Introduction

Agricultural water use plays the most critical role in water resources management all around the world. Currently, 17% of the total agricultural land is irrigated and 40% of food and fiber demand are supplied from this irrigated land with the use of 70% of available water resources (Anonymous 1996). In Turkey, 36% of available water resources are under control and nearly 74% of this water is used by 25% of irrigable agriculture land. To increase the percentage of controlled water resource requires greater investment. Moreover, improving irrigation efficiency will enable an increase in the proportion of irrigated land with the same water. However, Howell (2001) stated that, in some cases, increasing irrigation efficiencies alone may not result in new water for allocation unless the consumptive use part of the diverted water is actually reduced.

Monitoring the water use in the field is important to develop effective precautions and for this purpose some indicators are required. Yield, Water Use Efficiency (WUE) and Soil Water Deficit Index (SWDI) are some of the principal indicators related to water use and productivity of irrigated agriculture. Sinclair et al. (1984) described WUE as crop yield per unit of water use. Bos (1980, 1985) recommended that WUE for irrigation be based on the yield production above the rainfed or dryland yield divided by the net crop evapotanspiration (ETc) difference for the irrigated crop, which he called the yield/ET ratio. Irrigation remains vitally important worldwide as a means to enhance production and increase WUE (Howell 2001). In addition to irrigation water, some agricultural practices affect the WUE. Cotton furrow operation system (Ünlü et al. 2007), wheat nitrogen fertilization (Pala et al. 2007; Angus and van Herwaarden 2001; Oweis et al. 2000), potatoes irrigation frequency (Kang et al. 2004), genetic structure of seed for barley (Sivamani et al. 2000), can be given as examples. In addition, Howell (2001) stated that, many agronomic, engineering, and management technologies can reduce nonproductive water use in irrigated agriculture. SWDI, which depends on the ratio between the actual and potential ETc, is essentially an indicator of crop water use (Colaizzi et al. 2003a,b). It is necessary to define the potential and actual ETc for calculation of the SWDI. Actual ETc can be determined through soil water balance or estimated based on related methods (Allen et al. 1998).

WUE and SWDI calculations could be very hard due to the difficulty of actual and potential ETc measurement and/or estimation for large irrigated areas. Because irrigation water applications can vary within the growing period and from site to site. Thus actual ETc is dynamic on a spatial and temporal basis. A very large body of research, spanning almost four decades, has demonstrated that much of the required agricultural information can be derived from remotely sensed data (Pinter et al. 2003). The ability to accurately assess water stress symptoms in vegetation using spectral reflectance measurements is an important goal for remote sensing research (Jackson 1986). Vegetation and water stress indices calculated based on remotely sensed data, such as spectral reflectance and infrared surface temperature, can provide spatial and temporal information.

There are numerous studies addressing the use of spectral reflectance to obtain the leaf/canopy water status and water stress (Luquet et al. 2003; Ceccato et al. 2001; Penuelas et al. 1993, 1997; Shibayama et al. 1993; Danson et al. 1992). Spectral reflectance at a given wavelength is the ratio of radiance to the irradiance. Under sunny conditions a target receives both direct and diffuse radiation and this incoming radiation called irradiance (W m−2). The radiation that is reflected from the target is called radiance (W m−2). In order to measure detect reflectance, hand-held radiometers can be used such as spectroradiometer or infrared thermometer receiving radiation reflected from a target with in the field of view. According to this, radiance of a crop canopy could be measure by means of a spectroradiometer. In order to detect irradiance, a panel (e.g., spectralon or BaSO4) which is able to reflect the incoming radiation completely could be used (Jackson et al. 1980). Figure 1 shows the radiance and reflectance data of sugar beet crops which growth under fully irrigated and rainfed conditions. The spectral reflectance values of the region between 300–1100 nm were given at 2 nm intervals. From this data, reflectance of each individual wavelength can be obtained such as 900, 800, 680 nm. On the other hand spectral data related to bands (e.g., Red and Near Infra Red) could be calculated (Köksal 2006).

