Agronomy Journal Grow Your Career With ASA
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published in Agron J 100:114-121 (2008)
DOI: 10.2134/agrojnl2006.0363
© 2008 American Society of Agronomy
677 S. Segoe Rd., Madison, WI 53711 USA
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Supplemental Data
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Flynn, E. S.
Right arrow Articles by Wendroth, O.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Flynn, E. S.
Right arrow Articles by Wendroth, O.
Agricola
Right arrow Articles by Flynn, E. S.
Right arrow Articles by Wendroth, O.
Related Collections
Right arrow Grazing Management
Right arrow Geostatistics
Right arrow Remote Sensing

PASTURE MANAGEMENT

Assessment of Pasture Biomass with the Normalized Difference Vegetation Index from Active Ground-Based Sensors

E. Scott Flynna,*, Charles T. Doughertyb and Ole Wendrothb

a Dep. of Agronomy, Iowa State Univ., Ames, IA 50011-1010
b Dep. of Plant and Soil Science, Univ. of Kentucky, Lexington, KY 40546-0312. Submitted with the approval of the Director, Kentucky Agric. Exp. Stn. as publication 06-06-119. Supported jointly by Kentucky Agric. Exp. Stn. and USDA CSREES Special Grant Animal Health and Grazing Systems

* Corresponding author (esflyn2{at}iastate.edu).


   ABSTRACT
TOP
 NOTES
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Calculating forage availability is challenging for managers of grazing systems due to the spatial heterogeneity of swards. Remote sensing applications may help to overcome this problem through estimates of biomass made with reflectance data. The objectives of this study were to (i) estimate herbage mass using an active, on-the-go, ground-based, narrow band sensor to calculate the normalized difference vegetation index (NDVI), (ii) determine if NDVI may be used to assess spatial variability of herbage mass of grasslands, and (iii) determine if NDVI may be used to evaluate management of grazing systems. The NDVI was measured using an active ground-based sensor, the GreenSeeker (Ntech Industries, Ukiah, CA). In tall fescue [Schedonorus arundinaceus (Schreb.) Dumort], NDVI was correlated with biomass determined by destructive harvesting (r2 = 0.68) and also with a calibrated rising plate meter (RPM) (r2 = 0.54). Semivariograms revealed that NDVI sampling intervals of 0.76 m adequately described the spatial variability structure of grazed swards. The frequency distributions of sward biomass derived from NDVI may reflect the foraging strategies of cattle. Negative skewness and high kurtosis are consistent with selective grazing, while positive skewness and low kurtosis are consistent with less diet selectivity. Frequency distributions also improved definition of available forage within each field. We concluded that spatial properties of grassland biomass may be derived from high resolution NDVI and RPM data and could be used to evaluate conditions of grassland landscapes and to aid decision-making of managed grazing systems.

Abbreviations: CSSH, compressed sward surface height • DM, dry matter • GIS, global information system • GPS, global positioning system • LAI, leaf area index • NDVI, normalized difference vegetation index • RPM, rising plate meter • SR, stocking rate

Assessment of Pasture Biomass with the Normalized Difference Vegetation Index from Active Ground-Based Sensors

E. Scott Flynna,*, Charles T. Doughertyb and Ole Wendrothb

a Dep. of Agronomy, Iowa State Univ., Ames, IA 50011-1010
b Dep. of Plant and Soil Science, Univ. of Kentucky, Lexington, KY 40546-0312. Submitted with the approval of the Director, Kentucky Agric. Exp. Stn. as publication 06-06-119. Supported jointly by Kentucky Agric. Exp. Stn. and USDA CSREES Special Grant Animal Health and Grazing Systems

* Corresponding author (esflyn2{at}iastate.edu).

Received for publication December 22, 2006.
Calculating forage availability is challenging for managers of grazing systems due to the spatial heterogeneity of swards. Remote sensing applications may help to overcome this problem through estimates of biomass made with reflectance data. The objectives of this study were to (i) estimate herbage mass using an active, on-the-go, ground-based, narrow band sensor to calculate the normalized difference vegetation index (NDVI), (ii) determine if NDVI may be used to assess spatial variability of herbage mass of grasslands, and (iii) determine if NDVI may be used to evaluate management of grazing systems. The NDVI was measured using an active ground-based sensor, the GreenSeeker (Ntech Industries, Ukiah, CA). In tall fescue [Schedonorus arundinaceus (Schreb.) Dumort], NDVI was correlated with biomass determined by destructive harvesting (r2 = 0.68) and also with a calibrated rising plate meter (RPM) (r2 = 0.54). Semivariograms revealed that NDVI sampling intervals of 0.76 m adequately described the spatial variability structure of grazed swards. The frequency distributions of sward biomass derived from NDVI may reflect the foraging strategies of cattle. Negative skewness and high kurtosis are consistent with selective grazing, while positive skewness and low kurtosis are consistent with less diet selectivity. Frequency distributions also improved definition of available forage within each field. We concluded that spatial properties of grassland biomass may be derived from high resolution NDVI and RPM data and could be used to evaluate conditions of grassland landscapes and to aid decision-making of managed grazing systems.

