Abstract
Estimating evaporation from standard meteorological data continues to be an active area of research and practical application. Here we report on recent progress in using standard meteorology data to estimate potential, reference and actual evaporation from terrestrial landscapes as well as evaporation from lakes and reservoirs. We also address recent enhancements to standard methodologies through use of remote sensing and data-driven procedures. From our report we observe that remote sensing offers significant potential for mapping spatial variations in evaporation. There has been limited progress in estimating actual evaporation via the complementary relationship, whereas applications of the Penman-Monteith and related equations incorporating actual surface resistance term(s) have dominated the recent literature.
Keywords
I Introduction
Evaporation is the process whereby liquid water moves from an evaporating surface to the atmosphere as water vapour. Evaporating surfaces include open bodies of water (e.g. lakes, reservoirs, ponds and streams), saturated soil surfaces and water intercepted on vegetation. Transpiration is the process of water vapour transfer from leaves of a plant to the atmosphere. Many authors refer to the combination of evaporation and transpiration as evapotranspiration (ET). In this report we adopt the term evaporation to cover all these processes but retain the term evapotranspiration where we refer to literature that specifically uses this term. The difference between potential and actual evaporation requires explanation. Potential evaporation (PE) ‘is the rate at which evapotranspiration would occur from a large area completely and uniformly covered with growing vegetation which has access to an unlimited supply of soil water, and without advection or heating effects’ (Dingman, 1992: 308). This is in contrast to actual evaporation which occurs from water bodies and terrestrial environments experiencing a wide range of conditions.
Here we report progress in estimating evaporation based on standard meteorological data from the peer-reviewed literature since 2007. Recent developments using remote sensing that enhance at-site standard meteorological measurements are also reviewed. We do not report on field measurement of evaporation by Bowen Ratio (BR) (Shuttleworth, 2007), eddy covariance (EC) methods (Shuttleworth, 2007) or remotely sensed energy balance methods (e.g. SEBS; see Kalma et al., 2008), nor do we deal with sublimation from snow.
This report is organized in six parts. The first section provides a background, where key historical developments in estimating evaporation using at-site meteorological data are briefly presented along with reviews since 2007. The two following sections cover recent developments in, and applications of, models for estimating terrestrial evaporation (potential, reference crop and actual evaporation) and lake and reservoir evaporation. Progress in enhancing evaporation estimates through (1) remote sensing and (2) data-driven techniques are covered in separate sections, and the key points from this progress report are highlighted in a concluding section.
II Background
1 Past developments in estimating evaporation
To appreciate the significance of recent developments in estimating evaporation, an understanding of historical developments in the science is required. Two key papers were published in 1948: Thornthwaite (1948) introduced the term ‘potential evapotranspiration’ (PET) and Penman (1948) was first to combine an aerodynamic approach, based on Dalton (1802), with an energy equation to estimate evaporation from a wet surface. In 1965, Monteith (1965) (P-M) modified the Penman (1948) equation by introducing a surface resistance variable to estimate ET from vegetation. This model is the basis of the UN Food and Agriculture Organization Reference Crop Evapotranspiration model (FAO-56) (Allen et al., 1998) for estimating crop water requirements. In 1972, Priestley and Taylor (1972) (P-T) replaced the aerodynamic term of Penman with a coefficient known as the P-T constant. In 1963, Bouchet (1963) proposed that PET and actual ET depend on each other in a complementary relationship (CR) through land and atmosphere feedbacks in regions where advection of heat and moisture are minimal. Morton developed three models, based on the CR, to estimate actual evaporation from land and lakes (Morton, 1983a, 1983b, 1986). The Advection-Aridity (A-A) model was developed by Brutsaert and Stricker (1979) by combining the Penman and P-T equations through the CR to estimate catchment evaporation. Combining energy and water availability, Budyko (1974) developed an equation to estimate mean annual actual catchment ET.