Fig. 1
figure 1

Radiance and reflectance characteristics of fully watered and water stressed sugar beet at 300–1,100 nm spectral region (Köksal 2006)

Full size image

A spectral vegetation index is a quantity obtained directly or rationing, differencing, or otherwise transforming spectral reflectance data to represent plant canopy characteristics (Jackson et al. 1980). Numerous spectral indexes have been developed for detection of crop properties such as water stress, yield, LAI and biomass. The most known spectral indexes which depends on the spectral reflectance of Red and Near Infra Red Regions are SR (Tucker 1979; Aparicio et al. 2000), NDVI (Tucker 1979; Penuelas et al. 1997), SAVI (Huete 1988) and WI (Penuelas et al. 1997).

The objective of this study is to determine the relationship between ETc, SWDI, WUE and total yield and to define the use possibilities of the spectral vegetation indexes for estimation of total yield, SWDI and WUE either at the end of the growing season or during the irrigation water management period.

Materials and methods

Field experiments

This study was conducted at the Ankara Soil and Water Resources Research Institute (39″ 53′ N and 32″ 45′ E, altitude 924 m) under semi arid climatic conditions of Ankara, Turkey, during the cropping season of 2004 and 2005. The dwarf green beans (Phaseolus vulgaris, humilis) were planted at 0.70 m row spacing on May 10th in, 2004 and on May 12th in 2005. Final plant density was nearly 65 crops m−2. Some physical properties of soils were analyzed by the Soil and Water Laboratory of Institute. The average sand, loam and clay percentages were 37.3%, 25.3% and 37.3%, respectively. The average gravimetric-based field capacity, wilting point and volumetric mass values were 27.6%, 16.9% and 1.17 g cm−3, respectively. The total available soil moisture (TAM) was 125.2 mm/m.

The experimental design was a completely randomized block design with three replications. Due to the difficulty of soil water content based scheduling under drip irrigated conditions, the irrigation treatments were based on US Weather Bureau Class A pan evaporation. According to the results of previous studies completed at the same region on beans, optimum percentages for Class A pan evaporation were differed from 60 to 120% and irrigation intervals were varied between 5 to 10 days (Ayla 1993; Üstün and Ayla 1993; Üstün et al. 1997). In this study, the irrigation interval was set at one week. Fixed irrigation water was applied to all treatments in the experiment to sprout and grow to a height of nearly 15 cm. The irrigation treatments of experiment were received 120% (B1), 90% (B2), 60% (B3), 30% (B4) and 10% (B5) of Class A pan evaporation and B6 was rainfed (Table 1).

Table 1 Description of irrigation treatments
Full size table

Irrigation water was supplied via drip irrigation method and the amount was measured by water meter. The pressure of each plot was stabilized according to pressure meters mounted on the pipe of each plot. A lateral drip line was designed for each row. The diameter of the lateral pipes was 16 mm with 4 l/h in-line drippers spaced at 50 cm. Fertilizer was applied based on soil analysis recommendations, as before sowing with 60 kg ha−1 P2O5 and 23.4 kg ha−1 N and five times during the growing season with 5.3 kg ha−1 N. The yield was harvested at 1 week intervals and the first and last harvests were carried out on 15 June 2004, 19 June 2005 and 01 October 2004 and 29 September 2005, respectively.

Field measurements

During the growing period, spectral reflectance, climatic parameters and soil water content were measured. Climatic parameters and soil water contents were observed throughout the growing season and spectral measurements were carried out from mid July to mid September.

