Calculated and Measured Potential Evapotranspiration across Land Uses in Florida
Ellen M. Douglas1, Jennifer M. Jacobs2, and David Sumner3
2Environmental Earth and Ocean Sciences, University ofMassachusetts, Boston, MA, USA
2Department of Civil and Environmental Engineering, University of New Hampshire, Durham, NH03824, USA
3United States Geological Survey, 224 W. Central Pkwy., Suite 1006, Altamonte Springs, FL 32714, USA
Corresponding Author:
Ellen M. Douglas, PE, PG, PhD
Environmental, Earth and Ocean Sciences
University of Massachusetts, Boston
100 Morrissey Blvd
Boston, MA02125
Phone: 617-287-7437
Fax: 617-287-7474
Email:
Key Words: Potential Evapotranspiration, Experiment, Florida
Abstract (or why bother to read/cite this article)
1. Introduction
Groundwater and surface water models typically require precipitation and evapotranspiration surface forcings to determine the upper boundary condition. During the past few decades, many models have been developed to simulate water flow in the unsaturated zone, utilizing different techniques to couple the evapotranspiration process with water flow in unsaturated zone. A commonly used approach to determine the water lost to the atmosphere from the surface soil layers is to force the model using potential evapotranspiration (PET) and use soil moisture and canopy characteristics to determine the actual evapotranspiration. While numerous methods exist to estimate PET, the best method is not clear.
Previous modeling studies have shown that ET does not affect streamflow simulations as much as precipitation and in fact may be adequately estimated using average annual ET values [Burnash, 1995; Fowler, 2002]. Additionally, Oudin et al.’s (2005a) comparison of numerous PET methods impact on four rainfall-runoff models for 308 watershed models suggests that temperature-based PET estimates perform as well as or better than more physically-based PET methods. However, Oudin et al.’s study removed systematic biases by scaling by the Penman ETp estimates prior to use in the rainfall-runoff models. Other studies have shown that ET estimates can significantly impact model simulations [Nandakumar and Mein, 1997; Andreassian et al., 2004], particularly when biases exist among ETp methods [Federer et al., 1996; Vorosmarty et al., 1998].
Experimental data have been widely used to compare among PET methods. In the southeastern United States, several studies have compared methods. Yoder et al.’s (2005)grass lysimeter studyin the humid Southeast found that FAO-56 Penman-Monteith equation gave the best results, but that the Turc equation was a reasonable, less complex alternative. Sumner and Jacobs study of nonirrigated pasture site in Florida, USA found that both Penman-Monteith (PM) and a modified Priestley-Taylor (PT) methods required seasonal calibration parameters. Jacobs et al.’s (2002, 2004) study of a wet prairie community in Central Florida, USA found that a calibrated Penman-Monteith model gave good results for the potential evapotranspiration, the Priestley-Taylor and the Penmanmodels overestimated PET, and the uncalibrated, simpler Turc and Makkink methods performed nearly as well as the Penman-Monteith method. Abtew and Obeysekera (1995) found that the Penman-Monteith method was well suited to estimate evapotranspiration from cattails (Typha domingensis), mixed marsh vegetation, and an open water/algae system, but that calibrated simpler models also provided reasonable estimates (Abtew, 1996).
While these site specific studies provide insight to individual landuses and climates, a challenge to conducting ET intercomparison studies for heterogeneous regions is that coincident ET measurements seldom available across a region for representative landuses. The recent emergence of eddy covariance instruments has significantly expanded the breadth of evapotranspiration measurements. Temporal dynamics of water and energy fluxes are measured across seasons and years by routinely deploying one or more eddy covariance towers at numerous sites including the AmeriFlux and FLUXNET networks which includes over 120 separate sites that measure fluxes in the United States (Law et al. 2002). Additionally, a number of experiments have provided evapotranspiration measurements across heterogeneous landscapes including FIFE results, OASIS (Observations At Several Interacting Scales) (Leuning et al. 2004), and SMACEX (2002 Soil Moisture-Atmosphere Coupling Experiment) (Crow et al., 2005)among others, these data sets are typically for short periods (seasonal), under non-potential conditions, and have not been analyzed using commonly available PET estimation methods. Even so, analyses that are conducted at spatial and temporal scales necessary to characterize errors at a regional scale are still very limited.
