• Allen, R. G., L. S. Pereira, D. Raes, and M. Smith, 1998: Crop evapotranspiration—Guidelines for computing crop water requirements. FAO Irrigation and Drainage Paper 56, 2d ed., Food and Agriculture Organization of the United Nations, Rome, Italy, 300 pp.

  • Arnell, N. W., and Coauthors, 2001: Hydrology and water resources. Climate Change 2001: Impacts, Adaptation, and Vulnerability, J. J. McCarthy et al., Eds., Cambridge University Press, 191–233.

    • Search Google Scholar
    • Export Citation
  • Bouchet, R. J., 1963: Evapotranspiration réelle evapotranspiration potentielle, signification climatique. Int. Assoc. Sci. Hydrol., 62 , 134142.

    • Search Google Scholar
    • Export Citation
  • Brutsaert, W., and M. B. Parlange, 1998: Hydrologic cycle explains the evaporation paradox. Nature, 396 , 30.

  • Bureau of Meteorology, 2002: NCCSOL—Australian solar radiation data. CD-ROM, 2.209.

  • Bureau of Meteorology, 2005a: Australian daily evaporation data. CD-ROM, IDCJDC05.200506.

  • Bureau of Meteorology, 2005b: Australian daily wind data. CD-ROM, IDCJDC06. 200506.

  • Bureau of Meteorology, 2005c: Australian hourly temperature humidity and pressure. CD-ROM, IDCJHCO2.200506.

  • Bureau of Meteorology, 2005d: Australian hourly wind data. CD-ROM, IDCJHC01. 200506.

  • Chattopadhyay, N., and M. Hulme, 1997: Evaporation and potential evapotranspiration in India under conditions of recent and future climate change. Agric. For. Meteor., 87 , 5573.

    • Search Google Scholar
    • Export Citation
  • Cohen, S., A. Ianetz, and G. Stanhill, 2002: Evaporative climate changes at Bet Dagan, Israel: 1964–1998. Agric. For. Meteor., 111 , 8391.

    • Search Google Scholar
    • Export Citation
  • Collins, D. A., P. M. Della-Marta, N. Plummer, and B. C. Trewin, 2000: Trends in annual frequencies of extreme temperature events in Australia. Aust. Meteor. Mag., 49 , 277292.

    • Search Google Scholar
    • Export Citation
  • Della-Marta, P., D. Collins, and K. Braganza, 2004: Updating Australia’s high-quality annual temperature dataset. Aust. Meteor. Mag, 53 , 7593.

    • Search Google Scholar
    • Export Citation
  • Forgan, B. W., 2005: Australian solar and terrestrial network data. Proc. Pan Evaporation: An Example of the Detection and Attribution of Trends in Climate Variables, Canberra, Australia, Australian Academy of Science, 50–53.

  • Gilgen, H., M. Wild, and A. Ohmura, 1998: Means and trends of shortwave irradiance at the surface estimated from global energy balance archive data. J. Climate, 11 , 20422061.

    • Search Google Scholar
    • Export Citation
  • Grayson, R. B., R. M. Argent, R. J. Nathan, T. A. McMahon, and R. G. Mein, 1996: Hydrological Recipes: Estimation Techniques in Australian Hydrology. Cooperative Research Centre for Catchment Hydrology, 125 pp.

    • Search Google Scholar
    • Export Citation
  • Henderson-Sellers, A., 1992: Continental cloudiness changes this century. GeoJournal, 27 , 255262.

  • Hobbins, M. T., J. A. Ramírez, and T. C. Brown, 2004: Trends in pan evaporation and actual evapotranspiration across the conterminous U.S.: Paradoxical or complementary? Geophys. Res. Lett., 31 .L13503, doi:10.1029/2004GL019846.

    • Search Google Scholar
    • Export Citation
  • Jeffrey, S. J., J. O. Carter, K. B. Moodie, and A. R. Beswick, 2001: Using spatial interpolation to construct a comprehensive archive of Australian climate data. Environ. Modell. Software, 16 , 309330.

    • Search Google Scholar
    • Export Citation
  • Jovanovic, B., D. Jones, and D. Collins, 2007: A high quality monthly pan evaporation dataset for Australia. Climatic Change, in press.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437471.

  • Kendall, M. G., 1948: Rank Correlation Methods. 1st ed. Charles Griffin, 160 pp.

  • Kohler, M. A., T. J. Nordenson, and W. E. Fox, 1955: Evaporation from pans and lakes. U.S. Dept. of Commerce Research Paper 38, Washington, D.C., 21 pp.

  • Lagarias, J. C., J. A. Reeds, M. H. Wright, and P. E. Wright, 1988: Convergence properties of the Nelder–Mead simplex method in low dimensions. SIAM J. Optim., 9 , 112147.

    • Search Google Scholar
    • Export Citation
  • Liu, B., M. Xu, M. Henderson, and W. Gong, 2004: A spatial analysis of pan evaporation trends in China, 1955–2000. J. Geophys. Res., 109 .D15102, doi:10.1029/2004JD004511.

    • Search Google Scholar
    • Export Citation
  • Lucas, C., B. Trewin, and N. Nicholls, 2004: Development of an historical humidity database for Australia. Proc. Climate and Water: 16th Australian New Zealand Climate Forum, Lorne, Victoria, Australia, ANZCF 2004 Conference Committee, 59.

  • Mann, H. B., 1945: Nonparametric tests against trend. Econometrica, 13 , 245259.

  • Morton, F. I., 1983: Operational estimates of areal evapotranspiration and their significance to the science and practice of hydrology. J. Hydrol., 66 , 176.

    • Search Google Scholar
    • Export Citation
  • Nicholls, N., B. Trewin, and M. Haylock, 2000: Climate extremes indicators for state of the environment monitoring. Australia: State of the Environment, Second Tech. Paper Series (The Atmosphere) No. 1, Department of the Environment and Heritage, Canberra, Australia, 20 pp.

  • Penman, H. L., 1948: Natural evaporation from open water, bare soil and grass. Proc. Roy. Soc. London, 193A , 120145.

  • Peterson, T. C., V. S. Golubev, and P. Y. Groisman, 1995: Evaporation losing its strength. Nature, 377 , 687688.

  • Quintana-Gomez, R. A., 1998: Changes in evaporation patterns detected in northernmost South America: Homogeneity testing. Proc. Seventh Int. Meeting on Statistical Climatology, Whistler, BC, Canada, 97. [Available online at http://cccma.seos.uvic.ca/imsc/proceedings/7IMSC.pdf.].

  • Ramanathan, V., P. J. Crutzen, J. T. Kiehl, and D. Rosenfeld, 2001: Aerosols, climate, and the hydrological cycle. Science, 294 , 21192124.

    • Search Google Scholar
    • Export Citation
  • Ramírez, J. A., M. T. Hobbins, and T. C. Brown, 2005: Observational evidence of the complementary relationship in regional evaporation lends strong support for Bouchet’s hypothesis. Geophys. Res. Lett., 32 .L15401, doi:10.1029/2005GL023549.

    • Search Google Scholar
    • Export Citation
  • Rayner, D. P., K. B. Moodie, A. R. Beswick, N. M. Clarkson, and R. L. Hutchinson, 2004: New Australian daily historical climate surfaces using CLIMARC. Queensland Department of Natural Resources, Mines, and Energy, 76 pp.

  • Roderick, M. L., and G. D. Farquhar, 2002: The cause of decreased pan evaporation over the past 50 years. Science, 298 , 14101411.

  • Roderick, M. L., and G. D. Farquhar, 2004: Changes in Australian pan evaporation from 1970 to 2002. Int. J. Climatol., 24 , 10771090.

  • Roderick, M. L., and G. D. Farquhar, 2005: Changes in New Zealand pan evaporation since the 1970s. Int. J. Climatol., 25 , 20312039.

  • Sen, P. K., 1968: Estimates of regression coefficient based on Kendall’s tau. J. Amer. Stat. Assoc., 63 , 13791389.

