Assessing reference evapotranspiration at regional scale based on remote sensing, weather forecast and GIS tools

International Journal of Applied Earth Observation and Geoinformation 55 (2017) 32–42

Contents lists available at ScienceDirect

International Journal of Applied Earth Observation and Geoinformation journal homepage: www.elsevier.com/locate/jag

Assessing reference evapotranspiration at regional scale based on remote sensing, weather forecast and GIS tools J.M. Ramírez-Cuesta, M. Cruz-Blanco, C. Santos, I.J. Lorite ∗ IFAPA-Centro “Alameda del Obispo”, Regional Government of Andalusia, Avda. Menéndez Pidal s/n, P.O. Box 3092, 14080 Córdoba, Spain

a r t i c l e

i n f o

Article history: Received 2 July 2016 Received in revised form 24 September 2016 Accepted 7 October 2016 Keywords: Reference evapotranspiration Remote sensing Weather forecast GIS Interpolation Weather stations

a b s t r a c t Reference evapotranspiration (ETo ) is a key component in efficient water management, especially in arid and semi-arid environments. However, accurate ETo assessment at the regional scale is complicated by the limited number of weather stations and the strict requirements in terms of their location and surrounding physical conditions for the collection of valid weather data. In an attempt to overcome this limitation, new approaches based on the use of remote sensing techniques and weather forecast tools have been proposed. Use of the Land Surface Analysis Satellite Application Facility (LSA SAF) tool and Geographic Information Systems (GIS) have allowed the design and development of innovative approaches for ETo assessment, which are especially useful for areas lacking available weather data from weather stations. Thus, by identifying the best-performing interpolation approaches (such as the Thin Plate Splines, TPS) and by developing new approaches (such as the use of data from the most similar weather station, TS, or spatially distributed correction factors, CITS), errors as low as 1.1% were achieved for ETo assessment. Spatial and temporal analyses reveal that the generated errors were smaller during spring and summer as well as in homogenous topographic areas. The proposed approaches not only enabled accurate calculations of seasonal and daily ETo values, but also contributed to the development of a useful methodology for evaluating the optimum number of weather stations to be integrated into a weather station network and the appropriateness of their locations. In addition to ETo , other variables included in weather forecast datasets (such as temperature or rainfall) could be evaluated using the same innovative methodology proposed in this study. © 2016 Elsevier B.V. All rights reserved.

1. Introduction Water shortages caused by climate change and by increased competition for water between sectors have led to a growing interest in improving agricultural water management, especially in arid and semiarid regions. Thus, a number of actions such as the promotion of irrigation advisory services for the accurate assessment of irrigation water requirements (Lorite et al., 2012) or the development of deficit irrigation strategies to reduce the volume of water applied without affecting production (Fereres and Soriano, 2007) have gained prominence in Mediterranean agricultural systems. Estimating crop water demands requires an accurate assessment of the actual evapotranspiration (ET), defined as the sum of soil surface evaporation and crop transpiration, and the reference evapotranspiration (ETo ), defined as the ET from a hypothetical

∗ Corresponding author. E-mail address: [email protected] (I.J. Lorite). http://dx.doi.org/10.1016/j.jag.2016.10.004 0303-2434/© 2016 Elsevier B.V. All rights reserved.

grass reference crop without water stress (Allen et al., 1998). However, lysimetry under reference conditions is the only way to accurately measure these variables (Vaughan et al., 2007). As an alternative, methods based on field measurements of soil water balances (Chávez et al., 2009; De Bruin et al., 2010), energy balances (Todd et al., 2000; Villa-Nova et al., 2007) or micrometeorology techniques (Er-Raki et al., 2009; Chávez et al., 2009) have been developed. Focusing on reference evapotranspiration (ETo ), simpler estimation methods use numerical approaches based on measured weather data; many such equations have been developed (e.g. Penman-Monteith, Makkink or Hargreaves) and some of which require prior local/regional calibration (Gavilán et al., 2006; Cruz-Blanco et al., 2014a). Due to the wide range of available approaches, and in order to standardize ETo estimation in different climates, Allen et al. (1998) established the Penman-Monteith equation as the reference method for ETo assessment. This method has been used with very satisfactory results around the world (Berengena and Gavilán, 2005; Temesgen et al., 2005; Allen et al., 2006). However, weather data measurement must comply with

J.M. Ramírez-Cuesta et al. / International Journal of Applied Earth Observation and Geoinformation 55 (2017) 32–42

strict rules concerning the location and the physical conditions of the field where the weather station is located, which is not an easy task in semi-arid and arid conditions (Temesgen et al., 1999). Unfortunately, these requirements are not always fulfilled in modern weather station networks, thus generating uncertainties and overestimations in ETo assessment (Cruz-Blanco et al., 2014a). In addition, weather stations provide local measures, and in many cases, these data are representative of only a very limited area, even at times providing non-valid data for fields located near the weather station. Lastly, often there is a limited number of weather stations and vast areas are located far from well-managed weather stations (Voogt, 2006; Collins, 2011). In order to overcome such limitations, the interpolation of data from weather stations emerges as a promising approach for assessing ETo (Alves et al., 2013). There are numerous interpolation methods and many have already been applied to the assessment of weather variables with very satisfactory results from around the world (Martínez-Cob, 1996; Nalder and Wein, 1998; Hart et al., 2009; Li and Heap, 2011). The use of interpolation methods has increased as result of the technological advances made in recent decades, mainly related with the development of Geographic Information Systems (GIS) (Irmak et al., 2010). Numerous studies have been conducted to identify the most accurate interpolation methods for climatic variables estimation (Mardikis et al., 2005; Li and Heap, 2008, 2011; Di Piazza et al., 2011; Keblouti et al., 2012). These studies revealed limitations in the validation process due to the low number of checkpoints and highlighted the difficulty in boosting their numbers. Thus, the use of remote sensing techniques to evaluate the quality of the interpolation methods emerges as a promising alternative, although the number of studies taking such an approach is still very limited (Hart et al., 2009; Wentz et al., 2010). Remote sensing techniques have contributed significantly to the improvement of water management by providing an accurate estimation of crop evapotranspiration (Bastiaanssen et al., 1998; Allen et al., 2007a,b; Santos et al., 2008, 2010), and reference evapotranspiration (De Bruin et al., 2010; Cruz-Blanco et al., 2014a). This study used the geostationary satellite Meteosat Second Generation (MSG), integrated in the Land Surface Analysis Satellite Application Facility (LSA SAF) tool, which provided accurate ETo values for vast areas (De Bruin et al., 2010, 2012; Cruz-Blanco et al., 2014a). Thus, the LSA SAF estimated spatially distributed ETo values throughout the Andalusian region (comprising more than 6000 cells), representing the first time that this tool has been used to validate an interpolation approach. However, the use of LSA SAF data is not limited to validation purposes; it has also helped provide excellent results in studies related to uncertainty in ETo estimation (Cruz-Blanco et al., 2015) or to irrigation scheduling at the regional scale (Cruz-Blanco et al., 2014b). There is thus a clear benefit to combining weather data from weather station networks with data from LSA SAF and GIS; they are innovative tools that provide support to technicians and farmers attempting to improve water management at field and regional scale (Lorite et al., 2012). In this study, those approaches were developed and evaluated under semi-arid conditions in the Andalusian region (southern Spain).

