Introduction

The availability of water for irrigation will probably decrease in the future due to increased demands from other sectors such as municipal, tourism, recreation, and the environment. In Spain, fresh water demand is around 35,000 hm3/year, with about 70% devoted to irrigation, and the rest to other uses (MIMAM 1998). In this scenario, assessments of irrigation performance for improving water management in irrigated areas are of high importance. These activities are needed not only to improve water productivity (Hsiao et al. 2007) but also to increase the sustainability of irrigated agriculture.

The improvement in water management in an irrigation scheme requires the assessment of irrigation performance as a point of departure. Robust models for calculating consumptive use requirements now exist (Doorenbos and Pruit 1977; Allen et al. 1998). Additionally, to facilitate data acquisition and to carry out spatial analyses, recently developed tools such as remote sensing (Kite 2000; Kite and Droogers 2000) and GIS (Hartkamp et al. 1999) have been combined with hydrologic models to assess the behavior of irrigation schemes (Droogers and Bastiaanssen 2002; Er-Raki et al. 2007; Santos et al. 2008; Immerzeel et al. 2008). One of the inputs for irrigation performance assessment is the consumptive use or ET. This input may be calculated with the standard FAO methodology (Allen et al. 1998) or may be derived from remote sensing techniques recently developed for estimating actual ET (Bastiaanssen et al. 1998; Allen et al. 2007a). METRIC (Mapping EvapoTranspiration with high Resolution and Internalized Calibration) is an ET estimation model developed by the University of Idaho, USA (Allen et al. 2007a) and based on the SEBAL model of Bastiaanssen et al. (1998). The SEBAL model has been applied and tested at a large number of locations around the world (Bastiaanssen et al. 2005), while METRIC has been applied in the western United States (Allen et al. 2005a, 2007b) to produce high resolution ET maps. Estimates of ET by METRIC have been compared favorably with a series of lysimeter ET measurements at two locations in the northwest US (Tasumi et al. 2005; Allen et al. 2007b).

Several authors (Molden and Gates 1990; Kalu et al. 1995; Malano and Burton 2001; Bos et al. 2005) have defined sets of indicators that characterize irrigation system performance intending to evaluate current practices and to recommend improvements in irrigation efficiency and water productivity. These indicators have been used for different purposes, such as resources optimization (Molden and Gates 1990), assessing trends in performance (Sarma and Rao 1997; Droogers and Kite 1999; Droogers et al. 2000) or for comparing performance among irrigation schemes (Burt and Styles 1999). Also, Kalu et al. (1995) used these indicators for determining values of equity and efficiency within an irrigation area.

In working with performance indicators, important uncertainties influence the accuracy of these indicators such as the actual water applied and water consumed, and the actual and maximum yields. Actual water applied may be derived from water meter records, and water consumed may be estimated from remote sensing techniques; however, all this information is not available in many irrigation schemes. For estimating actual and maximum yields, different models are used, with those providing yield estimates (e.g., Doorenbos and Kassam 1979; Stockle et al. 2003) being much more uncertain than those used for calculating ET.

Attempts to obtain performance indicators by remote sensing techniques have been proposed (Bastiaanssen and Bos 1999; Bastiaanssen et al. 2000; Ray et al. 2002; Karatas et al. 2009). Consoli et al. (2006), using a similar index to the annual relative irrigation supply (ARIS; Malano and Burton 2001), quantified the performance of an irrigation area with orange groves. Roerink et al. (1997) applying the SEBAL model introduced new performance indicators focused on uniformity aspects, while Bandara (2003) determined water productivity and relative water supply (RWS) for three large irrigation systems in Sri Lanka using the NOAA satellite. Bastiaanssen et al. (2001) analyzed the evolution of several performance indicators also with the NOAA satellite in an irrigation scheme located in Brazil, and Panigrahy et al. (2005) using the Indian remote sensing satellite (IRS) and Radarsat synthetic aperture radar (SAR) proposed three cropping system performance indices: multiple cropping index (MCI), area diversity index (ADI), and cultivated land utilization index (CLUI) to prioritize the districts for crop intensification and diversification in the state of West Bengal (India). However, the integration of remote sensing techniques with simulation models has not been commonly used to determine performance indicators, as these indicators are calculated directly from ET estimations (Zwart and Bastiaanssen 2004, 2007; Bandara 2006; Teixeira et al. 2009a, b).

Traditionally, the use of simulation models is limited by the lack of reliable and accurate data for calibration and validation (Bastiaanssen et al. 2007). Thus, the analysis of irrigation performance is often done by applying water balance models to irrigation schemes that develop performance indicators based on average values for the whole irrigation area (Kloezen and Garcés-Restrepo 1998; Molden et al. 1998; Burt and Styles 1999). However, the average values may not reflect accurately the actual irrigation practices in the area since they do not capture the degree of variation in irrigation management among farmers (Lorite et al. 2004a, b; Fernández et al. 2007). Thus, large variability in performance among farmers was considered as an indicator to detect a deficient water management, even if the average performance values are reasonable (Lorite et al. 2004b). A prerequisite to characterize the farm to farm variability in irrigation and its effects is the need to have access to information at the individual field level, implying a considerable effort in data acquisition. At this point, a remote sensing technique that would provide accurate information at the field scale would facilitate greatly data collection.

Estimating water productivity using remote sensing techniques requires an approach for yield estimation. Bastiaanssen and Ali (2003) combined the photosynthetically active radiation (PAR) model of Monteith (1972) with the light-use efficiency model of Field et al. (1995) and the surface energy balance model of Bastiaanssen et al. (1998) to estimate crop growth and to forecast crop yield. Similar procedures were used by Bandara (2006) applied to grain yield estimations for rice, by Patel et al. (2006) for regional distribution of wheat yield estimates, and by Zwart and Bastiaanssen (2007) also for wheat. Despite the popularity and simplicity of radiation-use efficiency models, this approach has several sources of uncertainty (Gower et al. 1999; Mo et al. 2005). Others have used process-based models to simulate biomass accumulation and yield (Maas 1988; Plummer 2000). In other applications, remotely sensed reflectance and thermal radiance data were included into crop growth models for validation (Olioso et al. 1999). Vicente-Serrano et al. (2006) combined high temporal resolution remote sensing images and drought indices to predict cereal production 4 months prior to harvest in a semi-arid region of northeast Spain. Also meteorological satellites have been used to provide crop yield forecasts at national level in Europe. In fact, meteorological indicators derived from visible and infrared Meteosat data, i.e., global solar radiation and relative evapotranspiration, were processed by Roebeling et al. (2004) to predict winter wheat and spring barley yields for Spain and the United Kingdom. However, the use of remotely sensed evapotranspiration to estimate crop yields is not common.

