On the attribution of changing pan evaporation

Authors

  • Michael L. Roderick,

    1. Cooperative Research Centre for Greenhouse Accounting, Environmental Biology Group, Research School of Biological Sciences, Australian National University, Canberra, ACT, Australia
    Search for more papers by this author
  • Leon D. Rotstayn,

    1. Marine and Atmospheric Research, CSIRO, Aspendale, Victoria, Australia
    Search for more papers by this author
  • Graham D. Farquhar,

    1. Cooperative Research Centre for Greenhouse Accounting, Environmental Biology Group, Research School of Biological Sciences, Australian National University, Canberra, ACT, Australia
    Search for more papers by this author
  • Michael T. Hobbins

    1. Cooperative Research Centre for Greenhouse Accounting, Environmental Biology Group, Research School of Biological Sciences, Australian National University, Canberra, ACT, Australia
    Search for more papers by this author

Abstract

[1] Evaporative demand, measured by pan evaporation, has declined in many regions over the last several decades. It is important to understand why. Here we use a generic physical model based on mass and energy balances to attribute pan evaporation changes to changes in radiation, temperature, humidity and wind speed. We tested the approach at 41 Australian sites for the period 1975–2004. Changes in temperature and humidity regimes were generally too small to impact pan evaporation rates. The observed decreases in pan evaporation were mostly due to decreasing wind speed with some regional contributions from decreasing solar irradiance. Decreasing wind speeds of similar magnitude has been reported in the United States, China, the Tibetan Plateau and elsewhere. The pan evaporation record is invaluable in unraveling the aerodynamic and radiative drivers of the hydrologic cycle, and the attribution approach described here can be used for that purpose.

1. Introduction

[2] Measurements of evaporation from pans have traditionally been used to represent the evaporative demand of the atmosphere when estimating crop water requirements [Doorenbos and Pruitt, 1977; Stanhill, 2002]. Averages over many pans show declines over the last 30 to 50 years with typical rates of −2 to −5 mm a−2 (mm per annum per annum) reported in the USA and across many parts of the former Soviet Union [Peterson et al., 1995; Golubev et al., 2001; Groisman et al., 2004], China [Liu et al., 2004; Chen et al., 2005; Wu et al., 2006], Canada [Burn and Hesch, 2007], Australia [Roderick and Farquhar, 2004], New Zealand [Roderick and Farquhar, 2005] and on the Tibetan plateau [Shenbin et al., 2006; Zhang et al., 2007]. That range is not universal and data from India [Chattopadhyay and Hulme, 1997] and Thailand [Tebakari et al., 2005] show declines of −10 to −12 mm a−2.

[3] Evaporation pans are useful in agro-ecology and hydrology because they are simple robust instruments that integrate the relevant physical factors, namely radiation, temperature, humidity and wind speed, into a single measure of evaporative demand. However, understanding the observed decline in evaporative demand requires that integration be unraveled. Previous efforts to do this have been based on simplified arguments [Roderick and Farquhar, 2002] or calculations that assume a surface of short green grass [Thomas, 2000; Chen et al., 2005; Shenbin et al., 2006] instead of a pan. The preferred approach is to calculate pan evaporation using physical variables and we developed a generic model, called PenPan, for that purpose [Rotstayn et al., 2006]. Here we test the use of that model to attribute changes in pan evaporation to changes in the underlying physical variables.