2. Materials and methods 2.1. Study area description The study area was the region of Andalusia in southern Spain (Fig. 1), characterized by a Mediterranean climate, with an annual precipitation ranging from 506 to 725 mm, and an average air tem-

33

perature ranging from 10 ◦ C in winter to 27 ◦ C in summer. The period under study was from 2007 to 2009. 2.2. The Agroclimatic Information Network of Andalusia (RIA) The Agroclimatic Information Network of Andalusia (RIA) was established with the main objective of improving irrigation water management in southern Spain by providing accurate daily weather data. The network consists of 100 automated weather stations that cover most of the irrigated regions of Andalusia (Gavilán et al., 2006) although in light of the quality analysis performed by Cruz-Blanco et al. (2014a), only 57 of those weather stations were used in this study (Fig. 1). A quality control procedure was applied to the RIA weather data, involving range, step and persistence tests, as well as internal and spatial consistency tests (Meek and Hatfield, 1994; Shafer et al., 2000). Daily ETo data were estimated using the Penman-Monteith equation (Allen et al., 1998) for each weather station for 2007, 2008 and 2009. All the weather stations included in the study are located on bare soil (except one located near Cordoba which is installed on near-reference conditions). Under these non-reference conditions, ETo values calculated by PM-FAO56 could be affected by overestimations, especially in arid and semi-arid conditions (Allen, 1996; Temesgen et al., 1999; Allen et al., 2002; Cruz-Blanco et al., 2014a), which significantly increases the uncertainty in ETo assessment when using data from weather station networks (Cruz-Blanco et al., 2015). Weather stations in regions neighbouring Andalusia are managed differently as they form part of weather station networks owned by different institutions, and even from different countries (to the East, Portugal). This implies differences in instrumentation, quality control tests and methodologies for ETo assessment. 2.3. ETo calculated from EUMETSAT LSA SAF The European Organization for the Exploitation of Meteorological Satellites (EUMETSAT) created the Land Surface Analysis Satellite Application Facility (LSA SAF) to allow a more effective use of the collected data from the Meteosat Second Generation (MSG) satellite. The LSA SAF tool estimates ETo from remotelysensed daily solar radiation data provided by MSG and from the forecasts for air temperature at 2 m provided by the European Centre for Medium-range Weather Forecasts (ECMWF) (Persson, 2013). Thus, the LSA SAF estimated spatially distributed daily ETo values for the region of Andalusia with pixel sizes ranging from 3.2 km by 4.2 km (13.44 km2 ) to 3.2 km by 4.5 km (14.4 km2 ) for southern and northern Andalusia, respectively, for 2007, 2008 and 2009. In order to adjust these ETo values to the regional conditions of southern Spain, a calibration process was used involving a weighing lysimeter under near-reference weather conditions. This process considered the Makkink equation developed by De Bruin et al. (2012) and is described in Cruz-Blanco et al. (2014a). Solar radiation (Rs ) and near-surface air temperature (T2m ) estimated by MSG and ECMWF, respectively, were validated under the semi-arid conditions of southern Spain. The comparison of measured data by the Cordoba station under near-reference conditions with estimations provided by MSG and ECMWF gave root mean square error (RMSE) values of 1.56 MJ m−2 d−1 and 1.36 ◦ C, R2 values of 0.97 and slopes of the linear regressions equal to 0.98 and 0.99, for Rs and T2m , respectively (Cruz-Blanco et al., 2014a). Validation results were similar for the 57 weather stations located throughout the Andalusian region (see Section 2.2), with RMSE values of 1.47 MJ m−2 d−1 and 1.53 ◦ C for Rs and T2m , respectively (Cruz-Blanco et al., 2015). Following these satisfactory validation results for Rs and T2m , validation of reference evapotranspiration (ETo ) estimated by the LSA SAF shows a similarly accurate per-

34

J.M. Ramírez-Cuesta et al. / International Journal of Applied Earth Observation and Geoinformation 55 (2017) 32–42

Fig. 1. Study area and location of the weather stations included in this study. The Guadalquivir River is shown in blue. For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.

Table 1 Interpolation methods used in the study. Abbreviation

Method

IDW TPS SK OK UK SCKa OCKa

Inverse Distance Weighting Thin Plate Splines Simple Kriging Ordinary Kriging Universal Kriging Simple Cokriging Ordinary Cokriging

a With these methods, the Digital Elevation Model (DEM), the Slope (SLP) and Aspect (ASP), and all of those taken together, were additionally used as covariates.

formance: considering weighing lysimeter data, RMSE, slope and R2 of linear regression were 0.50 mm d−1 , 0.97 and 0.96, respectively (Cruz-Blanco et al., 2014a), while considering data from 57 RIA weather stations, RMSE, slope and R2 of linear regression of 0.69 mm d−1 , 0.94 and 0.92, respectively (Cruz-Blanco et al., 2015). As ETo values were regionally calibrated and calculated with the same methodology and equipment for the whole analysed region, the results were consistent and reliable. In addition, due to the excellent validations results, and in spite of the slight underestimations in ETo assessment caused by the omission of the aerodynamic components (Cruz-Blanco et al., 2014a, 2015), ETo values assessed by the LSA SAF were taken as a reference.

2.4. Interpolation approaches In order to estimate spatially distributed ETo values for the Andalusian region, some commonly-used interpolation methods (Table 1; Li and Heap, 2011) were employed. These interpolation methods used local data collected in each weather station of the RIA and provided ETo maps with a spatial resolution of 1 km. These maps were rescaled to the spatial resolution of the LSA SAF dataset in order to be able to carry out the ETo validation procedure for each interpolation approach.