2 Recent reviews (2007 to present)
Shuttleworth (2007: 210) provided a historical overview of developments in ‘the science of natural evaporation from land surfaces’ over the previous 35 years. He noted the ‘big leaf’ representation of whole-canopy exchanges is favoured over a multi-layer model (see Raupach and Finnigan, 1988, for further discussion) and recommended using stomatal resistance rather than crop factors (Allen et al., 1998) for estimating crop water requirements. Farahani et al. (2007) discussed progress in adopting the one-step P-M model (Shuttleworth, 2006) using crop stomatal resistance to estimate crop water use. Verstraeten et al. (2008) discussed ET assessments across a range of scales using remote sensing. Wang and Dickinson (2012) presented a wide-ranging review of evaporation theories, observational methods, satellite algorithms and land surface models for estimating long-term trend and variability in ET. In a report to the World Meteorological Organization Commission for Hydrology, Finch and Calver (2008) assessed the advantages and limitations of methods to estimate evaporation from water bodies. They saw the equilibrium temperature method as an attractive means of determining evaporation. Blyth and Harding (2011) examined methods to identify the proportions of total evaporation due to interception, soil and transpiration, and suggested a procedure, based on Kim et al. (1996), to model these components as a time series. The most recent review is ‘a pragmatic synthesis’ by McMahon et al. (2013) who assessed techniques to estimate actual, potential, reference crop and pan evaporation using standard meteorological data.
III Estimating terrestrial evaporation
1 Potential and reference crop evaporation
a New developments or modifications
The one-step P-M model, known as the Matt-Shuttleworth (M-S) model (Shuttleworth, 2006; Shuttleworth and Wallace, 2009), uses crop aerodynamic and canopy resistances to estimate evaporation without a crop factor as an alternative to the two-step crop factor FAO-56 model (Allen et al., 1998). Shuttleworth and Wallace (2009) demonstrated techniques for estimating crop aerodynamic and canopy resistances at five Australian locations. They compared M-S estimates against FAO-56 and noted similar results under humid conditions, but more realistic higher evaporation from the M-S model under arid windy conditions. According to Shuttleworth and Wallace (2009), because agricultural crops are generally taller than the reference crop, their aerodynamic resistance is lower, especially under windy conditions, which results in higher ET rates. Lovelli et al. (2008) described how the canopy resistance term of the one-step model can be estimated using a simplified leaf area index (LAI) model.
Lascano and van Bavel (2007) and Lascano et al. (2010) demonstrated the recursive combination method of Budyko (1956), which estimates the evaporating surface temperature via iteration. A simplifying assumption behind Penman-based methods allows air temperature rather than the temperature of the evaporating surface to be used. However, this assumption breaks down as the difference between surface and air temperatures increases (McArthur, 1992). Lascano and van Bavel (2007) compared the recursive Budyko and P-M ET models and observed that on hot summer days P-M under-estimated crop ET relative to Budyko, which accurately reproduced weighting lysimeter values of evapotranspiration for well-watered alfalfa (Lascano et al., 2010).
Other recent developments for reference crop ET procedures include: (1) a wind adjustment factor for the Turc equation (Trajković and Stojnić, 2007); (2) two alternative temperature expressions for the Blaney-Criddle method in high-elevation locations (Juday et al., 2011); and (3) modifications to FAO-56 for a windbreak effect (Campi et al., 2012).
b Recent applications and evaluations
We identified 36 recent applications (42 different models) of PET and reference crop ET. Of the 36, 92% included an application of P-M-like models (P-M, FAO-56 and ASCE reference crop (ASCE, 2005)). The key observations from this literature are: (1) only four of the 36 studies independently evaluated their evaporation estimates against observed BR (Bowen, 1926), energy balance (EB), EC or lysimeter data (see Benli et al., 2010; Douglas et al., 2009), and the other 32 studies were inter-comparison studies of different ET methodologies; (2) almost universally, FAO-56 and/or ASCE reference crop are regarded as providing the most accurate estimates of reference crop ET (see Aguilar and Polo, 2011: 2495); and (3) the Hargreaves-Samani model (Hargreaves and Samani, 1985) is the reference crop technique of choice when only temperature data are available.
2 Actual evaporation
a New developments or modifications
Several authors have modified the A-A model of Brutsaert and Stricker (1979) to estimate actual evaporation. Szilagyi and Jozsa (2008) developed a new method to estimate the equilibrium surface temperature via BR. Developments by Crago et al. (2010) included: (1) replacing the Penman wind function with a Monin-Obukhov similarity theory approach; and (2) estimating humidity via minimum temperature in order to eliminate the need for humidity observations. Drawing upon the A-A and Granger-Gray (Granger and Gray, 1989) models, Han et al. (2011) developed a new model for estimating actual ET that maintained the key features of both models and performed well in initial testing.