This research’s objective is to characterize the relative strengths and weaknesses of available PET models across a range of land covers common in the southeastern United States. The approach is to use existing models and model parameters as determined from the literature to model PET ET. A unique aspect of this research is that 18 validations sites having measured evapotranspiration and ancillary climate data over comparable study periods were available for this PET intercomparison study. This allows modeled PET ET values to be compared to measured ET values and for the model errors to be considered by site, land uses and overall.
2. Methods
2.1 Validation sites
18 validations sites having measured evapotranspiration and ancillary climate data were used for the PET intercomparison. The sites were distributed throughout the State of Florida and represent a variety of land cover types: open water (3 sites), marshland (4 sites), grassland/pasture (4 sites), citrus (2 sites) and forest (5 sites). Figure 1 shows the locations of these sites. Table 1 lists each site, its location, dominant land cover, and measurement period. All stations were measured using the eddy correlation approach. Measurements were made at 30-minute increments.DAVE: please add information about data collection and processing, instrument maintenance and any details that would be useful to know for this analysis.
2.2 Potential Evapotranspiration Models
Potential evapotranspiration (PET) models routinely use solar radiation directly or in combination with longwave radiation to provide a measure of the net available radiation. Combination equations, including the Priestley-Taylor (PT) method (Priestly and Taylor, 1972) and the Penman-Monteith (PM) method (Penman, 1948; Monteith, 1965), typically give the best PET estimates for a variety of vegetated surfaces and climates, but their application is suitable only at locations where in-situ measurements of temperature, wind, water vapor, and sunshine duration or solar radiation are available. For data limited regions, numerous empirical models exist to determine PET. Two methods that apply solar radiation directly, the South Florida Water Management District’s Simple method (Abtew, 1996), method and the Turc method (Turc, 1961).The SFWMD Simple Method provides estimates of PET with only measured solar radiation using the following equation
(1)
where ETo is the wet marsh potential evapotranspiration (mm d-1), the latent heat of vaporization (MJ kg-1), w the density of water (kg m-3), K1 the coefficient (0.53 for mixed marsh, open water and shallow lakes), and Rs solar radiation received at the land surface (MJ m-2 d-1). The method has been validated for cattails and wet marsh vegetation (Abtew, 1996). Here, daily values were computed as
.(2)
The Turc radiation method, developed in Western Europe for regions where the relative humidity is greater than 50%, expresses PET as
(3)
where LE is the mean daily latent heat flux (W m-2), Rs the daily solar radiation (W m-2), and Ta the mean daily air temperature (oC). In this study, we used the Turc method to represent a solar radiation-based approach to estimating PET.
The Priestley-Taylor method uses the concept of the theoretical lower limit of evaporation from a wet surface as the “equilibrium” evaporation to estimate PET where
(4)
where ETo is the potential evapotranspiration (mm day-1), the latent heat of vaporization (MJ kg-1), w the density of water (kg m-3), the slope of the saturation vapor pressure temperature curve, the psychrometric constant, Rn the net radiation (W m-2), and G the soil heat flux (W m-2). Priestly and Taylor (1972) showed that for conditions of minimum advection with no edge effects, = 1.26. Here G is assumed to equal zero over the course of a day. The parameters (in kPa ºC), (MJ kg-1) and (in kPa ºC) were computed as
(5)
(6)
where es is the saturated vapor pressure (in kPa), cp is the specific heat of moist air (=1.013 kJ kg-1 ºC-1), P is atmospheric pressure (set equal to 101.3 kPa) and T is the minimum daily temperature (in ºC). Saturated vapor pressure was as
.(7)
The Penman-Monteith model is an extension of the Penman equations that allows the equation to be applied to a range of surface vegetation through the introduction of plant specific resistance factors and is given as
(8)
where ETo is the potential evapotranspiration (mm day-1), the latent heat of vaporization (MJ kg-1), w the density of water (kg m-3), the slope of the saturation vapor pressure temperature curve, the psychrometric constant, Rn the net radiation (W m-2) and G the soil heat flux (W m-2). D is the vapor pressure deficit of the air (in kPa). a is the mean air density at constant pressure, cp the specific heat of air (1.013 kJ kg-1 ºC-1), rs the bulk surface resistance (s m-1), and ra the aerodynamic resistance (s m-1). The mean air density, a, was using
(9)
where P was set equal to constant 101.3 kPa and Ts was the average daily temperature (in ºC). The vapor pressure deficit, D, was computed as es – e, where e is the observed daily vapor pressure. The aerodynamic resistance was computed using Monin-Obukhov similarity
(10)
where u is the wind speed (in m s-1) and zu is the height at which the wind speed was measured,ze is the height of the vapor pressure/relative humidity instrument, d the displacement height (approximated as 0.67hc, where hc is the average vegetation height), zom the roughness height for momentum, zov is the roughness height for water vapor (approximated as 0.1zom) and k is the Von Karmen’s constant (0.41). We used literature values for zom, because using a relationship between zom and canopy height is not appropriate for all land cover types. Height of wind measurement (zu), height of vapor pressure/relative humidity measurement (ze) and average canopy height (hc) were obtained from the metadata for each site or from the personnel responsible for collecting the data. zu and ze were assumed to be equal unless otherwise noted.