The Inverse Distance Weighting (IDW) approach interpolates using data from surrounding weather stations, weighting by the distance between each weather station and the point under analysis. The Thin Plate Splines (TPS) approach generates a minimally bended smooth surface without requiring prior estimation of the spatial auto-covariance structure. Kriging approaches are based on a semivariogram which defines variation as a function of distance. In Simple Kriging (SK) the trend component is a known constant mean. Ordinary Kriging (OK) uses local means instead of a single constant mean as SK does. Finally, Universal Kriging (UK) is an extension of OK and assumes that there is a local trend that varies from one location to another. Cokriging approaches use secondary information as covariates in order to minimize the variance of the estimation error by exploiting the cross-correlation between several variables. Simple Cokriging (SCK) uses stationary means for the estimation of both primary variables (ETo ) and secondary variables (covariates such as elevation or slope), while Ordinary Cokriging (OC) uses local means. The secondary information used as covariates was the digital elevation model (SCDEM and OCDEM), the slope model (SCSLP and OCSLP), the aspect model (SCASP and OCASP), and all of these taken together (SCDSA and OCDSA). A full description of the interpolation approaches can be found in Li and Heap (2008, 2011). 2.5. Alternative approaches for ETo assessment 2.5.1. Spatial analysis to select the closest (NN) and the most similar (TS) weather station The closest weather station to each location (NN) was selected by means of the Thiessen polygon method (Thiessen, 1911). The most similar weather station (TS) was selected according to temporal signatures of reference evapotranspiration (TS-ETo ). TS-ETo describes the daily temporal evolution of ETo during the 2007–2009 period estimated by the LSA SAF tool for the 6768 cells that make up the Andalusian region, including the cells where RIA weather stations are located. The TS-ETo value for each cell was compared with the TS-ETo from the cells where each weather station is located,

J.M. Ramírez-Cuesta et al. / International Journal of Applied Earth Observation and Geoinformation 55 (2017) 32–42

35

Fig. 2. Diagram of the procedures used to estimate the differences in ETo assessment caused by the use of different methods (DA), the total difference maps (TD) and the interpolation error maps (IE) based on MSG, ECMWF and RIA data. In addition, procedures to determine the closest weather station (NN), the most similar weather station (TS), and the correction indexes with NN (CINN) and TS (CITS) are shown. Circles indicate that interpolation approaches were employed.

thus identifying the weather station that recorded the most similar TS-ETo for each cell. To carry out this comparison, RMSE was used as an indicator of similarity. 2.5.2. Correction factor assessment based on the LSA SAF dataset (CINN and CITS) A correction factor for each cell was calculated as the ratio between the annual/seasonal ETo provided by the LSA SAF in the cell under analysis and in the cell where the reference weather station is located. There were two options for the reference weather station: the closest weather station to the cell under analysis (CINN) and the weather station recording the most similar ETo trend (CITS). Using this approach, ETo was derived for each cell by taking the ETo value estimated in the closest/most similar weather station and multiplying it by the described correction factor (CINN/CITS). 2.6. Performance assessment The reference evapotranspiration map generated by each interpolation approach (Table 1) with data from 57 RIA weather stations (ETo RIAi ) was rescaled to the same spatial resolution as that of the LSA SAF and was compared, cell by cell, with the ETo map provided by the LSA SAF (ETo LSASAF). The total difference (TD) for each cell was calculated as: TD = (

ET o RIAi − ET o LSASAF ) × 100 ET o LSASAF

(1)

TD incorporates two components: the differences in ETo caused by the use of different methodology for ETo assessment (PenmanMonteith for RIA vs. Makkink for LSA SAF) (DA) and the errors generated by the interpolation of RIA values (IE). DA was derived from the interpolation of the differences in seasonal/annual ETo obtained by RIA and the LSA SAF for the 57 weather stations included in the study, using the same interpolation approach as that employed to generate the ETo -RIAi map. After calculating TD and DA terms, the interpolation error (IE) was calculated as: IE = TD − DA

(2)

Fig. 3. Interpolation errors for each interpolation method and year (n = 6473). Different letters indicate significant differences at p < 0.05. Whiskers indicate the standard error of the mean.

Then, for a cell containing a weather station the interpolation error will be null and all the detected TD will be caused by DA. A diagram of the procedure followed for calculating TD, DA and IE is shown in Fig. 2. For the alternative approaches described in Section 2.5, errors in annual/seasonal ETo assessment were calculated taking the LSA SAF dataset as a reference. Thus, seasonal/annual ETo values provided by the LSA SAF for the locations of the nearest weather station (NN) and the most similar weather station (TS), and those calculated by means of the three-year average correction index using NN (CINN) and TS (CITS) were compared with the LSA SAF ETo value for each cell (Fig. 2). The average difference in ETo in absolute values for all the cells with the same NN or TS was taken as the average error. Similarly, taking the ETo values collected by the RIA as a reference, the results provided by the proposed approaches were compared with the ETo values collected in each weather station adjusted by the annual CITS maps, in order to correct the previously described differences in ETo assessment between approaches (Fig. 2). Finally, a daily ETo assessment was carried out for four representative days per year based on the LSA SAF and RIA values. An ANOVA test was used to calculate the significance level of the differences between approaches for ETo assessment (interpolation methods, NN, TS, CINN and CITS). 3. Results and discussion 3.1. Evaluation of the different interpolation methods for seasonal ETo assessment The comparison of the LSA SAF dataset with the interpolated maps based on RIA values detected significant differences of around 9.7%. These differences were due to the different approaches employed for ETo assessment (hereafter, DA) and to the interpolation error (IE). An average DA value of 8.7% was recorded, with the

36

J.M. Ramírez-Cuesta et al. / International Journal of Applied Earth Observation and Geoinformation 55 (2017) 32–42 Table 2 Average and coefficient of variation (CV) of the difference between approaches (DA) and interpolation error (IE) by season and year (n = 6473) using Thin Plate Splines (TPS) method. Different letters indicate significant differences at p < 0.05 between year/seasons. Year

Season

2007

Autumn Spring Summer Winter Year Autumn Spring Summer Winter Year Autumn Spring Summer Winter Year

2008

2009

DA (%)

IE (%)

Avg.

CV

Avg.