Other developments include Lhomme et al. (2012, 2013), who investigated existing models for estimating evaporation from sparse and heterogeneous canopies, including that of Shuttleworth and Wallace (1985), and proposed a new two-component model, which they generalized into a multi-component model. Ding et al. (2013) modified the P-T coefficient to include leaf area, soil moisture, mulching fraction and leaf senescence on ET.
b Recent applications and evaluations
There are at least four approaches to estimate actual terrestrial evaporation using standard meteorological data. These are based on: (1) the complementary relationship; (2) adjusting the P-M canopy resistance to represent the actual plant moisture stress; (3) modifying PE by a variable or variables that represents the reduction from the potential to actual rate; and (4) inputting PET into a rainfall-runoff model in which actual evaporation is a function of soil moisture and PET. During our review period we identified 22 models for estimating actual terrestrial ET. The ET estimates were validated against EC in 12 studies, BR in four and the water balance in three. P-M type models were assessed in more than half the papers, with a key scientific issue being estimation of surface resistance values to allow P-M to be used to estimate actual ET. Armstrong et al. (2008) observed in a mixed grassland setting that resistance formulations need to be improved for more reliable estimation of evaporation. They noted that the minimum reference resistance may not always be applicable due to changing plant state (e.g. during and after flowering). Agam et al. (2010) reassessed the P-T coefficient and suggested that P-T can be used for heat flux estimates of canopy transpiration over a range of vegetation and moisture conditions that are not saturated.
IV Estimating lake and reservoir evaporation
In this section we report on techniques that use standard meteorological data to estimate evaporation from open water and lakes partially covered by vegetation or artificial material, but not river evaporation.
1 Recent reviews
Jensen (2010) provided one of the few reviews since 2007 of lake and reservoir evaporation. In it he concentrates on shallow water bodies and notes the importance of taking into account the energy stored in the water body and net advected energy from inflows and outflows. Jensen (2010) briefly outlined the equilibrium temperature approach applied by Finch (2001) and the finite difference approach of Finch and Gash (2002). In addition, Jensen (2010) considered applications of pan evaporation coefficients and the application of the FAO-56 equation modified by a coefficient (Allen et al., 1998). McMahon et al. (2013) provided a synthesis of shallow and deep lake evaporation and detailed techniques with worked examples.
2 Recent applications and evaluations
During our review period we identified 17 applications where standard meteorological data were used to estimate evaporation from open water lakes, reservoirs, wetlands or covered water. Additional variables used in some applications were storage depth and temperature profiles. Most studies validated evaporation models against BR, EC or scintillometer (McJannet et al., 2013) observations. Lake heat storage effects are regarded as important and, therefore, advection of energy from rainfall, river and groundwater inflows need to be accounted for (Elsawwaf et al., 2010b). Three approaches for dealing with heat storage effects were: (1) lake water temperature profiles (Elsawwaf et al., 2010a); (2) average lake temperatures (Gallego-Elvira et al., 2010); and (3) equilibrium temperature (McJannet et al., 2008). To estimate open water evaporation, McJannet et al. (2012) reviewed the literature and developed a wind function related to water body area for use in a mass transfer model. Masoner et al. (2008) investigated differences in pan evaporation between floating and land-based pans, while Masoner and Stannard (2010) demonstrated the utility of a floating pan to measure open water evaporation from small limited fetch water bodies.
V Enhancing evaporation estimates with remote sensing applications
1 Recent reviews
Kalma et al. (2008) provided an extensive review of methods to estimate land surface evaporation using remote sensing of surface temperature. Verstraeten et al. (2008) reviewed the theory of plant water uptake, remote sensing and data assimilation to estimate ET, and soil moisture retrieval techniques. Their Table 3 lists 13 evaporation models, five of which are P-M-based and use remotely sensed data. As a general rule, to apply these models at regional scales, meteorological data are required along with a range of land surface temperatures, albedo values and vegetation index measures.
Glenn et al. (2010, 2011b) reviewed the use of satellite derived vegetation indices (VI) in combination with ground measurements to estimate actual ET. In a synthesis of 17 studies in which ET was estimated for natural ecosystems, Glenn et al. (2010) noted that all but one used MODIS as the satellite sensor system and that the Enhanced Vegetation Index (EVI) was preferred to the Normalized Difference Vegetation Index (NDVI) because EVI was found to be better correlated with measured ET than was NDVI. Glenn et al. (2011a) reviewed the Australian experience in estimating actual ET over large regions using ground-based and remote sensing techniques. Wang and Dickinson (2012) observed that most satellite algorithms relate land-based variables to ET using either P-M or Monin-Obuklov similarity theory (Monin and Obukhov, 1954).
