wheat irrigation management using multispectral crop ... - naldc - USDA

1 downloads 0 Views 2MB Size Report
D. J. Hunsaker, G. J. Fitzgerald, A. N. French, T. R. Clarke, M. J. Ottman, P. J. Pinter Jr. ABSTRACT ...... square error (RMSE), mean absolute error (MAE; Legates.
WHEAT IRRIGATION MANAGEMENT USING MULTISPECTRAL CROP COEFFICIENTS: I. CROP EVAPOTRANSPIRATION PREDICTION D. J. Hunsaker, G. J. Fitzgerald, A. N. French, T. R. Clarke, M. J. Ottman, P. J. Pinter Jr.

ABSTRACT. A method widely used for irrigation management determines crop evapotranspiration (ETc ) from reference evapotranspiration (ETo ) calculations and estimated crop coefficients. However, standard time‐based crop coefficients may fail to represent the actual crop water use, for example, when deviations in weather or agronomic constraints appreciably change crop development patterns from typical conditions. In this study, the FAO‐56 dual crop coefficient procedures were applied during experiments with wheat to calculate the estimated ETc for irrigation scheduling. The objective of this research was to determine whether basal crop coefficients (Kcb ) determined from a normalized difference vegetation index (NDVI treatment) improve the prediction of ETc over a standard application with a locally developed time‐based Kcb curve (FAO treatment). The experiments conducted for two seasons in central Arizona included subtreatments, equally replicated within the NDVI and FAO treatments, of three plant densities (typical, dense, and sparse) and two nitrogen levels (high and low) to provide a range of crop development and water use conditions. The effects of plant density and N level resulted in significant differences in measured seasonal ETc . Large variations that occurred in the observed Kcb and ETc trends between subtreatments were better correlated with the NDVI than the FAO treatment. The mean absolute percent difference for predicted ETc was significantly smaller for NDVI than FAO during both seasons. The treatment difference was 5% for the first season, but 10% for the second season when an unexpected early decline in ETc and Kcb was effectively predicted by the NDVI treatment but not by the FAO treatment. NDVI appears to be a robust approach for Kcb estimation of wheat, able to reliably predict actual ETc for both typical and abnormal water use conditions. Keywords. Crop canopy reflectance, Irrigation water requirements, NDVI, Normalized difference vegetation index, Soil water balance.

L

imited water availability due to the rising demand from urban growth and environmental restoration requires farmers in the western U.S. to apply irriga‐ tion water to crops more efficiently. A key feature for efficient irrigated crop management is scheduling the proper timing and applying the correct quantity of irrigation water. Scientific irrigation scheduling technologies have been available for several decades, and grower usage of these methods has increased recently in some irrigated regions (Leib et al., 2002). However, as Howell (1996) suggested, significant advances in irrigation scheduling are needed to

Submitted for review in September 2006 as manuscript number SW 6690; approved for publication by the Soil & Water Division of ASABE in August 2007. Mention of company or trade names does not imply endorsement by the USDA. The authors are Douglas J. Hunsaker, ASABE Member Engineer, Agricultural Engineer, USDA‐ARS U.S. Arid Land Agricultural Research Center, Maricopa, Arizona; Glenn J. Fitzgerald, Senior Scientist (Remote Sensing), Department of Primary Industries, Horsham, Victoria, Australia; Andrew N. French, Physical Scientist, and Thomas R. Clarke, Physical Scientist, USDA‐ARS U.S. Arid Land Agricultural Research Center, Maricopa, Arizona; Michael J. Ottman, Extension Agronomist, Plant Sciences Department, University of Arizona, Tucson, Arizona; and Paul J. Pinter Jr., Former Research Biologist, USDA‐ARS U.S. Water Conservation Laboratory, Phoenix, Arizona. Corresponding author: Douglas J. Hunsaker, USDA‐ARS U.S. Arid Land Agricultural Research Center, 21881 N. Cardon Ln., Maricopa, AZ 85238; phone: 520‐316‐6372; fax: 520‐316‐6330; e‐mail: [email protected].

meet the limited water challenges facing growers today. Spe‐ cific research priorities recommended include developing improved crop evapotranspiration (ETc) estimation methods and capabilities to assess the spatial needs for water (Howell, 1996). Modern scientific irrigation scheduling, as practiced in the western U.S., combines ETc and irrigation data within soil water balance calculations of the crop root zone to deter‐ mine the quantity and frequency of irrigation water applica‐ tions (Martin and Gilley, 1993). Various methods are used to estimate ETc, but the most popular and widely used technique relies on empirical crop coefficients (Kc) (Jensen and Allen, 2000). In this methodology, described in detail by the Food and Agriculture Organization (FAO) of the UN, Paper 56 (FAO‐56; Allen et al., 1998), the evapotranspiration of a ref‐ erence crop (a uniform, well‐watered grass, or ETo) is calcu‐ lated from climatic data (solar radiation, air temperature, humidity, and wind speed) by the Penman‐Monteith equa‐ tion, and the ETc of all other crops is related as a proportion to this reference by Kc. More precise estimates of ETc are needed for real‐time irrigation scheduling and require sepa‐ rating the single Kc into the transpiration and evaporation components of ETc (Allen et al., 1998). The combined ex‐ pression for calculating daily ETc is given as: ETc = (Kcb Ks + Ke) ETo

(1)

where the basal crop coefficient (Kcb) represents the transpi‐ ration portion of ETc, Ke is the wet soil evaporation coeffi‐

Transactions of the ASABE Vol. 50(6): 2017-2033

2007 American Society of Agricultural and Biological Engineers ISSN 0001-2351

2017

cient, Ks is the water stress coefficient (Ks < 1 when the available soil water is insufficient for full ETc, and Ks = 1 when there is no soil water limitation on ETc), and ETo is ref‐ erence evapotranspiration. The procedures for determining the coefficients and calculation of equation 1, referred to as the dual crop coefficient procedures, are fully described in FAO‐56 (Allen et al., 1998). Ground‐based weather station networks, such as the California Irrigation Management In‐ formation System (CIMIS; www.cimis.water.ca.gov/cimis) and the Arizona Meteorological Network (AZMET; www.ag.arizona.edu/azmet), have been established in many western U.S. states and around the world and provide daily ETo data in support of crop coefficient irrigation scheduling. Sets of empirical crop coefficients have been derived or estimated for many agricultural crops and are provided in many sources, e.g., Allen et al. (1998). However, Doorenbos and Pruitt (1977) and Allen et al. (1998) have emphasized that these tabulated crop coefficients need localized adjust‐ ments and are specifically intended to only describe ETc for disease‐free crops grown under standard management prac‐ tices, having no growth and yield restrictions imposed by fac‐ tors such as soil water or fertility limitations or high soil salinity. Standard application of crop coefficients requires the use of an independent parameter (or driver) to describe canopy development during the growing season. Primary pa‐ rameters that are typically used in irrigation scheduling in‐ clude the elapsed time from planting (Doorenbos and Pruitt, 1977; Allen et al., 1998), percentage of time from planting to full cover and from full cover to harvest (Wright, 1982), and accumulated growing degree days (GDD) from planting (Stegman, 1988). However, even when crops are grown un‐ der excellent agronomic conditions, actual crop coefficient curves are subject to shifting along any of these indices from year to year. Profound shifts can occur particularly during early crop development when actual climate conditions vary from average or expected conditions (Bausch, 1995). In Phoenix, Arizona, Hunsaker et al. (2002) demonstrated the need to carefully adjust time‐based Kcb curves during each cutting cycle of alfalfa to account for the changes in climatic conditions that occur during the year in that region. Variabili‐ ty in optimum Kcb curves can also be increased when new or non‐standard crop varieties are grown (Martin and Gilley, 1993). Moreover, cropped fields, or portions within cropped fields, often deviate significantly from the optimum standard agronomic practice. In some cases, variations from standard conditions are forced due to the management strategy, such as imposing mild water stress to stimulate crop reproduction. On the other hand, there is an assortment of other cultural, managerial, and environmental factors that cause apprecia‐ ble changes from normal crop development rates and hence crop coefficient curves. For example, Hunsaker et al. (2005a) reported that increasing or decreasing the planting density and fertilizer application rates for cotton significantly altered the seasonal patterns of Kcb and water use during controlled experiments. At the field and farm‐scale level, non‐ uniformity of applied water and nutrients, variability in soil water retention, occurrence of crop disease, and other envi‐ ronmental limitations can introduce considerable spatial and temporal variations in crop evapotranspiration. When a re‐ duction or deviation from normal growth is caused by any of the various factors mentioned above, the crop coefficients should be adjusted. One of the most common approaches is to adjust the Kcb in relation to observed growth. For example,

2018

Allen et al. (1996) suggest adjustments based on measured changes for leaf area index (LAI) or fractional crop cover rel‐ ative to that for optimal vegetation. Such adjustments should provide reasonably good results since the basal crop coeffi‐ cient value changes in proportion with the amount of actively transpiring canopy (Jensen et al., 1990). However, for most commercial farming situations, adjusting Kcb based on LAI or cover measurements is probably impractical due to the amount of manually collected data that are needed over large areas within a relatively narrow time‐frame. Remote sensing (RS) measurements could provide a prac‐ tical way to obtain timely data needed to make appropriate adjustments for Kcb (Pinter et al., 2003). Over the years, re‐ mote sensing scientists have clearly established relationships between various multispectral vegetation indices (VIs), such as the normalized difference vegetation index (NDVI), de‐ rived from optical visible and near‐infrared data, and various biophysical aspects of vegetation canopies, such as leaf area index (Moran et al., 1995), crop yield (Plant et al., 2000), and percent crop cover (Heilman et al., 1982). Additional re‐ search has shown that the multispectral VIs can provide real‐ time surrogates of crop coefficients for a variety of crops (Bausch, 1995; Neale et al., 2003; Hunsaker et al., 2005a; Johnson and Scholasch, 2005). Therefore, VIs obtained with remote sensing data potentially offer a means to infer in near real‐time the spatial distribution of Kcb across the landscape of a local field or on a broader farm‐scale basis. Such infor‐ mation would greatly reduce the need to conduct labor‐ intensive crop surveys to obtain appropriate crop coefficients. An increasing number of U.S. growers are al‐ ready buying and using NDVI‐based products to monitor spa‐ tial or temporal variations in plant vigor or stress; expansion of the remote sensing capability to irrigation scheduling would add significant value to this investment. Hunsaker et al. (2005b) developed a model for predicting the Kcb of wheat from NDVI data. In the present research, an application of the wheat Kcb model was implemented within the framework of the FAO‐56 dual crop coefficient proce‐ dures to determine crop evapotranspiration for guiding ir‐ rigation scheduling in field experiments in central Arizona. The experiments, conducted during the 2003‐2004 and 2004‐2005 wheat seasons, included high and low levels of ni‐ trogen application and three levels of wheat plant densities to achieve a wide range of crop growth conditions that may be potentially encountered on commercial fields. The primary objective of this research was to determine if the remote sensing‐NDVI approach provided a significant improvement in Kcb and ETc estimation compared to a “standard‐optimal” Kcb curve approach. A succeeding companion article (Hun‐ saker et al., 2007) evaluates the effects of the predicted irriga‐ tion scheduling on irrigation system performance, final grain yield, and water use efficiency.