Irrigated agriculture in Spain is undergoing a process of modernization to reduce its share of total water usage (Plan Nacional de Regadíos; MAPA 2002). This is required given the climatic constraints and the intense competition for scarce water resources, all leading to the need to accurately assess water use in the irrigation sector. The objective of this work was to analyze and compare different procedures to calculate irrigation performance indicators of individual fields within a scheme using a satellite-based energy balance (METRIC; Allen et al. 2007a) and a water balance simulation model (LORMOD; Lorite et al. 2004a) and to compare their values with those obtained directly with actual measurements or calculated with the standard methodology for assessing ET. The selection of the different methodologies for irrigation assessment has been based on current tools used by irrigation advisory services in Spain (based on the FAO methodology; Allen et al. 1998) and recent methodologies using remote sensing techniques. The area selected to carry out the analysis of performance indicators was the Genil–Cabra irrigation scheme (GCIS) located in Andalusia, southern Spain.

Materials and methods

Area description

The irrigation scheme is located within the Genil–Cabra irrigation scheme (GCIS; 4°51′W, 37°31′N), near Córdoba, Spain. The area evaluated encompasses about 6,800 ha of irrigated land developed around 1990 and became fully operative after 1995. The climate is Mediterranean continental with an annual average precipitation of 610 mm and a rainless summer. The average air temperature ranges from 10°C in winter to 27°C in summer. The predominant soils in the area are Chromic Haploxererts (35%) and Typic Xerorthent (34.7%) with average values of soil water content at field capacity of 0.30 m3 m−3, and 0.15 m3 m−3 for soil water content at wilting point.

The area is serviced by a modern pressurized irrigation delivery system, which allows complete flexibility of frequency, rate and duration of water delivery. The efficiency of the delivery system was estimated around 80% by the technicians of the area. The whole area is divided into command areas, each composed of one or more parcels. A plot is an administrative unit within a unique boundary that belongs to a single owner. A plot may be divided into several fields, which comprise the crop-management units. Water delivery is measured at the inlet of each parcel using cumulative water meter readings, which were taken four times during the irrigation season.

The 2004/2005 irrigation season (considered from September 2004 to August 2005) was very dry; rainfall was 271 mm, while the average for the last fifteen irrigation seasons was 529 mm. The average irrigation depth applied in the irrigation scheme in the 2004/2005 irrigation season was 417 mm, a value that was much greater than the average depth applied over the last ten irrigation seasons (261 mm).

In the 2004/2005 irrigation season, the most important crops were wheat, cotton, and olive, representing 23, 18 and 14% of the irrigated area, respectively. Other crops included, in order of importance, maize, sugar beet, beans, garlic, sunflower, and several vegetable crops. Wheat and sunflower are irrigated with hand-move sprinkler systems, while horticultural crops and the olive are mainly irrigated with drip systems. Maize is irrigated by sprinkling, and cotton is irrigated by drip or with sprinklers. Measured application uniformity (CUC) determined for fields irrigated with sprinkler systems ranged from 40 to 78%. For drip systems, the measured uniformity ranged from 58 to 97%. The observed low uniformities were caused mainly by combinations of inadequate system design and irrigation management.

The average irrigation depth applied by farmers in the area varied significantly. For 2004/2005, maize, cotton, and sugar beet received 631, 690, and 615 mm, respectively. Garlic was irrigated with 349 mm, and wheat and sunflower had the lowest applied water values (136 and 124 mm, respectively).

Water balance simulation model (LORMOD)

A daily mass-balance model was developed to simulate water management at field scale, as described in detail by Lorite et al. (2004a). The model considered multiple soil layers where rainfall and irrigation were inflows (capillary rise and lateral flow were not considered). Outflows were crop transpiration, soil evaporation, surface runoff, and deep percolation. The soil profile was divided into 10-cm layers up to the maximum root system depth. Surface runoff was calculated with the SCS curve number method using daily rainfall records (SCS 1972) adjusted to consider the effect on runoff of slope and soil moisture conditions (Williams 1991). Water extraction by roots was calculated for each soil layer as a function of its water content and root-length density (Coelho et al. 2003).

Soil water properties were considered in LORMOD with three soil water thresholds for each soil layer: the saturated water content, the drained upper limit or field capacity (FC), and the lower limit of plant extractable water or wilting point. Infiltrated water (precipitation minus runoff) is distributed within the soil profile following a cascading approach where the soil profile is divided into 20 layers. The amount of water above FC in any given layer is transferred to the layer immediately below. Deep percolation was computed as the excess water of the last layer of the root zone. To this term, non-uniformity effects of irrigation were included (Mantovani et al. 1995). The soil water deficit (SWD) was computed daily as the difference between actual water content and FC.

Crop water requirements were calculated using the dual crop coefficients and equations based on FAO (Allen et al. 1998):

$$ {\text{ET}}_{c} = {\text{ET}}_{o} \cdot \left( {K_{\text{s}} \cdot K_{\text{cb}} + K_{\text{e}} } \right) $$
(1)

ET o is the daily reference evapotranspiration calculated with the Penman–Monteith equation, K cb is the basal crop coefficient, and K e is the soil evaporation coefficient. Soil evaporation is especially relevant after wetting events and during initial crop developmental stages. K e is obtained by calculating the amount of energy available at the soil surface (Allen et al. 1998):

$$ K_{\text{e}} = K_{\text{r}} \cdot \left( {K_{{{\text{c}}\max }} - K_{\text{cb}} } \right) $$
(2)

Kr is a dimensionless evaporation reduction coefficient dependent on topsoil water depletion, and Kc max is the maximum value of Kc following rainfall or irrigation. Using this formulation, evaporation is daily field scale calculated and considers all the precipitation and irrigation events.