  • Stanhill, G., and S. Cohen, 2001: Global dimming: A review of the evidence for a widespread and significant reduction in global radiation with discussion of its probable causes and possible agricultural consequences. Agric. For. Meteor., 107 , 255278.

    • Search Google Scholar
    • Export Citation
  • Tanner, C. B., and T. R. Sinclair, 1983: Efficient water use in crop production: Research or re-search? Limitations to Efficient Water Use in Crop Production, H. M. Taylor, W. R. Jordan, and T. R. Sinclair, Eds., American Society of Agronomy, 1–27.

    • Search Google Scholar
    • Export Citation
  • Thom, A. S., J. L. Thony, and M. Vauclin, 1981: On the proper employment of evaporation pans and atmometers in estimating potential transpiration. Quart. J. Roy. Meteor. Soc., 107 , 711736.

    • Search Google Scholar
    • Export Citation
  • van Dijk, M. H., 1985: Reduction in evaporation due to the bird screen used in the Australian class A pan evaporation network. Aust. Meteor. Mag., 33 , 181183.

    • Search Google Scholar
    • Export Citation
  • Wang, Q. J., F. H. S. Chiew, F. L. N. McConachy, R. James, G. C. de Hoedt, and W. R. Wright, 2001: Climatic Atlas of Australia: Maps of Evapotranspiration. Australian Government Bureau of Meteorology, 11 pp.

    • Search Google Scholar
    • Export Citation
  • Weisstein, E. W., cited. 2006: Spearman rank correlation coefficient. MathWorld—A Wolfram Web Resource. [Available online at http://mathworld.wolfram.com/SpearmanRankCorrelationCoefficient.html.].

  • Whetton, P., 2001: Climate projections for Australia. CSIRO Marine and Atmospheric Research, Melbourne, Australia, 8 pp.

  • View in gallery

    Geographic distribution of stations used in the pan evaporation trends attribution study. The 13 stations for which there were statistically significant trends in both the observed and modeled pan evaporation are labeled and denoted with a star symbol. The names of the observing stations have been abbreviated for convenience.

  • View in gallery

    Pan evaporation trends (1975–2004) for attribution model runs. The meteorological climate variables used in each of the model runs are indicated in each panel; average climate data were used for the remaining variables. In each panel, a plotted point represents the modeled vs observed pan evaporation trend for one observing station. Statistically significant observed trends are overprinted with a plus (+) symbol; statistically significant modeled trends are overprinted with a cross (×). Points overprinted with both a plus and a cross, an “asterisk”, have statistically significant trends in both observed and modeled pan evaporation. Modeled pan evaporation trends were only consistent with the observed pan evaporation trends when the meteorological wind run uC was used (bottom right panel).

  • View in gallery

    Average monthly wind run anomalies and annual pan evaporation anomalies for the 13 stations for which there were statistically significant trends in both observed and modeled pan evaporation. Anomalies were calculated using the base period 1975–2004.

  • View in gallery

    Comparison of trends in observed and proxy wind run time series: (a) observed and NCEP–NCAR reanalysis trends; (b) observed and pressure gradient derived trends; (c) NCEP–NCAR reanalysis and pressure gradient derived trends. Statistically significant trends in observed wind run are overprinted with a plus symbol, statistically significant trends in the NCEP–NCAR reanalysis wind run are overprinted with a cross, and statistically significant trends in the pressure gradient derived wind run are overprinted with a triangle. Note the change in scale in (c).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 444 316 15
PDF Downloads 257 173 11

Wind Run Changes: The Dominant Factor Affecting Pan Evaporation Trends in Australia

View More View Less
  • 1 Queensland Department of Natural Resources, Mines, and Water, Indooroopilly, Australia
Full access

Abstract

The Class A pan evaporation rates at many Australian observing stations have reportedly decreased between 1970 and 2002. That pan evaporation rates have decreased at the same time that temperatures have increased has become known as the “pan evaporation paradox.”

Pan evaporation is primarily dependant on relative humidity, solar radiation, and wind. In this paper, trends in observed pan evaporation in Australia during the period 1975–2004 were attributed to changes in other climate variables using a Penman-style pan evaporation model. Trends in daily average wind speed (termed wind run) were found to be an important cause of trends in pan evaporation. This result is a significant step toward resolving the pan evaporation paradox for Australia.

Data inspection and interstation comparison revealed that some of the significant wind run trends were discontinuous or spatially uncorrelated. These analyses raised the possibility that some of the changes in observed wind run, and by implication some of the significant changes in pan evaporation, may represent changes in the local environment surrounding the observing stations.

Daily pressure gradients and NCEP–NCAR reanalysis wind surfaces were analyzed in an attempt to identify any climatological wind run trends associated with large-scale changes in atmospheric circulations. Unfortunately, the trends from the two data sources were not consistent, and the challenge remains to conclusively identify the cause or causes of the changes in observed station wind run in Australia.

Corresponding author address: David Rayner, Queensland Department of Natural Resources, Mines, and Water, QCCA Building, Gate 4, 80 Meiers Rd., Indooroopilly, Qld 4068, Australia. Email: David.Rayner@qld.gov.au

Abstract

The Class A pan evaporation rates at many Australian observing stations have reportedly decreased between 1970 and 2002. That pan evaporation rates have decreased at the same time that temperatures have increased has become known as the “pan evaporation paradox.”

Pan evaporation is primarily dependant on relative humidity, solar radiation, and wind. In this paper, trends in observed pan evaporation in Australia during the period 1975–2004 were attributed to changes in other climate variables using a Penman-style pan evaporation model. Trends in daily average wind speed (termed wind run) were found to be an important cause of trends in pan evaporation. This result is a significant step toward resolving the pan evaporation paradox for Australia.

Data inspection and interstation comparison revealed that some of the significant wind run trends were discontinuous or spatially uncorrelated. These analyses raised the possibility that some of the changes in observed wind run, and by implication some of the significant changes in pan evaporation, may represent changes in the local environment surrounding the observing stations.

Daily pressure gradients and NCEP–NCAR reanalysis wind surfaces were analyzed in an attempt to identify any climatological wind run trends associated with large-scale changes in atmospheric circulations. Unfortunately, the trends from the two data sources were not consistent, and the challenge remains to conclusively identify the cause or causes of the changes in observed station wind run in Australia.

Corresponding author address: David Rayner, Queensland Department of Natural Resources, Mines, and Water, QCCA Building, Gate 4, 80 Meiers Rd., Indooroopilly, Qld 4068, Australia. Email: David.Rayner@qld.gov.au

Keywords: Evaporation; Wind

1. Overview

a. Evapotranspiration, potential evaporation, and pan evaporation

Evapotranspiration is the transfer of water from the landscape to the atmosphere, a combination of evaporation from soil and plant transpiration. Evapotranspiration is a critical component of the water cycle in Australia, with over 90% of rainfall returned to the atmosphere through evapotranspiration (Wang et al. 2001).

Evapotranspiration is difficult to measure directly, however, and is usually calculated using a semiempirical equation that combines a climatic component, vegetation characteristics, and water availability. The climatic component––potential evapotranspiration––is often expressed as the evapotranspiration that would occur from a well-watered grass reference crop (Allen et al. 1998).

Accurately calculating potential evapotranspiration requires a suite of measured climate data, typically temperature, vapor pressure, solar radiation, and wind speed. A simple and inexpensive surrogate is the evaporation rate in an open water container, termed pan evaporation. However, there are important considerations when using pan evaporation to approximate potential evapotranspiration. First, the surface resistance of damp grass is smaller than the surface resistance of open water, necessitating a scaling factor (commonly taken as 0.7–0.8; Grayson et al. 1996). Second, pan evaporation is an estimate of point potential evaporation, meaning that the surrounding atmosphere is not modified in any significant way. In contrast, a large, watered crop surface may significantly change the vapor pressure in the air above, and so areal potential evapotranspiration estimates (the evaporation from an area sufficiently large that edge effects can be ignored; Allen et al. 1998; Morton 1983) are a better approximation.