2. Attribution Using the PenPan Model

[4] The PenPan model is based on Penman's combination equation [Penman, 1948]. It assumes a steady state energy balance, which for a Class A pan requires periods of at least a week and the applications described later use monthly input data. The radiative and aerodynamic components are based on the Linacre [1994] and Thom et al. [1981] models respectively. In brief, the evaporation rate from the pan (Ep, kg m−2 s−1) is expressed as the sum of radiative (Ep,R) and aerodynamic (Ep,A) components,

equation image

with s (Pa K−1) the change in saturation vapour pressure (es, Pa) with temperature evaluated at the air temperature (Ta, K) two metres above the ground, Rn (W m−2) the net irradiance of the pan, λ (J kg−1) the latent heat of vaporisation, a (= 2.4 here) the ratio of effective surface areas for heat and vapour transfer, γ (∼67 Pa K−1) the psychrometric constant, D (= esea, Pa) the vapour pressure deficit at two metres and fq(u) (kg m−2 s−1 Pa−1) the vapour transfer function [Thom et al., 1981],

equation image

where u (m s−1) is the mean wind speed at two metres above the ground. The net irradiance of the pan is,

equation image

The last two terms are the incoming (Rl,in) and outgoing (Rl,out) long-wave irradiance, with Rl,out calculated assuming the pan is a black body radiating at temperature Ta. The first term is the net short-wave irradiance, with Ap (= 0.14 here) the pan albedo and Rsp the incoming short-wave irradiance of the pan. Rsp is greater than the global solar irradiance (Rs) because of additional interception by the walls of the pan [Rotstayn et al., 2006].

[5] For attribution, the change in pan evaporation rate is given by differentiating equation (1),

equation image

The term dEp,A/dt is then partitioned into three components, denoted U*, D*, T* for changes due to changing wind speed, vapour pressure deficit and temperature respectively. The components are defined by,

equation image

3. Materials and Methods

[6] Data were collated from existing Australian Bureau of Meteorology (BoM) digital records: class A pan evaporation and wind speed (IDCJDC05.200506), temperature and humidity (IDCJHC02.200506) and radiation (NCCSOL Vers 2.209). We estimated monthly averages when 25 daily observations were flagged as validated by the BoM. Months not satisfying this criterion were omitted. Ta and humidity was measured in Stevenson screens, while u was measured using an anemometer 2 m above the pan. Starting with the 60 or so high quality sites previously identified [Roderick and Farquhar, 2004; Jovanovic et al., 2005], we identified a subset of 41 sites (auxiliary Table S1) having near-complete records of Ep and the observations needed to calculate Ep,A.

[7] As in many regions [Stanhill, 1997] the radiation database [Forgan, 2005] is the most heterogeneous of the meteorological databases. Of the 41 sites, 26 have some measurements of Rs (auxiliary Table S1) in the 1975–2004 period, but only seven have complete 30-year records. Observations of Rl,in are more restrictive with 11 sites having observations, the earliest from 1995. For 1975–2004, we estimated Rl,in at any site having Rs observations using the FAO56 approach [Allen et al., 1998],

equation image

with Ro (W m−2) the top of atmosphere solar irradiance, and z (m) the site elevation. Equation (6) includes water vapour but ignores other greenhouse gases (e.g., CO2) and aerosols (e.g., dust). For the 30-year period considered here, the effect of trends in the ignored greenhouse gases on Rl,in should be small compared to the known trends in Ep. Further, recent simulations with the CSIRO climate model suggest that changes in dust-loading over Australia were also small between the 1950s and 1990s [Rotstayn et al., 2007].

[8] Digital metadata (BoM) showed no site location changes at any of the 41 sites. We examined the observations for obvious problems, especially discontinuities due to, for example, unreported changes in site location. At Darwin Airport, there was an obvious problem with u measurements before 1977 (auxiliary Figure S3). All analyses at that site are for 1977–2004. In several other instances, we identified what initially appeared to be suspect Ep observations. For example, at Alice Springs the very low Ep during 1975–1978 look anomalous when viewed in isolation. However, they were quantitatively consistent with the concurrent low values of u, D and Rs (auxiliary Figure S3). The same was found when examining other apparently anomalous situations.

