and (i) air temperature (T air ).For (b) and (e) the interaction with site was not significant.

Figure 5 :Figure 6 :
Figure 5: Annual changes of entropy fluxes of (a-c) ground heat flux (J G ), and (d-f) entropy production (σ), and their changes in interaction with site and (a and b) year, (c and d) soil water content (SWC) and (e) soil temperature (T soil ) and (f) vapor pressure deficit (VPD).For (e) and (f) the interaction with site was not significant.

Figure 7 :
Figure 7: Differences in the entropy efficiencies by for radiation entropy efficiency (eff rad ) in interaction with site and (a) soil water content (SWC), (b) vapor pressure deficit (VPD), and (c) air temperature (T air ), as well as (d) metabolic entropy efficiency (eff M ) in interaction with site and SWC.