The Australian Government Bureau of Meteorology (BoM) maintains a network of Class A pan evaporimeters. The Class A pan evaporimeter is a circular pan made of galvanized iron, 4 ft (≈121 cm) in diameter and 10 in. (≈25 cm) deep, mounted on an open wooden platform. In Australia, the pans are protected with a wire bird guard to stop animals from affecting the measurements. BoM began systematically installing Class A pans in the 1960s, with the universal installation of bird guards following a few years later. By 1975, nearly all pans in the BoM evaporimeter network had consistent instrumentation (for details, see BoM 2005a).

b. Pan evaporation trends

A general decline in pan evaporation rates was first reported by Peterson et al. (1995) for the United States and the former Soviet Union, and subsequently for India (Chattopadhyay and Hulme 1997), Venezuela (Quintana-Gomez 1998), China (Liu et al. 2004), Australia (Roderick and Farquhar 2004), and New Zealand (Roderick and Farquhar 2005), with an increase reported at a single pan in Israel (Cohen et al. 2002). If all other atmospheric properties were static, the temperature increases that occurred in the twentieth century might be expected to lead to an increase in pan evaporation, on the grounds that the water-holding capacity of the atmosphere is related to temperature. This discrepancy between observation and supposed expectation has become known as the “pan evaporation paradox” (e.g., Brutsaert and Parlange 1998).

In reality, however, researchers and climate change scientists (e.g., Arnell et al. 2001) have been well aware that pan evaporation rates are not dominated by temperature, and studies have concentrated on attributing the changes in pan evaporation to changes in the contributing climate variables. Two main explanations have been proposed. The first explanation, advocated by Peterson et al. (1995), Cohen et al. (2002), and Roderick and Farquhar (2002), associates the pan trends primarily with the declining global solar irradiance (e.g., Gilgen et al. 1998; Stanhill and Cohen 2001) caused by changes in cloudiness (Henderson-Sellers 1992) or aerosol concentrations (Ramanathan et al. 2001). Conversely, Hobbins et al. (2004) and Ramírez et al. (2005) support the suggestion by Brutsaert and Parlange (1998) that the pan evaporation trends across the United States were a manifestation of the “complementary relationship” between actual evaporation and potential evaporation (Bouchet 1963).

The report by Roderick and Farquhar (2004) that the Class A pan evaporation rates decreased between 1970 and 2002 at many Australian observing stations has generated considerable debate within the Australian climate research community, because it appears to conflict with the highly publicized general circulation model projections for increasing potential evaporation in Australia. For example, Whetton (2001) examined eight models and concluded that areal potential evaporation would most likely increase by 0%–8% per degree of global warming over most of Australia. Such changes, combined with projected rainfall changes, could lead to an increase in aridity in Australia.

More recently, BoM has developed a homogenized pan evaporation dataset for Australia (Jovanovic et al. 2007). Although Jovanovic et al. did not find any significant trend in the Australian annual mean pan evaporation over the 1970–2005 period, they concluded that the need to reconcile the small trend in pan evaporation with the large projected changes in potential evaporation remained.

To investigate the Australian pan evaporation paradox, in this paper the trends in observed pan evaporation during the period 1975–2004 are attributed to changes in other climate variables using the Thom et al. (1981) Class A pan evaporation model.

2. Modeling daily pan evaporation

a. The Thom, Thony, and Vauclin Class A pan evaporation model

Thom et al. (1981) derived a Penman-style (Penman 1948) combination equation for Class A pan evaporation. Their model takes the form [Thom et al. 1981, their Eq. (12)]
i1520-0442-20-14-3379-e1
where (units and notation as used in this paper)
  • ECA is pan evaporation rate for Class A pan (mm day−1),

  • Δ is slope of the saturation vapor pressure curve at temperature T (kPa °C−1),

  • aRn is net radiation absorbed by the entire pan (mm day−1 equivalent),

  • S is change in pan heat storage (mm day−1 equivalent),

  • is the effective psychrometric constant of the pan (kPa °C−1),

  • δeC is the vapor pressure deficit at the pan rim height (kPa), and

  • fνC is the pan wind function for mass transfer as a function of wind run uC (mm day−1 kPa−1).

Equation (1) represents the daily evaporation rate assuming constant atmospheric properties. Ideally, one would start with a model for the instantaneous evaporation rate and integrate this over the day to determine the daily evaporation. However, because only limited subdaily climate data were available with which to model historical pan evaporation, the terms in Eq. (1) were approximated by “characteristic daily values.” Appendix A describes how the terms in Eq. (1) were approximated using functions of daily meteorological data.

b. Meteorological data

The meteorological datasets used were all either surface-observed climate data provided by BoM or derived products. The model used daily maximum and minimum temperatures, vapor pressures recorded at 0900 and 1500 local time, daily wind run, and daily solar radiation. Model calibration used observed pan evaporation records. The meteorological datasets are described in detail in appendix B.

Any pan evaporation modeling study has to deal with incomplete datasets. Although it is preferable to use meteorological data from the stations that observe pan evaporation, in practice there are very few climate stations in Australia that record pan evaporation, solar radiation, temperature, humidity, and wind run. Consequently, this analysis used a mixture of observed, spatially interpolated, and modeled meteorological datasets. The pan evaporation trends attribution study described in section 4 used

  • observed pan evaporation records,

  • a composite of observed and spatially interpolated daily maximum and minimum temperature data,

  • vapor pressure data from interpolated surfaces,

  • solar radiation data derived from observations of cloud amounts at 0900 and 1500 local time, sourced from interpolated surfaces, and

  • observed wind run recorded at 2-m height.

The transformations that relate meteorological data to the terms used in Eq. (1) are given in appendix A.

Note that, because the solar radiation data were derived from observations of cloud amounts (in oktas), they will not represent any changes in cloud properties or clear-sky optical properties. The solar radiation dataset, and the reasons for using it, are discussed in appendix B.

Usually, historical climate change analyses use homogenized datasets, where corrections have been applied to remove artificial discontinuities caused by changes in instrumentation, site location, or observational practices (e.g., Della-Marta et al. 2004). Here, the objective was to model the changes in the “raw” pan evaporation data, as were used by Roderick and Farquhar (2004), and then to use the model to attribute any changes in observed pan evaporation to changes in the observed meteorological data.

c. Model calibration

The Thom et al. pan evaporation model and the daily approximations used for its constituent terms have a total of five empirical parameters. These parameters were determined by minimizing the differences between 4-day cumulative pan evaporation and model evaporation using a simplex search (e.g., Lagarias et al. 1988). Four-day averages were used to minimize any bias caused by S, the pan heat storage term. A single set of parameters was derived for all observing stations. A detailed account of the model calibration procedure and comparison of the parameter values with those from Thom et al. is given in appendix C.

d. Trend calculations

Trends in annual pan evaporation were calculated for stations that had observed wind run and pan evaporation data for at least 80% of days in the period 1975–2004. The geographic distribution of these 67 stations is shown in Fig. 1; the observing stations are referenced by their BoM site numbers. The process used to aggregate daily time series with missing data into annual time series, and the statistical methodologies used to estimate trends and significance, are given in appendix D.

e. Model performance

Inevitably, because of differing climate regimes, equipment, and observer practices, the pan evaporation model represents observed pan evaporation more accurately at some stations than others. The correlations between the daily modeled pan evaporation anomalies1 and daily observed anomalies were r2 ≈ 0.45–0.65,2 and were higher for the summer months. The correlations between annual observed and modeled pan evaporations were slightly higher at r2 ≈ 0.45–0.8.

The correlation between the trends3 in observed pan evaporation and the trends in modeled pan evaporation for the stations was r2 = 0.61. Overall, a third of the spatial variance in observed pan evaporation trends over the period 1975–2004 could be attributed to changes in other climate variables using the Thom et al. model.