Spectral reflectance measurements and indexes

Spectral observations were made by using a spectroradiometer (Model LI-1800, Licor, Inc., Lincoln, NE, USA) with 150 field of view lenses. The radiance of the crops in the region ranging from 300 to 1100 nm was detected at 2 nm intervals from 1.5 m above the crop with nadir orientation of telescope body by means of a tripod. As a result, samples about 40 cm in diameter were obtained. Irradiance spectra of sunlight were taken before and after each crop measurement with a spectralon panel (Lapsphere, North Sutton, USA) and the average was used as the reflectance standard. All measurements were made three times and the LI-1800 computed the average of these measurements and stored them as a text file. Reflectance of individual wavelength was calculated by dividing the vegetation radiance measurements with the irradiance measurements which were made by spectralon panel. All spectral data collections were made at solar zenith angles of 45°–50° on cloudless days and at least two measurements were taken each week.

In this study, SR (Tucker 1979), NDVI (Penuelas et al. 1997), SAVI (Huete 1988) and WI (Penuelas et al. 1997) were used for estimation of total yield, SWDI and WUE. These indexes were calculated according to Eqs. 1, 2, 3 and 4. In order to evaluate the detection ability of these spectral indexes, some statistical analyses were made.

$${\text{SR}} = \frac{{{\text{R}}800}}{{{\text{R}}680}}$$
(1)
$${\text{NDVI}} = \frac{{{\text{R}}800 - {\text{R}}680}}{{{\text{R}}800 + {\text{R}}680}}$$
(2)
$${\text{SAVI}} = \frac{{{\text{R}}800 - {\text{R}}680}}{{{\text{R}}800 + {\text{R}}680 + L}}\left( {1 + L} \right)$$
(3)
$${\text{WI}} = \frac{{{\text{R}}900}}{{{\text{R}}970}}$$
(4)

where, R970, R900, R800 and R680 are the reflectance values at 970, 900, 800 and 680 nm wavelengths, respectively. L is the adjustment factor to reduce soil noise problems. Huete (1988) stated that the optimal L values exhibit differences according to vegetation density. In this study, according to Köksal (2008) L values were used for LAI under and over 4.0 as 0.4 and 0.08, respectively.

Leaf area measurements and leaf area index

The leaf area obtained from one representative plant was multiplied by the number of plants in 1 m2 area to obtain LAI (m2/m2; Thenkabail et al. 2000). Leaf area was measured by running the leaves over a leaf area meter Model LI-3000A, Licor, Inc., Lincoln, NE, USA). LAI values were used for calibration of L parameter of SAVI in Eq. 3.

Soil water content measurements and ETc calculation

ETc was calculated as the soil water balance residual for the time periods between two successive Soil Water Content (SWC) measurements dates (Jensen et al. 1990). SWC measurements were made in the centre of trial plots with a neutron probe at 0.16, 0.45, 0.75, 0.95 and 1.15 m depth before each irrigation water applications. In addition, gravimetric water content for 0.3 m soil depth was defined. Since, irrigation rate did not exceed the infiltration rate, there was no runoff in the experimental plots. Effective rooting depth of dwarf green beans was 60 cm. In order to consider the percolation, moisture measurements related to 120 cm soil profile were used for soil water budget calculations.

Water use efficiency and soil water deficit index

Most researchers describe WUE as the ratio of total yield to ETc (Sinclair et al. 1984; Howell 2001; Bos 1980, 1985; Hatfield et al. 2001; Pala et al. 2007). In this study WUE (kg m−3) of each irrigation treatment based on Eq. 5. SWDI was defined according to Colaizzi et al. (2003b) by using Eq. 6. ETc calculated for B1 treatment was used as potential ETc (ETp) for SWDI calculation in Eq. 6.

$${\text{WUE}} = \frac{{{\text{Yield}}}}{{{\text{ET}}_{\text{c}} }}$$
(5)
$${\text{SWDI}} = 1 - \frac{{{\text{ET}}_{\text{c}} }}{{{\text{ET}}_{\text{p}} }}$$
(6)

Statistical analyses

Statistical analyses were performed using Statistical Discovery Software (JMP) and MsExcell. At first, total yield, ETc, WUE, SWDI and seasonal average values of SR, NDVI, SAVI and WI were statistically analyzed. A wide range of data gathered from six irrigation treatments was used together. In order to determine the appropriate time to use spectral index for estimation purposes, daily values of SR, NDVI, SAVI and WI values of all irrigation treatments for each day of measurement were analyzed with total yield, WUE and SWDI.

