Glenn J. Fitzgerald

COTTON IRRIGATION SCHEDULING USING REMOTELY SENSED AND FAO-56 BASAL CROP COEFFICIENTS

Multispectral vegetation indices calculated from canopy reflectance measurements have been used to simulate nreal-time basal crop coefficients (Kcb), which have been validated to improve evapotranspiration (ETc) estimation for several ncrops. In this article, an application of the approach was evaluated for cotton using remote sensing observations of the nnormalized difference vegetation index (NDVI) to estimate Kcb as a function of NDVI. The dual crop coefficient procedures nof FAO Paper 56 (FAO-56) were used to calculate ETc and determine irrigation scheduling using Kcb estimates from remote nsensing (NDVI treatment) as well as from time-based Kcb curves (FAO treatment), which were developed locally for standard ncrop conditions using FAO-56 procedures. Two cotton experiments, conducted in 2002 and 2003 in central Arizona, included nsub-treatments of three levels of plant density and two levels of nitrogen management to impose a wide range of crop ndevelopment and water use. The NDVI-Kcb relationships used for 2002, developed from previous data for a different cotton ncultivar, row orientation, and soil type, substantially underestimated ETc, resulting in significantly less irrigation water napplied and lower lint yields for NDVI compared to the FAO treatment. The 2002 data were used to recalibrate the NDVI-Kcb nrelationships, which were then used for the NDVI treatments in 2003. The FAO Kcb curve used in 2002 described ETc and nirrigation scheduling reasonably well for sparse plots, but consistently underestimated water use and soil water depletion nfor the higher plant densities during the first half of the season. Consequently, an adjusted FAO Kcb curve, based on 2002 nresults, was employed for the FAO treatment in 2003. For the 2003 experiment, estimated cotton ETc for the NDVI treatment nresulted in a mean absolute error of 9% compared to 10% for the FAO treatment, where the difference was not significant nbetween treatments. However, the NDVI-Kcb relations used in 2003 greatly improved estimates for ETc compared to the nprevious year, where the mean absolute error for the NDVI treatment in 2002 was 22%. Predicted ETc using the FAO Kcb curve nof 2003 for typical planting density and high nitrogen conditions resulted in a mean absolute error of 10% compared to 15% nin 2002. Final lint yields for 2003 were not significantly different between the two Kcb methods. Although additional research nis needed to validate remote sensing Kcb estimation for other conditions than those in these experiments, this study did not nshow significant advantages for the NDVI approach over a carefully derived single FAO Kcb application. However, the NDVI napproach has the potential to further extend our present crop coefficient estimation capabilities when weather, plant density, nor other factors cause cotton canopy development and water use patterns to depart from typical conditions.

Wheat Irrigation Management Using Multispectral Crop Coefficients: I. Crop Evapotranspiration Prediction

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.

Energy Balance Estimation of Evapotranspiration for Wheat Grown Under Variable Management Practices in Central Arizona

Estimating and monitoring the spatial distribution of evapotranspiration (ET) over irrigated crops is becoming increasingly important for managing crop water requirements under water scarce conditions. The usual point-based approaches for estimating ET, however, do not provide enough data for precision farming applications, whereby irrigation schedules could be customized by crop conditions at sub-field scales. Needed in addition are spatially distributed ET modeling approaches, obtainable only through remote sensing, which can observe ET-related surface properties such as vegetation density and surface temperature. Although research using remote sensing to estimate ET has been pursued for many years, there are still few ground-validated, full-season ET studies at fine spatial scales. In this study, we assessed the ability of a remote sensing model to retrieve daily ET throughout an entire growing season for wheat. Image data with 0.5 m resolution were collected in 2005 over an irrigation scheduling research site in central Arizona (Maricopa). The 1.3 ha study area (FISE05) contained 32 leveled, flood-irrigated plots with treatments for irrigation scheduling, planting density, and fertilization. Daily ET was modeled using a two-source energy balance (TSEB) approach and airborne image observations in visible, near-infrared, and thermal infrared wavelengths for six dates throughout the growing season. Using independent soil water depletion observations, modeled daily ET values were accurate to within 0.4 mm d-1 for most of the FISE05 season and sensitive to changes in wheat canopy changes. Late-season ET estimates were less satisfactory, with accuracies to within 1.3 mm d-1. The significance of these results was supported by verifying agreement between ground and airborne estimates of vegetation indices, wheat canopy cover, and surface temperature. Results from FISE05 also showed that the value of this particular implementation of ET remote sensing was limited by inadequate temporal sampling, seasonal dependence of vegetation density estimators, and uncertain parameterization for the late-season. In each case, modeled and projected daily ET estimate errors could exceed 1 mm d-1. These ET/energy balance results can be improved, however, by revising the vegetation density estimators and by combining episodic high-spatial-resolution image data with continuous high-temporal-resolution ground data.

