Remote Sensing and GIS Techniques for Assessing Irrigation Performance: Case Study in Southern California

Colorado River water is diverted into PVID’s main canal using a small diversion dam. Since the district is relatively flat and the fields are mostly blocked-end furrows and borders, the surface runoff is not significant, and any generated runoff is directed toward the drains. Therefore, water outflow from PVID consists of two components, drainage and canal spills. The high density of deep open drains (about 0.5 km per every ${\mathrm{km}}^2$ of land), along with the medium texture of PVID’s alluvial soils, significantly enhances the movement of water from the root zone toward the drains. All of the drains merge and discharge into a main outfall drain at the downstream end of PVID. The United States Geological Survey (USGS) measures the flow rates of water diversion, drain discharge, and all canal spills on a daily basis and reports them online. These data were downloaded from the USGS (2010b).

PVID benefits from a network of 32 rain gauges distributed uniformly across the entire district. This means that there is a rain gauge for every $13.7{\mathrm{km}}^2$ of PVID’s cultivated land ($439{\mathrm{km}}^2$). The point measurements of rain gauges were interpolated in ArcGIS to develop precipitation maps for each event.

PVID has a network of 260 piezometers distributed uniformly across the entire district (one piezometer for every $1.7{\mathrm{km}}^2$ of PVID’s cultivated land). Depth to groundwater was measured at each piezometer once every month. These point measurements were collected, analyzed for quality, and then interpolated (Krigin method) in the ArcGIS environment to generate monthly maps of actual groundwater depth (${\mathrm{GWD}}_a$).

The surface energy balance algorithm for land (SEBAL) Idaho implementation was used to map actual evapotranspiration (${\mathrm{ET}}_{\mathit{a}}$) across the study area (Bastiaanssen et al. 1998b). The performance of SEBAL has been assessed and validated in numerous applications (Bastiaanssen et al. 1998a; Ramos et al. 2009), showing a high accuracy at field and catchment scales (Bastiaanssen et al. 2005). In SEBAL, net radiation (${\mathit{R}}_{\mathit{n}}$) is estimated through quantifying all of the incoming and outgoing short- and long-wave radiation components. Once ${\mathit{R}}_{\mathit{n}}$ is determined, soil heat flux (G) is modeled as a ratio of net radiation and a function of surface temperature and fraction of vegetation cover.

The innovation of SEBAL is in mapping sensible heat flux (H). This parameter is estimated for two extreme pixels and then interpolated for all other pixels. One extreme pixel (cold pixel) represents a well-watered crop at full cover, where the available energy is used mainly for changing the phase of water from liquid to gas, leaving a negligible amount of energy to be used in generating a temperature gradient (H about zero). The other extreme pixel (hot pixel) represents a dry, bare soil, where the available energy is used primarily in heating the atmosphere (maximum H). The final step of SEBAL involves estimating the latent heat flux (LE) as the remainder of the surface energy balance equation

Space- or airborne imagery—as input data to models such as SEBAL—provides only a snapshot of LE at the time of overpass. As a result, instantaneous estimates of ET need to be scaled up to longer periods (daily and seasonal) for most practical purposes. In the earlier versions of SEBAL, instantaneous ET was extrapolated to daily values using the evaporative fraction (EF or $\mathrm{\Lambda }$). This concept is based on the assumption that the ratio of instantaneous ET to instantaneous available energy (${\mathit{R}}_{\mathit{n}}\mathit{-}\mathit{G}$) is constant during the day (Brutsaert and Sugita 1992; Crago 1996), especially under cloud-free conditions (Zhang and Lemeur 1995). Once this ratio is determined, daily ET could be calculated by multiplying the EF ratio and the daily value of available energy.