3. Pan evaporation trends attribution study

To determine the relative contributions of the different climate variables to the trends in modeled pan evaporation, an attribution study was conducted in which the model was run substituting day-of-year average climate data for meteorological climate data.

a. Methodology

As a precursor, a sensitivity study was performed. The model was run six times, each time using meteorological climate data for only one of the following:

  • ea,1500 or ea,0900, the vapor pressures at 1500 and 0900 local time,

  • Tmax or Tmin, the daily maximum and minimum temperatures,

  • uC, the wind run expressed as a daily-average wind speed, or

  • Rs, the incoming solar (“shortwave”) radiation.

In each run, average climate data were used for the other variables. These runs were used to rank the climate variables in terms of their contribution to the trends in the modeled pan evaporation. The ranking was determined to be as follows: Rs (lowest), ea,0900, Tmin, ea,1500, Tmax, and uC (highest).

For the attribution study, the model was run 6 more times, the first time using meteorological data for only the climate variable with the least influence on the modeled trends (Rs), and subsequently adding variables in order of rank. Thus, the penultimate run used meteorological data for all climate variables except the wind run, and the final run used meteorological data for all climate variables.

b. Results

The attribution study results are shown in Fig. 2. Trends in the meteorological wind run are clearly important for modeling trends in observed pan evaporation. With meteorological wind run, the slope of the regression of modeled trend onto observed trend was 0.69 ± 0.14,4 with r2 = 0.61. Without meteorological wind run, the slope of the regression was only 0.14 ± 0.08, with r2 = 0.14.

Seventeen stations had statistically significant trends in observed pan evaporation. Of these, 4 stations had increasing trends and 13 had decreasing trends. The trends in observed and modeled pan evaporation for these 17 stations are summarized in Table 1. For the model run using the meteorological wind run, 3 stations had increasing trends and 14 had decreasing trends, similar to the observed. When using the average wind run in the model, however, the ratio was reversed; 14 stations had increasing trends and 3 had decreasing trends. In fact, for the model using the average wind run, only 9 of the 67 stations had declining pan evaporation trends, and none of these trends were statistically significant.

The model residuals (for the model run using meteorological data for all variables) contained statistically significant trends at many stations. Of the 67 stations, there were increasing trends in the residuals for 21 stations (7 significant), and decreasing trends for 46 stations (20 significant). The reason there were more statistically significant trends in the residuals than in the original observed data is that the model removes much of the interannual variability. Thus, it appears that there are underlying trends in the observed pan evaporation time series that are not represented using the Thom et al. model with available meteorological data.

c. Discussion

Changes in wind run were found to be the most important cause of trends in observed pan evaporation. This result has important implications for the study of historical pan evaporation in Australia, as discussed in section 6.

Figure 2 suggests that solar radiation changes have had very little influence on pan evaporation trends. However, as mentioned above, the solar radiation data are derived from twice-daily observations of cloud amounts, and there may be changes in global irradiance that are not captured by cloud observations. This is probably the most significant deficiency in the attribution study and could potentially explain the trends in the model residuals.

Interestingly, the model results shown in Fig. 2 also suggest that the effect of changes in temperature and vapor pressure has been to increase pan evaporation at most stations in Australia over the period 1975–2004. The model suggests that the reason widespread increases have not been observed is primarily because they are masked by the changes in wind run, possibly in association with changes in solar radiation.

4. Differentiating between climatological, artificial, and environmental change in the wind run measurements

The correlation between observed pan evaporation trends and trends modeled using observed wind run is itself sufficient to demonstrate that the observed evaporation trends were not entirely caused by artificial factors such as instrumental errors or observational procedure changes, but that they were related to changes in wind run at observing stations.

However, having made this point, the correlation does not establish whether the wind run changes represent large-scale climatological changes or whether they represent local environmental changes. Such environmental changes might be very local, such as construction or vegetation changes immediately adjacent to the observing station, or they could represent more regional changes such as large-scale land use change. This section describes some simple tests that were performed in an attempt to differentiate between climatological, local, and artificial changes.

a. Visual inspection

Monthly anomaly time series of observed pan evaporation, modeled pan evaporation, solar radiation, vapor pressure deficit, and wind run for all stations used in the attribution study are available online (see http://www.nrm.qld.gov.au/silo/pan/).

The monthly wind run anomalies and annual pan evaporation anomalies for the 13 stations that had statistically significant trends in both observed pan evaporation and modeled pan evaporation are shown in Fig. 3. Many of the changes in pan evaporation can be seen to reflect changes in wind run, although there are clearly exceptions such as Mount Bold (23734) and Rockhampton (39083). For a number of stations the changes in wind run and evaporation are rapid and dramatic, and are unlikely to be related to large-scale climatological change, for example, the changes at Alice Springs (15590) and Woomera (16001) before 1980. The well-matched and highly significant pan evaporation and wind run changes at Pemberton (9592) and Blanchetown (24564) occur over only ≈10 yr, which would be very rapid if these were climatological changes.

Of course, gradual change does not itself imply climatological change. For example, the decrease in wind run at Mount Gambier (26012) is relatively gradual, but it is possible that this decrease is related to land use change, as pine plantations are slowly replacing farmland in the surrounding area (D. Jones, BoM, 2006, personal communication).

b. Nearby stations

Comparing climate time series against nearby stations is an important consistency check. Wind run data for all stations near the 13 stations in Fig. 3 were assembled for visual inspection. A climatological explanation for the abrupt discontinuities in the wind run time series for Alice Springs (15590), Mount Bold (23734), and Caliph (25050) could be confidently rejected because nearby stations showed no such discontinuities.

The nearby stations used to assess the nature of abrupt discontinuities could have relatively short wind run records, provided that those records overlapped the discontinuities. However, assessing the nature of gradual changes in wind run using nearby stations was less useful, because this required long wind run records, and the number of stations with long records was fairly small. The nearest stations with comparable wind run data were usually hundreds of kilometers away, making them subject to significantly different geographical conditions (especially where a station was much closer to the coastline than its “neighbor”).

The exception was the region containing the cluster of 4 stations in South Australia, shown in the inset in Fig. 1, in which there were 17 stations with at least 15 yr of wind run data. Fourteen of these stations showed decreasing wind run trends (seven statistically significant) and only three showed increasing trends (none significant). However, three of the seven statistically significant trends for decreasing wind run involved visual discontinuities in the time series [Bolivar (23081), Mount Bold (23734), and Caliph (25050)].

c. Large-scale atmospheric circulation patterns

In parts of Australia, daily wind run is highly dependent on large-scale atmospheric circulations. Wind run proxies were created from two sources of information on large-scale atmospheric circulations: the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis wind product and Silo daily MSL pressure surfaces. Although the observed wind run trends were not observed in either proxy series, the trends in the two series were inconsistent.

1) NCEP–NCAR reanalysis wind run proxy

The NCEP–NCAR reanalysis project (Kalnay et al. 1996) produces a record of global atmospheric fields from 1958 to the present. The reanalysis data are produced by numerical weather simulation and incorporate extensive meteorological and satellite observations. Importantly, the reanalysis project uses a frozen forecast/analysis system to minimize jumps in the climate fields associated with changes in data assimilation or instrumentation.

The reanalysis project provides 6-hourly wind speed data on a 2.5° × 2.5° grid, and the comparatively coarse grid precludes local geographical factors affecting the wind field product. The wind product used is an “A class” field (Kalnay et al. 1996), indicating that it is strongly influenced by the assimilated data. Note that the BoM observed wind run data used in this report are not assimilated into the reanalysis, so the two datasets can be considered independent estimates of surface wind run.

A reanalysis wind run proxy time series was calculated from the 6-hourly, 0.995 sigma-level wind product. This level corresponds to a height of approximately 40 m, and so the reanalysis product was linearly rescaled to represent the 2-m wind run by comparing the reanalysis wind run and the observed wind run for each station. The rescaling factor varied between stations, with the median corresponding to a roughness factor5 of f = 0.29. This was consistent with a median roughness factor of f = 0.30 derived by comparing 10- and 2-m station wind run observations.

The correlations between daily reanalysis wind run anomalies and observed wind run anomalies were r2 ≈ 0.25–0.4 for the 67 stations in the attribution study but were higher (up to r2 ≈ 0.6–0.7) for stations in southern Australia. However, despite the correlation in daily anomalies, the correlation between the annual averages was low, and the observed wind run trends were not correlated with the reanalysis wind run trends (Fig. 4a). This is discussed further below.