CV

17.31b 8.14d 5.26e 21.98a 10.14c 17.21a 5.32d 8.34b 17.11a 7.46c 14.82b 9.79c 5.84e 21.30a 8.52d

0.63 0.71 0.71 0.60 0.63 0.58 0.73 0.72 0.61 0.71 0.54 0.59 0.71 0.45 0.61

5.04a 3.29c 2.85d 4.07b 2.56e 3.28a 1.76c 1.41e 2.73b 1.54d 6.19a 1.86e 3.75b 2.52d 3.26c

1.45 1.38 1.05 1.16 1.21 1.33 1.04 1.03 1.10 1.03 1.23 0.94 1.15 1.12 1.09

3.2. Identification of the nearest (NN) and the most similar (TS) weather station for ETo assessment

Fig. 4. Interpolation error using the Thin Plate Splines (TPS) approach for 2007, 2008 and 2009. Dots indicate the location of each weather station included in the study.

greatest differences detected during winter and autumn (average values of 20.1 and 16.4%, respectively) and the smallest differences during spring and summer (7.8 and 6.5%, respectively). Average IE for the three-year period depended on the interpolation approach used and ranged from 2.5% for Thin Plate Splines (TPS) to 5.0% for Simple Kriging (SK) (Fig. 3). The IE generated with SK was significantly reduced when slope and elevation were incorporated as covariates (SCSLP and SCDEM approaches), improving the average accuracy by around 23 and 33%, respectively (Fig. 3). Analysing the IE performance by season, maximum errors were found for autumn (ranging from 3.3 to 6.2%) and minimum for spring (ranging from 1.8 to 3.3%; Table 2). Similar trends were recorded for the rest of the interpolation approaches. Spatial analysis of the results provided by TPS revealed that the lowest section of the Guadalquivir Valley was the region showing the smallest errors (below 1% for the three years), while the largest errors were concentrated in mountainous areas in the east and south of the region (as high as 25% for some locations; Fig. 4). Equally, this spatial analysis quantified the increase in IE according to the distance to the nearest station; for example, for 2007, IE increased by 2.2% and by 3.3% for areas located 5–10 km and 35–40 km from the nearest weather station, respectively (data not shown).

The temporal signatures of ETo generated from the LSA SAF dataset, enabled the identification for each location of the weather station with the most similar ETo trend (TS). In many cases, that station was not in fact the closest one (NN; Fig. 5), with significant differences, especially in heterogeneous topographic areas such as the coastal and mountainous regions in the north and east of Andalusia. Compared with the LSA SAF dataset, NN provided an average annual error of 2.8%, ranging from 2.2 to 3.4% between years (Fig. 6), with maximum errors reaching 27% (Fig. 7). TS generated smaller errors, with an average error of 2.3%, ranging from 1.7 to 2.9% between years (Fig. 6) and with maximum errors reaching 25% (Fig. 7). Spatial analysis of the errors in ETo assessment (Fig. 7) identified greater errors in those locations farther away from the reference weather station, while errors were small and homogenously distributed in the lowest section of the Guadalquivir Valley, where homogenous climatic conditions and greater weather station density were found. Conversely, in the mountainous areas located in the south and east of the region, with heterogeneous weather conditions and low weather station density, the greatest errors were found. In general, the spatial pattern of the errors remained constant during the three years under study (Fig. 7), indicating that topography played a critical role in ETo assessment. Thus, for example, locations with mountainous coastline showed the largest average errors, with values for NN as high as 9.1% in 2009 (Fig. 7). Analysis of the error by season showed that autumn recorded the largest errors (4.6 and 4.0% for NN and TS, respectively; Table 3), while spring was the season with the lowest errors (2.6 and 2.1% for NN and TS, respectively; Table 3), with similar results for summer. 3.3. Correction factor assessment based on the nearest (CINN) and the most similar (CITS) weather station When average three-year correction factors based on the nearest weather station (CINN; Fig. 8) and on the most similar weather station (CITS; Fig. 9) were applied, a significant improvement was found; average errors decreased to 1.2 and 1.1% for CINN and CITS, respectively (Fig. 6), and maximum errors decreased to 17% for both approaches. Spatial analysis of the generated errors identified those locations farthest from the reference weather stations (nearest or most similar) or in mountainous areas as showing the

J.M. Ramírez-Cuesta et al. / International Journal of Applied Earth Observation and Geoinformation 55 (2017) 32–42

37

Fig. 5. Influence area for each weather station based on the temporal signature analysis (shown in a different color). Thiessen polygons based on the nearest weather stations are shown with black lines. Orange dots indicate the location of each weather station. For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article. Table 3 Seasonal average and maximum errors for ETo assessment and coefficient of variation (CV) considering the closest weather station (NN), the most similar weather station (TS) and the correction indexes with NN (CINN) and TS (CITS) (n = 6768). Year

Season

NN

2007

Autumn Spring Summer Winter Year Autumn Spring Summer Winter Year Autumn Spring Summer Winter Year

3.65d (28.43) 2.91c (17.50) 2.09a (18.21) 3.74d (24.52) 2.71b (16.72) 4.12e (37.03) 2.44c (17.42) 1.66a (12.58) 3.87d (31.42) 2.18b (15.71) 6.05e (59.55) 2.40a (15.66) 3.95d (42.86) 3.11b (28.55) 3.43c (26.57)

Avg. (max)

2008

2009

TS CV 1.05 1.05 1.00 0.97 1.02 1.13 1.07 0.98 0.99 1.06 1.15 0.98 1.13 1.05 1.07

Avg. (max) 3.18d (21.83) 2.37c (17.25) 1.86a (13.52) 3.27d (26.21) 2.22b (16.72) 3.36e (36.01) 1.90c (15.87) 1.41a (9.31) 3.20d (31.42) 1.70b (15.71) 5.38e (59.65) 2.04a (11.50) 3.38d (42.86) 2.72b (25.93) 2.87c (24.74)

CINN CV 0.96 1.05 1.00 0.95 1.02 1.14 0.97 0.94 0.99 1.02 1.18 0.91 1.14 1.00 1.07

Avg. (max) 2.03c (16.33) 1.07a (7.03) 1.20b (9.27) 1.20b (12.61) 1.01a (10.53) 1.74e (18.84) 0.77a (5.94) 1.26d (11.49) 1.18c (11.70) 0.92b (8.25) 2.88e (34.54) 0.97a (7.73) 2.24d (29.80) 1.19b (12.35) 1.53c (16.51)

CITS CV 1.14 0.90 0.98 1.20 1.01 1.19 0.94 1.09 1.09 1.11 1.20 0.91 1.23 1.02 1.17

Avg. (max) 1.90b (16.58) 1.07a (6.45) 1.12a (9.27) 1.10a (12.37) 0.97a (8.68) 1.53e (15.21) 0.71a (4.83) 1.17d (11.49) 1.02c (9.69) 0.82b (7.33) 2.88e (34.50) 0.96a (5.21) 2.00d (29.80) 1.09b (6.43) 1.48c (16.51)

CV 1.16 0.89 0.97 1.09 0.99 1.07 0.91 1.11 0.91 1.07 1.21 0.89 1.26 0.92 1.18

greatest errors (Fig. 7). Conversely, homogeneous areas in the lowest section of the Guadalquivir Valley showed errors lower than 1% (Fig. 7). An analysis of the error by season revealed that autumn was the season with the highest errors (2.2 and 2.1%, for CINN and CITS, respectively; Table 3). Conversely, spring and summer were the seasons with the lowest errors, with errors of 0.9% for CINN and CITS for spring (Table 3).