Results

In order to ensure a good plant stand, 112.0 mm and 128.0 mm irrigation water was applied to all treatments in 2004 and 2005, respectively. After that, applied amount of irrigation water was varied between 112.0 mm and 814.0 mm in 2004 and 128.0 mm and 805.0 mm in 2005. ETc values were calculated between 297.9 mm and 840.4 mm in 2004 and 345.7 mm and 874.2 mm in 2005. During the growing seasons of 2004 and 2005, 75.0 mm and 131.0 mm precipitation occurred, respectively (Table 2). Green beans were harvested once per weak and total yield were changed between 6.08–30.36 t ha−1 in 2004 and 6.73–25.21 t ha−1 in 2005, respectively. Maximum total yield was obtained from the B2 irrigation treatment which received the second highest irrigation water amount.

Table 2 Total applied amount of water, seasonal crop evapotranspiration, rainfall and yield
Full size table

Figure 2 shows the soil water content course of irrigation treatments for 60 cm soil profile, in 2004 and 2005 growing season. Using 120% of A pan evaporation as irrigation water, the soil water content was varied over the field capacity between DAP 50 and 100. After DAP 100 with this irrigation program soil water varied between field capacity and 50% of TAM. Higher than 60% and percentage of A pan evaporation prove a moisture variation over permanent wilting point. As a consequence, it is very hard to obtain an economical yield with an irrigation schedule which is based on a percentage of A pan evaporation under 30.

Fig. 2
figure 2

Soil water course of irrigation treatments in 2004 and 2005. FC Field capacity, TAM total available moisture and WP wilting point

Full size image

The highest correlations were obtained from the polynomial (quadratic equation) relationships between total yield, seasonal ETc, SWDI and WUE (Fig. 3). ETc over nearly 750 mm did not increase the total yield. According to this, when dwarf green beans use 750 mm water, 25,5 tha−1 yield can be produced. Under these conditions SWDI will be nearly 0.3. Because WUE is the ratio between total yield and ETc, it considers the relative increase of total yield per unit water. Although B1 and B2 irrigation treatments were received more irrigation water than B3, WUE values of B3 were higher than that of B1 and B2. The correlation of the polynomial relationship between WUE and total yield were greater than linear. According to WUE, it is reasonable to offer the 60% of A pan evaporation for irrigation scheduling of dwarf green beans. In addition to this, Duncan analyzes (not given) related to total yield and water treatments, showed similar results.

Fig. 3
figure 3

Relationship between ETc, SWDI, WUE and yield

Full size image

Linear regression analyzes were made between total yield, seasonal SWDI, WUE, and daily spectral vegetation indexes (NDVI, SAVI, SR and WI). The path of correlation coefficients for these relationships throughout the measurement period was given graphically in Figs. 4, 5 and 6. In these graphs, 0.5 and 0.01 probability level were shown.

Fig. 4
figure 4

Correlation path of relationships between yield and NDVI, SAVI, SR and WI throughout the 2004 and 2005 growing seasons

Full size image
Fig. 5
figure 5

Correlation path of SWDI and spectral vegetation indexes relationships throughout the 2004 and 2005 growing seasons

Full size image
Fig. 6
figure 6

Correlation path of WUE and spectral vegetation indexes relationships throughout the 2004 and 2005 growing seasons