2) Pressure gradient derived wind run proxy

Atmospheric pressure provides a source of climatological wind information that should have fewer systematic errors than the station wind run records (Nicholls et al. 2000). Surface pressure measurements are reliable, and the long spatial-correlation lengths make MSL pressure easy to interpolate.

A simple pressure-derived model wind run was calculated using regression of observed daily wind run against pressure gradients calculated from the Silo (http://www.nrm.qld.gov.au/silo/) 0900 local time MSL pressure surfaces (Jeffrey et al. 2001). The pressure gradients were calculated for each day and for each station by fitting a plane to the pressure surface at the station location.

The correlations between daily pressure-derived wind run anomalies and observed wind run anomalies were similar to the NCEP–NCAR reanalysis wind run discussed above, with r2 ≈ 0.3–0.45. And again, the correlations between the annual averages were low, and the observed wind run trends were not correlated with the pressure-derived wind run trends (Fig. 4b). In addition, despite a number stations having statistically significant trends in both wind run proxy time series, the correlation between the pressure-derived wind run trends and the reanalysis wind run trends was very low (r2 = 0.11; Fig. 4c).

The low correlations between the observed and proxy annual wind run time series, and the low correlations between the wind run proxy trends, suggest that neither wind run proxy can be considered a reliable indicator of climatological wind run changes for the stations in this study. This is perhaps not surprising: the proxy wind run time series do not correlate well with the observed wind run for many of the northern stations in the study sample, especially in austral summer, presumably implying that daily wind run at these stations is not dominated by large-scale atmospheric circulations. A separate analysis of wind run trends for 23 stations in southeast Australia, using April to September averages, showed high correlations between the proxy time series (r2 ≈ 0.39–0.60), which is consistent with this explanation.

d. Other tests

Other potential proxies for historical wind run data were investigated, but none were found to be useful.

BoM is in the process of computerizing their station metadata records and helpfully provided the records they have computerized so far. The metadata sheets warn explicitly that there are “large gaps in the information contained.” There were many instances of rapid, systematic changes in wind run and pan evaporation that had no explanation in the metadata, and it was regrettably concluded that the published metadata are not yet complete enough to assist in identifying local or artificial wind run changes.

Eight of the 13 stations examined in detail had short periods with concurrent observations from two anemometers at different heights. However, the higher-level wind run records were all too short to investigate 1975–2004 trends, and the periods of concurrent observation did not cover any of the large discontinuities mentioned in section 5a.

Instantaneous wind speed observations are made using a different anemometer to daily wind run observations, and there is a very large dataset of 0900 and 1500 local time wind speed observations (BoM 2005d). Unfortunately, there were no correlations between wind speed trends and either wind run trends or pan evaporation trends.

Many areas of Australia have seen large-scale land use change in recent decades. The Statewide Landcover and Trees Study (SLATS; http://www.nrm.qld.gov.au/slats/index.html) 30-m resolution Foliage Projective Cover (FPC) product was investigated to see whether changes in station wind run could be related to changes in vegetation cover in the regions surrounding the observing stations. However, the FPC time series and the wind run time series were uncorrelated.

5. Summary and concluding remarks

The Thom et al. (1981) model for Class A pan evaporation has been recalibrated using Australian meteorological data. The model trends correlated well with the trends in observed pan evaporation reported by Roderick and Farquhar (2004). A sensitivity study showed that trends in daily average wind speed (termed wind run) were an important cause of trends in pan evaporation. This also shows that the BoM wind run measurements generally represent the wind run experienced by the evaporation pans.

A number of important questions concerning historical pan evaporation, and by implication potential evapotranspiration, remain unanswered. The most immediate question is whether the trends in observed station wind run represent climatic changes or changes specific to the location of the observing stations. The initial tests reported here raised the possibility that some of the changes in observed wind run may represent changes in the local environment surrounding the observing stations, but no wind run proxy was found that could be considered a reliable indicator of climatological wind run changes for the stations in this study.

This study raises a practical problem for understanding historical pan evaporation in Australia. It would be valuable to apply the Thom et al. pan evaporation model to longer time series of climate variables, to see whether recent changes in pan evaporation are historically unprecedented or whether they are within the realm of natural climate variability. Given the importance of wind in modeling pan evaporation trends, clearly any such analysis requires historical wind data. However, there are very few BoM wind run records prior to 1970. Because the alternative sources of pre-1970 wind information—the NCEP–NCAR reanalysis, pressure gradients, and hourly wind speed observations—do not represent the changes in observed pan evaporation, they cannot be used to model long-term historical pan evaporation changes in Australia.

Clearly, modeling historical pan evaporation and potential evapotranspiration in Australia for large regions, such as river catchments or even for the whole continent, will require high-quality, homogenized climate datasets. BoM has produced homogenized daily minimum and maximum temperature datasets (Collins et al. 2000) and are working on a vapor pressure dataset (Lucas et al. 2004). Hopefully there will be many stations common to all datasets.

Finally, the recently released homogenized pan evaporation dataset (Jovanovic et al. 2007) has helped to clarify the regional significance of recent changes in observed pan evaporation. However, evaporation models, such as the Thom et al. pan evaporation model used here, will be the primary tools for investigating potential evaporation changes over the last century, and for determining the climate variables responsible for any potential evaporation trends.

Acknowledgments

I am very grateful to David Jones (Australian Government Bureau of Meteorology), Michael Roderick (Australian National University), and Greg McKeon and Steven Crimp (Queensland Department of Natural Resources, Mines, and Water) for their valuable suggestions and constructive reviews. The NCEP–NCAR reanalysis data were provided by the NOAA–CIRES ESRL/PSD Climate Diagnostics branch (from their Web site at http://www.cdc.noaa.gov/).

REFERENCES

  • Allen, R. G., L. S. Pereira, D. Raes, and M. Smith, 1998: Crop evapotranspiration—Guidelines for computing crop water requirements. FAO Irrigation and Drainage Paper 56, 2d ed., Food and Agriculture Organization of the United Nations, Rome, Italy, 300 pp.

  • Arnell, N. W., and Coauthors, 2001: Hydrology and water resources. Climate Change 2001: Impacts, Adaptation, and Vulnerability, J. J. McCarthy et al., Eds., Cambridge University Press, 191–233.

    • Search Google Scholar
    • Export Citation
  • Bouchet, R. J., 1963: Evapotranspiration réelle evapotranspiration potentielle, signification climatique. Int. Assoc. Sci. Hydrol., 62 , 134142.

    • Search Google Scholar
    • Export Citation
  • Brutsaert, W., and M. B. Parlange, 1998: Hydrologic cycle explains the evaporation paradox. Nature, 396 , 30.

  • Bureau of Meteorology, 2002: NCCSOL—Australian solar radiation data. CD-ROM, 2.209.

  • Bureau of Meteorology, 2005a: Australian daily evaporation data. CD-ROM, IDCJDC05.200506.

  • Bureau of Meteorology, 2005b: Australian daily wind data. CD-ROM, IDCJDC06. 200506.

  • Bureau of Meteorology, 2005c: Australian hourly temperature humidity and pressure. CD-ROM, IDCJHCO2.200506.

  • Bureau of Meteorology, 2005d: Australian hourly wind data. CD-ROM, IDCJHC01. 200506.

  • Chattopadhyay, N., and M. Hulme, 1997: Evaporation and potential evapotranspiration in India under conditions of recent and future climate change. Agric. For. Meteor., 87 , 5573.

    • Search Google Scholar
    • Export Citation
  • Cohen, S., A. Ianetz, and G. Stanhill, 2002: Evaporative climate changes at Bet Dagan, Israel: 1964–1998. Agric. For. Meteor., 111 , 8391.

    • Search Google Scholar
    • Export Citation
  • Collins, D. A., P. M. Della-Marta, N. Plummer, and B. C. Trewin, 2000: Trends in annual frequencies of extreme temperature events in Australia. Aust. Meteor. Mag., 49 , 277292.