Although the EF technique has provided reliable results in many studies (Gowda et al. 2008), its accuracy decreases in arid regions because of the occurrence of afternoon advection. Trezza (2002) modified the EF ratio by replacing available energy with alfalfa reference ET (${\mathrm{ET}}_r$), which encompasses the effect of energy imported from dry neighboring areas. Upscaled daily ET estimates using this new method (${\mathrm{ET}}_{r}F$) have been shown to be significantly improved (Romero 2004; Allen et al. 2007a, b). Alternatively, grass-based reference ET (${\mathrm{ET}}_o$) has been used in extrapolating instantaneous ET values (${\mathrm{ET}}_{o}F$ method). Colaizzi et al. (2006) compared estimates of five different upscaling techniques with measurements of precision weighing lysimeters at Bushland, Texas, where strong advection of heat is common. For cropped surfaces, the ${\mathrm{ET}}_{\mathrm{o}}\mathrm{F}$ method worked better than ${\mathrm{ET}}_{r}F$ and EF methods. Chávez et al. (2008) also indicated that, under advective conditions, ${\mathrm{ET}}_{o}F$ outperforms ${\mathrm{ET}}_{r}F$ and EF methods. Since PVID is surrounded by dry desert regions, ${\mathrm{ET}}_{o}F$ was selected as the up-scaling method in this study.

All cloud-free Landsat TM5 scenes collected between January 2008 and January 2009 were downloaded from the website of the USGS Global Visualization Viewer (GLOVIS) (USGS 2010a). These scenes were all processed using the LPGS processing system, which results in 60-m resolution for the thermal band and 30-m resolution for other bands. PVID is located on the overlap zone of two Landsat paths (38 and 39; Row 37). This allowed the acquisition of six extra scenes from Path 39 in addition to 15 scenes from Path 38 (total of 21 scenes). Required weather parameters (including ${\mathrm{ET}}_o$) were acquired from a California Irrigation Management Information System (CIMIS) weather station located within PVID (Station 135).

The concept of potential ET (${\mathrm{ET}}_p$) has been used in estimating some performance indicators. However, the exact definition of this parameter and the approach to estimating it have been debated among researchers. In this study, ${\mathrm{ET}}_p$ was considered as the level of water consumption that a specific crop could potentially reach. This is similar to the concept of standard (nonstressed) crop ET (${\mathrm{ET}}_c$) in that both are estimated for stress-free conditions, when the amount of crop water consumption is limited by energy availability. Two distributed ${\mathrm{ET}}_p$ estimation methods have been used before in evaluating performance over large and heterogeneous irrigation schemes: available energy (${\mathit{R}}_{\mathit{n}}\mathit{-}\mathit{G}$) method, implemented by Roerink et al. (1997) and Bastiaanssen et al. (1996), and the Priestley–Taylor approach (Priestley and Taylor 1972), utilized by Bastiaanssen et al. (2001), Bandara (2003), and Karatas et al. (2009). A major concern with both methods is that only radiative energy is considered and the advective energy is neglected (Glenn et al. 2007). As explained before, irrigated areas in arid/semiarid regions frequently experience advective conditions. When water is available, the effect of advection on enhancing ET would be large in these regions, resulting in significant errors if neglected.

In this study, ${\mathrm{ET}}_p$ was estimated on a pixel-by-pixel basis using the Priestley–Taylor (PT) equation (Priestley and Taylor 1972)where $\alpha$ = Priestley–Taylor parameter; $\mathrm{\Delta }$ = slope of the water vapor saturation curve; $\gamma$ = psychrometric constant; and other terms are defined as previously. To account for the effect of advection, the above equation was calibrated for the local condition of PVID through a process similar to what was implemented by Diaz-Espejo et al. (2005). In this process, the original $\alpha$ value of 1.26 was modified to force the PT equation to estimate ET rates similar to what was estimated over a reference grass surface, using the Penman-Monteith equation (Allen et al. 1998).