4. Results

4.1. Evaluation of the PenPan Model

[9] We first used the PenPan model to calculate Ep using complete (post-1995) observations (Rs, Rl,in, Ta, u, es, ea). The agreement between modelled and observed Ep at the 11 sites (auxiliary Figure S1, R2 = 0.95, n = 903, RMSE = 22 mm mth−1) was excellent. Next, we used the available Rl,in observations to evaluate the FAO56 equation. There was no evidence of a change in the slight bias (∼6 W m−2) over time (results not shown) and we concluded that equation (6) was satisfactory for the intended purpose (auxiliary Figure S2, R2 = 0.97, n = 916). Finally, we used equation (6) to estimate Rl,in and thereby calculated Ep at the 26 sites for any month with observations of Rs, Ta, u, es and ea. The comparison with Ep observations was excellent (Figure 1).

Figure 1.

Comparison of observed and calculated pan evaporation rates. The PenPan model was forced with observations (Rs, Ta, es, ea, u) with Rl,in calculated using equation (6). Locations (n = 26 sites) shown in the inset where sites denoted Δ are the seven “elite” sites (listed in Table 1). Best fit regression; y = 1.01 x + 7.7, R2 = 0.95, n = 5071 (1:1 line shown). The RMSE is 24 mm mth−1. (Note that 1 mm = 1 kg m−2.)

4.2. Applying the Attribution Approach

[10] In order to test whether the approach was feasible, we first used the seven “elite” sites having complete observations for 1975–2004, where we calculated Ep,A and Ep,R and thereby closed the energy balance. The trends in those two components should sum to the observed Ep trend (equation 4) within error limits. The estimated uncertainty in the model calculations (24 mm mth−1, Figure 1) is equivalent to an uncertainty in the trend estimate over the 30-year period of 1.8 mm a−2 (±1 sd). With that uncertainty, the observed and calculated trends were within 95% confidence intervals at all seven sites (Table 1).

Table 1. Observed (OBS) and Model-Calculated (CALC) Trends in Pan Evaporation Rate (dEp/dt, in mm a−2) at 7 Sites Having Near-Continuous Data for 1975–2004a
SiteOBSCALC = Rad + AeroRadAeroAero Partition
dEp/dtdEp/dtdEp,R/dtdEp,A/dtU*D*T*
  • a

    OBS, observed trends; CALC, model-calculated trends. Pan evaporation rate (dEp/dt) is given in mm a−2. The modelled trend is the sum of the radiative (Rad = dEp,R/dt) and aerodynamic (Aero = dEp,A/dt) components per equation 4. The aerodynamic component is partitioned into components due to changing wind speed (U*), vapour pressure deficit (D*) and temperature (T*) per equation 5. See auxiliary Figure S3 for data and calculations at the sites.

  • b

    Data for 1977–2004.

GERALDTON AIRPORT−4.1−2.20.0−2.2−1.6−0.80.3
DARWIN AIRPORTb−17.0−15.3−6.0−9.3−8.9−0.30.1
ALICE SPRINGS AIRPORT25.821.42.019.416.95.4−1.8
MOUNT GAMBIER AERO−6.1−8.4−0.2−8.2−7.4−1.40.3
ROCKHAMPTON AERO11.07.73.24.50.34.3−0.2
WAGGA WAGGA AMO−1.81.40.50.9−0.51.40.1
MILDURA AIRPORT−8.8−11.90.6−12.5−13.20.60.3

[11] While most pan evaporation sites considered here, or anywhere, do not have radiative observations, we did have near-complete records of the aerodynamic component at each site. To apply the approach at all 41 sites, we estimated the trend in the radiative component as the difference between the trends in observations and aerodynamic components (i.e., dEp,R/dt = dEp/dtdEp,A/dt). The results, including the separation of the aerodynamic component into individual components (U*, D*, T* per equation 5) are shown in Figure 2.

Figure 2.