Basal crop coefficients (K cb) were obtained from tabulated FAO-56 values (Allen et al. 1998) and customized (maturity dates and stages of development) to the general conditions of the area, following local observations collected from technicians and farmers of the area. However, these adjustments do not include possible changes from field to field due to specific problems such as poor stands and inadequate irrigation scheduling. Nevertheless, these general K c curves have been commonly used in irrigation assessment (Kite and Droogers 2000; Er-Raki et al. 2007; García-Vila et al. 2008) as they are usually the only information available. One of the aims of this work is to compare the irrigation schedules based on these general K c curves with those obtained via remote sensing.

Finally, water stress is computed in Eq. 3 using a water stress coefficient as follows (Allen et al. 1998):

$$ K_{\text{s}} = {\frac{{{\text{WHC}} - D_{\text{r}} }}{{\left( {1 - p} \right) \cdot {\text{WHC}}}}} $$
(3)

WHC is the water-holding capacity of the root zone, D r is the root zone depletion, and p is the fraction of the WHC below which the root zone water content limits crop transpiration.

The daily application of this methodology for each field located in the Genil–Cabra Irrigation Scheme provided the optimal irrigation schedules required for the assessment of irrigation performance.

Satellite-based energy balance estimation of crop ET

Eleven Landsat 5 TM images were used in this work, covering Landsat path 201 and row 34. The Landsat image dates were November 13 of year 2004 and March 5, April 22, May 8, June 9, June 25, July 11, August 12, August 28, September 13, September 29 of year 2005. The images were processed using the METRIC energy balance computation procedure (2006 version) of Allen et al. (2007a) to obtain daily ET for each image date.

METRIC estimates ET as a residual of the energy balance at the surface, where energy consumed by the ET process is calculated as a residual of the surface energy equation:

$$ {\text{LE}} = R_{n} - G - H $$
(4)

where LE is the latent energy consumed by ET, R n is net radiation, G is sensible heat flux conducted into the ground, and H is sensible heat flux to the air. Details of the METRIC model are given by Allen et al. (2007a).

Net radiation is computed by subtracting all outgoing radiant fluxes from all incoming radiant fluxes and includes solar and thermal radiation. Soil heat flux is estimated as a function of R n , T s , and vegetation indexes, while sensible heat flux is estimated by deriving a near-surface air-temperature gradient (dT) and aerodynamic resistance between two near-surface heights (0.1 and 2 m above the zero plane displacement), assuming dT is linear to radiometric T s . Calibration of the dT function is accomplished by selecting two extreme “calibration pixels” representing very dry and very wet agricultural surfaces, as described by Allen et al. (2007a). Site-specific coefficients (e.g., surface roughness length by land use type) were locally validated by checking calculated values against local estimates. It was found that most values were very similar to those used by Allen et al. (2007a) in their original formulation.

To characterize field ET, pixels located near the centers of fields were selected to avoid edge effects. In fields with enough size, more than one pixel was chosen to assess the within-field variability in ET. In some plots, up to 15 pixels were characterized, and the coefficient of variation of the ET estimate (CVw-ET) was computed.

Actual crop coefficient (K c act) values were obtained by dividing the ET values derived from METRIC by the grass reference ET (ET o ) calculated with the ASCE version of the Penman–Monteith method (ASCE-EWRI 2005), which is identical for 24-h calculation timesteps to the FAO-56 Penman–Monteith method (Allen et al. 1998). The use of the ASCE version of the Penman–Monteith method for METRIC was required due to METRIC requires the determination of instantaneous ET o calculation. Weather data for calculating ET o were provided by five automatic weather stations included in the Agroclimatic Information Network of Andalusia (Gavilán et al. 2006) and located close to the GCIS. The weather information was evaluated and quality controlled following standardized procedures recommended by ASCE-EWRI (2005). To obtain the K c act curves, the values obtained from METRIC were interpolated between each image date (every 16 days at a minimum) using a spline function to characterize the temporal evolution. A full description of the model may be found in Allen et al. (2007a, b) and in Santos et al. (2008). As an example, K c act curves for cotton and maize are shown in Fig. 1, where the effects of the plot to plot variation in irrigation management are shown, including water stress. The general increase in K c act late in the season, clearly shown in the ripening crop of maize, was due to a rainfall event (24 mm) on September 7. The use of remote sensing techniques and the model METRIC allowed the determination of specific and accurate K c act curves for each field, which includes the effects caused by rainfall, irrigation, or stress on ET.

Fig. 1
figure 1

Crop coefficient (K c act) curves defined to develop the ARIS indicators for cotton and maize in the GCIS during 2004–2005 irrigation season. Individual curves are for specific fields as derived from METRIC. ENV is the upper envelope of these curves. Similar procedure was developed for the others crops

Full size image

Integration of METRIC ET with LORMOD for irrigation scheduling

To integrate estimates of actual ET derived from METRIC with the LORMOD simulation model (Lorite et al. 2004a), the following procedure was used with the aim of using actual K c values (K c act values obtained with METRIC following the procedure described in Sect. “Satellite-based energy balance estimation of crop ET”) over time to provide adjusted irrigation schedules for each field. These actual K c act values include E and T and were used as a single K c (Allen et al. 1998). The integration of LORMOD with METRIC required the modification of LORMOD to use a single K c instead of the dual approach that uses K cb after crop establishment.