The WCU is a measure of irrigation equity (Bastiaanssen and Bos 1999), assessed through estimating the coefficient of variation for ${\mathrm{ET}}_a$ at two levels, namely, within (${\mathrm{CV}}_w$) and among (${\mathrm{CV}}_s$) irrigation units. In order to extract required statistical parameters for each field in PVID, a crop classification layer was used in ArcGIS as a mask layer. Developed from Landsat TM5 imagery by the U.S. Bureau of Reclamation (USBR), this classification layer defined field boundaries and provided information on the cropping pattern in 2008. Since the thermal band of Landsat TM5, used in mapping ${\mathrm{ET}}_a$, has a coarser resolution than the other six bands, an inner buffer of 60 m was applied to field boundaries to avoid including edge pixels in analysis. The mean and standard deviation of ${\mathrm{ET}}_a$ were obtained for every field that had an area larger than 4.0 ha after applying the buffer. This reduced the number of studied units from more than 2,000 fields to 1,485.

RET was estimated as the ratio of ${\mathrm{ET}}_a$ to ${\mathrm{ET}}_p$. This indicator was used as a measure of irrigation adequacy to investigate the presence and severity of water shortages. The interpretation of results was accomplished at two levels: including all fields and limiting to fields at full canopy cover. Elimination of the fields that were not close to full cover was achieved by setting a threshold value of 0.65 on the soil adjusted vegetation index [SAVI (Huete 1988)]. All fields with an average SAVI less than this threshold were excluded from the second level of analysis.

DF is a rather new term for an old concept that has been given several names such as irrigation efficiency, water efficiency, water application efficiency, and consumptive use coefficient. All of these terms provide information on the fraction of water that has been depleted from available resources (Jensen 2007). According to Bos et al. (2005), DF is defined aswhere ${\mathit{P}}_{\mathit{g}}$ = gross precipitation over the study area; and V = volume of applied (${\mathit{V}}_{\mathit{a}}$) or diverted (${\mathit{V}}_{\mathit{d}}$) water. For the sake of developing benchmarks and comparing irrigation schemes, it is important to differentiate between applied and diverted water, because operational and maintenance constraints may impose a large variation in the amount of canal spills. Estimated DF would be gross ($\mathrm{D}{\mathrm{F}}_g$) or net ($\mathrm{D}{\mathrm{F}}_n$), if ${\mathit{V}}_{\mathit{d}}$ or ${\mathit{V}}_{\mathit{a}}$ were used in the above equation, respectively.

The DR is another performance indicator that provides information on what portion of applied water has left the study area in the form of drainage. DR is inversely related to ${\mathrm{DF}}_n$ (Bos et al. 2005)

The leaching fraction (LF) necessary for maintaining a favorable salt balance can be taken as a critical value of DR. If estimated DR is less than required LF, soil salinization may become problematic in the future. A DR value greater than LF means that applied water can be reduced without affecting the current level of agricultural production. Adjusting irrigation management to reduce DR may not have any effect on the quantity of water available to downstream users, but the effect on preserving its quality could be significant.

The DDU was introduced in this study for the first time. DDU is assessed by evaluating among and within field coefficient of variation (${\mathrm{CV}}_s$ and ${\mathrm{CV}}_w$) of the actual groundwater depth. A low DDU is only an indication of uniform depth to groundwater and does not necessarily mean that water is at a depth that allows an adequate root respiration. Therefore, this indicator should be studied along with other drainage performance indicators, such as relative groundwater depth (explained subsequently). In areas where the crop water requirement is met by controlling the level of groundwater, DDU could serve as a measure of traditional irrigation distribution uniformity.