Trends in observed pan evaporation rate and its components at 41 sites for the period 1975–2004. (a) Observed pan evaporation rate. (b) Radiative component of pan evaporation rate calculated as the difference between Figures 2a and 2c. (c) Aerodynamic component of pan evaporation rate. The trend in the aerodynamic component is further partitioned (equation 5) into the change due to changing (d) wind speed, (e) vapour pressure deficit, and (f) air temperature. The change in each panel, averaged across all 41 sites is (a) −2.0 mm a−2, (b) +0.6 mm a−2, (c) −2.6 mm a−2, (d) −2.7 mm a−2, (e) 0.0 mm a−2, and (f) 0.0 mm a−2. Details and trends are available for each site in auxiliary Table S1. (Note: The magnitude of the trend is scaled to the dot area per the legend.)

[12] Much of the trend in Ep observations (Figure 2a) was due to changes in the aerodynamic component (Figure 2c), and the majority of that was due to changes in wind speed (Figure 2d) with generally minor changes due to changes in both vapour pressure deficit and air temperature (Figures 2e and 2f). However, as expected [Roderick and Farquhar, 2004], there was spatial variation in the results. A notable feature is the decrease in the radiative component shown in the northwest (Figure 2b). Also of note are the two sites showing relatively large increases in Ep in the centre (Alice Springs, 133.89°E, 23.80°S) and south (Woomera, 136.81°E, 31.16°S). At Woomera there were no obvious problems with the data. At Alice Springs, the trend was very sensitive to the starting date because of very low Ep values during 1975–1978 (auxiliary Figure S3).

[13] To put the changes in perspective, the trend in D averaged over all 41 sites was −0.2 Pa a−1 (auxiliary Figure S4) compared to a background average of 1205 Pa, or less than 1% over the 30 years. In contrast, the trend in u averaged over all 41 sites was −0.01 m s−1 a−1 (auxiliary Figure S4) against a background average of 2.3 m s−1: a reduction of 13% over the same period. The change in u was occurring more or less equally in all seasons (auxiliary Figure S5).

5. Discussion

[14] Previous research reported a trend in pan evaporation rate, averaged over 61 Australian sites for 1975–2002, of −3.3 mm a−2 [Roderick and Farquhar, 2004]. This was later updated (an addendum is available from the authors) to −3.2 mm a−2 to account for the installation of bird guards. The trend for 1975–2004 over the same 61 sites is lower at −2.4 mm a−2 (results not shown) because of the high pan evaporation rates during the drought conditions prevailing over much of southeast Australia since 2002. For the 41 sites used here, the averaged trend for 1975–2004 was similar at −2.0 mm a−2.

[15] Improvements could be made to the PenPan model, particularly in the calculation of the pan albedo and the treatment of incoming and outgoing long-wave irradiance. Similarly, the meteorological databases are subject to ongoing improvements [Coughlan et al., 2005]. With those caveats, the model performed satisfactorily (Figure 1, Table 1) given the well-known difficulties in making long-term measurements of, and modelling, micrometeorological phenomena. According to the attribution analysis (Figure 2), the reasons for changing pan evaporation differed between sites: there was an indication of a decrease in the radiative component in northwest Australia consistent with increased rainfall and cloud cover in that region [Smith, 2004; Rotstayn et al., 2007]. However, decreases in the aerodynamic component were more important and primarily due to decreasing wind speed. These results are consistent with recent research [Roderick and Farquhar, 2006; Rayner, 2007]. The importance of decreasing wind speed and/or radiation as a reason for decreasing pan evaporation has also been found in the USA [Hobbins, 2004], parts of China [Xu et al., 2006a] and the Tibetan Plateau [Shenbin et al., 2006; Zhang et al., 2007].

[16] Whether the “stilling” reported here is local, i.e., attributable to changes in the immediate environment of the pans (e.g., growing trees or other obstacles progressively obstructing the air flow), or a more regional phenomenon is difficult to assess. Rayner [2007] investigated that by comparing the BoM wind observations against two alternative sources, (1) wind fields in the NCEP reanalysis, and (2) wind calculated using BoM surface air pressure observations. The results were inconclusive because the trends derived from (1) and (2) were inconsistent, and neither result was consistent with the BoM surface observations.