    • Search Google Scholar
    • Export Citation
  • Della-Marta, P., D. Collins, and K. Braganza, 2004: Updating Australia’s high-quality annual temperature dataset. Aust. Meteor. Mag, 53 , 7593.

    • Search Google Scholar
    • Export Citation
  • Forgan, B. W., 2005: Australian solar and terrestrial network data. Proc. Pan Evaporation: An Example of the Detection and Attribution of Trends in Climate Variables, Canberra, Australia, Australian Academy of Science, 50–53.

  • Gilgen, H., M. Wild, and A. Ohmura, 1998: Means and trends of shortwave irradiance at the surface estimated from global energy balance archive data. J. Climate, 11 , 20422061.

    • Search Google Scholar
    • Export Citation
  • Grayson, R. B., R. M. Argent, R. J. Nathan, T. A. McMahon, and R. G. Mein, 1996: Hydrological Recipes: Estimation Techniques in Australian Hydrology. Cooperative Research Centre for Catchment Hydrology, 125 pp.

    • Search Google Scholar
    • Export Citation
  • Henderson-Sellers, A., 1992: Continental cloudiness changes this century. GeoJournal, 27 , 255262.

  • Hobbins, M. T., J. A. Ramírez, and T. C. Brown, 2004: Trends in pan evaporation and actual evapotranspiration across the conterminous U.S.: Paradoxical or complementary? Geophys. Res. Lett., 31 .L13503, doi:10.1029/2004GL019846.

    • Search Google Scholar
    • Export Citation
  • Jeffrey, S. J., J. O. Carter, K. B. Moodie, and A. R. Beswick, 2001: Using spatial interpolation to construct a comprehensive archive of Australian climate data. Environ. Modell. Software, 16 , 309330.

    • Search Google Scholar
    • Export Citation
  • Jovanovic, B., D. Jones, and D. Collins, 2007: A high quality monthly pan evaporation dataset for Australia. Climatic Change, in press.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437471.

  • Kendall, M. G., 1948: Rank Correlation Methods. 1st ed. Charles Griffin, 160 pp.

  • Kohler, M. A., T. J. Nordenson, and W. E. Fox, 1955: Evaporation from pans and lakes. U.S. Dept. of Commerce Research Paper 38, Washington, D.C., 21 pp.

  • Lagarias, J. C., J. A. Reeds, M. H. Wright, and P. E. Wright, 1988: Convergence properties of the Nelder–Mead simplex method in low dimensions. SIAM J. Optim., 9 , 112147.

    • Search Google Scholar
    • Export Citation
  • Liu, B., M. Xu, M. Henderson, and W. Gong, 2004: A spatial analysis of pan evaporation trends in China, 1955–2000. J. Geophys. Res., 109 .D15102, doi:10.1029/2004JD004511.

    • Search Google Scholar
    • Export Citation
  • Lucas, C., B. Trewin, and N. Nicholls, 2004: Development of an historical humidity database for Australia. Proc. Climate and Water: 16th Australian New Zealand Climate Forum, Lorne, Victoria, Australia, ANZCF 2004 Conference Committee, 59.

  • Mann, H. B., 1945: Nonparametric tests against trend. Econometrica, 13 , 245259.

  • Morton, F. I., 1983: Operational estimates of areal evapotranspiration and their significance to the science and practice of hydrology. J. Hydrol., 66 , 176.

    • Search Google Scholar
    • Export Citation
  • Nicholls, N., B. Trewin, and M. Haylock, 2000: Climate extremes indicators for state of the environment monitoring. Australia: State of the Environment, Second Tech. Paper Series (The Atmosphere) No. 1, Department of the Environment and Heritage, Canberra, Australia, 20 pp.

  • Penman, H. L., 1948: Natural evaporation from open water, bare soil and grass. Proc. Roy. Soc. London, 193A , 120145.

  • Peterson, T. C., V. S. Golubev, and P. Y. Groisman, 1995: Evaporation losing its strength. Nature, 377 , 687688.

  • Quintana-Gomez, R. A., 1998: Changes in evaporation patterns detected in northernmost South America: Homogeneity testing. Proc. Seventh Int. Meeting on Statistical Climatology, Whistler, BC, Canada, 97. [Available online at http://cccma.seos.uvic.ca/imsc/proceedings/7IMSC.pdf.].

  • Ramanathan, V., P. J. Crutzen, J. T. Kiehl, and D. Rosenfeld, 2001: Aerosols, climate, and the hydrological cycle. Science, 294 , 21192124.

    • Search Google Scholar
    • Export Citation
  • Ramírez, J. A., M. T. Hobbins, and T. C. Brown, 2005: Observational evidence of the complementary relationship in regional evaporation lends strong support for Bouchet’s hypothesis. Geophys. Res. Lett., 32 .L15401, doi:10.1029/2005GL023549.

    • Search Google Scholar
    • Export Citation
  • Rayner, D. P., K. B. Moodie, A. R. Beswick, N. M. Clarkson, and R. L. Hutchinson, 2004: New Australian daily historical climate surfaces using CLIMARC. Queensland Department of Natural Resources, Mines, and Energy, 76 pp.

  • Roderick, M. L., and G. D. Farquhar, 2002: The cause of decreased pan evaporation over the past 50 years. Science, 298 , 14101411.

  • Roderick, M. L., and G. D. Farquhar, 2004: Changes in Australian pan evaporation from 1970 to 2002. Int. J. Climatol., 24 , 10771090.

  • Roderick, M. L., and G. D. Farquhar, 2005: Changes in New Zealand pan evaporation since the 1970s. Int. J. Climatol., 25 , 20312039.

  • Sen, P. K., 1968: Estimates of regression coefficient based on Kendall’s tau. J. Amer. Stat. Assoc., 63 , 13791389.

  • Stanhill, G., and S. Cohen, 2001: Global dimming: A review of the evidence for a widespread and significant reduction in global radiation with discussion of its probable causes and possible agricultural consequences. Agric. For. Meteor., 107 , 255278.

    • Search Google Scholar
    • Export Citation
  • Tanner, C. B., and T. R. Sinclair, 1983: Efficient water use in crop production: Research or re-search? Limitations to Efficient Water Use in Crop Production, H. M. Taylor, W. R. Jordan, and T. R. Sinclair, Eds., American Society of Agronomy, 1–27.

    • Search Google Scholar
    • Export Citation
  • Thom, A. S., J. L. Thony, and M. Vauclin, 1981: On the proper employment of evaporation pans and atmometers in estimating potential transpiration. Quart. J. Roy. Meteor. Soc., 107 , 711736.

    • Search Google Scholar
    • Export Citation
  • van Dijk, M. H., 1985: Reduction in evaporation due to the bird screen used in the Australian class A pan evaporation network. Aust. Meteor. Mag., 33 , 181183.

    • Search Google Scholar
    • Export Citation
  • Wang, Q. J., F. H. S. Chiew, F. L. N. McConachy, R. James, G. C. de Hoedt, and W. R. Wright, 2001: Climatic Atlas of Australia: Maps of Evapotranspiration. Australian Government Bureau of Meteorology, 11 pp.

    • Search Google Scholar
    • Export Citation
  • Weisstein, E. W., cited. 2006: Spearman rank correlation coefficient. MathWorld—A Wolfram Web Resource. [Available online at http://mathworld.wolfram.com/SpearmanRankCorrelationCoefficient.html.].

  • Whetton, P., 2001: Climate projections for Australia. CSIRO Marine and Atmospheric Research, Melbourne, Australia, 8 pp.

APPENDIX A

Terms in the Pan Evaporation Model for Application to Daily Data

In the absence of continuous climate data, the following approximations were used for the components in Eq. (1).

Slope of the saturation vapor pressure curve, Δ

The slope of the saturation vapor pressure curve was calculated using the method of Allen et al. [1998, their Eq. (13)]:
i1520-0442-20-14-3379-ea1
which uses the mean daily temperature:
i1520-0442-20-14-3379-ea2
An attempt to improve the approximation of Δ by allowing the weights of Tmax and Tmin in Eq. (A1) to vary in the calibration did not improve the pan evaporation model residuals.