3.4. Reference evapotranspiration assessment based on available weather data and LSA SAF

Fig. 6. Errors in annual ETo assessment considering the closest weather station (NN), the most similar weather station (TS) and the correction indexes with NN (CINN) and TS (CITS). Different letters indicate significant differences at p < 0.05. Whiskers indicate the standard error of the mean. LSA SAF dataset was used as a reference.

ETo estimates with SK and TPS interpolation approaches, NN, CINN, TS and CITS were assessed using LSA SAF and available RIA data. Average annual errors ranged from 1.1% for CITS to 6.4% for the SK interpolation approach, with a limited inter-annual variability (Table 4). Analysing seasonal ETo values, a clear temporal pattern was found; spring was the season with the lowest errors (ranging from 0.9 to 5.0% for CITS and SK, respectively) while autumn showed the highest errors (from 2.1 to 12.3% for CITS and SK, respectively; Table 4), with similar performance between approaches and years.

38

J.M. Ramírez-Cuesta et al. / International Journal of Applied Earth Observation and Geoinformation 55 (2017) 32–42

Fig. 7. Spatial distribution of the errors considering the closest weather station (NN), the most similar weather station (TS), the correction indexes with NN (CINN) and TS (CITS) for each year. LSA SAF dataset was used as a reference.

Table 4 Seasonal and annual average errors in ETo assessment and coefficient of variation (CV) based on RIA and LSA SAF data, using Simple Kriging (SK), Thin Plate Splines (TPS), the closest weather station (NN), the most similar weather station (TS) and the correction indexes with NN (CINN) and TS (CITS). Year

Season

2007

Winter Spring Summer Autumn Year Winter Spring Summer Autumn Year Winter Spring Summer Autumn Year

2008

2009

SK Avg. (CV)

TPS Avg. (CV)

NN Avg. (CV)

CINN Avg. (CV)

TS Avg. (CV)

9.99 (0.86) 5.58 (0.92) 6.04 (0.95) 13.32 (1.50) 5.56 (0.83) 8.47 (1.05) 4.52 (0.86) 7.24 (0.87) 11.22 (0.94) 6.34 (0.74) 8.75 (1.29) 4.91 (0.82) 7.78 (0.87) 12.48 (1.06) 7.26 (0.88)

8.62 (0.91) 5.47 (0.91) 6.61 (0.96) 13.15 (1.11) 5.20 (0.85) 5.26 (0.99) 2.66 (0.88) 5.86 (1.11) 7.80 (0.97) 4.07 (0.94) 6.46 (1.14) 3.11 (0.87) 6.15 (0.97) 9.22 (1.02) 5.27 (0.95)

5.36 (1.28) 3.61 (1.22) 3.84 (1.40) 7.63 (1.92) 3.45 (1.08) 5.18 (1.22) 2.97 (1.09) 4.24 (1.82) 6.77 (1.25) 3.50 (1.25) 5.53 (1.46) 3.17 (1.10) 5.47 (1.18) 8.92 (1.22) 4.82 (1.14)

3.72 (1.75) 2.24 (1.70) 3.47 (1.63) 6.44 (1.88) 2.33 (1.44) 3.54 (1.54) 2.11 (1.42) 4.11 (1.90) 4.79 (1.54) 2.88 (1.54) 4.22 (1.87) 2.31 (1.47) 4.21 (1.38) 6.45 (1.45) 3.64 (1.43)

3.25 (0.96) 2.40 (1.05) 1.80 (1.00) 3.13 (0.95) 2.23 (1.03) 3.15 (1.01) 1.90 (0.98) 1.39 (0.94) 3.34 (1.16) 1.68 (1.05) 2.69 (1.00) 2.02 (0.92) 3.45 (1.13) 5.51 (1.17) 2.92 (1.06)

Spatial analysis of the errors in ETo assessment revealed clear differences between approaches (Fig. 10). Using the CITS approach, huge areas of the lowest section of the Guadalquivir Valley showed negligible errors, with higher but still small errors in the mountainous areas in the east of the region. Conversely, TPS and SK showed high errors without any spatial pattern (Fig. 10).

CITS Avg. (CV) 1.09 (1.07) 1.09 (0.88) 1.11 (0.99) 1.90 (1.09) 0.98 (0.99) 1.02 (0.91) 0.72 (0.91) 1.18 (1.11) 1.55 (1.07) 0.82 (1.07) 1.07 (0.93) 0.96 (0.89) 2.02 (1.27) 2.93 (1.22) 1.49 (1.18)

8.1%, respectively). When annual correction factors were applied, a similar pattern of accuracy was found, but with slightly worse performance (2.0 and 2.3% for spring and summer, and 7.4 and 8.2% for winter and autumn; Table 5).

4. Discussion 3.5. Daily reference evapotranspiration assessment Daily ETo assessment using the CITS approach was carried out for four representative days for 2007, 2008 and 2009. The application of seasonal correction factors showed clear differences between seasons, with small average errors for spring and summer (1.6 and 2.0%, respectively) and much higher for winter and autumn (7.1 and

The use of remote sensing techniques and weather forecast data, integrated into the LSA SAF tool, has enabled the development of alternative tools for ETo assessment in areas with limited or inaccurate weather data from traditional weather stations. Average errors in annual ETo assessment were reduced significantly, with errors of around 1.1% achieved (Table 4).

J.M. Ramírez-Cuesta et al. / International Journal of Applied Earth Observation and Geoinformation 55 (2017) 32–42

39

Fig. 8. Annual and seasonal correction factors taking the nearest weather station (defined with Thiessen polygons) as a reference, based on LSA SAF dataset.