Full size image

In general, the significance of the relationships was less than 0.01 probability level and correlations were high. After around DAP 80 total yield estimation by means of these indexes can give reliable results (Fig. 3). SAVI, which calibrated based on LAI, has the highest correlation. After DAP 70, spectral vegetation indexes and seasonal SWDI have significant linear relationship (P < 0.05). Especially, NDVI, SAVI and SR have potential for SWDI estimation (Fig. 5). While, the highest correlations (from 0.84–0.94) were defined between SWDI and SAVI (P < 0.01), the SWDI-WI has the lowest. Some significant relationships (P < 0.05) were found between WUE and spectral indexes (Fig. 5). The correlations were poorer than that of total yield and SWDI. These relationships become significant after DAP 100 except for SAVI. After DAP55, the variation in WUE could explained by SAVI significantly (P < 0.05). As a result, SAVI has the highest correlations with three of total yield, seasonal SWDI and WUE.

Total yield, seasonal average SWDI, WUE and SAVI have statistically linear relationships and the correlation coefficients were 0.98, 0.91 and 0.74, respectively (P < 0.01). Based on these results, use of SAVI for total yield, SWDI and WUE estimation can be attributed to a formulation as Eqs. 7, 8 and 9, respectively. In these equations, “a” is the slope and “b” is the offset of the linear regressions between total yield, SWDI, WUE and SAVI.

$${\text{Yield}} = a_{{\text{Yield}} - {\text{SAVI}}} \left( {{\text{SAVI}}} \right) + b_{{\text{Yield}} - {\text{SAVI}}} $$
(7)
$${\text{SWDI}} = a_{{\text{SWDI}} - {\text{SAVI}}} \left( {{\text{SAVI}}} \right) + b_{{\text{SWDI}} - {\text{SAVI}}} $$
(8)
$${\text{WUE}} = a_{{\text{WUE}} - {\text{SAVI}}} \left( {{\text{SAVI}}} \right) + b_{{\text{WUE}} - {\text{SAVI}}} $$
(9)

Figure 7 shows the course of slope and offset values of linear regression equations between total yield, seasonal average SWDI, WUE and daily SAVI. As can be seen from the graphs, these coefficients were varied in the time regularly. Statistical analysis showed that there was a significant difference (P < 0.01) between the slope and offset values based on DAP. Regression analysis between DAP and these values indicated that the best fit was polynomial. According to this, it is possible to calculate the slope and offset values of Eqs. 7, 8 and 9, based on DAP, by means of the quadratic equations given in Fig. 7.

Fig. 7
figure 7

Variation of slope and offset values of yield, SWDI, WUE and SAVI regression and quadratic equation of relationship between DAP, slope and offset values

Full size image

Consequently, estimation of total yield, SWDI and WUE with the use of SAVI can be made in two steps. The first step is the calculation of slope and offset values of Eqs. 7, 8 or 9. The second step will be the use of Eqs. 7, 8 or 9 with the calculated slope and offset values for estimation. As an example, for DAP 113, the slope and offset values could be determined as −10.91 and 43.14, respectively, from the total yield and SAVI graphs in Fig. 7. After that, for DAP 113 Eq. 7 will be determined as Yield = 43.14 SAVI-10.91. If the SAVI value of DAP113 measured as 0.87, total yield will be estimated as 26.6 t ha−1. Another example could be solved for DAP 126 and SAVI value of 0.3 to estimate the SWDI. For this reason the offset and the slope values could be calculated as −1.09 and 0.87 and Eq. 8 become SWDI = −1.09 SAVI + 0.87. As a result, SWDI could be calculated a value of 0.55, for DAP 126. The reliability of the approximation can be improved by using more than one SAVI value measured in various DAP. Because, only one SAVI value represents one time of day and/or period and the conditions of the measurement time can negatively affect the accuracy of estimation.

The comparison of estimated and measured total yield, calculated SWDI and WUE were illustrated in Fig. 8. In these statistics, measured total yield, calculated SWDI and WUE were the original values obtained from the field experiments. Estimated total yield, SWDI and WUE were the seasonal average of daily values. All the interception points of estimated and measured values were close to the 1/1 line. High correlation (r = 0.98) was indicate that total yield could be estimated during the growing season with SAVI by means of this approximation. On the other hand some previous studies have shown that total yield estimation trough spectral indexes are possible (Raun et al. 2001; Kleman and Fagerlund 1987; Moran et al. 1989).