METHODS AND MATERIALS EXPERIMENTAL SITE AND PREPLANT FIELD OPERATIONS Wheat irrigation scheduling experiments were conducted for two growing seasons (December through May) in 2003‐2004 and 2004‐2005 on a 1.3 ha field site at The Uni‐ versity of Arizona Maricopa Agricultural Center (MAC) (33° 04′ N, 111° 58′ W, 361 m MSL), in central Arizona. The field soil is mapped as a Casa Grande sandy loam (reclaimed

TRANSACTIONS OF THE ASABE

Ν 90 m

NDL B1

Border plot

FTL B1

NTH B1

FDH B1

FTH B1

NTL B1

NDH B1

NDL B2

NTH B2

FSH B2

NSL B2

NTL B2

FTL B2

FTH B2

FSL B2

Planted and non-planted test area NSL B3

FSL B3

NDH B4

FTL B4

NTH B4

FDH B4

NSH B3

NTH B3

FDL B3

FTH B3

FDL B4

FTH B4

NSH B4

NTL B4

Border plot

FTL B3

Border plot

7m

NTL B3

Boardwalks

91 m

Border plot

Inlet valve

Border plot

Gated pipe

21 m

Border plot

Gated pipe

Border plot

Inlet valve

FSH B1

Border plot

11.2 m

8m

Neutron access tubes

Figure 1. Design of field experiment showing subtreatment assignments within blocks (B1, B2, B3, and B4), border plot areas, layout of irrigation sys‐ tem, and location of neutron access tubes and boardwalks. Full description of subtreatments is in table 1.

fine‐loamy, mixed, superactive, hyperthermic, Typic Natrar‐ gid). The volumetric soil water contents at field capacity (qFC) and wilting point (qWP) for the upper 1.0 m of the soil profile were determined to be 0.24 ±0.04 m3 m-3 and 0.12±0.01 m3 m-3, respectively (Post et al., 1988). Prior to each experiment, the field site was laser‐leveled to zero grade. On 8 December 2003, a preplant fertilizer application of 16:20:0 ammonium phosphate was uniformly incorporated into the soil surface at rates of 36 kg N ha-1 and 45 kg P ha-1. Between the first and second wheat seasons, a sudangrass cover crop was grown during summer 2004 to remove avail‐ able soil N and reduce field variability. Prior to the second wheat season, residual nutrient contents in the upper 0.3 m soil layer were determined in multiple composite samples to guide preplant soil fertility management for the second wheat study. On 21 December, a dry preplant fertilizer mixture was uniformly incorporated into the soil, providing nitrogen, phosphorous, and sulfur at the prescribed rates of 37 kg N ha-1, 89 kg P ha-1, and 444 kg S ha-1, respectively. Sulfur was added to help reduce the soil pH. TREATMENT STRUCTURE AND STATISTICAL DESIGN Thirty‐two treatment plots (each 11.2 × 21 m) were estab‐ lished in the field (fig. 1) and assigned to 12 different experi‐ mental treatments (table 1). The main treatment consisted of two Kcb estimation methods designated as the FAO (F) treat‐ ment and the NDVI (N) treatment, each treatment consisting of 16 plots. Additional subtreatment variables, equally em-

Vol. 50(6): 2017-2033

Table 1. Summary of subtreatments for the 2003‐2004 and 2004‐2005 wheat experiments. Experimental Variables Subtreatment Abbreviation

Kcb Method

Plant Density

Nitrogen Level

No. of Replicates

FSH FSL FTH FTL FDH FDL

FAO (F) FAO (F) FAO (F) FAO (F) FAO (F) FAO (F)

Sparse (S) Sparse (S) Typical (T) Typical (T) Dense (D) Dense (D)

High (H) Low (L) High (H) Low (L) High (H) Low (L)

2 2 4 4 2 2

NSH NSL NTH NTL NDH NDL

NDVI (N) NDVI (N) NDVI (N) NDVI (N) NDVI (N) NDVI (N)

Sparse (S) Sparse (S) Typical (T) Typical (T) Dense (D) Dense (D)

High (H) Low (L) High (H) Low (L) High (H) Low (L)

2 2 4 4 2 2

bedded within the two main treatments, were imposed to create differences in wheat growth and water use responses among plots during the season. Subtreatments included three plant densities designated as typical (T; 150 plants m-2), sparse (S; 75 plants m-2), and dense (D; 300 plants m-2), and two nitrogen fertilization levels designated as high (H) and low (L). The 12 subtreatments were assigned to plots within four equal‐sized blocks (42 × 45 m) within the field site, each block consisting of eight plots (fig. 1). The subtreatment plot locations were unchanged from the first to second wheat sea‐ sons. Within the FAO and NDVI main treatments, each of the

2019

three plant density levels contained an equal number of high and low N plot replicates. The typical (T) density subtreat‐ ments had four randomly assigned replicates (one in each block), whereas the dense (D) and sparse (S) subtreatments had only two replicates within each combination (table 1). Sparse and dense subtreatment replicates were assigned at random to the remaining plots (i.e., after assigning the T plots) with the constraint that any subtreatment combination could not appear twice in the same block. The statistical design was a complete random design with incomplete blocking. Experimental data for seasonal crop evapotranspiration, seasonal irrigation water applied, and fi‐ nal grain yield were analyzed using the General Linear Mod‐ els (GLM) procedure of SAS (SAS, 1998) to determine statistical effects for main treatments, subtreatments, and interactions. Average percent sand content in the upper 1.2 m soil profile was determined for each plot using procedures de‐ scribed by Hunsaker et al. (2005a). The average sand content was included as a covariate in the GLM analyses. The least‐ significant difference option of GLM was used to determine significant differences among means for Kcb method, nitro‐ gen level, and plant density. WHEAT PLANTING, POST‐PLANT OPERATIONS, AND CROP EMERGENCE Hard red spring wheat (Triticum aestivum L., cv. Yecora Rojo) was planted with a 2.0 m wide Wintersteiger grain drill (Wintersteiger AG, Ried im Innkreis, Austria) in rows spaced 0.20 m apart, in a dry soil surface, on 10‐12 December 2003 and 22 December 2004. Target densities for each subtreat‐ ment plot were achieved by separately planting a specified weight of wheat seed to individual plots. Border plots, lo‐ cated on the east and west sides of the field, and certain areas in the center of the field (fig. 1) were planted to wheat at the typical density. Neutron access tubes were then installed to a depth of 3.0m in a central area of the plot at a distance approximately 1.0 m from the plot center (fig. 1). A 0.3 m long time‐domain reflectometry (TDR) probe was installed 0.5 m away from the access tube. Irrigation border dikes were then formed on the four sides of each plot. Raised boardwalks on concrete blocks across the center of the plots provided non‐destructive access (fig. 1). Two gated pipe irrigation systems, 152 mm in diameter, were installed in the E‐W direction extending the length of the field (fig. 1). Irrigation water was controlled by an alfalfa valve located at the west end of each gated pipe system, and gated ports spaced 1.02 m along the pipe were used to control water delivery to individual plots. The irrigation volume for each irrigation event was measured with calibrated in‐line propeller‐type water meters placed at the head of each gated pipe system. Irrigation water was gravity‐fed to the alfalfa valves from a nearby storage reservoir at MAC. On 19 December 2003 and 30 December 2004, light ir‐ rigations (50 to 66 mm) were applied to all subtreatment plots to initiate germination. For the 2003‐2004 experiment, another light irrigation (48 mm) was applied to all plots on 31 December 2003. Additional emergence irrigation was not needed for the second experiment, since heavy rainfall on 3and 4 January 2005 (89 mm) inundated the plots with stand‐ ing water. Standing water was drained from the plots on 5 Jan‐ uary by breaking the plot border dikes and pumping the excess water to a sump. Wheat emergence counts were made

2020

in all plots approximately every other day through 9 January 2004 and 18 January 2005. Plant populations were deter‐ mined by counting emerged seedlings in ten, 1.25‐m long rows, located in a designated final harvest area of all plots. Average dates of 50% crop emergence for subtreatments were about 29 December 2003 and 10 January 2005. Average final plant populations and standard errors for sparse, typical, and dense subplots were 89 ±4, 164 ±3, and 291 ±10 plants m-2 for 2003‐2004, respectively. For 2004‐2005, they aver‐ aged 79 ±4, 156 ±4, and 305 ±10 plants m-2, respectively. WHEAT CANOPY REFLECTANCE MEASUREMENTS Beginning at crop emergence, canopy reflectance factors were measured two to four times per week for all subtreat‐ ment plots (including FAO plots) until harvest (39 dates dur‐ ing 2003‐2004, and 49 dates during 2004‐2005). Measurements were made using a 4‐band Exotech hand‐held radiometer (model BX‐100, Exotech, Inc., Gaithersburg, Md.) equipped with 15° field‐of‐view optics, held in a nadir orientation, 1.5 to 2.0 m above the soil surface. Data were col‐ lected at a morning‐time period corresponding to a nominal solar zenith angle of 57°. For each plot, 24 reflectance ob‐ servations were averaged across a 6 m transect along the north edge of the final harvest area (south of boardwalks, fig.1). Reflectance factors in the red (0.61 to 0.68 mm) and near‐infrared (NIR, 0.79 to 0.89 mm) wavebands were com‐ puted as the ratio of target radiance to time‐interpolated val‐ ues of solar irradiance inferred from frequent measurements of a calibrated, 0.6 × 0.6 m, 99% Spectralon reference panel (Labsphere, Inc., North Sutton, N.H.). The NDVI was com‐ puted as: NDVI = (NIR - red)/(NIR + red) (2) Reflectance measurements obtained on days when there was cloud interference with the direct beam solar insolation or when soils were wet from irrigation or rainfall were not used in computations. The NDVI data for each plot were in‐ terpolated linearly, generating a daily NDVI curve up to the most recent acceptable measurement. A weighted linear re‐ gression model based on the four most recent NDVI measure‐ ments was used for projecting daily NDVI for days past the last measurement for irrigation scheduling. FAO‐56 DUAL CROP COEFFICIENT PROCEDURES AND COEFFICIENT ESTIMATES FOR TREATMENTS Estimated daily ETc was calculated with equation 1 and related FAO‐56 dual crop coefficient procedures using (1) a uniform Kcb curve applied to all FAO subtreatment plots and (2) individual NDVI‐based Kcb for each NDVI subtreatment plot. Measured daily meteorological data, including solar radiation, air temperature, wind speed, humidity, and rain‐ fall, were used to compute daily values for the grass‐ reference evapotranspiration (ETo) using the FAO‐56 Penman‐Monteith (P‐M) equation (Allen et al., 1998). The data were provided by a University of Arizona, AZMET weather station (Brown, 1989) that was located approximate‐ ly 200 m from the field site. The weather station was centrally located within a large turf area isolated from large obstacles. The turf grass was kept under well‐watered conditions and maintained between a height of 0.08 and 0.12 m. During the wheat seasons, an AZMET technician inspected the weather station to ensure it was operating properly. The wheat grow‐ ing degree days (GDD) were calculated by the sine curve