In the procedure followed here, initially, LORMOD developed preliminary irrigation schedules for each field using the envelope curve (ENV curve in Fig. 1) calculated as the maximum K c act values that were obtained by METRIC over all the fields for each crop (actual values for cotton and maize are presented in Fig. 1). Such values provide an upper limit for the K c act for each crop over the scheme, indicating the upper boundary of the crop coefficient for those fields (while any K c act reflects possible water use limitations related to water deficits, poor establishment, pests/diseases, etc.) and were used to generate preliminary schedules instead of using crop coefficient values from the literature, as given by Lorite et al. (2004a) or rather than using the field-specific K c act values from METRIC. Using the maximum calculated K c act values (envelope curve, ENV; Fig. 1) ensured that these irrigation schedules preliminarily generated provided sufficient water to meet the full crop demand for all fields. However, irrigation schedules obtained in this fashion would recommend watering in excess of actual needs, in almost all of the cases. To prevent excessive irrigation, the preliminary schedules were updated for each case using field-specific K c act information determined by METRIC.

Soil water deficit (SWD) was initialized at the beginning of the season (1 September), considering that the soil was completely dry and was daily calculated by LORMOD until the sowing date. For each satellite image available, a new daily ET rate was back-calculated using the K c act obtained via METRIC, and the SWD of each field was recalculated for each day up to the previous image date. In that way, the SWD was updated using actual K c act values for the plot (and actual ET) that was provided by METRIC, instead of the maximum K c act values (envelope K c curve) used initially. The LORMOD model adjusted the next irrigation date based on the updated SWD. Once the SWD was updated, the next irrigation event was delayed accordingly, if necessary.

The procedure described previously was designed to develop an optimum irrigation schedule considering the specific field conditions by using the K c act determined by remote sensing (although for some fields and periods, the crop management was not the optimal and K c values were not at their maximum; Fig. 1). The irrigation events determined in these schedules were assumed to occur under field conditions. The actual irrigation dates and amounts applied by farmers were not included in this procedure, because such information was not available until the end of the irrigation season. If actual surveys are performed during the season, such information can be fed back into the model and the SWD adjusted, leading to revised irrigation schedules.

Performance indicators

Two indicators were used in this work considering the parcel with a unique crop as the reference unit (see Sect. “Area description”). One is the annual relative irrigation supply (ARIS), defined by Malano and Burton (2001) as:

$$ {\text{ARIS}} = {\frac{{{\text{Annual}}\;{\text{Volume}}\;{\text{of}}\;{\text{Irrigation}}\;{\text{Water}}\;{\text{Inflow}}}}{{{\text{Annual}}\;{\text{Volume}}\;{\text{of}}\;{\text{Crop}}\;{\text{Irrigation}}\;{\text{Demand}}}}} $$
(5)

Similarly, Bos et al. (2005) defined the Field Application Ratio as the ratio between the volume of irrigation water needed to avoid undesirable stress in the crops throughout the growing cycle and the volume of water delivered to field, i.e., the inverse of ARIS.

The choice of ARIS as the relevant indicator instead of another performance indicator such as the annual relative water supply (ARWS; Molden et al. 1998; Burt and Styles 1999; Malano and Burton 2001; Bandara 2003) was made in order to eliminate the effects of rainfall on this indicator, as ARIS is focused exclusively on the irrigation management. When the rainfall is eliminated, the information provided by the indicator is more useful, especially when indicators for years with different rainfall conditions are compared. Equally, ARIS allows the definition of thresholds to indicate deficit irrigation valid for different rainfall conditions.

Among the several indicators defined to evaluate the water productivity, we used irrigation water productivity (IWP; Lorite et al. 2004a) that represents the benefits obtained per unit of water applied,

$$ {\text{IWP}} = {\frac{{{\text{Increase}}\;{\text{in}}\;{\text{Value}}\;{\text{of}}\;{\text{Production}}\;{\text{due}}\;{\text{to}}\;{\text{Irrigation}}}}{{{\text{Volume}}\;{\text{of}}\;{\text{Irrigation}}\;{\text{Water}}\;{\text{Inflow}}}}}\,({\sf C}\hbox{\kern-8pt}\raisebox{1.2pt}{=}{\kern-1pt}/{\text{m}}^{3} )$$
(6)

Here, the numerator is computed as the difference between actual yield under irrigation and the rain fed yield. It is assumed that management does not change when the grower shifts from rain fed to irrigated conditions, which is likely the case for the GCIS.

IWP is a modification of the traditional water productivity (as proposed by Bos et al. 2005) used here to avoid the effects of rainfall on this indicator. This is very important for water productivity comparisons among winter/spring and summer crops.

The use of irrigation water inflow instead of irrigation consumption (ET) in the denominator was decided to determine the productivity at field scale of the applied irrigation water. The water productivity considering irrigation applied is the appropriate indicator to detect differences in irrigation management among farmers in the area that could not be detected if uniquely ET was considered.

Following is a description on how the numerator and denominator of both indicators were calculated using remote sensing techniques and field data measurements.

Annual relative irrigation supply (ARIS)

The irrigation volume (numerator) was determined in two ways:

  1. a.

    Directly from water meter records. Every plot outlet in the GCIS has a water meter that provides cumulative water delivery records. These records were compiled by the staff of the irrigation scheme, who read the water meters at least four times throughout the irrigation season.

  2. b.

    Estimated by METRIC. Irrigation water inflow was determined by dividing the seasonal ET estimated with METRIC for each field less effective rainfall, by the average irrigation efficiency, which was previously determined for each crop and irrigation system in the GCIS (Santos et al. 2008). Field irrigation efficiency (IE) was defined by Burt et al. (1997) as the ratio of the volume of irrigation water beneficially used (ET) and the volume of irrigation water applied and was estimated for selected fields using the ET determined by METRIC, the effective rainfall, and the actual plot water applied obtained from water meter records. Irrigation efficiency was computed using the ET estimates and water meter records of 20% of all the fields (randomly chosen) for which there were records.