[17] Some of the wind speed decreases reported here are no doubt due to local effects. Alternatively, the very widespread nature of the stilling is by itself some evidence of a more robust regional effect. Indeed, the changes reported here are very similar to those reported elsewhere (Table 2). Whilst largely unanticipated in the climate change impacts community, previous analyses have predicted a slowing in the overall circulation rate in tropical regions and, presumably, a reduction in averaged wind speed in those regions with greenhouse warming [Betts, 1998; Held and Soden, 2006; Vecchi et al., 2006]. Although not strictly comparable to surface winds, the summary compiled by Lorenz and DeWeaver [2007] shows that climate models generally predict changes in zonally averaged mid-latitude wind speeds (at 850 hPa) of about −0.5 to −1.5 m s−1 over the 21st Century with largely complementary increases nearer the poles. This would qualitatively fit the pattern in the observations (Table 2, increase in Antarctica, decrease elsewhere). The model projections are equivalent to trends of −0.005 to −0.015 m s−1 a−1 and are of the same order as the observed trends (Table 2). In contrast to the terrestrial-based anemometer records (Table 2), recently reported satellite retrievals indicate increases in oceanic wind speed averaging 0.008 m s−1 a−1 for 1987–2006 [Wentz et al., 2007]. This emphasises the urgent need for research on the wind measurements and the modelling given the scientific importance as well as the widespread interest in wind power generation.

Table 2. Summary of Observed Changes (Represented as a Linear Trend) in Near-Surface Wind Speed (du/dt)a
du/dt, m s−1 a−1LocationDetailsRef.
  • a

    All studies are based on terrestrial anemometer records.

−0.010Australia1975–2004, 41 sitesThis study
−0.005USA1962–1990, 207 sites across the 48 conterminous statesHobbins [2004]
−0.004USA1960–1990, 176 sites across the 48 conterminous statesKlink [1999]
−0.008Yangtze River Catchment, China1960–2000, 150 sitesXu et al. [2006a]
−0.020China1969–2000, 305 sitesXu et al. [2006b]
−0.010Loess Plateau, China1980–2000, 52 sitesMcVicar et al. [2005]
−0.013Tibetan Plateau1960–2000, 101 sitesShenbin et al. [2006]
−0.017Tibetan Plateau1966–2003, 75 sitesZhang et al. 2007]
−0.013Italy∼1955–∼1996, 17 sites on Italian coast. Break point in ∼1975. Trend of ∼−0.026 m s−1 a−1 before and ∼−0.002 m s−1 a−1 after 1975Pirazzoli and Tomasin [2003]
−0.011New Zealand1975–2002, 5 sitesM. L. Roderick (unpublished data, 2005)
−0.017Canada∼1950–∼1990, 4 sites on west coastTuller [2004]
+0.006Antarctica∼1960–∼2000, 11 sitesTurner et al. [2005]

6. Conclusion

[18] When forced with radiation, temperature, humidity and wind observations, the PenPan model simulated the pan evaporation observations well. Over Australia, that approach revealed differences between sites, but on the whole, decreasing wind speed was found to be the main reason for decreasing pan evaporation. The observed decrease in wind speed, was similar to the decreases reported over other terrestrial surfaces. Our results show that the extensive world-wide network of pan evaporimeters could be used to recover information about changes in the radiative and aerodynamic drivers of evaporative demand. This would be extremely useful because there are many more pan evaporimeters than radiometers.

Acknowledgments

[19] We thank Alan Beswick, John Carter, Edward Linacre, David Rayner and Blair Trewin for helpful discussions and Alison Saunders for expert assistance with acquiring and processing the data. We acknowledge funding from the Managing Climate Variability Program managed by Land and Water Australia (MLR, GDF), the Australian Greenhouse Office (LDR) and a Gary Comer Award (GDF). We acknowledge the BoM and especially the numerous BoM observers whose work formed the ultimate basis of this study.

Ancillary