TRANSACTIONS OF THE ASABE

method (Brown, 1991) using upper and lower air temperature thresholds of 30°C and 4.4°C, respectively. The Kcb curve (denoted as the FAO Kcb curve), applied uniformly to all FAO plots, was constructed following FAO‐56 procedures, including Kcb adjustments for the his‐ torical local climate and growth stage lengths developed lo‐ cally for Yecora Rojo wheat based on data collected during three years of previous experimental studies at MAC (Hun‐ saker et al., 2005b). The crop coefficients and expected lengths of the four wheat growth stages for the FAO Kcb curve are provided in table 2. For the NDVI plots, separate daily Kcb curves for each NDVI subtreatment plot were generated from NDVI data. A relationship that described Kcb as a func‐ tion of normalized NDVI (NDVIn), also developed from the previous experiments at MAC (Hunsaker et al., 2005b), was used to calculate Kcb (table 2). Normalizing the NDVI helps reduce the effects of different soil backgrounds on NDVI val‐ ues prior to full ground cover. NDVIn is expressed as: NDVIn = (NDVI - NDVImin)/(NDVImax - NDVImin) (3) where NDVIn is the normalized NDVI, NDVI is the mea‐ sured or interpolated daily value, and NDVImin and NDVImax are the minimum and maximum NDVI values for the crop, respectively. As applied in these experiments, NDVImin was estimated as the average NDVI measured for all plots at approximately 50% crop emergence. This time was chosen to define NDVImin since the vegetative signal was minimal and the soil surface conditions had stabilized (soil roughness changes between planting and emergence inducing slight changes in NDVI). A value of 0.927, estimated for full cover wheat by Hunsaker et al. (2005b), was used for NDVImax. Calculation of equation 1 required daily values of the soil evaporation coefficient (Ke), which were estimated using FAO‐56 procedures. Soil parameters used in calculating Ke were based on survey data for the soil type (table 3). All other soil and crop parameters required in the calculations (table3), with the exception of estimated crop height (hc),

Parameter

Kcb as a function of NDVIn (all growth stages) Kcb = 0.18 + 1.63X - 2.57X2 + 1.93X3, where X = NDVIn[b] Coefficient of Mean Absolute Determination Error (%) RMSE (%) 0.90 11.0 8.7 [a]

Not applicable because Kcb values are interpolated between initial and mid‐season and between mid‐season and end‐of‐season values. [b] K function is valid for all values of NDVI . cb n

followed various estimation procedures described in FAO‐56. Crop height was estimated as a function of accumu‐ lated GDD, developed from weekly crop height measure‐ ments made during the previous wheat experiments with Yecora Rojo. A daily soil water balance of the surface soil layer (Ze) subject to drying by evaporation was computed separately for each subtreatment plot, where the FAO‐56 pa‐ rameter estimation procedures were uniformly applied to all subtreatment plots (table 3). Note that canopy coverage (fc) with the FAO‐56 estimation technique employed was depen‐ dent upon the estimated Kcb value. Consequently, fc values, and hence Ke, were subject to considerable variability among NDVI subtreatment plots for a given day, since Kcb estima‐ tion for NDVI plots were a function of the measured NDVIn during the experiment. This, of course, was not the case for

Table 3. Soil and crop parameters used in the FAO‐56 dual crop coefficient procedures (Allen et al., 1998) to estimate the soil evaporation coefficient (Ke) during the 2003‐2004 and 2004‐2005 wheat experiments. FAO‐56 Abbreviation Value and Unit Source

Soil water content at field capacity Soil water content at wilting point Depth of soil surface evaporation layer Total evaporable water Readily evaporable water Fraction of soil surface covered by vegetation Fraction of soil surface wetted by irrigation Fraction of soil surface wetted by rainfall Crop height

Parameter

Table 2. Parameters describing the locally developed FAO wheat basal crop coefficient (Kcb) curve and a locally developed relationship describing wheat Kcb as a function of normalized NDVI (NDVIn) that were used to estimate the Kcb for subtreatments during the 2003‐2004 and 2004‐2005 experiments. FAO Kcb curve Growth Stage Intervals (days past emergence) Growth Stage Kcb Initial stage 0.15 1 to 14 Development na[a] 15 to 63 Mid‐season 1.17 64 to 109 Late season na 110 to 144 End of season 0.22 na

θFC θWP Ze TEW REW fc fw fw hc

0.24 m3 m-3 0.12 m3 m-3 0.11 m 20 mm 9 mm 0 to 0.98 (unitless) 1.0 (unitless) 1.0 (unitless) 0.05 to 0.7 m

Post et al. (1988) survey data for soil type Post et al. (1988) survey data for soil type FAO‐56 (pg. 144, Allen et al., 1998) FAO‐56 (eq. 73, Allen et al., 1998) FAO‐56 (table 19, Allen et al., 1998) FAO‐56 (eq. 76, Allen et al., 1998) FAO‐56 (table 20, Allen et al., 1998) FAO‐56 (table 20, Allen et al., 1998) Function of GDD (from Hunsaker et al., 2005b)

Table 4. Soil and crop parameters used for all plots in the FAO‐56 dual crop coefficient procedures (Allen et al., 1998) to estimate the water stress coefficients (Ks) during the 2003‐2004 and 2004‐2005 wheat experiments. FAO‐56 Abbreviation Value and Unit Source

Soil water content at field capacity Soil water content at wilting point Effective crop root depth Total available water Soil depletion fraction[a]

θFC θWP Zr TAW p

0.24 m3 m-3 0.12 m3 m-3 0.3 to 1.3 m 36 to 156 mm 0.55 (unitless)

Post et al. (1988) survey data for soil type Post et al. (1988) survey data for soil type Function of GDD (from Hunsaker et al., 2000) FAO‐56 (eq. 82, Allen et al., 1998) FAO‐56 (table 22, p.163, Allen et al., 1998)

[a] 0.55 is the baseline p value for wheat; the FAO‐56 numerical approximation was used to adjust the baseline p value daily for variations in atmospheric conditions.

Vol. 50(6): 2017-2033

2021

FAO subtreatment plots, since all FAO plots were estimated by the same Kcb time‐based curve. Calculation of equation 1 also required daily values of the water stress coefficient (Ks), which were estimated using FAO‐56 procedures. Soil parameters used in calculating Ks were based on survey data for the soil type (table 4). All other soil and crop parameters required in the calculations for Ks (table 4), with the exception of the effective crop root depth (Zr), followed various estimation procedures described in FAO‐56. Effective crop root depth was estimated as a func‐ tion of accumulated GDD, developed using measurements of root biomass extension made during the previous wheat ex‐ periments with Yecora Rojo (Hunsaker et al., 2000). A daily root zone water balance was computed separately for each plot to predict daily soil water depletion of the effective root zone (Dr). To initiate the root zone water balance, soil water depletion was estimated from volumetric soil water content (qv) data for the plots made at the beginning of the seasons. Parameters Dr and Zr (expressed in units of mm), when com‐ bined with field capacity and wilting point of the soil, deter‐ mine the total available water (TAW). A soil depletion fraction (p), which is limited between 0.1 and 0.8, represents the fraction of TAW that can be depleted from the rooting depth before water stress occurs. TREATMENT IRRIGATION SCHEDULING For both years and all subtreatment plots, irrigations were scheduled for the day after Dr from the soil water balance cal‐ culation exceeded 45% TAW of the effective rooting depth. The 45% of TAW was selected as an allowable soil water depletion percentage (SWDp) expected to minimize crop wa‐ ter stress for all treatments. Calculation of SWDp for any day i was expressed as: SWDp,i = 100 × (Dr,i /TAWi )

(4)

Irrigation amounts replaced 100% of the estimated Dr at the time of irrigation, plus an additional 10% to account for the inefficiency of the irrigation system. For both seasons, all 16 FAO subtreatment plots were irrigated on the same days and with approximately equal amounts of water. For the NDVI treatment in both seasons, irrigations were scheduled individually for each of the NDVI subtreatment plots, on the day after the allowable SWDp was exceeded. A regression model described by Hunsaker et al. (2000) was used to estimate the date of initial crop senescence for each plot. The estimated crop senescence date was then used to govern end‐of‐season irrigation scheduling for individual plots. The regression model used the measured NDVI data to estimate the biologically active fraction of absorbed photo‐ synthetically active radiation (fAPAR*). The date at which fAPAR* declined to 25% of the given plot's mid‐season maxi‐ mum was assumed as the initial senescence date for the plot. An end‐of‐season irrigation was given to a plot if its fAPAR* was above 25% of its mid‐season maximum, or if its fAPAR* was 25% or less of its mid‐season maximum, but its estimated SWDp was 30% or greater. The final irrigation protocol was determined individually for each NDVI subtreatment plot. However, for the FAO treatment, all subtreatment plots were irrigated when the criteria were met for three or more sub‐ treatment plots.

TREATMENT NITROGEN MANAGEMENT Following wheat establishment, nitrogen fertilizer was applied to subtreatment plots by injecting soluble urea am‐ monium nitrate (32% N) through the gated pipe system dur‐ ing irrigation. Calibrated injector pumps connected at the head end of the pipe systems were used to meter the fertilizer to the plot. The irrigation system was flushed with non‐ amended water after each plot's fertilization, to avoid con‐ tamination of the next plot. Nitrogen applications for the high N treatments followed locally recommended practices for wheat grown on sandy loam (Ottman and Pope, 2000). The H treatment received a total of 205 kg N ha-1 and 224 kg N ha-1 during the 2003‐2004 and 2004‐2005 studies, respec‐ tively, which was applied in four split applications (table 5). Nitrogen was purposely withheld from the low N treatment when the first application to the H treatment was made fol‐ lowing emergence, and then the L treatment received half the amount applied to the H treatment during the three subse‐ quent applications. Total N applied to L treatments after emergence was 74.5 and 84 kg N ha-1 during 2003‐2004 and 2004‐2005, respectively (table 5). The first N application given during the 2004‐2005 season occurred later than that given during 2003‐2004 (table 5). This was primarily due to abnormally high precipitation throughout late January and February 2005. Consequently, the recommended first in‐season application of N at the five‐ leaf stage (approx. 10 February in 2005) could not be made due to the high water content within the soil profile during this period. Subsequent nitrogen applications during 2004‐2005 also occurred somewhat later than recommended timing as a result of the initial delay of the first N application. SUPPORTING MEASUREMENTS Wheat canopy height (hc) was obtained every 2 to 3 weeks using a meter stick to measure hc in six rows in each plot. Crop growth parameters including green leaf area index (GLAI) were determined from biweekly sampling of six plants per plot. The destructive sampling was made in the northern half of the plots within different areas that had been pre‐assigned for each sampling date. The fraction of ground cover (fc) for plots was determined photographically from analyses of weekly digital camera pictures taken 3.0 m di‐ rectly overhead around solar noon for all plots. Data for fc were obtained by performing supervised classification using iterative self‐organizing data analysis techniques (Tou and Gonzalez, 1974) and the Imagine 8.3 image‐processing pack‐ age (ERDAS, 1997). Volumetric soil water contents (qv) were measured in two ways: neutron probes (model 503, Campbell Pacific Nuclear, Martinez, Cal.) were used to measure the 0.3 to 2.9 m profile Table 5. Fertilizer application dates and amounts for high N and low N treatments during the 2003‐2004 and 2004‐2005 wheat experiments. 2003‐2004 Growing Season 2004‐2005 Growing Season Treatment (kg N ha-1) Date High N Low N (2004) 2 to 6 Feb. 56 0 18 Feb. to 4 Mar. 56 28 12 to 19 Mar. 56 28 24 to 31 Mar. 37 18.5 Total