The definition of ARIS required the determination of optimal irrigation schedules in order to obtain the irrigation demand (denominator of Eq. 5). Three alternative approaches were used. Irrigation schedules were calculated with LORMOD using three different crop ET inputs: (a) crop ET calculated with the traditional K cb values obtained from FAO-56 (Allen et al. 1998) and modified as described in Sect. “Water balance simulation model (LORMOD)”; (b) ET using the actual K c act values determined with METRIC, following the procedure described in Sect. “Satellite-based energy balance estimation of crop ET”; and (c) using the K c envelope curve obtained from plotting all the K c act curves from individual fields, as shown in Fig. 1 (ENV curve). This last alternative provided a conservative irrigation scheduling adapted to local/regional conditions. This strategy is useful to be considered in those fields where a regional METRIC analysis has been carried out, but due to the pixel size of satellite data (around 4 ha for Landsat 5), it is not applicable to small fields. This ENV curve would be equivalent to the crop coefficient curves provided by FAO, which are formulated to achieve potential or maximum ET c values, without considering specific field limitations.

Table 1 describes the information required for the FAO-56 and METRIC approaches to determine ARIS.

Table 1 Information required for each of the two approaches to determine ARIS and yield
Full size table

Irrigation water productivity

To estimate the yield increase from rain fed to irrigation, we used rain fed yield estimates obtained from farmers’ experience in the area, and then we computed the expected yields under irrigation using locally adjusted production functions (Doorenbos and Kassam 1979), as discussed by Lorite et al. (2004a). To apply the production function model, both actual and maximum ET values (obtained by the ET METRIC maps in those fields with maximum yield; Table 2) were used, together with maximum yields, estimated from records of the best plots in the area (Table 2). Alternative yield estimation models based on radiation utilization (Monteith 1972) and using remote sensing techniques were described by Bastiaanssen and Ali (2003) and by Lobell et al. (2003).

Table 2 ETmax (mm) and Y max (kg ha−1) for the main crops in GCIS
Full size table

Two approaches were used to estimate actual ET; one is the standard FAO-56 methodology (Allen et al. 1998), using the K cb curves and K s reduction factors described in Sect. “Water balance simulation model (LORMOD)”. In our case, we combined the simulation of soil water deficits using the model LORMOD (Lorite et al. 2004a) with the records of applied water of each parcel to assess the periods of water stress and their severity. The second procedure used actual ET from direct estimates with the METRIC model to assess the periods of water deficits. Thus, K c act curves (Fig. 1) and daily ET o values were combined to compute the seasonal ET that was used in the Doorenbos and Kassam (1979) methodology. The denominator of IWP has been described in Sect. “Annual relative irrigation supply (ARIS)”. Table 1 describes the information required in each of two approaches to determine yield.

To validate these results, the yields of 60 irrigated fields of various crops grown in the 2004/2005 irrigation season were obtained. These field-specific yield values were provided by farmers under the supervision of the technical staff of the GCIS.

ET deficit assessment

To assess the occurrence of water deficits, the actual K c values (K c act) of every field obtained by METRIC were compared to the optimal K c values (maximum K c values defining the ENV curve; Fig. 1). These optimal values were derived from the curves of Fig. 1, removing the effects of the water stress using the envelope curves as reference. The ratio of both K c’s has been defined by Allen et al. (2005b) as the water stress coefficient (ETDef), and it was previously described in Sect. “Water balance simulation model (LORMOD)”. For every crop, the temporal variability throughout the crop cycle of this coefficient (relative ET) was also computed (CVt-ETDef) and was considered as an indicator of stress.

Results and discussion

Annual relative irrigation supply

Figure 2 shows the results of applying the three different procedures previously described (see Sect. “Annual relative irrigation supply (ARIS)”) to obtain the annual volume of crop irrigation demand, needed for the calculation of ARIS. The integration of METRIC and LORMOD yielded irrigation requirements which were lower than those obtained with the standard ET calculation procedure (FAO-56; Allen et al. 1998), except for garlic and sugar beet. In those crops, the FAO-56 had lower values than the actual conditions obtained with METRIC. The differences were particularly significant for sugar beet, where irrigation requirements increased from 730 to 882 mm with the METRIC + LORMOD method (Fig. 2), due to the expansion of the normal season length for sugar beets caused by winter frosts that forced to replant several times. This long season caused a significant increase in total sugar beet ET which was detected by METRIC but was not computed with the FAO-56 procedure. Sunflower gave opposite results, with irrigation requirements of 598 mm obtained with FAO-56, compared with 403 mm calculated with METRIC + LORMOD (Fig. 2). Deficit irrigation is a common practice for sunflower in the area, as previously was documented by Lorite et al. (2004b), and this is the reason why the actual requirements determined with METRIC were less than the theoretical needs. The irrigation requirements determined using the maximum envelope of the K c act curves compared with METRIC + LORMOD were always the highest. Additionally, in general METRIC + LORMOD results had the highest variability compared with the two other procedures (Fig. 2) given that it is the only one reflecting actual field conditions.

Fig. 2
figure 2

Crop irrigation requirements following different procedures to determine the optimum irrigation schedules. Bars indicate standard deviation

Full size image

As it was described in Sect. “Annual relative irrigation supply (ARIS)”, the volume of water applied is required for ARIS calculation. METRIC has also been used to obtain these values, and in Fig. 3, average volume of irrigation applied using METRIC is compared with the actual irrigation volume measured with water meters in each field. Differences between both procedures were small (around 5%) and were caused by uncertainty in the irrigation efficiency estimations (because of the use of average values of irrigation efficiency instead of specific field values; see Sect. “Annual relative irrigation supply (ARIS)”). METRIC estimations were very similar to field measurements in all the crops, except for wheat (Fig. 3). The differences for this crop were possibly caused by assuming a lower irrigation efficiency than the actual value. To this underestimation contributed the fact that farmers commonly use deficit irrigation in wheat in the area (Lorite et al. 2004b). Estimated irrigation applied for sunflower was not possible to be determined due to the very limited number of fields with available data.