Fig. 8
figure 8

Comparison of estimated and measured yield, calculated SWDI and WUE

Full size image

Estimation of SWDI through SAVI has a correlation of 0.93 and Standard Error (SE) was calculated as 0.13. As stated in previous studies, water stress and spectral vegetation indexes have indirect relationships based on the vegetation amount (Penuelas et al. 1993; Thomas et al. 1971). This significant relationship between calculated and estimated SWDI could be explained by the same concept. SWDI calculation depends on actual and potential ETc, and consists of some experimental errors. Estimated SWDI values were lower than the measured for all six irrigation treatments. On the other hand, according to 1/1 line, differences between estimated and calculated values of various water and vegetation level were similar. The highest differences were determined for B3 and B4 treatment which received 60 and 30% of A pan evaporation. In these treatments, vegetation of green beans was not so poor and the irrigation program of the B3 was evaluated as viable based on the statistical analysis between total yield and ETc. According to these result, estimated values of SWDI through SAVI up to 0.12 indicates non-water stressed conditions. Some previous studies propose estimation of water stress by means of spectral indexes (Pinter 1983; Saha et al. 1986; Luquet et al. 2003; Cure et al. 1989; Cohen 1991; Riggs and Running 1991; Danson et al. 1992; Penuelas et al. 1993, 1997).

WUE is an indicator of the productivity of water. Because of this, it depends on both total yield and ETc. However, numerically WUE values could be high for lower irrigation conditions or low for non-water stressed situations. The results of this study indicated that, for dwarf green bean, maximum WUE values could be obtained with the application of an irrigation schedule based on 60% of A pan evaporation. Excessive (higher ETc) or deficit water application (more or less than 60% of A pan) will cause low WUE value, because of either high ETc or less yield. As can be seen from Fig. 8 the relationship between calculated and estimated WUE is significant (P < 0.01). The interception points of the two data sets were close to the 1/1 line. Estimated value of WUE between 3 and 3.5 indicates a condition of optimum total yield and unimportant level of irrigation water deficiency.

Conclusion

In this field experiment, different levels of irrigation water based on Class A pan evaporation were applied to dwarf green beans. Based on this, crop water use, water stress, vegetation amount and total yield were varied from one irrigation treatment to another. All the indicators investigated in this study were achieved by these variations.

Total yield has significant quadratic relationships with ETc, SWDI and WUE. These polynomial relationships showed that the optimal irrigation water level for dwarf green been under a semi-arid climate condition is 60% of A pan evaporation. In this situation ETc, SWDI and WUE should be nearly 580 mm, 0.12 and 3.2 kg/m3. The application of this irrigation amount can produce nearly 25.5 t ha−1 green beans.

Seasonal average values of spectral indexes (NDVI, SAVI, SR and WI) have significant relationships with total yield, seasonal average SWDI and WUE. However, the relationship between daily values of spectral indexes and observed parameters becomes statistically significant after DAP 80. Generally, the highest correlations were obtained with SAVI. According to this, the estimation of total yield, seasonal average SWDI and WUE through spectral indexes could be reliable after DAP 80 and the most appropriate index is SAVI.

Slope and offset values for the equation of linear regression between SAVI and total yield, SWDI and WUE have polynomial relationships with DAP. Based on this, estimation procedure of observed parameters by means of SAVI could be run in two steps as (1) calculation of slope and offset values based on related quadratic equation and DAP, (2) estimation of total yield, SWDI or WUE with the linear equation. As a conclusion, spectral indexes are useful estimation tools for yield, SWDI and WUE of dwarf green beans. Through these indexes, water use and productivity could be monitored and used as a decision support tool for water management in irrigated agriculture.