2022

205

74.5

Treatment (kg N ha-1) Date High N Low N (2005) 1 to 3 Mar. 56 0 21 to 23 Mar. 75 37.5 4 to 08 Apr. 56 28 18 to 25 Apr. 37 18.5 Total

224

84

TRANSACTIONS OF THE ASABE

in 0.20 m increments, and time‐domain reflectometry (TDR, Trase1, Soil‐Moisture Equipment Corp., Santa Barbara, Cal.) was used for the shallow 0 to 0.3 m soil surface, where neutron probe accuracy decreases due to atmospheric losses. The neutron probes and TDR system were calibrated to the field soil with gravimetric soil samples, achieving volumet‐ ric soil water content accuracies of ±0.02 m3 m-3. The water content measurements were begun for all plots the day before the first post‐planting irrigation in each season. Subsequent readings for plots occurred approximately weekly and in‐ cluded measurements made one day prior to or the morning before irrigation application and then two to four days after the irrigation. The measurements for all plots were made in morning hours between 0600 and 0900 hours. Grain yields were obtained on 26 May 2004 and 27 May 2005 with a Hege plot combine (Wintersteiger AG, Ried im Innkreis, Austria) equipped with a 1.5 m cutter bar. Samples were harvested within designated final harvest areas measur‐ ing approximately 24 m2 in the south half of each plot (south of the boardwalk in fig. 1). Grain yields were adjusted to 12% water content on a wet weight basis. MEASURED CROP EVAPOTRANSPIRATION AND DERIVED BASAL CROP COEFFICIENTS Field data were used to evaluate the performance of the crop coefficient and evapotranspiration estimation employed during the irrigation scheduling experiments. Soil water con‐ tent measurements provided data to calculate ETc rates for each plot. The procedures used were identical to those de‐ scribed by Hunsaker et al. (2005a). Briefly, ETc that occurred between each successive soil water content measurement made during the seasons was calculated as the residual of the soil water balance equation for the crop root zone. The cal‐ culations used measured irrigation volumes, rainfall, and soil water contents. An effective maximum rooting depth of 1.3m was assumed for all plots based on measured soil water depletion patterns observed during the two seasons. Deep percolation following irrigation or heavy rainfall was esti‐ mated from soil water content data measured below the root zone, as described by Hunsaker et al. (2005a). Soil water bal‐ ance calculations of ETc were begun at 50% crop emergence for both seasons and continued through 14 May 2004 and 18May 2005, dates that corresponded to approximately complete crop senescence for the first and second experi‐ ments, respectively. Statistical analyses were made to evaluate agreement be‐ tween ETc estimated using the FAO‐56 procedures and the water balance ETc determined for 23 to 25 successive inter‐ vals during the two seasons. Statistical evaluation parameters included the coefficient of determination (r2), root mean square error (RMSE), mean absolute error (MAE; Legates and McCabe, 1999), and mean absolute percent difference (MAPD; Kustas et al., 1999). Measured ETc was also used to derive Kcb data for each of the 32 experimental plots for comparison with the esti‐ mated Kcb used in the irrigation scheduling experiments. The derivation of Kcb for the wheat plots followed the same pro‐ cedures described by Hunsaker et al. (2005a). Briefly, a Kcb value was derived for each successive interval with back‐ calculations of the FAO‐56 dual crop coefficient procedures and the “known” ETc determined from the soil water balance. Back‐calculations of the FAO‐56 dual crop coefficient proce‐ dures were employed to separate the soil water balance ETc

Vol. 50(6): 2017-2033

into the basal and evaporation contributions, while consider‐ ing the effects of water stress on the basal ETc. The parameter estimates used in the Kcb derivations for determining the soil evaporation coefficients (Ke) and the water stress coeffi‐ cients (Ks) for individual plots differed from those used dur‐ ing the experiments (tables 3 and 4), as noted below. For the derivation of Kcb, separate field capacity (qFC) and wilting point (qWP) values were used for each subtreatment plot, as determined for each plot in the previous cotton exper‐ iments at the site (Hunsaker et al., 2005a). Based on the pre‐ vious qFC and qWP data for the surface layer, resultant values of the calculated total evaporable water (TEW) varied from 18 to 22 mm over all plots. Based on the soil profile data for qFC and qWP obtained in the previous experiments with cot‐ ton, the average and standard deviation of the field capacity and wilting point for the 1.3 m crop root depth were computed as 306 ±20 mm (0.234 ±0.015 m3 m-3) and 150 ±22 mm (0.115 ±0.016 m3 m-3), respectively, and the calculated maximum total available water (TAW) for the profile aver‐ aged 159 ±12 mm over all plots. Also for the derivation of Kcb, crop height (hc) and vegetation cover (fc) were based on measured values for each plot instead of the modeled values given in table 3. Daily values for both hc and fc were esti‐ mated by linear interpolation of measured values.

RESULTS AND DISCUSSION GROWING SEASON CLIMATES AND EFFECTS ON CROP GROWTH Monthly rainfall, reference evapotranspiration (ETo), and GDD data are shown for the 2003‐2004 and 2004‐2005 grow‐ ing seasons in table 6. Comparison with the long‐term aver‐ age data (1989 to 2004) at MAC (table 6) indicates that the 2003‐2004 wheat experienced generally average climatic conditions during early growth and development in January 2004, lower than normal GDD accumulation during Febru‐ ary, and above‐average heat unit accumulation and ETo dur‐ ing a hot, dry month of March. The climatic data during the remainder of the 2003‐2004 growing season (April and May) were similar to the long‐term averages at MAC, with the ex‐ ception of above‐average rainfall during April. For the over‐ all growing season of 2003‐2004, total rainfall and ETo were about the same as long‐term data, whereas total cumulative GDD were about 136 degree C‐days (or about 8%) above av‐ erage for the site. Rainfall during wheat development periods for 2004‐2005 was abnormally high for the area, when a total of 174 mm of rain occurred during January and February (table 6). The cloudy, wet climate during this period suppressed reference eva‐ potranspiration compared to ETo during the same two months the previous year. On the other hand, GDD totals during January and February 2005 were increased above those in January and February 2004 because temperatures were higher during those months in 2005. During March, April, and May 2005, the ETo and GDD data were about average, but were somewhat below those during 2003‐2004, particularly during March. Seasonal total ETo and GDD for the 2004‐2005 growing season were 84 mm (13%) and 188 degree C‐days (10%) less than those in 2003‐2004, respectively. The time course of measured GLAI, green plant cover de‐ termined photographically, and normalized NDVI (NDVIn) derived from radiometer measurements are shown for

2023

Table 6. Growing season rainfall, reference evapotranspiration (ETo), and growing degree days (GDD) for 2003‐2004, 2004‐2005 wheat experiments, and the long‐term average at the Maricopa Agricultural Center, Maricopa, Arizona. Rainfall (mm) ETo (mm) GDD (degree C‐day)

[a] [b] [c]

Period

2003‐ 2004

2004‐ 2005

Long‐term Average [a]

2003‐ 2004

2004‐ 2005

Long‐term Average

2003‐ 2004

2004‐ 2005

Long‐term Average

Dec.[b] Jan. Feb. Mar. Apr. May[c]

2 18 23 6 25 0

1 89 85 8 3 0

8 17 20 20 9 2

27 63 76 153 191 138

2 55 67 125 186 129

25 65 82 132 184 138

82 230 190 469 463 367

16 241 249 334 445 326

80 212 238 346 448 340

Total

74

186

76

648

564

626

1800

1612

1664

Average data for indicated periods from 1989 to 2004, AZMET weather station, Maricopa Agricultural Center. December includes data starting from crop germination (19 December for 2003‐2004, and 30 December for 2004‐2005). May includes data through 18 May, approximate complete senescence of wheat for both 2003‐2004 and 2004‐2005. 8

8

(a)

FTH FSL FDH

6 5 4 3

1

5 4 3 2 1

0

0

100

20

40

60

80

100

120

140

(b)

160

0 100

FTH FSL FDH

Green crop cover (%)

0

Green crop cover (%)

NTH NSL NDH

6

2

80 60 40 20 0

20

40

60

80

100

120

140

160

(e)

NTH NSL NDH

(f)

NTH NSL NDH

80 60 40 20 0

FTH FSL FDH

(c)

1.0

1.0 0.8

NDVIn

0.8

NDVIn

(d)

7

GLAI (m2 m-2)

GLAI (m2 m-2)

7

0.6 0.4

0.6 0.4

0.2

0.2

Green crop cover (%) 0.0

0.0 0

20

40

60

80

100

120

140

160

Days Past Emergence, DPE [2003-04]

0

20

40

60

80

100

120

140

160

Days Past Emergence, DPE [2003-04]

Figure 2. (a) Measured green leaf area index (GLAI), (b) green crop cover, and (c) normalized NDVI (NDVIn) for FAO subtreatments: FAO‐typical‐ high N (FTH), FAO‐sparse‐low N (FSL), and FAO‐dense‐high N (FDH), and (d) measured green leaf area index (GLAI), (e) green crop cover, and (f)normalized NDVI (NDVIn) for NDVI subtreatments: NDVI‐typical‐high N (NTH), NDVI‐sparse‐low N (NSL), and NDVI‐dense‐high N (NDH) for the 2003‐2004 experiment.

typical‐high N, sparse‐low N, and dense‐high N subtreatments within each main treatment for 2003‐2004 and 2004‐2005 (figs. 2 and 3, respectively). The figures show the differences that oc‐ curred in wheat development among the subtreatments and also demonstrate how well the remotely sensed NDVIn data repre‐ sent the seasonal subtreatment trends for GLAI and green cover. For the 2004‐2005 experiment, the effects of the higher temper‐ atures and heat units during the early season growing period cor‐

2024

responded to accelerated GLAI and cover relative to early season growth for 2003‐2004. However, the effects of sub‐ optimal timing of N applications due to early season weather conditions for 2004‐2005 were apparent on the mid‐season wheat development, resulting in lower maximum GLAI and cover than for subtreatments in 2003‐2004. This also caused earlier wheat senescence and a shorter total growing season for 2004‐2005 than 2003‐2004.