Table 5 Average differences between estimated daily ETo using the CITS approach with seasonal and annual correction factors for representative days during 2007, 2008 and 2009. Year

2007

2008

2009

Season

Winter Spring Summer Fall Winter Spring Summer Fall Winter Spring Summer Fall

DOY

41 131 230 323 39 120 230 323 40 136 228 323

Seasonal correction factor

Annual correction factor

Avg. (max)

CV

Avg. (max)

CV

8.01 (263.70) 1.75 (14.70) 1.66 (34.30) 11.31 (139.16) 4.01 (191.63) 1.36 (13.61) 1.50 (12.34) 3.70 (23.23) 9.49 (71.83) 1.80 (43.59) 2.82 (33.27) 9.31 (126.12)

1.48 1.05 1.38 0.99 2.28 1.01 1.10 0.98 1.03 1.28 1.23 1.04

8.10 (269.19) 2.17 (15.93) 1.96 (41.22) 11.42 (124.38) 4.21 (191.24) 1.79 (14.02) 1.63 (13.78) 3.42 (19.37) 10.02 (74.39) 2.04 (42.76) 3.27 (39.42) 9.68 (129.52)

1.47 0.98 1.32 0.99 2.24 1.08 0.98 0.92 1.01 1.21 1.15 1.07

The use of the LSA SAF tool to evaluate the accuracy of some interpolation methods based on measured ETo values from the RIA network in southern Spain found that the Thin Plate Splines (TPS) and Inverse Distance Weighting (IDW) methods performed best (Fig. 3), with average errors ranging from 1.5 to 4.5%. Previous studies confirm the good results produced with TPS and IDW; Jarvis and Stuart (2001), Irmak et al. (2010) and Keblouti et al. (2012) obtained satisfactory results for air temperature and rain-

fall assessment using the two approaches. Conversely, the kriging technique produced a sub-optimal performance, although results were improved by the incorporation of covariates; in particular, there was an improvement of around 33% when elevation was used as a covariate. Similar improvement with the use of elevation as a covariate was found by Di Piazza et al. (2011), Li et al. (2006) and Luo et al. (2008) for rainfall, temperature and wind speed estimations.

40

J.M. Ramírez-Cuesta et al. / International Journal of Applied Earth Observation and Geoinformation 55 (2017) 32–42

Fig. 9. Annual and seasonal correction factors taking the most similar weather station (defined with the temporal signatures procedure) as a reference, based on LSA SAF dataset.

The performance of the interpolation approaches was not equal for the whole analysed region. The mid and lower sections of the Guadalquivir Valley showed the smallest interpolation errors due to the homogenous weather conditions resulting from the flat topography and the high density of weather stations located in these sections. However, in the mountainous areas located in the north and east of the region, the lower density of weather stations and the abrupt changes in ETo generated large errors in ETo assessment. Luo et al. (2008) also observed this effect in England by using the TPS method for interpolating wind speed. Along the same lines, the error was shown to increase as the distance from the weather station increased, following a similar pattern to that obtained by Melvin et al. (2008), Johansson and Glass (2008) or Glahn and Im (2013). Although the described interpolation procedures provided satisfactory results for ETo assessment, they require the implementation and use of specific tools that are usually technically and economically inaccessible for farmers or technicians. Accordingly, farmers have traditionally relied on the information provided by the closest weather station to their fields, and have achieved satisfactory results (Cruz-Blanco et al., 2014b). However, the nearest weather station may not be representative of the local weather conditions in the field, with discrepancies primarily due to the observed differences in altitude (Fig. 1), affecting temperature, rela-

tive humidity and wind speed. Though it may be possible to account for the effect of altitude on temperature, it is not possible to estimate the corresponding changes in relative humidity and wind speed (Lorite et al., 2015), and this can lead to significant potential errors. These limitations underline the need to examine new methods. The LSA SAF dataset was used to establish the most appropriate weather station for each location in Andalusia (Fig. 5); in many cases it was not in fact the closest one. Consequently, the use of weather data from the most similar weather station generated much smaller errors than those produced with the use of the closest one, ranging from 1.4 to 5.5%. A further step was the development of correction indexes that improved the ETo estimation, resulting in small errors ranging between 0.7 and 2.9% when using the CITS approach. In addition, the proposed methodology helps to identify areas where additional weather stations should be installed (those with large correction indexes) and the areas which already have a sufficient number of weather stations (those with small correction indexes). Errors in ETo assessment are not constant throughout the year: all the approaches detected the highest errors during winter and autumn (ranging from 5.3 to 13.2% for TPS and from 1.0 to 2.9% for CITS, during the three years under study). This variance is caused by the higher spatial heterogeneity of weather conditions (and consequently ETo ) as unstable atmospheric conditions do not usually

J.M. Ramírez-Cuesta et al. / International Journal of Applied Earth Observation and Geoinformation 55 (2017) 32–42

41

platforms (Lorite et al., 2012). In addition to these tools for ETo assessment, similar tools could be developed for other variables such as temperature, radiation or even rainfall, all key elements in computing chilling requirements or irrigation scheduling. 5. Conclusions

Fig. 10. Errors in ETo assessment based on values measured by RIA for 2008 when using the CITS (a) and the TPS (b) method. The delimitation of the areas was based on the most similar weather station.

occur at the same time throughout the entire region. However, during the late spring/summer period, stable conditions for the whole region are very common, thus resulting in better performance of the proposed approaches (ranging from 2.7 to 6.6% for TPS and from 0.7 to 2.0% for CITS, during the three years under study). Similar conclusions were reached by Cruz-Blanco et al. (2014a) when evaluating ETo assessment for southern Spain, or Mardikis et al. (2005), who compared interpolation methods for estimating ETo in Greece. Comparing ETo estimations at the annual, seasonal and daily scale, it was the last of these three that produced the poorest performance, although the errors were not high in any of these cases. This demonstrates that the proposed approaches are useful even for daily ETo assessment. In addition, the highest errors in ETo assessment coincided with the period when the impact on irrigation scheduling is minimal (winter and autumn), as the semiarid climate conditions of southern Spain mean that irrigation is mainly concentrated during spring and summer. In spite of the excellent results obtained with the new approaches, many uncertainties affect their application to ETo assessment, mainly related to the difficulty in assessing real ETo under arid and semi-arid conditions (Allen et al., 1998; Temesgen et al., 2005; Cruz-Blanco et al., 2014a,b). Such difficulties include the non-reference conditions where the weather stations are usually located (Allen et al., 1998) and the inability of alternative tools (such as the LSA SAF) to account for all the components involved in ETo (Cruz-Blanco et al., 2015). Other uncertainties are related with the different spatial resolution provided by the LSA SAF in different locations, which could generate variations in the quality of the ETo assessment when evaluating large areas far from the satellite nadir position. Accordingly, additional studies focusing on ETo assessment under semi-arid and arid conditions are required, in order to promote new, accurate and user-friendly approaches to help farmers improve their water management. The easy integration of the proposed approaches within a GIS allows the development of specific websites that provide ETo values at the field scale, complementing the limited available weather data. These tools will form the basis for innovative tools for irrigation management improvement, included in advisory services