Bos (1997) proposed the RGD as the ratio of actual groundwater depth ($\mathrm{G}\mathrm{W}{\mathrm{D}}_a$) to critical groundwater depth ($\mathrm{G}\mathrm{W}{\mathrm{D}}_c$). Although it was defined over a decade ago, RGD has never actually been estimated over any irrigated area, to the best of the authors’ knowledge. Since for most crops, groundwater should be kept at levels below the root zone to avoid any negative effects caused by waterlogging and/or soil salinization, it is assumed that $\mathrm{G}\mathrm{W}{\mathrm{D}}_c$ is equal to the effective root depth for every crop group. Table 22 in Allen et al. (1998) suggests a range of effective depths for each crop. In this study, the upper limit of this range was selected as $\mathrm{G}\mathrm{W}{\mathrm{D}}_c$. These values were assigned to each field in PVID, using the crop classification layer. Then, the minimum value of $\mathrm{G}\mathrm{W}{\mathrm{D}}_a$ was obtained for each field from the interpolated maps of depth to groundwater for the months of February and September, when groundwater is farthest from and closest to the surface, respectively. RGD estimates based on minimum $\mathrm{G}\mathrm{W}{\mathrm{D}}_a$ and maximum $\mathrm{G}\mathrm{W}{\mathrm{D}}_c$ (as opposed to average values) are more conservative because they represent the portion of the field that has the highest potential for waterlogging issues. Values less than unity indicate that the water table is within the crop root zone, a situation that irrigation managers should avoid for most crop types.

The volume of water diverted from the Colorado River for irrigating PVID lands between January 2008 and January 2009 was $\mathrm{1,088}\mathrm{million}{\mathrm{m}}^3$ or 2,479 mm when divided by the total cultivated area. The monthly average flow rates ranged between 14.5 and $49.3{\mathrm{m}}^3/\mathrm{s}$ in January and June, respectively. With a volume of about $125\mathrm{million}{\mathrm{m}}^3$, canal operational spills were 11.5% of the total annual water diversion. Monthly average canal spills measured at discharge structures varied from 2.3 to $4.5{\mathrm{m}}^3/\mathrm{s}$ in January and June, respectively. It is very important to take the amount of canal spills into account when evaluating system performance, since this water returns to the river with little to no changes in quality. Monthly average flow rates of drainage discharge measured at the downstream end of the outfall drain ranged between $10.2{\mathrm{m}}^3/\mathrm{s}$ in January and $16.2{\mathrm{m}}^3/\mathrm{s}$ in July, which shows a lag in comparison with the peak flow rate of diversion. With a volume of about $438\mathrm{million}{\mathrm{m}}^3$, the total annual amount of drainage was 40% of diversion. This portion increases to over 45% if only applied water (diverted minus spills) is considered.

Twenty-five precipitation events were recorded during the study period, all in forms of rainfall. Out of all events, only two had an average depth greater than 10 mm. The total annual precipitation depth was 71 mm, underscoring the aridity of this region.

Depth to groundwater was affected by irrigation intensity. The deepest levels were observed in February 2008, after irrigations had been ceased for some time. Diversion to the main canal was also shut down for maintenance purposes during the first week of January. The average ${\mathrm{GWD}}_a$ was 3.3 m in February, which was equal to the average depth of open drains. After February, groundwater levels rose until they reached 2.8 m from the surface in September and October 2008, which was toward the end of the main irrigation season. From this point, ${\mathrm{GWD}}_a$ declined until it reached 3.2 m in January 2009. Fig. 2 demonstrates cumulative frequency curves developed from interpolated maps of ${\mathrm{GWD}}_a$ for the 3 months of February, June, and September 2008. These months were selected to show three different levels of groundwater from lowest to highest. According to Fig. 2, no PVID groundwater levels were closer than about 2.0 m to the soil surface, an indication of a successfully functioning drainage system.

On average, the annual ${\mathrm{ET}}_a$ over PVID’s cultivated land ($439{\mathrm{km}}^2$) was 1,286 mm. This was similar to the annual ET estimate of 1,268 mm, reported by Taghvaeian and Neale (2011) over the same period based on a district-wide water balance approach. Annual field-level ${\mathrm{ET}}_a$ ranged from less than 70 mm for fallow fields to more than 2,000 mm for alfalfa fields. Water consumption of other dominant crops was less than alfalfa at 1,063 and 1,320 mm for Sudan grass and cotton, respectively. Fig. 3 presents a SEBAL-ET map on September 17, 2008, as an example of the 21 processed images.