Wheat Irrigation Management Using Multispectral Crop Coefficients: II. Irrigation Scheduling Performance, Grain Yield, and Water Use Efficiency

Current irrigation scheduling is based on well-established crop coefficient and reference evapotranspiration procedures to estimate daily crop evapotranspiration (ETc). Effective irrigation scheduling and efficient irrigation water use can occur when ETc is calculated with crop coefficients representative of actual crop water use conditions. The objective of this research was to evaluate irrigation scheduling using two approaches to estimate the basal crop coefficient (Kcb) during wheat experiments conducted in 2003-2004 and 2004-2005 at Maricopa, Arizona. Each Kcb approach (main treatment) included six subtreatment combinations (three plant densities and two N managements) imposed to create spatial and temporal variations in water use among experimental plots. The first approach (NDVI treatment) estimated Kcb separately for each plot based on normalized difference vegetation index (NDVI) data obtained by frequent canopy reflectance measurements. The second approach (FAO treatment) estimated Kcb uniformly for all plots based on a Kcb curve developed for standard wheat conditions. The Kcb estimates were incorporated within the FAO-56 dual crop coefficient procedures to calculate daily ETc and root zone soil water depletion (Dr). Plot irrigations were provided when the predicted Dr reached 45% of the available soil water. During both wheat experiments, considerable variations in measured soil water depletion were observed for subtreatments due to differences in crop water use rates. For the FAO treatment, mean absolute percent difference (MAPD) for predicted Dr was 27% and 40% for 2003-2004 and 2004-2005, respectively. Prediction of Dr was improved significantly for NDVI for both experiments where treatment MAPD was 17% (2003-2004) and 18% (2004-2005). Although mean irrigation application efficiency for NDVI (89%) and FAO (88%) was similar for 2003-2004, it was significantly higher for NDVI (86%) than FAO (77%) for 2004-2005. Differences for irrigation scheduling resulted in significantly lower seasonal irrigation water use for the NDVI than FAO treatment, 8% (2003-2004) and 13% (2004-2005), but did not result in appreciable treatment differences for seasonal ETc, final grain yield, and crop water use efficiency (yield per unit ETc). Consequently, a primary outcome for both experiments was significantly higher irrigation water use efficiency (yield per unit irrigation water) for NDVI than FAO. Incorporating Kcb estimates based on NDVI within existing crop coefficient algorithms provides an opportunity to improve wheat irrigation scheduling strategies for conserving irrigation water while maintaining grain yield potentials.

An Automated Nonrigid Registration for a Tunable Hyperspectral Imaging System

A method that uses the Phase Correlation and a geometric transformation is proposed to estimate nonrigid registration errors for a hyperspectral imaging system. It computes multiple correlations to find and correct for local registration errors.

Shadow fraction in spectral mixture analysis of a cotton canopy

Hyperspectral imagery is capable of providing detailed spectral reflectance information of agricultural fields for potential use in site-specific management operations. Analysis of these data are complicated by the large number of spectral bands, the many different components or endmembers (e.g. plant and soil), and the presence of shadows. Unlike simple unmixing approaches which compute the fraction of a fixed number of components, multiple endmember spectral mixture analysis (MESMA) also determines which components are present in each pixel. This study compared whether using different shadow endmembers (EM) in a 4-EM model (sunlit green leaf, sunlit soil, shadowed leaf, shadowed soil) would improve estimates of scene components compared to a 3-EM model (sunlit green leaf, sunlit soil, photometric shade). Results revealed that correlations with percent cover and height were improved when shadow or shade endmembers were included for both models compared to the green leaf fraction alone. The 3-EM model was superior for developing a direct relationship for estimating cover and height but was not able to estimate SPAD or chlorophyll a. The 4-EM model showed the best results for SPAD and chlorophyll a, with r2 values of 0.84 and 0.77, respectively.

Automated registration of hyperspectral images

Hyperspectral images of the Earth’s surface are increasingly being acquired from aerial platforms. The dozens or hundreds of bands acquired by a typical hyperspectral sensor are acquired either through a scanning process or by collecting a sequence of images at varying wavelengths. This latter method has the advantage of acquiring coherent images of a scene at different wavelengths. However, it takes time to collect these images and some form of co-registration is required to build coherent image cubes. In this paper, we present a method to register many bands acquired sequentially at different wavelengths from a helicopter. We discuss the application of the Phase Correlation (PC) Method to recover scaling, rotation, and translation from an airborne hyperspectral imaging system, dubbed PHyTIS. This approach is well suited for remotely sensed images acquired from a moving platform, which induces image registration errors due to along and across track movement. We were able to register images to within ± 1 pixel across entire image cubes obtained from the PHyTIS hyperspectral imaging system, which was developed for precision farming applications.