TRANSACTIONS OF THE ASABE

6

6

(a)

FTH FSL FDH

4 3 2 1

3 2

0

100

20

40

60

80

100

120

140

160 100

FTH FSL FDH

(b)

0

Green crop cover (%)

0

Green crop cover (%)

4

1

0

80 60 40 20 0

20

40

60

80

100

120

140

160 NTH NSL NDH

(f)

80 60 40 20 0

1.0

FTH FSL FDH

(c)

NTH NSL NDH

(g)

1.0 0.8

NDVIn

0.8

NDVIn

NTH NSL NDH

(d)

5

GLAI (m2 m-2)

GLAI (m2 m-2)

5

0.6

0.6

0.4

0.4

0.2

0.2

0.0

0.0 0

20

40

60

80

100

120

140

160

Days Past Emergence, DPE [2004-05]

0

20

40

60

80

100

120

140

160

Days Past Emergence, DPE [2004-05]

Figure 3. (a) Measured green leaf area index (GLAI), (b) green crop cover, and (c) normalized NDVI (NDVIn) for FAO subtreatments: FAO‐typical‐ high N (FTH), FAO‐sparse‐low N (FSL), and FAO‐dense‐high N (FDH), and (d) measured green leaf area index (GLAI), (e) green crop cover, and (f)normalized NDVI (NDVIn) for NDVI subtreatments: NDVI‐typical‐high N (NTH), NDVI‐sparse‐low N (NSL), and NDVI‐dense‐high N (NDH) for the 2004‐2005 experiment.

SEASONAL IRRIGATION AND EVAPOTRANSPIRATION AND FINAL GRAIN YIELDS The FAO treatment irrigation schedule resulted in all FAO subtreatments receiving essentially the same amount of sea‐ sonal irrigation water applied (Iw) during a given season (tables 7 and 8 for the 2003‐2004 and 2004‐2005 seasons, re‐ spectively). For 2003‐2004, mean values of seasonal ETc for FAO subtreatments were similar across all three planting densities within both high and low N managements, but mean seasonal ETc for the FAO high N subtreatments was 9% high‐ er than that for the FAO low N subtreatments. The later crop emergence date, less than optimum N management, and shorter growing season for 2004‐2005 led to an overall reduc‐ tion in seasonal ETc of about 14% compared to 2003‐2004. Unlike 2003‐2004, seasonal ETc for FAO subtreatments dur‐ ing 2004‐2005 tended to increase with plant density over both the high and low N managements, although the difference in seasonal ETc between FAO‐high N and FAO‐low N subtreat‐ ments was somewhat less (5%) than for the previous year's experiment. The focus for the NDVI‐Kcb approach was to schedule ir‐ rigations to subtreatments by monitoring the crop water needs for individual plots. Consequently, NDVI subtreat‐ ments received variable amounts of seasonal irrigation water

Vol. 50(6): 2017-2033

during both seasons, where seasonal irrigation depth in‐ creased considerably from sparse to dense plots within both N managements (10% to 23% over both years) and to a lesser extent from low N to high N plots (tables 7 and 8). The mea‐ sured seasonal ETc for the NDVI treatment also generally in‐ creased from sparse to dense plots. For 2003‐2004, seasonal ETc was 14% and 17% higher for dense than sparse for low N and high N managements, respectively, whereas it was only 2% and 13% for 2004‐2005, respectively. A primary result of different irrigation scheduling ap‐ proaches was that mean Iw for the NDVI treatment was sig‐ nificantly lower than that for the FAO irrigation treatment, 8% lower in 2003‐2004, and 13% lower in 2004‐2005. The difference for seasonal irrigation applied water did not corre‐ spond to a significant difference in the measured seasonal ETc between NDVI and FAO treatments for 2003‐2004, whereas mean seasonal ETc was about 5% lower for NDVI than FAO for the second season and the difference was signif‐ icant at p < 0.05. However, the disparity in seasonal irrigation amount or crop water use between the FAO and NDVI treat‐ ment was not related to significant differences for final grain yield between the two Kcb methods in either year (tables 7 and 8). The most significant effect on grain yield was nitro‐ gen level, where the mean yield for the low N treatment was

2025

Table 7. Experimental treatment means and subtreatment plot means for seasonal irrigation water applied, crop evapotranspiration (ETc), and final grain yield (Y) for the 2003‐2004 wheat experiment. Treatment Means[a] Kcb method FAO NDVI Nitrogen level High Low Plant density Sparse Typical Dense

Irrigation[b] (mm)

ETc[c] (mm)

Y (kg ha-1)

561 a 519 b

543 a 538 a

6550 a 6892 a

551 a 528 b

565 a 516 b

7244 a 6198 b

526 b 537 ab 557 a

527 b 538 ab 559 a

6270 a 6804 a 7006 a

Table 8. Experimental treatment means and subtreatment plot means for seasonal irrigation water applied and crop evapotranspiration (ETc), and final grain yield (Y) for the 2004‐2005 wheat experiment. Treatment Means[a] Kcb method FAO NDVI Nitrogen level High Low Plant density Sparse Typical Dense

Irrigation[b] (mm)

ETc[c] (mm)

Y (kg ha-1)

465 a 405 b

478 a 453 b

5117 a 4881 a

448 a 422 b

486 a 445 b

5434 a 4564 b

425 b 428 b 458 a

449 b 465 ab 484 a

4482 b 5081 a 5353 a

Subtreatment Plot Means Subtreatment FSH NSH FTH NTH FDH NDH FSL NSL FTL NTL FDL NDL

Subtreatment Plot Means

Irrigation (mm)

ETc (mm)

Y (kg ha-1)

559 515 562 540 562 569 559 472 559 490 564 535

577 513 566 571 559 600 522 495 520 496 517 562

7044 5946 7305 7425 7175 8330 5906 6187 6268 6219 5135 7385

Subtreatment FSH NSH FTH NTH FDH NDH FSL NSL FTL NTL FDL NDL

Irrigation (mm)

ETc (mm)

Y (kg ha-1)

465 387 464 430 468 476 464 386 465 355 466 421

455 449 500 482 511 509 457 436 464 413 473 443

4502 4695 5657 5550 6222 5607 4236 4500 4758 4341 5110 4470

[a]

Treatment means followed by different letters for irrigation method, planting density, or nitrogen level are significantly different at the 0.05 level according to the least‐significant difference test. [b] Includes emergence irrigations. [c] Data from 29 December 2003 (crop emergence) through 14 May 2004.

[a]

14% and 16% below that of the high N treatment in the first and second seasons, respectively. Although mean grain yield was not significantly different for plant density in 2003‐2004, the yield of the sparse treatment was significantly lower than those of the typical and dense treatments for 2004‐2005. Growing season and N management differences corre‐ sponded to an overall mean decrease in grain yield of 34% for 2004‐2005 compared to the previous year. However, the yield for 2004‐2005 was only 24% less than the first year, when considering just subtreatments under typical density and high N management.

curred between subtreatments (e.g., fig. 4a) as affected by variations in late season crop senescence rates. The effects of nitrogen management on the derived Kcb trends can be ob‐ served by comparing figure 4a with figure 4b. By about 60DPE, nutrient stress due to limited fertilizer for the low N subtreatments had begun to affect plant water use. At this point during the season, lower Kcb values for low N compared to high N subtreatments were first apparent. The derived Kcb for the low N subtreatments remained lower than those for the high N subtreatments during mid‐season, and also generally declined (senesced) more rapidly than the high N subtreat‐ ments during the late season. Considering the data of figure 4a, agreement between the single FAO Kcb curve used to predict Kcb and the derived Kcb appears to be reasonable for all FAO high N subtreatments during 2003‐2004. However, the FAO Kcb curve did not pre‐ dict the Kcb trends well for low N subtreatments during much of the season (fig. 4b). Comparison of the predicted versus derived Kcb for NDVI subtreatments of 2003‐2004 suggests that the NDVI approach improved Kcb predictions over FAO under the high N management (fig. 4c), whereas it appeared to greatly improve predictions under low N management (fig.4d). This point can be more effectively illustrated by comparing derived versus predicted Kcb over all plots within each subtreatment condition (fig. 5). For 2003‐2004, the FAO subtreatments show a substantial amount of scatter about the regression lines (figs. 5a, 5b, and 5c for typical, dense, and sparse, respectively). The scatter is

DERIVED AND PREDICTED KCB The derived Kcb values for FAO subtreatments for 2003‐2004 show the effects of plant density on Kcb, particu‐ larly from about 20 to 60 days past emergence, DPE (figs. 4a and 4b, FAO‐high N and FAO‐low N subtreatments, respec‐ tively). The early season differences for derived Kcb due to density correspond to the observed density differences for GLAI and canopy cover that also occurred during that period (fig. 2). The subtreatment Kcb variations became smaller starting about 60 DPE as canopy cover levels begun to con‐ verge for all plant densities within each nitrogen manage‐ ment. Starting about 110 DPE, Kcb values decline for all subtreatments coinciding with the reduction in green leaf area and the beginning of crop senescence. During the late season (110 to 133 DPE), some differences for Kcb also oc-

2026

Treatment means followed by different letters for irrigation method, planting density, or nitrogen level are significantly different at the 0.05 level according to the least‐significant difference test. [b] Includes emergence irrigations. [c] Data from 10 January 2005 (crop emergence) through 18 May 2005.

TRANSACTIONS OF THE ASABE

1.6

1.6

(a)

Basal Crop Coefficient, Kcb

1.4 1.2

1.2

1.0

1.0

0.8

0.8

0.6

0.6

0.4 0.2

NTH pred. NTH avg. NSH pred. NSH avg. NDH pred. NDH avg.

0.4

FAO pred. FTH avg. FSH avg. FDH avg.

0.2

0.0

0.0 0

20

40

60

80

100

120

140

160

1.6

Basal Crop Coefficient, Kcb

(c)

1.4

0

20

40

60

80

100

120

140

160

120

140

160

1.6

(b)

1.4

(d)

1.4

1.2

1.2

1.0

1.0

0.8

0.8

0.6

0.6

0.4 0.2

NTL pred. NTL avg. NSL pred. NSL avg. NDL pred. NDL avg.

0.4

FAO pred. FTL avg. FSL avg. FDL avg.

0.2

0.0

0.0 0

20

40

60

80

100

120

140

160

0

20

Days Past Emergence, DPE [2003-04]

40

60

80

100

Days Past Emergence, DPE [2003-04]

Figure 4. Seasonal progression of predicted Kcb along with derived Kcb for three plant densities: (a) FAO‐high N subtreatments, (b) FAO‐low N sub‐ treatments, (c) NDVI‐high N subtreatments, and (d) NDVI‐low N subtreatments for the 2003‐2004 experiment.