Crop coefficients (${\mathit{K}}_{\mathit{c}}$) were determined by dividing SEBAL-${\mathrm{ET}}_a$ by the grass-based reference ET estimated at the CIMIS weather station. The frequency distribution and cumulative frequency of ${\mathit{K}}_{\mathit{c}}$ values over PVID are demonstrated in Fig. 4 for two satellite overpass dates of April 8 and July 13, 2008. On both dates, the most frequent ${\mathit{K}}_{\mathit{c}}$ occurred at about 1.1, which is about the ${\mathit{K}}_{\mathit{c}}$ of alfalfa and cotton at peak water use. Since alfalfa has a year-round growing season, the April 8 ${\mathit{K}}_{\mathit{c}}$ distribution was mainly controlled by alfalfa. The frequency distribution changed on July 13 for ${\mathit{K}}_{\mathit{c}}$ values smaller than 0.2 and larger than 1.0, but it had minimal differences for values in between these two limits. This evolution can be attributed to the growth of other crops such as cotton. Cotton is planted in mid-March in PVID, so its water consumption is still small in early April (${\mathit{K}}_{\mathit{c}}$ values less than 0.25). By mid-July, however, most of the cotton fields are at full cover, consuming water at the highest rates. As a result, the low-${\mathit{K}}_{\mathit{c}}$ frequencies move to a range between 1.0 and $1.2.{\mathit{K}}_{\mathit{c}}$ values larger than 1.2 appeared to belong to recently irrigated fields.

In estimating ${\mathrm{ET}}_p$, the PT parameter, $\alpha$, was adjusted to include the effect of advective energy on enhancing and/or suppressing water consumption. Adjusted values were estimated through dividing the Penman-Monteith ${\mathrm{ET}}_o$ by the equilibrium ET. Fig. 5 shows adjusted daily $\alpha$ values for the period from March 1 to October 1, a time frame that represented the main agricultural growing season and when advective energy was expected to be significant.

Out of 215 days of analyzed data, only one day had an α value less than 1.26. On this date (May 24) PVID rain gauges recorded an average precipitation depth of 14.4 mm, the most intensive rainfall during the study period. Therefore, it is highly probable that moist and cool air from surrounding deserts converged over PVID, resulting in water consumptions lower than what was predicted by the available energy. Except for 3 days, adjusted $\alpha$ values were all above 1.4, suggesting that advection was a major contributor to water consumption of PVID crops. Several previous studies reported $\alpha$ values greater than 1.4 (Diaz-Espejo et al. 2005; Pereira and Nova 1992) and 1.5 (Li and Yu 2007), offering that such values indicate the occurrence of enhanced advective conditions.

Within field coefficient of variability (${\mathrm{CV}}_w$) of annual ${\mathrm{ET}}_a$ ranged from 1.0 to 80%, with an average and median of 7.0 and 3.4%, respectively. Although variabilities as high as 80% were detected, 85% of the fields had a ${\mathrm{CV}}_w$ lower than 10%. Using maps of ${\mathrm{CV}}_w$, PVID managers can locate the remaining 15% of the fields with ${\mathrm{CV}}_w$ values larger than 10% and focus their attention on these flagged fields rather than the entire district. This is an example of how the distributed nature of remotely sensed indicators can save time and energy in evaluating and improving scheme-wide irrigation performance. The low values of average ${\mathrm{CV}}_w$ suggest that, overall, water application at the field level was uniform across PVID. A primary reason could be that almost all PVID fields are precisely leveled, using modern laser grade-control systems. Among the fields, the coefficient of variation (${\mathrm{CV}}_s$) for PVID fields was 38.2%, but this large value does not translate into poor irrigation equity since PVID fields are highly diverse in crop type, growing season, and water requirement. Table 1 compares the CV estimates found in this study with the reported values in the literature.