Fig. 3
figure 3

Comparison between irrigation applied determined by water meters installed at field and estimated using METRIC. Bars indicate standard deviation

Full size image

Figure 4 presents the ARIS determined with the three different procedures for the GCIS during 2004/2005. For cotton, the standard methodology (ARISFAO) gives ARISFAO values very close to 1.0 (0.98), thus indicating optimum irrigation management [values close to 1 are considered here as an indication of optimum irrigation management (Lorite et al. 2004a)]. However, with the integration of METRIC and LORMOD, the ARISM+L values, which should be closer to reality, give an average value of 1.27, indicating overirrigation for this crop. For sugar beet and garlic, the results were the opposite in that while the standard approach yielded an ARISFAO value of 1.07 which suggested good performance, the calculated value of ARISM+L for sugar beet was 0.81, clearly indicating deficit irrigation. The third approach (ARISENV-M) always showed lower values than ARISM+L because the ET estimates are obtained with a K c that reflects the maximum observed, as shown in Fig. 1.

Fig. 4
figure 4

ARIS determination using different procedures to obtain the optimum irrigation requirements (see Fig. 2). Bars indicate standard deviation

Full size image

To develop a method that determines performance indicators at the scheme level without having information on applied irrigation volumes, a common situation in many schemes, a procedure was developed to determine ARIS considering only METRIC ET estimates and the information of irrigation efficiency by crop. With respect to the previous ARIS determination procedure, here to estimate the annual volume of irrigation water inflow (the numerator of Eq. 5), we used the ET determined by METRIC reduced by the effective rainfall and divided by the irrigation efficiency. Figure 5 shows the ARIS values obtained using two different procedures to determine the optimum irrigation requirements (using METRIC + LORMOD and using the K c envelope curve; see Sect. “Annual relative irrigation supply (ARIS)”), considering irrigation water applied estimated by METRIC (marked with an asterisk to separate them from those values calculated with measured irrigation volumes). The results differed among crops depending on the method; for instance, ARIS using irrigation water applied estimated by METRIC for cotton varied from 1.24 for ARIS(M+L)* to 0.88 for ARIS(ENV-M)* (Fig. 5).

Fig. 5
figure 5

ARIS determination using the different procedures to obtain the optimum irrigation requirements (see Fig. 2) and using the volume of irrigation applied estimated by METRIC. Values for sunflower were omitted due to very scarce number of fields with valid data

Full size image

Although the accuracy of the procedure used in Fig. 5 is less than when irrigation volume data are available, the comparison between the two approaches shown in Fig. 6 indicates that there was good agreement (RMSE = 0.124), with the exception of wheat, due to uncertainty in estimating applied water in this particular crop. The difficulties in obtaining applied water information at the field or farm scales in irrigated agriculture (Lorite et al. 2004b, 2005) make this estimation approach quite useful relative to those that require applied water data to assess performance.

Fig. 6
figure 6

Comparison of ARISM+L determined with actual volume of irrigation and using METRIC volume estimation (ARISM+L*)

Full size image

Plot to plot variability may be inferred from Fig. 7 which shows cumulative frequency ARIS values for maize and sugar beet. The lines for ARISENV-M and ARISM+L represent the lowest and upper limits with ARISFAO in between (Fig. 7a). The differences were especially important for sugar beet (Fig. 7b), where the two procedures based on METRIC had similar lines which differed clearly from the ARISFAO values, probably because of differences in season length between standard and actual values as previously was indicated.

Fig. 7
figure 7

Accumulated frequency for the three different ARIS values for a summer crop (maize; a) and a winter–spring crop (sugar beet; b), considering the estimated volume of irrigation applied based on METRIC

Full size image

Determining ET deficit using METRIC

Temporal evolution of irrigation performance has been analyzed in previous works considering performance indicators such as ARIS (Molden and Gates 1990; Lorite et al. 2004b), or by analyzing the temporal variability in water delivery as a reliability indicator (Sanaee-Jahromi and Feyen 2001). The estimation of actual ET with METRIC allows the assessment of ET deficits when they occur by comparing these ET estimations with optimal ET values, calculated using K c curves determined from the original K c act curves (Fig. 1) but removing the water stress effects by using as reference the envelope curves (see Fig. 1; Sect. “ET deficit assessment”). Figure 8 shows the evolution of the relative ET (ETDef) averaged over all fields for the main crops. It is possible then to detect the timing of ET deficits in relation to the crop critical periods (Robins and Domingo 1953; Denmead and Shaw 1960; Barrett and Skogerboe 1978; Allen et al. 1998; Allen 2000), and it could be of help to analyze and to develop deficit irrigation strategies. Figure 8 shows that there was significant stress near the end of the season in cotton (management induced to promote early ripening), while for garlic, a stress period was detected around mid-April. In the case of sunflower, the water stress is well below the standard value, indicating significant water deficits throughout the season (Fig. 8).

Fig. 8
figure 8

Average water stress coefficient (relative ET; ETDef) evolution for the main crops in the GCIS

Full size image

Combining performance indicators

Several attempts to combine performance indicators in the past (Bastiaanssen et al. 2001; Lorite et al. 2004b; Bandara 2006) have shown that it is possible to generate recommendations to improve irrigation management. The temporal variation in ET deficit (water stress coefficient) for any given crop is also indicative of the scheme performance. By comparing the ARIS indicator with the temporal water stress coefficient variability (CVt-ETDef; Fig. 9a) and the ARIS indicator with the within-field ET variability (CVw-ET; Fig. 9b), areas in the graph that reflect differences in irrigation management could be delineated and it would be possible to make some improvement recommendations.

Fig. 9
figure 9

a Relation between ARISM+L and temporal variability in water stress (CVt-ETDef). b Relation between ARISM+L and ET variability within field (CVw-ET)

Full size image

Figure 9a helps analyze if irrigation management was adequate for the whole season. The comparison of ARIS values with CVt-ETDef provides information not only about the irrigation depth applied but also if irrigation scheduling was appropriate to avoid water stress. Thus, high values of ARIS with also high values for CVt-ETDef indicate that there were periods with water deficits, even under sufficient water supply, as it happened in GCIS during 2004/2005 with cotton and maize (Fig. 9a). The combination of information obtained in Figs. 8 and 9a allows the assessment of the adequacy of irrigation scheduling by farmers in the scheme, aiming at avoiding water deficits which are detrimental to yield.