1.6

1.6 FTH data FTH regr. 2 r =0.90 FTL data FTL regr. r2 =0.85 1:1 line

1.4

Derived Kcb

1.2

1.4 1.2

1.2

1.0

1.0

0.8

0.8

0.8

0.6

0.6

0.6

0.4

0.4

0.4

(a)

(b)

0.2

0.0 0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.6

0.0 0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.6 NTH data NTH regr. 2 r =0.94 NTL data NTL regr. 2 r =0.94 1:1 line

1.4 1.2

0.0

NDH data NDH regr. 2 r =0.94 NDL data NDL regr. 2 r =0.95 1:1 line

1.2

1.0

0.8

0.8

0.8

0.6

0.6

0.6

0.4

0.4

0.4

(e)

0.2

0.2

0.4

0.6

0.8

1.0

Predicted Kcb

1.2

1.4

1.6

0.8

1.0

1.2

1.4

1.6

(f)

0.2

0.0 0.0

0.6

NSH data NSH regr. 2 r =0.93 NSL data NSL regr. 2 r =0.89 1:1 line

1.2

1.0

0.0

0.4

1.4

1.0

(d)

0.2

1.6

1.4

0.2

(c)

0.2

0.0 0.0

FSH data FSH regr. 2 r =0.93 FSL data FSL regr. 2 r =0.89 1:1 line

1.4

1.0

0.2

Derived Kcb

1.6 FDH data FDH regr. 2 r =0.84 FDL data FDL regr. 2 r =0.80 1:1 line

0.0 0.0

0.2

0.4

0.6

0.8

1.0

Predicted Kcb

1.2

1.4

1.6

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

Predicted Kcb

Figure 5. Comparison of derived versus predicted Kcb and resulting regression line for: (a) FAO‐typical, (b) FAO‐dense, (c) FAO‐sparse, (d) NDVI‐ typical, (e), NDVI‐dense, and (f) NDVI‐sparse subtreatments for the 2003‐2004 experiment.

Vol. 50(6): 2017-2033

2027

apparent for smaller derived Kcb values, indicating poor agreement during the early and late periods of the growing season. For the mid‐season period, derived Kcb values widely fluctuated above and below the single FAO Kcb prediction of 1.17. The best prediction agreement for the FAO subtreat‐ ments occurred for the FTH (typical‐high N) and FSH (sparse‐high N) subtreatments, whose resulting coefficients of determination (r2) were 0.90 and 0.93, respectively. Compared with the FAO treatment, the data are more evenly distributed about the regression line for the NDVI subtreat‐ ments of 2003‐2004 (figs. 5d, 5e, and 5f). For NDVI subtreat‐ ments, the r2 values were 0.93 or better for all but the NSL (sparse‐low N) subtreatment (r2 = 0.89), suggesting that the NDVI‐Kcb estimation technique more aptly captured the dif‐ ferences in Kcb associated with the various canopy develop‐ ment conditions throughout the entire season. The superiority of the NDVI‐Kcb estimation method was particu‐ larly evident for dense subtreatments (fig. 5e vs. 5b). For 2004‐2005, differences for derived Kcb between typical, sparse, and dense plantings within the FAO treatment were par‐ ticularly more pronounced than during the previous season (fig. 6a and 6b for high and low N subtreatments, respectively). Un‐ like the 2003‐2004 season, differences for Kcb trends among plant densities within a given nitrogen level persisted well be‐ yond the beginning of mid‐season (60DPE), eventually con‐ verging late in the season. As in 2003‐2004, however, differences for Kcb between the high and low N subtreatments became discernible after about 60DPE. For all subtreatments, Kcb values had begun to decline around 90 DPE, indicating ini‐ tial crop senescence, which occurred about 20 days sooner than in the previous season. The effects of climate and sub‐optimal nitrogen manage‐ ment for 2004‐2005 caused trends for derived Kcb to depart 1.6

1.6

(a)

Basal Crop Coefficient, Kcb

1.4

(c)

1.4

1.2

1.2

1.0

1.0

0.8

0.8

0.6

0.6

0.4 0.2

NTH pred. NTH avg. NSH pred. NSH avg. NDH pred. NDH avg.

0.4

FAO pred. FTH avg. FSH avg. FDH avg.

0.2

0.0

0.0 0

20

40

60

80

100

120

140

160

1.6

Basal Crop Coefficient, Kcb

considerably form the predicted FAO Kcb curve. Conse‐ quently, all FAO subtreatments were more poorly described by the FAO Kcb curve during the 2004‐2005 growing season. During the early to mid‐season period, the derived Kcb were highly underestimated by the FAO Kcb curve for dense and typical plant densities, although the FAO Kcb curve reason‐ ably described the Kcb for sparse plant densities during this period. Typical and dense plots under high N (fig. 6a) attained maximum Kcb values similar to the FAO curve from about 60 to 80 DPE, but the derived Kcb then sharply declined begin‐ ning about 90 DPE, long before the FAO curve declined from maximum. The derived Kcb for sparse‐high N subtreatment and all low N subtreatments were substantially overestimated by the FAO Kcb curve from about 60 DPE through the remain‐ der of the season. By comparison, the NDVI approach showed good agreement between predicted and derived Kcb for both high N (fig. 6c) and low N (fig. 6d) during the 2004‐2005 season. While the 2003‐2004 results suggest that the single FAO Kcb curve provided adequate Kcb estimations for the typical growth/climate conditions of that season, the abnormal weather/growth during 2004‐2005 clearly illustrates the shortcomings for the use of a single Kcb curve. As expected, correlations between predicted and derived Kcb values for the FAO treatment were much lower for 2004‐2005 (figs. 7a, 7b, and 7c) than for the previous season. For the FAO treatment of 2004‐2005, the r2 values ranged from only 0.63 to 0.75, there was considerable scatter about the regression lines for intermediate to high values of predicted Kcb, and the regres‐ sion slopes were not close to unity. In contrast, correlations between derived and predicted Kcb for the NDVI treatment of 2004‐2005 (figs. 7d, 7e, and 7f) were respectable over all

0

20

40

60

80

100

120

140

160

100

120

140

160

1.6

(b)

1.4

(d)

1.4

1.2

1.2

1.0

1.0

0.8

0.8

0.6

0.6

0.4 0.2

NTL pred. NTL avg. NSL pred. NSL avg. NDL pred. NDL avg.

0.4

FAO pred. FTL avg. FSL avg. FDL avg.

0.2

0.0

0.0 0

20

40

60

80

100

120

140

Days Past Emergence, DPE [2004-05]

160

0

20

40

60

80

Days Past Emergence, DPE [2004-05]

Figure 6. Seasonal progression of predicted Kcb along with derived Kcb for three plant densities: (a) FAO‐high N subtreatments, (b) FAO‐low N sub‐ treatments, (c) NDVI‐high N subtreatments, and (d) NDVI‐low N subtreatments for the 2004‐2005 experiment.

2028

TRANSACTIONS OF THE ASABE

1.6

1.6 FTH data FTH regr. 2 r =0.74 FTL data FTL regr. r2 =0.67 1:1 line

1.4

Derived Kcb

1.2

1.4 1.2

1.2

1.0

1.0

0.8

0.8

0.8

0.6

0.6

0.6

0.4

0.4

0.4

(a)

(b)

0.2

0.0 0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.6

0.0 0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.6 NTH data NTH regr. 2 r =0.90 NTL data NTL regr. 2 r =0.88 1:1 line

1.4 1.2

0.0

NDH data NDH regr. 2 r =0.92 NDL data NDL regr. 2 r =0.90 1:1 line

1.2

1.0

0.8

0.8

0.8

0.6

0.6

0.6

0.4

0.4

0.4

(e)

0.2

0.2

0.4

0.6

0.8

1.0

Predicted Kcb

1.2

1.4

1.6

0.8

1.0

1.2

1.4

1.6

(f)

0.2

0.0 0.0

0.6

NSH data NSH regr. 2 r =0.90 NSL data NSL regr. 2 r =0.93 1:1 line

1.2

1.0

0.0

0.4

1.4

1.0

(d)

0.2

1.6

1.4

0.2

(c)

0.2

0.0 0.0

FSH data FSH regr. 2 r =0.75 FSL data FSL regr. 2 r =0.75 1:1 line

1.4

1.0

0.2

Derived Kcb

1.6 FDH data FDH regr. 2 r =0.68 FDL data FDL regr. 2 r =0.63 1:1 line

0.0 0.0

0.2

0.4

0.6

0.8

1.0

Predicted Kcb

1.2

1.4

1.6

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

Predicted Kcb

Figure 7. Comparison of derived versus predicted Kcb and resulting regression line for: (a) FAO‐typical, (b) FAO‐dense, (c) FAO‐sparse, (d) NDVI‐ typical, (e), NDVI‐dense, and (f) NDVI‐sparse subtreatments for the 2004‐2005 experiment.

subtreatments (r2 = 0.88 to 0.93). This latter finding demon‐ strates a potentially significant aspect of the NDVI approach, i.e., that it can be a reliable estimator of Kcb even when crop and climate conditions depart significantly from average conditions. TREATMENT COMPARISON FOR EVAPOTRANSPIRATION PREDICTION The expected end product of Kcb estimation is reliable quantification of the actual crop evapotranspiration that oc‐ curs during the season. In the following analyses, the mea‐ sured ETc, determined for 23 to 25 intervals during each season, is compared with the predicted ETc determined for the same intervals for each subtreatment (2003‐2004 in fig.8 and table 9, and 2004‐2005 in fig. 9 and table 10). The ETc calculated from the single FAO Kcb curve de‐ scribed measured ETc well for both FTH and FSH subtreat‐ ments during 2003‐2004 (figs. 8a and 8c, respectively). For both of these subtreatments, the predicted total seasonal ETc was within a few millimeters of the seasonal measured ETc (table 9). The mean absolute percent difference (MAPD) for the FSH subtreatment (10.7%) was the lowest among all sub‐ treatments, whereas it was 13.6% for FTH (table 9). Although the resultant regression slope for the FDH subtreatment was near 1:1 (fig. 8b), the r2 was poor (0.79) and the MAPD was high (22.2%), indicating that measured ETc for FDH was poorly predicted during the season. Moreover, the measured and predicted ETc did not agree well for any of the three low N subtreatments within FAO, where the MAPD were 21% to 25%, and the measured seasonal ETc was underestimated by about 50 mm (table 9). However, as seen in figs. 8d to 8f and