Net daily solar radiation absorbed by the pan, aRn

The two components of this term were as follows:

  1. Rn, the net daily solar radiation onto a hypothetical grass reference surface: Rn was calculated from the Silo (http://www.nrm.qld.gov.au/silo/) solar (shortwave) radiation Rs (Jeffrey et al. 2001) using the method of Allen et al. (1998, 45–53). The vapor pressure used in Allen et al. [1998, their Eq. (39)] was the 0900 local time vapor pressure—this was found to give the highest correlation with observed pan evaporation.

  2. a, an empirical parameter relating the radiation absorbed by a Class A pan to the radiation absorbed by a grass reference surface: this and other empirical parameters were determined through optimization (see appendix C).

Change in pan heat storage, S

The change in pan heat storage was not modeled, so S was treated as zero and changes in pan heat storage result in model errors. An attempt to model S using daily pan water maximum and minimum temperature records, which were available for some pans, did not decrease the model residuals. This was presumably because changes in pan heat storage were distributed between both evaporative and convective heat flux, and the distribution depends on the prevailing climate conditions.

As described below, 4-day pan averages were used in the calibration process to minimize systematic errors due to changes in pan heat storage.

Effective psychrometric constant of the pan, cγ

The effective psychrometric constant of the pan has two components:

  1. γ, the psychrometric constant, calculated using Allen et al. [1998, their Eq. (8)]:
    i1520-0442-20-14-3379-ea3
    where P is the station-level pressure.
  2. c is an empirical parameter, interpreted “as the ratio of the total effective area of the pan participating in sensible heat exchange to its water surface area” (Thom et al. 1981).

Vapor pressure deficit, δeC

The vapor pressure deficit term approximates the effects of subdaily changes in air temperature and vapor pressure as
i1520-0442-20-14-3379-ea4
where
  • e0(T) is the saturation vapor pressure at air temperature T (°C), given by (Allen et al. 1998):
    i1520-0442-20-14-3379-ea5
  • ea,1500 and ea,0900 are the vapor pressures at 1500 and 0900 local time (kPa),

  • r is an empirical parameter representing the diurnal variation in vapor pressure deficit, and

  • Tmax and Tmin are the daily maximum and minimum temperatures (°C).

Pan wind function, fυC

The function form of fυC is totally empirical. A suitable function was estimated by examining plots of pan evaporation against wind run for sets of days with similar solar radiation and vapor pressure deficit. The pan wind function used was
i1520-0442-20-14-3379-ea6
where
  • uC is the wind run expressed as a daily average wind speed (m s−1), and

  • b1 and b2 are empirical parameters.

This is different to the wind function used by Thom et al. (1981), which was a linear function of uC. However, the wind speeds in the Thom et al. experiment were only 0–2 m s−1, and over this range Eq. (A6) is approximately linear. By contrast, the Australian historical wind record has days with believable average wind speeds of over 10 m s−1, and for such wind speeds it appears that the linear approximation of Thom et al. is no longer valid.

APPENDIX B

Meteorological Data

Pan evaporation

This study used observed daily pan evaporation data. The data were sourced from BoM (BoM 2005a). Only nonaggregated data, thought to be from a Class A pan fitted with a bird guard, and with the quality code “quality controlled and acceptable” were used; pans without bird guards are known to have higher evaporation rates (van Dijk 1985). Interstation comparison showed that evaporation records greater than 40 mm day−1 represented errors, and these were discarded.

Individual station pan evaporation time series were inspected, and stations 38003, 70263, 74128, 75050, and 76064 were removed because they had anomalously large discontinuities. Of these, only Boulia Airport (38003) was included in the Roderick and Farquhar (2004) study.

Daily wind run

Only observed wind run data were used. Wind data were sourced from BoM (BoM 2005b). The “wind-run-below-3 m” field was taken to represent wind run at the evaporation pan. Inspection suggested that most records greater than 1200 km day−1 were highly anomalous, and these were discarded.

Daily maximum and minimum temperature

The daily temperature data were sourced from the Silo Patched Point Dataset (http://www.nrm.qld.gov.au/silo/ppd/) and Silo Data Drill (http://www.nrm.qld.gov.au/silo/datadrill/). The resulting time series contained observed station records where these were available, with missing data patched using data from interpolated surfaces (Jeffrey et al. 2001).

Vapor pressure at 0900 and 1500 local time

The 0900 local time vapor pressure records were sourced from the Silo Data Drill. The vapor pressure records in the Data Drill are derived from 0900 wet-bulb/dry-bulb thermometer readings (Jeffrey et al. 2001). These station-based vapor pressure records are then spatially interpolated.

Vapor pressure observations for 1500 local time were obtained from BoM (BoM 2005c) and were interpolated using the same algorithm that was used for the 0900 vapor pressure records.

All vapor pressure data used in the pan evaporation model were from the interpolated surfaces.

Daily solar radiation

Solar radiation data were sourced from the Silo Data Drill. The Silo solar radiation data are derived from observations of cloud oktas at 0900 and 1500 local time, using a model calibrated against pyronometer readings (Jeffrey et al. 2001). These station-based solar radiation estimates are then spatially interpolated. The solar radiation data used in the model were all from the interpolated surfaces.

It is important to be aware of the advantages and limitations of using Silo cloud-derived solar radiation data. The most important advantages were that cloud-derived solar radiation time series have relatively uniform quality and are easily available for all observing locations, and complete time series are available back to 1949. Ground-based solar radiation records are available for only a much sparser station set (Forgan 2005), and the records are not complete over the 1975–2004 study period.

Surprisingly, the accuracy of the cloud-derived solar radiation data was not a major limitation for this study. When compared against BoM high-quality post-1993 surface network data (BoM 2002), the error in the cloud-derived solar radiation was typically 10%–20%. More importantly, however, a comparison of modeled evaporation for these stations showed that surface network solar radiation data usually provided only a marginal improvement in the ability of the modeled evaporation to represent observed daily pan evaporation.

The main limitation of using okta-derived solar radiation is that it will not reflect changes in clear-sky optical properties, or in cloud properties other than cloud amount. If fact, because the solar radiation is derived from only two observations per day, only changes in the number of cloudy days are likely to have a significant effect on the long-term solar-radiation trends. Unfortunately, there are no datasets available to assess the reliability the cloud-derived solar radiation in representing long-term solar-radiation trends. The earlier records in the most commonly used historical ground-based solar radiation time series for Australia (BoM 2002) are affected by problems of instrumental variation, lack of calibration, and correction of data to modeled climatology (Forgan 2005). In an analysis of the eight best solar radiation records by Forgan (2005), the only statistically significant trend was related to an instrumental site change.

The cloud-derived solar radiation dataset has also not been homogenized. Because this dataset is derived from observations of cloud oktas, local site effects, and station relocations to nearby sites should not cause significant discontinuities. However, different observers may record cloud amounts differently, and this issue has not been investigated. There is also a known discontinuity in the solar radiation time series around 1948 (Rayner et al. 2004), when cloud observation recording changed from tenths to oktas (Henderson-Sellers 1992).

Station-level pressure

Station-level pressure was derived from interpolated MSL pressure (Pmslp) surfaces (Jeffrey et al. 2001) by modifying Allen et al.’s (1998) Eq. (7) to
i1520-0442-20-14-3379-ea7
where z is the station elevation (meters).

APPENDIX C

Model Calibration

In total, the Thom et al. (1981) pan evaporation model and the daily approximations used for its constituent terms have five empirical parameters; a, b1, b2, c, and r. These parameters were determined by minimizing the differences between cumulative observed and modeled pan evaporation.