For open water sites, the Penman-Monteith method’s (eqn 8) advection term was estimated following Shuttleworth (1993)
Open water(11)
which incorporates the ra formulation for open water as
(12)
where zm is a standardized measurement height of 2 m and zo = 0.00137 m.
The Penman-Monteith parameters used in this analysis are presented in Table 3. Also included in Table 3, for comparison purposes, are “at-site” rs values made available for the AlachuaCounty and KennedySpaceCenter forested sites. A range of bulk canopy resistance (rs) estimates for wetlands and for pine forest sites were available from published studies in Florida (Abtew, 1996; Abtew et al., 1995; Jacobs et al., 2002; Powell et al., 2005). Breuer et al.’s (2003) extensive compilation of published vegetation parameters was used to estimate canopy resistance. For grass/pasture sites, we computed rs using the relationships developed by Sumner and Jacobs (2004):
(13)
(14)
(15)
where gs is bulk surface conductance (in m s-1), D is vapor pressure deficit (in kPa) and gmax is the maximum bulk surface conductance. Bulk surface resistance for grass (rs, in s m-1) is the reciprocal of gs. Average bulk surface resistance for the grass/pasture validation sites, calculated for each site, ranged from 284 to 319 s m-1 (see Table 1) which is consistent with published values. The published value of rs for marsh/wetland vegetation is 55 s m-1 and rs for open water is zero. For marsh and wetland sites, rs was computed as a weighted average based on the proportion of vegetated area and open water area.
3. Results and discussion (need to add discussion about other intercomparison studies)
3.1 Observed Evapotranspiration
Observations for net and solar radiation, temperature, humidity and evapostranspiration (ET) were made at 30-minute increments at all sites.Daily values were calculated by averaging the 30-minute data over a 24-hour period. When eddy flux measurements were not available for a particular 30-minute increment, ET was estimated using the USGS approach of gap filling using Priestley-Taylor method (4). Only those days having ET measurements for 80% or more of the half-hour increments were used for this study.These were considered “good” observations. Most missing values occurred during nighttime or rainfall when ET values would be low (David Sumner, USGS, personal communication). Table 3summarizes the total number of days for which ET was measured and the number of days considered to be good. It was necessary to relax this criterion for some of the sites (i.e., the Everglades and Disney Wilderness sites) where ideal measuring conditions were difficult to maintain. For instance, at the Disney Wilderness site, more than 30% of the half-hour measurements were gap-filled due to wind direction with inadequate fetch, excessive misalignment of sonic anemometer, or obscured hygrometer windows. The remote location of the Everglades sites made instrument maintenance difficult (Ed German, USGS, personal communication). Also presented in Table 3 are the number of day for which the bowen ratio ( = sensible heat (H) divided by sensible latent heat (LE)) was less than unity, signifying days in which more energy was partitioned to LE (and hence, ET) than to H. We used this threshold as an indicator of potential evapotranspiration conditions; this will be discussed in greater detail in Section 3.2.