Vol. 50(6): 2017-2033

table 9, NDVI‐based prediction of measured ETc for low N plots was improved considerably over FAO low N plots, where the r2 for NTL, NDL, and NSL were 0.92 or higher, and the MAPD values were 15% to 16%. Values for the RMSE, MAE, MAPD, and r2 indicate that the measured and pre‐ dicted ETc was in better overall agreement for NDVI than FAO. Considering all subtreatments in the study, the mean MAPD was 5% lower for NDVI than for FAO (table 9), which was significant at p < 0.05. The advantage of wheat ETc prediction for the NDVI method was more marked during the anomalous 2004‐2005 season than during 2003‐2004 (fig. 9 and table 10). This was not because the NDVI predictions were better than they were during the previous season; rather, the performance of the FAO Kcb curve was poorer. For example, the measured ETc for the FTH subtreatment was poorly described during 2004‐2005, where the r2 was only 0.81, and the regression slope fell well below 1:1 (fig. 9a). Comparison of table 9 with table 10 shows that RMSE, MAE, and MAPD values for FAO subtreatments were increased considerably from 2003‐2004 to 2004‐2005, whereas the r2 values were decreased. For all FAO subtreatments of 2004‐2005, MAPD ranged from 20% to 26%, with a treatment mean of 23.5%. All of the NDVI subtreatments for 2004‐2005 were better predicted than FAO subtreatments, where the MAPD for NDVI varied from 11% to 17%. The NDVI treatment mean for MAPD during 2004‐2005 was 13.3%, which was significantly lower than the FAO treatment mean. The prevailing climate during 2003‐2004 was near aver‐ age for the area, and water and nutrient management during the season could be considered close to optimum for the high

2029

12

12 FTH data FTH regr. 2 r =0.91 FTL data FTL regr. r2 =0.82 1:1 line

11

Measured ETc (mm/d)

10 9 8

10 9 8

10 9 8

7

7

6

6

6

5

5

5

4

4

4

3

3

3

2

2

(a)

(b)

1

0

0 0

1

2

3

4

5

6

7

8

9

10 11 12

12 NTH data NTH regr. 2 r =0.95 NTL data NTL regr. 2 r =0.94 1:1 line

10 9 8

2

1

2

3

4

5

6

7

8

9

10 11 12

0

NDH data NDH regr. 2 r =0.95 NDL data NDL regr. 2 r =0.92 1:1 line

10 9 8

8 7

6

6

5

5

5

4

4

4

3

3

3

2

2

(e)

1 0

1

2

3

4

5

6

7

8

9

10 11 12

Predicted ETc (mm/d)

4

5

6

7

8

9

10 11

12

NSH data NSH regr. 2 r =0.94 NSL data NSL regr. 2 r =0.94 1:1 line

9

6

0

3

10

7

0

2

11

7

(d)

1

12

11

1

(c)

1 0

0 12

11

FSH data FSH regr. 2 r =0.92 FSL data FSL regr. 2 r =0.85 1:1 line

11

7

1

Measured ETc (mm/d)

12 FDH data FDH regr. 2 r =0.79 FDL data FDL regr. 2 r =0.80 1:1 line

11

2

(f)

1 0

0

1

2

3

4

5

6

7

8

9

10 11 12

0

Predicted ETc (mm/d)

1

2

3

4

5

6

7

8

9

10 11

12

Predicted ETc (mm/d)

Figure 8. Comparison of measured versus predicted ETc and resulting regression line for: (a) FAO‐typical, (b) FAO‐dense, (c) FAO‐sparse, (d) NDVI‐ typical, (e), NDVI‐dense, and (f) NDVI‐sparse subtreatments for the 2003‐2004 experiment. Table 9. Measured evapotranspiration (ETc) determined from soil water balance calculations for 23 to 25 intervals compared to predicted ETc from FAO‐56 procedures during the 2003‐2004 wheat experiment. Statistics used to evaluate interval ETc predictions include the mean, standard deviation (SD), root mean square error (RMSE), mean absolute error (MAE), mean absolute percent difference (MAPD), and the coefficient of determination (r2). Total Seasonal ETc Interval ETc Determinations Measured (mm) Predicted (mm) Measured (mm d-1) Predicted (mm d-1) RMSE MAE MAPD Mean SD Mean SD Mean SD Mean SD (mm d-1) (mm d-1) (%) Subtreatment FSH 577 18.8 572 0[a] 4.9 2.6 4.8 2.5 0.69 0.48 10.7 NSH 513 18.1 516 17.3 4.2 2.4 4.2 2.2 0.54 0.38 15.2 FTH 566 14.2 572 0 4.8 2.5 4.8 2.5 0.74 0.47 13.6 NTH 571 13.2 538 2.9 4.6 2.4 4.3 2.1 0.46 0.48 12.8 FDH 559 33.6 572 0 4.7 2.6 4.8 2.5 1.12 0.76 22.3 NDH 600 45.5 556 14.6 4.8 2.5 4.4 2.1 0.46 0.51 13.9 FSL 524 31.8 572 0 4.4 2.5 4.8 2.5 0.94 0.63 21.2 NSL 496 18.8 504 0 4.1 2.4 4.1 2.1 0.53 0.42 16.5 FTL 522 29.1 572 0 4.3 2.3 4.8 2.5 1.03 0.68 22.9 NTL 496 28.4 499 24.0 4.0 2.3 4.0 2.0 0.49 0.37 15.8 FDL 519 10.3 572 0 4.3 2.2 4.8 2.5 1.11 0.71 25.1 NDL 563 2.5 520 8.6 4.4 2.4 4.1 2.0 0.56 0.59 15.3

r2 0.92 0.94 0.91 0.95 0.79 0.95 0.85 0.94 0.82 0.94 0.80 0.92

All FAO[b] All NDVI[b]

0.85 0.94

[a] [b]

543 538

31.3 45.2

572 521

0 23.9

4.5 4.3

2.5 2.4

4.8 4.2

2.5 2.1

0.95 0.51

0.61 0.45

19.0 14.7

There was no difference in the estimated seasonal ETc for FAO subtreatments. The subtreatment data was combined within both irrigation methods.

N treatments. As indicated by the results for 2003‐2004, a lo‐ cally developed single FAO Kcb curve can be expected to es‐ timate wheat ETc reliably when these norms are met and when plant density is not vastly different from the standard density conditions. One of the primary causes of the poor es‐ timation of ETc for FAO during 2004‐2005 was the unex‐ pected downturn of Kcb starting about 90 DPE, as pointed out earlier. However, such shifts in crop water use conditions are

2030

not easily observed and therefore are difficult to accommo‐ date in practice using standard crop coefficients. Findings for 2004‐2005 suggest that the NDVI method has the capability to account for such crop water use peculiarities that may arise during a given season. Furthermore, the NDVI approach was found to be a robust method, able to effectively estimate actu‐ al ETc for both optimum and sub‐optimum crop growth con‐ ditions.

TRANSACTIONS OF THE ASABE

12

12 FTH data FTH regr. 2 r =0.81 FTL data FTL regr. r2 =0.77 1:1 line

11

Measured ETc (mm/d)

10 9 8

10 9 8

10 9 8

7

7

6

6

6

5

5

5

4

4

4

3

3

3

2

2

(a)

(b)

1

0

0 0

1

2

3

4

5

6

7

8

9

10 11 12

12 NTH data NTH regr. 2 r =0.93 NTL data NTL regr. 2 r =0.88 1:1 line

10 9 8

2

1

2

3

4

5

6

7

8

9

10 11 12

0

NDH data NDH regr. 2 r =0.94 NDL data NDL regr. 2 r =0.93 1:1 line

10 9 8

8 7

6

6

5

5

5

4

4

4

3

3

3

2

2

(e)

1 0

1

2

3

4

5

6

7

8

9

10 11 12

4

5

6

7

8

9

10 11

12

NSH data NSH regr. 2 r =0.93 NSL data NSL regr. 2 r =0.93 1:1 line

9

6

0

3

10

7

0

2

11

7

(d)

1

12

11

1

(c)

1 0

0 12

11

FSH data FSH regr. 2 r =0.79 FSL data FSL regr. 2 r =0.82 1:1 line

11

7

1

Measured ETc (mm/d)

12 FDH data FDH regr. 2 r =0.81 FDL data FDL regr. 2 r =0.77 1:1 line

11

2

(f)

1 0

0

Predicted ETc (mm/d)

1

2

3

4

5

6

7

8

9

10 11 12

0

1

Predicted ETc (mm/d)

2

3

4

5

6

7

8

9

10 11

12

Predicted ETc (mm/d)

Figure 9. Comparison of measured versus predicted ETc and resulting regression line for: (a) FAO‐typical, (b) FAO‐dense, (c) FAO‐sparse, (d) NDVI‐ typical, (e), NDVI‐dense, and (f) NDVI‐sparse subtreatments for the 2004‐2005 experiment. Table 10. Measured evapotranspiration (ETc) determined from soil water balance calculations for 23 to 25 intervals compared to predicted ETc from FAO‐56 procedures during the 2004‐2005 wheat experiment. Statistics used to evaluate interval ETc predictions include the mean, standard deviation (SD), root mean square error (RMSE), mean absolute error (MAE), mean absolute percent difference (MAPD), and the coefficient of determination (r2). Total Seasonal ETc Interval ETc Determinations Subtreatment

Measured (mm d-1)

Estimated (mm d-1)

Measured (mm)

Estimated (mm)

Mean

Mean

SD

Mean

SD

Mean

SD

0[a]

SD

FSH NSH FTH NTH FDH NDH FSL NSL FTL NTL FDL NDL

455 449 500 482 511 509 457 436 464 413 473 443

8.2 25.3 26.9 54.7 4.9 6.6 14.1 22.5 11.7 27.5 8.6 53.9

565 468 565 484 565 504 565 462 565 444 565 459

17.2 0 21.3 0 5.7 0 18.4 0 8.3 0 44.8

3.9 3.7 4.3 4.2 4.3 4.3 3.9 3.6 4.0 3.3 4.1 3.6

1.8 2.0 2.2 2.1 2.0 2.2 1.8 1.9 1.9 1.7 1.9 1.9

4.7 3.9 4.7 4.1 4.7 4.2 4.7 3.8 4.7 3.6 4.7 3.7

2.4 1.9 2.4 2.0 2.4 2.1 2.4 1.9 2.4 1.6 2.4 1.7

All FAO[b] All NDVI[b]

478 453

25.7 46.2

565 468

0 26.4

4.2 3.8

2.0 2.0

4.7 3.9

2.4 1.9

[a] [b]

RMSE MAE MAPD (mm d-1) (mm d-1) (%) 1.14 0.94 25.9 0.49 0.33 13.3 1.07 0.70 20.4 0.54 0.42 12.8 1.07 0.74 21.8 0.52 0.40 11.7 1.02 0.96 23.6 0.48 0.33 11.4 1.16 0.90 25.8 0.57 0.41 16.9 1.18 0.91 24.5 0.47 0.32 10.9 1.13 0.53

0.85 0.38

23.5 13.3

r2 0.79 0.93 0.81 0.92 0.81 0.94 0.82 0.93 0.77 0.88 0.77 0.93 0.79 0.92

There was no difference in the estimated seasonal ETc for FAO subtreatments. The subtreatment data was combined within both irrigation methods.