Crop-specific water consumption uniformity was also studied over cotton and alfalfa fields. For 22 large cotton fields, ${\mathrm{CV}}_w$ and ${\mathrm{CV}}_s$ were 3.2 and 8.5%, respectively. The uniformity was also promising for 45 large alfalfa fields in PVID, with ${\mathrm{CV}}_w$ and ${\mathrm{CV}}_s$ values of 3.1 and 9.4%, respectively. According to Molden and Gates (1990), a ${\mathrm{CV}}_s$ of less than 10% is considered good uniformity. Santos et al. (2008) reported a higher variability for 13 cotton fields in the Genil-Cabra Irrigation Scheme (GCIS) in southern Spain (${\mathrm{CV}}_w=5\%$ and ${\mathrm{CV}}_s=12\%$). GCIS fields were under a modern pressurized system with an on-demand delivering regime, while PVID fields were under surface irrigation and a modified-demand water delivery.

Average RET for all PVID fields was 0.97 when an $\alpha$ value of 1.26 was used. This is a high RET, suggesting that on average all studied fields were consuming water at rates similar to their potential. When considering only full-cover fields, RET increased further to 1.49, meaning that these fields had water consumption about 50% larger than their potential. Adjusting α for advective conditions reduced the average RET of all fields and full-cover fields to more realistic estimates of 0.69 and 1.06, respectively (Fig. 6). Estimating ${\mathrm{ET}}_a$ values that are 6% larger than ${\mathrm{ET}}_p$ on average is acceptable since $\alpha$ was calibrated with grass-based ${\mathrm{ET}}_o$, while the ET of alfalfa (the most dominant crop in PVID) can be up to 20% larger than grass under the same agroclimatological conditions.

Assuming that ${\mathrm{ET}}_p$ was equal to available radiative energy (${\mathit{R}}_{\mathit{n}}\mathit{-}\mathit{G}$), Roerink et al. (1997) reported average RET values of about 0.6 for several secondary and tertiary units in the Rio Tunuyan irrigation scheme in Argentina. Roerink et al. suggested that RET values of 0.75 and higher were satisfactory for irrigated agriculture. Using the $\alpha$ value of 1.26 (PT original), Karatas et al. (2009) estimated an average RET of 0.7 for the period of May to September over the Lower Gediz Basin (LGB) in Turkey, which is similar to PVID in terms of irrigation method and main crops. It may seem appropriate to compare RET estimates of LGB with the RET estimate of PVID before adjusting α. Such a comparison would favor PVID, with an average RET of 0.97. However, the effect of advection on ET may have been different for these two irrigation schemes. Although the amounts of precipitation during the study periods were similar for both regions (about 30 mm), advective conditions may have been less prevalent under the Mediterranean climate of LGB. If that was the case, comparing the average RET of LGB with the average RET of PVID after adjusting α may be more reasonable (0.7 versus 0.69). The same logic applies to comparing the RET estimate of PVID to that of the Nilo Coelho (0.76), a pressurized irrigation scheme in Brazil (Bastiaanssen et al. 2001). The results of the present study were also similar to those reported for irrigated winter wheat in central Morocco, with an average RET of 0.68 (Kharrou et al. 2013), as well as a flood-irrigated region in Uzbekistan where RET decreased from over 1.0 to less than 0.75 as the distance to the canal intake increased from less than 15 km to over 95 km (Conrad et al. 2013).

Gross and net DF were estimated on a daily basis and averaged for each month in 2008. $\mathrm{D}{\mathrm{F}}_g$ ranged between 0.33 in December and 0.58 in May, with an annual average of 0.49. As expected, $\mathrm{D}{\mathrm{F}}_n$ values were larger, varying from 0.41 in December to 0.64 in May, with an annual average of 0.55. Fig. 7 demonstrates the intermonthly variation of $\mathrm{D}{\mathrm{F}}_g$ and $\mathrm{D}{\mathrm{F}}_n$ in 2008.