Accurate reference evapotranspiration (ETo ) assessment is key to the sustainable use of irrigation water. However, traditional approaches to ETo estimation that only use data from weather stations suffer from a number of limitations such as the relative scarcity of such stations or the inaccuracies generated in the measurement. The integration of remote sensing techniques, weather forecast data and GIS has enabled the evaluation of a number of interpolation models and the development of innovative approaches for ETo assessment. Use of the LSA SAF dataset to evaluate several interpolation methods for estimating ETo revealed that the Thin Plate Splines approach produced the best results, with an average interpolation error of 3.2%. Two new ETo estimation approaches were evaluated as alternatives to farmers’ common practice of using the information provided by the closest weather station. The selection of the most similar weather station resulted in average errors lower than 2.3%, while the use of correction coefficients generated even better results (1.1%). These errors were not homogeneous throughout the entire region; the lowest section of the Guadalquivir Valley registered the smallest errors due to its homogenous weather conditions and higher density of weather stations. The greatest errors were found in mountainous areas, with abrupt changes in ETo values and a limited number of weather stations. Equally, the seasonal performance also varied significantly, with autumn and winter revealed as the seasons when the greatest errors were found, caused by the higher spatial heterogeneity of weather conditions (and consequently ETo ) during these seasons. Accurate ETo assessment and the evaluation of the adequacy of the current number and location of weather stations are some of the applications of the proposed methodologies. These contributions will help improve water management at the field scale, with the development of accurate irrigation scheduling adapted to local conditions, a key component in the sustainability of irrigation schemes under semi-arid conditions. Acknowledgements This study has been financially supported by the projects RTA2014-00030-00-00 and RTA2011-00015-00-00 funded by INIA and FEDER 2014–2020 “Programa Operativo de Crecimiento Inteligente”. The contributions of Dr. de Bruin and Dr. Trigo are greatly appreciated. References Allen, R.G., 1996. Assessing integrity of weather data for reference evapotranspiration estimation. J. Irrig. Drain. Eng. ASCE 122, 97–106. Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration: guidelines for computing crop water requirements. In: FAO Irrigation and Drainage Paper No 56. FAO, Roma. Allen, R.G., Droogers, P., Hargreaves, G., 2002. Predicting reference crop evapotranspiration with arid weather data. In: 18th International Congress on Irrigation and Drainage, Montreal, Canada. Allen, R.G., Pruitt, W.O., Wright, J.L., Howell, T.A., Ventura, F., Snyder, R., Itenfisu, D., Steduto, P., Berengena, J., Yrisarry, J.B., Smith, M., Pereira, L.S., Raes, D., Perrier, A., Alves, I., Walter, I., Elliott, R., 2006. A recommendation on standardized surface resistance for hourly calculation of reference ETo by the FAO56 Penman-Monteith method. Agric. Water Manag. 81, 1–22. Allen, R.G., Tasumi, M., Trezza, R., 2007a. Satellite-based energy balance for Mapping Evapotranspiration with Internalized Calibration (METRIC)- model. J. Irrig. Drain. Eng. 133 (4), 380–394. Allen, R.G., Tasumi, M., Morse, A., Trezza, R., Wright, J.L., Bastiaanssen, W., Kramber, W., Lorite, I., Robison, C.W., 2007b. Satellite-based energy balance for Mapping

42

J.M. Ramírez-Cuesta et al. / International Journal of Applied Earth Observation and Geoinformation 55 (2017) 32–42

Evapotranspiration with Internalized Calibration (METRIC)- applications. J. Irrig. Drain. Eng. 133 (4), 395–406. Alves, M.C., Carvalho, L.G., Vianello, R.L., Sediyama, G.C., Oliveira, M.S., Sá Junior, A., 2013. Geostatistical improvements of evapotranspiration spatial information using satellite land surface and weather stations data. Theor. Appl. Climatol. 113, 155–174. Bastiaanssen, W.G.M., Menenti, M., Feddes, R.A., Holtslag, A.A.M., 1998. A remote sensing surface energy balance algorithm for land (SEBAL). 1. Formulation. J. Hydrol. 212–213, 198–212. Berengena, J., Gavilán, P., 2005. Reference evapotranspiration estimation in a highly advective semiarid environment. J. Irrig. Drain. Eng. 131 (2), 147–163. Chávez, J.L., Howell, T.A., Copeland, K.S., 2009. Evaluating eddy covariance cotton ET measurements in an advective environment with large weighing lysimeters. Irrig. Sci. 28, 35–50. Collins, J.M., 2011. Temperature variability over Africa. J. Clim. 24, 3649–3666. Cruz-Blanco, M., Gavilán, P., Santos, C., Lorite, I.J., 2014a. Assessment of reference evapotranspiration using remote sensing and forecasting tools under semi-arid conditions. Int. J. Appl. Earth Obs. Geoinf. 33, 280–289. Cruz-Blanco, M., Lorite, I.J., Santos, C., 2014b. An innovative remote sensing based reference evapotranspiration method to support irrigation water management under semi-arid conditions. Agric. Water Manag. 131, 135–145. Cruz-Blanco, M., Santos, C., Gavilán, P., Lorite, I.J., 2015. Uncertainty in estimating reference evapotranspiration using remotely sensed and forecasted weather data under the climatic conditions of Southern Spain. Int. J. Climatol. 35, 3371–3384. De Bruin, H.A.R., Trigo, I.F., Jitan, M.A., Temesgen Enku, N., van der Tol, C., Gieske, A.S.M., 2010. Reference crop evapotranspiration derived from geo-stationary satellite imagery: a case study for the Fogera flood plain, NW-Ethiopia and the Jordan Valley, Jordan. Hydrol. Earth Syst. Sci. 14, 2219–2228. De Bruin, H.A.R., Trigo, I.F., Lorite, I.J., Cruz-Blanco, M., Gavilán, P., 2012. Reference Crop Evapotranspiration obtained from the geostationary satellite MSG (METEOSAT). Geogr. Res. Abs. 14, EGU 2012-11453. Di Piazza, A., Lo Conti, F., Noto, L.V., Viola, F., La Loggia, G., 2011. Comparative analysis of different techniques for spatial interpolation of rainfall data to create a serially complete monthly time series of precipitation for Sicily, Italy. Int. J. Appl. Earth Obs. Geoinf. 13, 396–408. Er-Raki, S., Chehbouni, A., Guemouria, N., Ezzahar, J., Khabba, S., Boulet, G., Hanich, L., 2009. Citrus orchard evapotranspiration: comparison between eddy covariance measurements and the FAO-56 approach estimates. Plant Biosyst. 143, 201–208. Fereres, E., Soriano, M.A., 2007. Deficit irrigation for reducing agricultural water use. J. Exp. Bot. 58, 147–159. Gavilán, P., Lorite, I.J., Tornero, S., Berengena, J., 2006. Regional calibration of Hargreaves equation for estimating reference ETo in a semiarid environment. Agric. Water Manage. 81, 257–281. Glahn, B., Im, J.S., 2013. Error estimation of objective analysis of surface observations. J. Oper. Meteorol. 1, 114–127. Hart, Q.J., Brugnach, M., Temesgen, B., Rueda, C., Ustin, S.L., Frame, K., 2009. Daily reference evapotranspiration for California using satellite imagery and weather station measurement interpolation. Civ. Eng. Environ. Syst. 26, 19–33. Irmak, A., Ranade, P.K., Marx, D., Irmak, S., Hubbard, K.G., Meyer, G.E., Martin, D.L., 2010. Spatial interpolation of climate variables in Nebraska. Am. Soc. Agric. Biol. Eng. 53, 1759–1771. Jarvis, C.H., Stuart, N., 2001. A comparison among strategies for interpolating maximum and minimum daily air temperature. Part II: the interaction between number of guiding variables and the type of interpolation method. J. Appl. Meteorol. 40, 1075–1084. Johansson, M.A., Glass, G.E., 2008. High-resolution spatiotemporal weather models for climate studies. Int. J. Health Geogr. 7, 52. Keblouti, M., Ouerdachi, L., Boutaghane, H., 2012. Spatial interpolation of annual precipitation in Annaba-Algeria—comparison and evaluation of methods. Energy Procedia 18, 468–475.