Although Thom et al. calibrated Eq. (1), their calibration was derived from continuous measurements of climate variables, whereas this study used a combination of daily integrated, daily extreme, and fixed-time climate data. Thus, a recalibration was required to minimize biases arising from the approximations represented by Eqs. (A1)(A6). For example, in this study the net solar radiation Rn was derived from solar radiation Rs, which was itself derived from 0900 and 1500 observations of cloud oktas. In contrast, the Thom et al. calibration used a net radiometer. Systematic differences between the two estimates of net radiation are inevitable.

Differences between the BoM Class A pans and the pan used by Thom et al. also called for recalibration. First, the pan used by Thom et al. did not have a bird guard, whereas all the pans used in this study did. A bird guard is expected to reduce the contribution from the second term in Eq. (1). Second, Thom et al. described their pan as “brand new.” Clearly, pans used in a study of evaporation changes over a 30-yr period will not be brand new. Although the pans operated by BoM are cleaned regularly, they will generally have higher turbidity and sediment levels than a newly filled pan. The albedo of the walls of an old pan is also probably lower than that of a new pan.

The calibration run used stations with 10 or more years of overlapping pan evaporation and wind run records. This selection process yielded 1.5 million station days of observations, which was more than adequate to constrain five model parameters. A single set of parameters was derived for all observing stations.

The model was always calculated at a daily time step, but the calibration was performed by minimizing the difference between 4-day cumulative pan evaporation and model evaporation using a simplex search (e.g., Lagarias et al. 1988). Four-day accumulations were used to minimize any bias caused by S, the pan storage term. Interestingly, tests showed that single-day calibration produced essentially the same results.

To test the integrity of the calibration, the model was calibrated independently for each year from 1975 to 2004, and very similar parameters were derived each time. Similarly, calibration using only the wettest three months of each year for each station gave nearly identical results to calibration using only the driest three months for each station. An interesting future study would be to calibrate each station independently and investigate whether differences in parameters could be related to differences in diurnal meteorology or site conditions.

Note that Thom et al. included provision in their model for some fraction of daily rainfall to be lost as splash-out. When an unknown splash-out factor was included in the model calibration, the optimization process produced very small splash-out values, which did not reduce the final model residuals. The splash-out factor was dropped in the final model.

Parameters derived from optimization

The parameters derived from the optimization are given below.

  • a = 1.32. This is virtually identical to the Thom et al. value of 1.31.

  • b1 = 4.1, b2 = 0.32. These give a pan wind function, which for low wind speeds (<4 m s−1) is consistent with the linear wind function derived by Thom et al. (1981). For higher wind speeds, the Thom et al. wind function is significantly higher. This is not surprising, however, because Thom et al. did not encounter wind speeds higher than 4 m s−1. Also, the pans used in this study had bird guards, while the one used by Thom et al. did not.

  • c = 2.75. This is higher than the value of 2.1 used by Thom et al., although it is closer to the value of 2.4 derived by Kohler et al. (1955). Note that Thom et al. did not derive c empirically, but estimated it from the pan geometry. It is likely that the calibration process used here distributes some of the effect of the bird guard from the wind function fνC to c.

  • r = 0.46. This is similar to the commonly used 0.5 (Allen et al. 1998), although it differs substantially from the 0.75 suggested by Tanner and Sinclair (1983) and used by Jeffrey et al. (2001). It should be noted, however, that this later value was derived for crop evapotranspiration rather than pan evaporation. The diurnal temperature response of the water surface of a pan will be quite different from the diurnal temperature response of a crop surface, because the pan has substantial internal heat capacity. Also, again, the optimization algorithm may transfer significance between parameters.

APPENDIX D

Notes on Statistical Methodologies

Annual cumulative pan evaporation and wind run

Aggregating daily pan and wind run data to annual time series has to account for missing data. The procedure followed for each station was as follows:

  • For each month with at least 15 valid observations, take the median daily value.

  • Convert the median daily value to a monthly total by multiplying by number of days in the month.

  • For years with 12 months of data, just take the sum.

  • Consider years with 9 or less months of data as missing.

  • For each year with 10 or 11 months, estimate the monthly data for the missing months, and then take the annual sum. The estimate is the long-term average for those months for that station, scaled by the average fractional anomaly of the observed months. For example, if the evaporation for the 10 months with valid medians was 20% above average, assume that the two missing months were also 20% above average. This procedure produced very similar final results to just excluding all years with less than 12 valid monthly medians.

When comparing trends in modeled and observed data, the same aggregation was applied to both the observed and modeled data; modeled data were treated as missing if the observed data for that day were missing, and vice versa.

Estimation of trends and significance

Trends were estimated using the Theil–Sen robust slope estimator (Sen 1968). Trends were identified as significant or not at the 0.05 confidence level using the nonparametric Mann–Kendall test (Kendall 1948; Mann 1945). Correlations were identified as statistically significant or not at the 0.05 confidence level using the nonparametric Spearman rank correlation coefficient test (e.g., Weisstein 2006).

The observed pan evaporation trends were consistent with the trends reported by Roderick and Farquhar (2004), with slight differences occurring because of the different data periods (Roderick and Farquhar used 1975–2002; this study uses 1975–2004), because this study used the Theil–Sen robust slope estimator, and because of differences in handling missing data. When the trends were recalculated using the Roderick and Farquhar data period and a regression slope, the trends were very similar (r2 = 0.95), and all outlier stations were those with significant periods of missing data.

Fig. 1.
Fig. 1.

Geographic distribution of stations used in the pan evaporation trends attribution study. The 13 stations for which there were statistically significant trends in both the observed and modeled pan evaporation are labeled and denoted with a star symbol. The names of the observing stations have been abbreviated for convenience.

Citation: Journal of Climate 20, 14; 10.1175/JCLI4181.1

Fig. 2.
Fig. 2.

Pan evaporation trends (1975–2004) for attribution model runs. The meteorological climate variables used in each of the model runs are indicated in each panel; average climate data were used for the remaining variables. In each panel, a plotted point represents the modeled vs observed pan evaporation trend for one observing station. Statistically significant observed trends are overprinted with a plus (+) symbol; statistically significant modeled trends are overprinted with a cross (×). Points overprinted with both a plus and a cross, an “asterisk”, have statistically significant trends in both observed and modeled pan evaporation. Modeled pan evaporation trends were only consistent with the observed pan evaporation trends when the meteorological wind run uC was used (bottom right panel).

Citation: Journal of Climate 20, 14; 10.1175/JCLI4181.1

Fig. 3.
Fig. 3.

Average monthly wind run anomalies and annual pan evaporation anomalies for the 13 stations for which there were statistically significant trends in both observed and modeled pan evaporation. Anomalies were calculated using the base period 1975–2004.

Citation: Journal of Climate 20, 14; 10.1175/JCLI4181.1

Fig. 4.
Fig. 4.

Comparison of trends in observed and proxy wind run time series: (a) observed and NCEP–NCAR reanalysis trends; (b) observed and pressure gradient derived trends; (c) NCEP–NCAR reanalysis and pressure gradient derived trends. Statistically significant trends in observed wind run are overprinted with a plus symbol, statistically significant trends in the NCEP–NCAR reanalysis wind run are overprinted with a cross, and statistically significant trends in the pressure gradient derived wind run are overprinted with a triangle. Note the change in scale in (c).

Citation: Journal of Climate 20, 14; 10.1175/JCLI4181.1

Table 1.

Pan evaporation trends for the 17 stations that had statistically significant trends in observed pan evaporation. The increasing trend and decreasing trend columns show the number of stations with increasing and decreasing trends; the numbers in parentheses show how many of the trends were statistically significant. The “model using meteorological wind run” used meteorological data for all climate variables; the “model using average wind run” used meteorological data for all climate variables except wind run. The trend summary columns indicate the magnitudes of the pan evaporation trends for these stations.

Table 1.

1

Anomalies were computed independently for each station with respect to the monthly average.

2

Ranges indicate interquartile (25th to 75th percentile) ranges across all observing stations.

3

In this paper, “the correlation between the trends” indicates a spatial correlation (i.e., across observing stations).

4

Quoted errors represent the 0.05 confidence interval.

5

The increase in wind run u with height z is conventionally related by a roughness factor f as (u1/u2) = (z1/z2)f.

Save