A general premise of this research is that accurate ETc es‐ timation is essential for effective irrigation scheduling and efficient irrigation water use. Although a through examina‐ tion of factors related to irrigation effectiveness is not pre‐ sented in this article, these results would imply improved irrigation scheduling and higher water application efficien‐ cies for the NDVI over the single Kcb curve approach. How‐

Vol. 50(6): 2017-2033

ever, more detailed evaluations on the effectiveness of the irrigation scheduling for the NDVI and FAO approaches are reported in the companion article (Hunsaker et al., 2007). These evaluations address treatment irrigation scheduling ef‐ fects on soil water status, irrigation performance parameters, and deep percolation, and their implications on yield and wa‐ ter use efficiency results.

2031

NDVI DATA ACQUISITION Ground sensors can allow detailed RS measurements at spatial scales from individual plants to portions of plots or fields, while airborne and satellite imagery can provide a syn‐ optic view of entire fields and regions. With the increased availability of high‐resolution imagery, within‐field details may be resolved to provide information for subfield manage‐ ment of crop and soil variability. However, routine imagery acquisitions at frequent intervals, e.g., three‐day intervals, necessary for the NDVI‐Kcb approach are either infeasible or operationally expensive. Retrieval frequency from satellite platforms is typically no better than 16 days under favorable conditions, while high‐frequency imagery from airborne platforms is logistically challenging and costly. Two alterna‐ tive RS approaches, however, may be able to overcome these problems, provided they are tested and verified in real‐world agricultural settings. One approach collects remote sensing data along transects from small, uninhabited aerial vehicles (UAV). Equipment and deployment costs for UAVs can be comparatively low, while data quality has been found to be sufficiently good for vegetation index retrieval (Herwitz et al., 2004). A second approach, possibly implemented in tan‐ dem with UAV remote sensing, collects radiometric image data from a network of fixed sensors. The fixed sensors can take continuous samples from strategically representative locations. Combining these approaches would potentially re‐ turn an extremely valuable vegetation data set, not achieved elsewhere, at moderate costs.

CONCLUSIONS One of the great challenges in crop coefficient estimation of evapotranspiration is determining coefficients that accu‐ rately reflect the actual crop development and water use. In this study, irrigation scheduling experiments were conducted for two years to evaluate and compare the prediction of wheat ETc using a locally developed time‐based Kcb curve and an NDVI‐based crop coefficients determined from canopy re‐ flectance. The following conclusions were obtained from the research: S Wheat growth stage lengths and the maximum Kcb val‐ ues that define the time‐based Kcb curve were clearly changed when different plant density and nutrient man‐ agements were imposed. S A locally developed time‐based Kcb curve may provide close estimates of crop evapotranspiration for optimum conditions during one year, but may not in another year. S Appropriate real‐time Kcb adjustments that account for the variable conditions of wheat development and wa‐ ter use can be attained using the NDVI‐based Kcb. S When used in irrigation management, the NDVI‐Kcb method could provide the spatial crop water use infor‐ mation needed to adjust irrigation schedules for actual plant density conditions. This could potentially result in significantly lower irrigation water use, particularly for sparse wheat canopies or fields with less vigorous growth due to nutrient stress. ACKNOWLEDGEMENTS The authors gratefully express their appreciation and thanks to the following people whose support during field op‐ erations and data collection made this research possible:

2032

C.Arterberry, A. Ashley, M. Conley, R. Jackson, S. Johnson, W. Luckett, S. Maneely, D. Powers, S. Richards, and R. Rokey from the USDA‐ARS U.S. Arid Land Agricultural Center; and M. Hartman, D. Langston, C. Jones, G. Main, C. O'Brien, and S. Hahnke from the University of Arizona Mari‐ copa Agricultural Center.

REFERENCES Allen, R. G., W. O. Pruitt, J. A. Businger, L. J. Fritschen, M. E. Jensen, and F. H. Quinn. 1996. Chapter 4: Evaporation and transpiration. In ASCE Handbook of Hydrology, 125‐252. New York, N.Y.: ASCE. Allen, R. G., L. S. Pereira, D. Raes, and M. Smith. 1998. Crop Evapotranspiration. FAO Irrigation and Drainage Paper 56. Rome, Italy: Food and Agriculture Organization of the UN. Bausch, W. C. 1995. Remote sensing of crop coefficients for improving the irrigation scheduling of corn. Agric. Water Mgmt. 27(1): 55‐68. Brown, P. W. 1989. Accessing the Arizona Meteorological Network (AZMET) by computer. Ext. Rep. No. 8733. Tucson, Ariz.: University of Arizona. Brown, P. W. 1991. Normal values of heat unit accumulation for southern Arizona. Ext. Rep. No. 190041. Tucson, Ariz.: University of Arizona. Doorenbos, J., and W. O. Pruitt. 1977. Crop Water Requirements. FAO Irrigation and Drainage Paper 24. Rome, Italy: Food and Agriculture Organization of the UN. ERDAS. 1997. ERDAS Field Guide. Atlanta, Ga.: ERDAS. Heilman, J. L., W. E. Heilman, and D. G. Moore. 1982. Evaluating the crop coefficient using spectral reflectance. Agron. J. 74(6): 967‐971. Herwitz, S. R., L. F. Johnson, S. E. Dunagan, R. G. Higgins, D. V. Sullivan, J. Zheng, B. M. Lobitz, J. G. Leung, B. A. Gallmeyer, M. Aoyagi, R. E. Slye, and J. A. Brass. 2004. Imaging from an unmanned aerial vehicle: Agricultural surveillance and decision support. Computers and Electronics in Agric. 44(1): 49‐61. Howell, T. A. 1996. Irrigation scheduling research and its impact on water use. In Proc. Intl. Conf. on Evapotranspiration and Irrigation Scheduling, 21‐33. C. R. Camp, E. J. Sadler, and R. E. Yoder, eds. St. Joseph, Mich.: ASAE. Hunsaker, D. J., B. A. Kimball, P. J. Pinter, Jr., G. W. Wall, R. L. LaMorte, F. J. Adamsen, S. W. Leavitt, T. L. Thompson, A. D. Matthias, and T. J. Brooks. 2000. CO2 enrichment and soil nitrogen effects on wheat evapotranspiration and water use efficiency. Agric. Meteorol. 104(2): 85‐105. Hunsaker, D. J., P. J. Pinter, Jr., and H. Cai. 2002. Alfalfa basal crop coefficients for FAO‐56 procedures in the desert regions of the southwestern U.S. Trans. ASAE 45(6): 1799‐1815. Hunsaker, D. J., E. M. Barnes, T. R. Clarke, G. J. Fitzgerald, and P. J. Pinter, Jr. 2005a. Cotton irrigation scheduling using remotely sensed and FAO‐56 basal crop coefficients. Trans. ASAE 48(4): 1395‐1407. Hunsaker, D. J., P. J. Pinter, Jr., and B. A. Kimball. 2005b. Wheat basal crop coefficients determined by normalized difference vegetation index. Irrig. Sci. 24(1): 1‐14. Hunsaker, D. J., G. J. Fitzgerald, A. N. French, T. R. Clarke, M. J. Ottman, and P. J. Pinter, Jr. 2007. Wheat irrigation management using multispectral crop coefficients: II. Irrigation scheduling performance, grain yield, and water use efficiency. Trans. ASABE 50(6): 2035-2050. Jensen, M. E., and R. G. Allen. 2000. Evolution of practical ET estimating methods. In Proc. 4th Natl. Irrig. Symp., 52‐65. R. G. Evans, B. L. Benham, and T. P. Trooien, eds. St. Joseph, Mich.: ASAE. Jensen, M. E., R. D. Burman, and R. G. Allen. 1990. Evapotranspiration and Irrigation Requirements. ASCE

TRANSACTIONS OF THE ASABE

Manuals and Reports on Engineering Practices No. 70. New York, N.Y.: ASCE. Johnson, L., and T. Scholasch. 2005. Remote sensing of shaded areas in vineyards. HortTech. 15(4): 859‐863. Kustas, W. P., X. Zhan, and T. J. Jackson. 1999. Mapping surface energy flux partitioning at large scales with optical and microwave remote sensing data from Washita '92. Water Resour. Res. 35(1): 265‐277. Legates, D. R., and G. J. McCabe. 1999. Evaluating the `goodness‐of‐fit' measures in hydrologic and hydroclimatic model validation. Water Resour. Res. 35(1): 233‐241. Leib, B. G., M. Hattendorf, T. Elliott, and G. Matthews. 2002. Adoption and adaptation of scientific irrigation scheduling: Trends from Washington, USA, as of 1998. Agric. Water Mgmt. 55(2): 105‐120. Martin, D. L., and J. R. Gilley. 1993. Chapter 2, Part 623: Irrigation water requirements. In National Engineering Handbook. Washington, D.C.: USDA‐SCS. Moran, M. S., S. J. Mass, and P. J. Pinter, Jr. 1995. Combining remote sensing and modeling for estimating surface evaporation and biomass production. Remote Sensing Reviews 12(4): 335‐353. Neale, C. M. U., H. Jayanthi, and J. L. Wright. 2003. Crop and irrigation water management using high‐resolution airborne remote sensing. In Proc. ICID Workshop Remote Sensing of ET for Large Regions, CD‐ROM. New Delhi, India: International Commission on Irrigation and Drainage.

Vol. 50(6): 2017-2033

Ottman, M. J., and N. V. Pope. 2000. Nitrogen fertilizer movement in the soil as influenced by nitrogen rate and timing in irrigated wheat. SSSA J. 64(5): 1883‐1892. Pinter Jr., P. J., J. L. Hatfield, J. S. Schepers, E. M. Barnes, M. S. Moran, C. S. Daughtry, and D. R. Upchurch. 2003. Remote sensing for crop management. Photogrammetric Eng. Remote Sensing 69(6): 647‐664. Plant, R. E., D. S. Munk, B. R. Roberts, R. L. Vargas, D. W. Rains, R. L. Travis, and R. B. Hutmacher. 2000. Relationships between remotely sensed reflectance data and cotton growth and yield. Trans. ASAE 43(3): 535‐546. Post, D. F., C. Mack, P. D. Camp, and A. S. Sulliman. 1988. Mapping and characterization of the soils on the University of Arizona Maricopa Agricultural Center. In Proc. Hydrology and Water Resources in Arizona and the Southwest, 49‐60. Tucson, Ariz.: University of Arizona. SAS. 1998. SAS User's Guide: Statistics. Ver. 7.0. Cary, N.C.: SAS Institute, Inc. Stegman, E. C. 1988. Corn crop curve comparison for the central and northern plains of the U.S. Applied Eng. in Agric. 4(3): 226‐233. Tou, J. T., and R. C. Gonzalez. 1974. Pattern Recognition Principals. Reading, Mass.: Addison‐Wesley. Wright, J. L. 1982. New evapotranspiration crop coefficients. J. Irrig. Drain Div., ASCE 108(1): 57‐74.

2033

2034

TRANSACTIONS OF THE ASABE