Winter crops had lower values for CVt-ETDef, even though irrigation management may not be adequate, because rainfall reduced the variability in ET deficit during the rainfall period, giving adequate values for CVt-ETDef (0.07 for wheat) but low ARISM+L values (0.51). Sugar beet growers had problems in the area that year with stand establishment (Fig. 9a), but irrigation management afterward was correct, although the depth of irrigation should have been increased in some fields.

Based on previous works (Roerink et al. 1997) and considering the irrigation method of the area, a limit of 10% for CVw-ET and a range of 0.8–1.0 for ARIS (Lorite et al. 2004b) may be considered acceptable values. Thus, six zones are defined in Fig. 9b. Fields or crops located in zone A indicate overirrigation affecting the entire field. The opposite case is zone F, which indicates deficit irrigation with high ET within-field variability. For the A situation, the obvious recommendation is to reduce the depth of irrigation while keeping the same application uniformity. However, if the value falls in zone F, the situation is not so clear, because the very low ARIS values suggest the need to increase the irrigation depth, although a low water productivity would counteract this need (Lorite et al. 2004b). In GCIS, for 2004/2005, maize and cotton were located in zone A and sunflower in zone F. Here, we consider that the zone C, with an ARIS between 0.8 and 1.0, is near-optimal, assuming that ARIS values below 1—but no less than 0.8—may be acceptable, given the compensatory effects of stored soil water at planting, and economic considerations (maximization of gross or net margin), all suggesting that optimal irrigation amounts are somewhat lower than the maximum amounts needed to avoid soil water deficits (Stewart and Hagan 1973; English et al. 2002; Fereres and Soriano 2007). During 2004/2005, garlic and sugar beet were included in this zone, although some fields were located outside of this zone. Finally, wheat data fell in zone E, implying deficit irrigation but with reduced field ET variability (caused by the winter/spring rainfall). The high within-field ET variability encountered in the case of sunflower (Fig. 9b) reflects the lack of interest in improving sunflower irrigation management in the area and suggests that increasing the seasonal irrigation depth in this crop may not be economical. Even though no crops fell in zone D during 2004/2005, a point in that zone would indicate an adequate ARIS with ET uniformity problems which would leave field areas under deficit irrigation. The low values for ET uniformity (CVw-ET) in summer crops (such as cotton) when compared to winter crops (such as garlic) are probably caused to the comparatively lower uniformity of irrigation relative to that of rainfall (Fig. 9b).

Yield and water productivity estimations

Reliable yield estimates are essential to determine the water productivity of a scheme. Previously, yield values were estimated using simulation models (Droogers and Kite 1999; Droogers et al. 2000; Lorite et al. 2004b, 2007) but had a high level of uncertainty.

It is possible to assess yield values based on ET estimates provided by METRIC, and these are shown in Table 3 and compared to actual values obtained from farmers in the area. An acceptable correlation was found between estimated and actual values, with average error in estimated yield ranging from 8.7% for cotton up to 16.5% for sugar beet. More than 40% of the analyzed fields were sugar beet. For this crop, the yield estimate based on METRIC exceeded actual yield, probably because the poor stand establishment influenced the seasonal ET and its relation to yield. If farmers had to replant, the E losses during the prolonged initial stages must have increased seasonal ET without contributing to biomass accumulation and yield. In addition, anomalous high relative ET levels at the end of the crop cycle were detected by METRIC in some fields. Figure 8 shows a relatively high ET value at the end of the season, when normal practices include the development of water deficits near the end of the season to increase sugar concentration and to reduce transportation costs. For other crops, errors in yield estimation are similar to previous works. Thus, wheat yield error obtained (RMSE = 802 kg ha−1; Table 3) were similar to those described by Lobell et al. (2003) using also Landsat imagery or by Bastiaanssen and Ali (2003) (RMSE = 525 kg ha−1) and Mo et al. (2005) (RMSE = 1,124 kg ha−1). Similar results were obtained for maize (RMSE = 1,610 kg ha−1; Table 3) compared with previous works (RMSE = 1,359 kg ha−1; Mo et al. 2005) and for cotton (RMSE = 480 kg ha−1 in GCIS vs RMSE = 551 kg ha−1 described by Bastiaanssen and Ali 2003). Some authors have indicated that the spatial resolution is an additional source of error (Bastiaanssen and Ali 2003; Moriondo et al. 2007), although in our case this error did not affect the results due to the selection of fields with enough size for Landsat 5 resolution requirements (Santos et al. 2008).

Table 3 Yield estimations using METRIC ET compared with actual data from specific fields
Full size table

Another approach for yield estimation is to use field data on applied irrigation water with LORMOD to calculate ET using a FAO-56 type of daily soil water balance and deriving yield from the same production function (Doorenbos and Kassam 1979) used in Table 3 with METRIC ET. To compare both methods, we normalized the yield values using maximum estimates, and the results of both methods are shown in Fig. 10. Normalized yields (Y N) were around 0.8 for garlic, maize, wheat, cotton, sugar beet, and onion, and around 0.4 for sunflower. These values were similar to those determined by Lorite et al. (2004b) in the same area in the period 1996–2000. Higher variability was found for the Y N as determined with LORMOD. Also, the yield estimates for sugar beet using METRIC ET were closer to actual yields than those obtained with LORMOD and applied water measurements, in spite of the high METRIC ET values described previously for this crop. Figure 11 shows that for sugar beet the RMSE was much less in the case of METRIC and was closer to the 1:1 line.