The temporal pattern of monthly $\mathrm{D}{\mathrm{F}}_n$ was different for irrigated areas in Uzbekistan, with values decreasing from April to June (minimum of about 0.4) and then increasing until September when they exceeded 0.8 (Conrad et al. 2013). For the Nilo Coelho irrigation scheme, Bastiaanssen et al. (2001) reported $\mathrm{D}{\mathrm{F}}_n$ values ranging between 0.4 in April and 0.85 in November, with an annual average of 0.61. For the condition of this irrigation scheme (perennial orchards under pressurized irrigation), $\mathrm{D}{\mathrm{F}}_n$ values beyond the operational range of 0.7–1.0 were found to result in at least 10% reduction in the yield. Three irrigation districts in central Morocco had January–February ${\mathrm{DF}}_g$ values similar to those estimated in the present study, but significantly larger values (above 0.8) in March and April (Kharrou et al. 2013). For the semiarid LGB in Turkey, Karatas et al. (2009) estimated ${\mathrm{DF}}_g$ values larger than 3.0 in May and September. The authors claimed that a depletion that was three times larger than diverted water was achieved through consumption of soil moisture stored in the root zone. ${\mathrm{DF}}_g$ values dropped to as low as 0.28 in July, resulting in an average ${\mathrm{DF}}_g$ of 0.69 for the whole study period (May to September) over the LGB. Another study found similar average ${\mathrm{DF}}_g$ values of 0.60 and 0.72 for cotton and grape fields, respectively, in a subdivision of the LGB (Droogers and Bastiaanssen 2002).

The average annual DR over PVID was 0.45, ranging from 0.36 in May to 0.59 in December. This was smaller than the average seasonal DR reported for three irrigation subdivisions in Uzbekistan, ranging from 0.55 to 0.65 (Conrad et al. 2013). Since no soil salinity study has been carried out in PVID, there is no information on leaching requirements to be used in assessing DR estimates. However, it may be safely assumed that the leaching requirement was not greater than 10–15% in PVID, since its water source had a low salinity (electrical conductivity of the Colorado River water measured by USGS above the diversion dam never exceeded $1.08\mathrm{dS}·{\mathrm{m}}^{-1}$ in 2008). If this was the case, the fact that drainage was 45% of applied water may be an indication of overirrigation in this area. Reducing the amount of applied water may have significant effects on downstream ecosystems because of the elevated levels of nutrients in irrigation return flows.

The ${\mathrm{CV}}_s$ and ${\mathrm{CV}}_w$ of the actual groundwater depth were estimated for PVID fields for the 2 months of February and September, representing periods when groundwater was farthest from and closest to the soil surface, respectively. The ${\mathrm{CV}}_s$ was 13.4% in February and 15.5% in September. The ${\mathrm{CV}}_w$ was significantly smaller, with average values of 0.9 and 1.1% for February and September, respectively. The uniformity of $\mathrm{G}\mathrm{W}{\mathrm{D}}_a$ within and across PVID fields suggests that the extent and density of the existing drainage network is appropriate for removing excess water from the top soil. However, the adequacy of drainage depth cannot be assessed based on DDU and requires estimating the next performance indicator.

Similar to DDU, RGD was estimated for the 2 months of February and September. Not a single field had a RGD value less than unity in either month, meaning that the groundwater level was always below the maximum depth that crop roots could reach. However, the distribution of RGD values was different between the 2 months (Fig. 8). For example, only 2.5% of the fields had a RGD between 1.0 and 1.5 in February, but this increased to about 27% in September. No significant correlation was found between RGD and distance to drain, estimated in ArcGIS as the distance between the centroid of each field and the closest drain. This means that expanding the existing drainage network will not have any major effect on the depth of drained soil.

Spatially distributed information on drainage performance indicators generated by GIS techniques can significantly assist irrigation managers with locating the fields that are at higher sustainability risks. Focusing only on these fields would result in a more efficient use of resources to improve drainage performance. Fig. 9 demonstrates two examples of such maps, one based on DDU (${\mathrm{CV}}_w$ of $\mathrm{G}\mathrm{W}{\mathrm{D}}_a$) and the other based on RGD, where the red color specifies fields that may have drainage issues. Any mathematical combination of two or more drainage performance indicators can also be generated (e.g., by assigning different weights) based on the importance of each factor in future sustainability of crop production.