Li, J., Heap, A.D., 2008. A Review of Spatial Interpolation Methods for Environmental Scientists. Geoscience Australia, Record 2008/23, 137 pp. Li, J., Heap, A.D., 2011. A review of comparative studies of spatial interpolation methods in environmental sciences: performance and impact factors. Ecol. Inform. 6, 228–241. Li, J., Huang, J.-F., Wang, X.-Z., 2006. A GIS-based approach for estimating spatial distribution of seasonal temperature in Zhejiang Province, China. J. Zhejiang Univ. Sci. A 7, 647–656. Lorite, I.J., García-Vila, M., Carmona, M.A., Santos, C., Soriano, M.A., 2012. Assessment of the irrigation advisory services’ recommendations and farmers’ irrigation management: a case study in Southern Spain. Water Resour. Manage. 26, 2397–2419. Lorite, I.J., Ramírez-Cuesta, J.M., Cruz-Blanco, M., Santos, C., 2015. Using weather forecast data for irrigation scheduling under semi-arid conditions. Irrig. Sci. 33, 411–427. Luo, W., Taylor, M.C., Parker, S.R., 2008. A comparison of spatial interpolation methods to estimate continuous wind speed surfaces using irregularly distributed data from England and Wales. Int. J. Climatol. 28, 947–959. Mardikis, M.G., Kalivas, D.P., Kollias, V.J., 2005. Comparison of interpolation methods for the prediction of reference evapotranspiration—an application in Greece. Water Resour. Manage. 19, 251–278. Martínez-Cob, A., 1996. Multivariate geostatistical analysis of evapotranspiration and precipitation in mountainous terrain. J. Hydrol. 174, 19–35. Meek, D.W., Hatfield, J.L., 1994. Data quality checking for single station meteorological databases. Agric. For. Meteorol. 69, 85–109. Melvin, M.J., Zygielbaum, A.I., Gutzmer, D., Rentschler, S., Bower, J., Hubbard, K.G., 2008. Network requirements for sensor accuracy and precision: a case study to assess atmospheric variability in simple terrain. Int. J. Climatol. 28, 267–272. Nalder, I.A., Wein, R.W., 1998. Spatial interpolation of climatic Normals: test of a new method in the Canadian boreal forest. Agric. For. Meteorol. 92, 211–225. Persson, A., 2013. User guide to ECMWF forecast products. Version 1.1. 23 July 2013. ECMWF, 129 pp. Santos, C., Lorite, I.J., Tasumi, M., Allen, R.G., Fereres, E., 2008. Integrating satellite-based evapotranspiration with simulation models for irrigation management at the scheme level. Irrig. Sci. 26, 277–288. Santos, C., Lorite, I.J., Tasumi, M., Allen, R.G., Fereres, E., 2010. Performance assessment of an irrigation scheme using indicators determined with remote sensing techniques. Irrig. Sci. 28, 461–477. Shafer, M.A., Fiebrich, C.A., Arndt, D.S., 2000. Quality assurance procedures in the Oklahoma Mesonetwork. J. Atmos. Ocean Technol. 17, 474–494. Temesgen, B., Allen, R.G., Jensen, D.T., 1999. Adjusting temperature parameters to reflect well-watered conditions. J. Irrig. Drain Eng. 125 (1), 26–33. Temesgen, B., Eching, S., Davidoff, B., Frame, K., 2005. Comparison of some reference evapotranspiration equations for California. J. Irrig. Drain Eng. 131 (1), 73–84. Thiessen, A.H., 1911. Precipitation for large areas. Mon. Weather Rev. 39, 1082–1084. Todd, R.W., Evett, S.R., Howell, T.A., 2000. The Bowen ratio-energy balance method for estimating latent heat flux of irrigated alfalfa evaluated in a semi-arid, advective environment. Agric. For. Meteorol. 103, 335–348. Vaughan, P.J., Trout, T.J., Ayars, J.E., 2007. A processing method for weighing lysimeter data and comparison to micrometeorological ETo predictions. Agric. Water Manage. 88, 141–146. Villa-Nova, N.A., Pereira, A.B., Shock, C.C., 2007. Estimation of reference evapotranspiration by an energy balance approach. Biosyst. Eng. 96, 605–615. Voogt, M.P., 2006. Meteolook, a Physically Based Regional Distribution Model for Measured Meteorological Variables. MSc Thesis. TU Delft, The Netherlands. Wentz, E.A., Peuquet, D.J., Anderson, S., 2010. An ensemble approach to space–time interpolation. Int. J. Geogr. Inf. Sci. 24, 1309–1325.