Table 4a and 4b presents thesite averageand coefficient of variability (CV), respectively, of observed net radiation (Rn), and DAET statistics by season for all 18 sites together. Rnrepresents the total available energy that is partitioned between LE and H. TheCV is computed as the standard deviation divided by the average which allows for the comparison of variability within and between sites. These statistics were also averagedover thefive general land cover classes: forest (4 sites), citrus (2 sites), grass (4 sites), marsh (4 sites) and open water (3 sites). LE and H measurements were not available for the Blue Springs Tract site in northern Florida, hence data from this site were not included. In order to eliminate the affect of erroneous latent heat measurements (which resulted in anomalously high bowen ratios and low values of ET), days in which observed latent heat was less than 5 Wm-2 were excluded. Seasons were demarcated by the julian day (JD) of their calendar start and end dates (winter: December 21 through March 20 (JD 355-79); spring: March 21 through June 20 (JD 80-171); summer: June 21 through September 20 (JD 172-263); and fall: September 21 through December 20 (JD 264-354)).
Table 4a shows that, in general, the highest daily average Rnwas observed during spring, while the lowest and highest observed ET occurred in summer. This suggests that in most cases, water, not energy, is the factor that limits ET acrossFlorida. As would be expected, the sites for which maximum Rnand ET occurred within the same season (spring) were the open water sites. At these sites, ET was presumably limited only by the available energy. This is further supported by the fact that the at the open water sites were consistently low (0.1 to 0.2) throughout the year. Not surprisingly, the highest ET values also occurred at the open water sites, ranging from an average of 3.3 mmd-1 in the winter to 5.3 mmd-1 in the spring. ET at the marsh sites was also high, ranging from 2.7 mmd-1 in winter to 4.4 mmd-1 in summer. Amongst the “water-limited” sites, the lowest ET occurred at the grass sites in winter (1.3 mmd-1) and fall (2.0 mmd-1) and at the forested sites in spring (2.8 mmd-1) and summer (2.3 mmd-1). From a measurement variability standpoint, Table 4b shows that marsh sites yielded the most consistent observations, with averaged CV ranging from 0.24 to 0.33 for Rn, 0.39 to 0.47 for and 0.21 to 0.30 for ET. At-site CV values for DAET at the marsh sites ranged from a high of 0.39 to a low of 0.18. At the open water sites, averaged CV was relatively low for Rn (0.27 to 0.37), moderate for DAET (0.28 to 0.46) and surprisingly high for (0.40 to 1.9). The high variability of at the water sites may be due to seasonal changes in both the magnitude and direction of heat flux across the water-air boundary. At-site CV values for DAET at open water sites ranged from 0.25 to 0.53. Forested sites yielded the most highly variable observations, with averaged CV ranging from 0.25 to 0.40 for Rn, 1.2 to 7.8 for and 0.44 to 0.59 for ET. At-site CV values for DAET ranged from 0.38 to 0.78. A large part of this variability is no doubt due to the fact that the forested sites covereda variety oftree types ranging from scrub oak (height = 1.5 m) tomature pine (height = 22 m). This is supported by the fact that, despite the smaller sample size, the citrus sites showed low to moderate variability, presumably due to a single tree type and a more uniform canopy height.
Figures 2a and 2b compare examples of daily observed LE for each general land cover class (forest, citrus, grass, marsh and open water). The vertical dashed lines indicate season breaks as defined previously. The wet season in Florida typically begins around JD 160 (early June). Observed LE peaked sooner at the open water sites than any of the others, further illustrating the influence of water-limiting versus energy-limiting conditions. The maximum LE (LEmax) at grass sites generally occurred between JD170 and 180, whereas at forest sites, LEmaxdid not occur until after JD 200. The lag at forest sites may have been due to deeper rooting depths. The timing of LEmaxat the marsh sites was surprisingly variable. At Blue Cypress, LEmax occurred at JD 168, at Everglade L1 and X1.5, LEmax occurred at JD 190, but at Everglade P33, it occurred at JD 210. The timing of LEmax at the open water sites was not easily defined (due to measurement variability) but tended to occur between JD 100 and 150.