Fig. 10
figure 10

Normalized yield (actual vs. maximum crop yield) using LORMOD and estimates based on METRIC-derived ET. Bars indicate standard deviation

Full size image
Fig. 11
figure 11

Comparison of normalized yield values for sugar beet fields using actual yield values versus using METRIC estimation. RMSE indicates the root mean square error for each estimation

Full size image

Irrigation water productivity (IWP) was determined based on ET, yield, and the volume of water applied. Table 4 shows IWP values for the main crops in the area during 2004/2005. Garlic was the crop with the highest values (around 3 €/m3), while maize had the lowest values (around 0.20 €/m3). These values agree with previous analyses in the same area (Lorite et al. 2004b) and with Howell (2001), who obtained values for IWP very similar to the observed in the GCIS. In general, differences between IWP obtained with both ET methods were small, although in some crops such as wheat, differences reach 32% (0.38 €/m3 with METRIC versus 0.29 €/m3 with LORMOD). Similar conclusions were reached when the volume of water estimated with METRIC was used instead of using actual water meter volumes (Table 4).

Table 4 IWP determined using either METRIC or LORMOD, (a) with actual water volumes and (b) with applied water estimated using METRIC (with asterisk)
Full size table

An alternative indicator for the IWP calculated here is to calculate the water productivity as the gross income per unit of water consumed (ET water productivity; WPET). One advantage of the latter indicator is that it requires less data (the volume of irrigation water applied is not needed) and this is why it is most widely used (Droogers et al. 2000; Zwart and Bastiaanssen 2004, 2007; Immerzeel et al. 2008; García-Vila et al. 2008; Teixeira et al. 2008, 2009a, b; Hellegers et al. 2009). The WPET indicator considers the effects of stored soil water and the effects of rainfall, but it is not appropriate in situations where the irrigation water management is incorrect and where excess irrigation is applied without influencing ET. WPET values calculated with ET c provided by METRIC in GCIS ranged from 0.20 €/m3 for maize to 2.63 €/m3 for garlic.

Conclusions

Irrigation scheme performance assessment is hindered by the need to have detailed on-farm water application records. We have shown here that using remote sensing approaches to determine actual ET such as METRIC, integrated with a water balance model, makes it possible to assess performance with minimum data. The proposed methodology incorporates the actual field characteristics improving other assessment approaches (Molden and Gates 1990; Molden et al. 1998; Droogers et al. 2000; Lorite et al. 2004a, b; Rodríguez-Díaz et al. 2008) and may be used with confidence in other irrigation schemes with current limited available data. In this work, the usefulness of remote sensing techniques for data collection is clearly demonstrated, as previously shown by Bastiaanssen et al. (2007).

The use of performance indicators comparing optimal strategies with actual management allows the detection of those fields with inadequate irrigation management, thus focusing on them potential irrigation improvement efforts. In addition, the use of simulation models considering the actual conditions provided accurate and individualized recommendations to improve irrigation water management in the area, including possible use of deficit irrigation strategies.

Traditionally, optimum irrigation schedules have sought to apply the required water to meet the crop demand without applying irrigation in excess, using general K c curves (Lorite et al. 2004a). However, under specific conditions, variability in sowing or harvest dates (Tasumi and Allen 2007) or deficit irrigation cause these curves do not correlate well with actual conditions (Santos et al. 2008). Then, errors in the irrigation assessment can result. The assumption that the ET of irrigated areas is close to the maximum ET must be verified and remote sensing techniques offer an adequate tool for that purpose.

In the comparison among ARIS values determined with different procedures, significant differences between ARISM+L and the standard approaches (ARISFAO) were detected (Fig. 4), highlighting the importance of considering actual field conditions for performance assessment. In areas where water deficits are common, it is possible to detect them using remote sensing (Sepulcre-Cantó et al. 2006; Santos et al. 2008). However, remote sensing does not detect overirrigation, but the use of simulation models as done here provides performance assessment values that would detect excess water applied (Fig. 4). For example, ARISFAO value for cotton was 0.98 (Fig. 4), indicating good irrigation management, while ARISM+L gave an average value of 1.27 which is indicative of overirrigation.

The approach used here also provides a quantification of both within- and between-field variability which is useful when combined with ARIS to make more specific recommendations based on the assessment of irrigation uniformity.

The estimation of yield using actual ET and a production function was improved when actual ET was calculated by METRIC (errors in RMSE were reduced from 0.259 to 0.106; Fig. 11). Such improvements in yield estimation suggest that the water productivity calculations for the area are equally improved with this methodology.

The estimations provided with the integration of METRIC and LORMOD allow the determination of performance indicators in irrigation schemes with very limited information with a high level of reliability and accuracy in the obtained results. The confidence of the results and conclusions presented in this work depends heavily on METRIC performance. Although the validation of METRIC is currently limited to several local analyses (Tasumi et al. 2005; Allen et al. 2007b), SEBAL (Bastiaanssen et al. 1998), the precursor model of METRIC has been applied with success in numerous locations around the world (Bastiaanssen et al. 2001, 2005; Bastiaanssen and Ali 2003; Teixeira et al. 2009a, b). At present, the validation of METRIC under Mediterranean conditions is being conducted by installing agrometeorological stations in commercial fields to assess the validity of METRIC ET estimates.

Improvement in irrigation management should be based on an accurate knowledge of the hydrology of irrigation districts, including knowledge of the actual ET of each field. Recent efforts in characterizing actual ET (Choi et al. 2009) should provide accurate ET estimations and also should aid in water stress detection. There is a need to improve the accuracy of the daily estimates of actual ET obtained from extrapolating instantaneous remote sensing values. Studies performed with imagery obtained from satellites (e.g., Bastiaanssen et al. 2005) and airplanes (e.g., Chavez et al. 2008) suggest that, although the methods have been validated under a variety of conditions, errors in the daily ET estimation may be significant in specific situations; thus, the need for further validation and calibration efforts of the methods presented here cannot be overemphasized.

At present, alternative procedures to remote sensing using satellites are not currently economically feasible for large areas (such as sensors installed aboard of airplanes). We have shown here that combining water balance models with remote sensing techniques that unable the estimation of actual ET is a feasible approach at characterizing some aspects of irrigation scheme performance at a reasonable cost. Nevertheless, even though the results presented here prove the validity of the approach, improvements in the methodology are possible and should be sought. Such improvements include higher data acquisition frequency and spatial resolution of satellite images, and the refinement of simulation models to incorporate soil spatial variability, the dynamics of crop performance, and various irrigation management strategies.