2 Recent applications and evaluations
In order to limit our review, we restricted our scope to practical applications of remote sensing that enhance the application of a standard procedure (e.g. Penman, P-M) applied to a large area. We identified 28 papers published since 2007 in this area. Eighty-six percent dealt with estimating actual ET. Penman-Monteith or FAO-56 were the most frequently applied models and they were used mainly to estimate actual ET. To do this, the surface resistance (rs) in P-M must be estimated for actual, rather than well-watered, conditions. Vegetation indices such as NDVI, Fraction of Absorbed Photosynthetically Active Radiation and Normalized Difference Water Index (Yi and Yang, 2011) have been found to be useful for this purpose. Using field data, rs can be estimated by inverting the P-M equation. Three studies used remote sensing data in combination with ground-based observations to estimate actual ET globally via P-M (Mu et al., 2007, 2011) or P-T (Fisher et al., 2008).
VI Enhancing evaporation estimates with data-driven models
The main reasons for using data-driven approaches, such as artificial neural networks (ANN), fuzzy genetic (FG) and genetic programming (GP) are: (1) to interpolate evaporation estimates across regions where data are unavailable to directly estimate evaporation (Adeloye et al., 2011); (2) to provide a statistical benchmark against which to assess process models (Whitley et al., 2009); and (3) to explore the underlying influence of evaporation variables (Karimaldini et al., 2012). In order to deal with the large number of papers under this heading we restricted our review to the 37 papers that use data-driven models to estimate PET, reference crop ET, actual ET or pan evaporation.
Kumar et al. (2011) provided a review of ANN theory, methodologies and applications. However, we were unable to find a comprehensive review of recent data-driven literature. El-Baroudy et al. (2010) provided a succinct description of several data-driven techniques including ANN, GP and evolutionary polynomial regression. Martí et al. (2011) used a ‘leave one out’ cross-validation procedure to assess the performance of an ANN across a wide range of training, cross-validation and testing data sets and observed that traditional ANN testing techniques can be misleading in terms of model performance and optimal configuration. We note many authors report that their data-driven approaches perform better than standard methods. These reports failed to heed the comments of writers like Whitley et al. (2009: 260) who pointed out that ‘ANN is a purely statistical-based response to the meteorological forcing’ and it ‘will always outperform mechanistic, conceptual models because the ANN effectively has up to 100 optimized parameters whereas most conventional models have less than 10. A direct comparison is therefore inappropriate’. Kumar et al. (2011: 19) also cautioned that ANN ‘developed for one location cannot be implemented to (an)other location without local training’.
In 37 published papers examined for this review, at least 21 different data-driven models were assessed. Of these, 73% incorporated some form of ANN, of which 56% were the traditional ANN or ANN with the fast converging Levenberg-Marquardt algorithm (Martí et al., 2011). The next largest group, accounting for approximately 22% of the ANN analyses, was the adaptive neuro-fuzzy inference system (ANFIS) model and its variations (grid partitioning (ANFIS-GP) and subtractive clustering (ANFIS-SC)). It was further noted that most model analyses considered potential, reference crop or Class-A pan evaporation; few dealt with actual ET. From the recent literature we make the following observations: (1) data-driven models were stated to be superior to climate-based or empirically based models which is an expected outcome from an inappropriate comparison (Whitley et al., 2009); (2) data-driven models estimating PET or reference crop ET were often found to be as accurate as a four-input variable climate model, e.g. Penman, even when fewer variables were used to train the data-driven model; and (3) ANFIS-SC, evolutionary neural networks, FG, linear genetic programming, neural network incorporating an autoregressive external input, support vector machine-neural network, and wavelet and neural network were generally more satisfactory than ANN, ANFIS, ANFIS-GP, gene expression programming, GP, generalized regression neural network and radial basis neural network.
VII Conclusions
From our review we note that remote sensing offers significant potential to enhance estimation of evaporation spatially. There has been limited progress in estimating actual evaporation via the complementary relationship, whereas applications of P-M-like equations incorporating actual surface resistance term(s) estimated from field and/or remote sensing have dominated the recent literature. Research into data-driven methods has primarily focused on identifying and testing new models. There has been limited practical application of these models.
Footnotes
Acknowledgements
Comments provided by an anonymous reviewer and the editor improved this manuscript.
Funding
This research was partially funded by the Australian Research Council (grant number FT120100130, LP100100756). ARC LP100100756 was co-funded by Melbourne Water and the Australian Bureau of Meteorology.