Abbreviations: CSSH, compressed sward surface height • DM, dry matter • GIS, global information system • GPS, global positioning system • LAI, leaf area index • NDVI, normalized difference vegetation index • RPM, rising plate meter • SR, stocking rate


   INTRODUCTION
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
MAINTAINING FORAGE AVAILABILITY at a level that maximizes profits is challenging for managers of grassland livestock systems, especially when they are working with spatially heterogeneous swards. Numerous methods of estimating pasture biomass have been developed, but most tend to be operator dependent, labor intensive, costly, and invariably require separate calibrations for different species, seasons, pasture management strategies, and geographical locations (Haydock and Shaw, 1975; Aiken and Bransby, 1992; Harmoney et al., 1997). The most complex problem in the measurement of grassland herbage biomass, however, is finding appropriate sampling procedures that account for spatial variability with appropriate spatial resolution. Although mean forage mass is a primary descriptor of grazing systems, information about spatial distribution of that biomass could be equally important. Maps depicting how yield is spatially distributed allow for the identification of chronically poor areas of productivity and may lead the way for more functional site-specific pasture and grazing management decisions (Hill et al., 1999).

Histograms are another important tool for characterizing variability. By examining the distribution of biomass, inferences may be made about the causes and effects of spatial heterogeneity of swards on grazing systems. For example, proportions of under- and overgrazed areas within a sward may be the results of selective grazing (Aiken et al., 1997; Cid and Brizuela, 1998; Correll et al., 2003). As swards are depleted, grazing animals may be forced to become less selective regarding the area in which they graze. The areas where they are forced to graze under such circumstances are usually less palatable and sometimes less digestible forages, such as those found in overmature patches of herbage and those found near patches of fecal material (Aiken et al., 1997; Cid and Brizuela, 1998; Correll et al., 2003). Therefore, the type of frequency distribution exhibited for biomass (normal, double normal, log normal, etc.) (Barthram et al., 2005) may allow characterization of diet selectivity and refinement of grazing management.

Remote sensing of grasslands at high resolution using the NDVI may be applied to the definition of the temporal and spatial variability of grasslands. The NDVI is correlated with biomass when the leaf area index (LAI) is less than 3 (Weiser et al., 1986; Serrano et al., 2000), and when collected with ground-based platforms, aircraft, or satellites, offers a nondestructive and minimally invasive method of sampling herbage mass. Correlations between NDVI and herbage biomass have been shown in barley (Wendroth et al., 2003), the shortgrass steppe (r2 = 0.66) (Todd et al., 1998); alfalfa (Medicago sativa L.) (r2 = 0.89) (Mitchell et al., 1990), and winter wheat (Triticum aestivum) (r2 = 0.60–0.78) (Moges et al., 2004).

New developments in sensor technology have led to the creation of active, on-the-go, ground-based sensors for collecting reflectance data and calculating NDVI. One such sensor, the Greenseeker (Ukiah, CA) has been widely accepted for mapping canopy reflectance in a variety of different crops. This sensor, which emits and records its own light and reflectance, filters out ambient light. This feature gives the sensor the ability to be used day or night without being influenced by current light conditions.

The objectives of this study were to (i) estimate herbage mass using NDVI, (ii) determine if NDVI may be used to assess spatial variability of herbage mass of grasslands, and (iii) determine if NDVI may be used to evaluate management of grazing systems.


   MATERIALS AND METHODS
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Sites and Botanical Composition
Two grassland sites were used. The first site (2004) was located on a 2.60-ha tall fescue (‘Select’) (Soreng et al., 2001) hayfield located on the Kentucky Agricultural Experiment Station Spindletop farm (38°10' N, 84°49' W). The cultivar ‘Select’ is not infected with Acremonium coenophialum Morgan-Jones and Gams. The sward was largely composed of the tall fescue, but did contain < 5% nimblewill (Muhlenbergia schreberi J.F. Gmel.), Kentucky bluegrass (Poa pratensis L.), and alfalfa (Medicago sativa L.).

The second site (2005) consisted of three 3.0-ha pastures of endophyte-infected (E+) tall fescue (‘Kentucky 31’) in a stocking rate (SR) experiment at the University of Kentucky Animal Research Center (ARC) (38°50' N, 84°44' W). Swards were composed of tall fescue with < 5% Carolina horsenettle (Solanum carolinense L.)

Weather
In 2004 and 2005, extremes in precipitation were recorded for the Commonwealth of Kentucky (http://wwwagwx.ca.uky.edu/cgi-public/farm_www.ehtml, verified 17 Oct. 2007): 2004 is one of the wettest years on record with 142 cm of precipitation, and 2005 is one of the driest with 94 cm (124 cm norm). During 2004, Spindletop Farm reported 133 cm of rainfall, of which 100 cm was received during the growing season (15 March–19 November) . The following year (2005), ARC reported only 69 cm of rainfall with only 46 cm received during the growing season.

Soils
The field used in Exp. 1 covered 2.60 ha, of which 78% was composed of a Maury silt loam (MiB) (fine, mixed, semiactive, mesic Typic Paleudalfs) and 22% a Lanton silty clay loam (dunning) (La) (fine-silty, mixed, superactive, thermic, Cumulic Epiaquolls. Exp. 2 consisted of three 3.0-ha fields: Field 1 consisted of 63% Donerail silt loam (Dob) (fine, mixed, active, mesic Oxyaquic Argiudolls), 15% MiB, and 22% Maury silt loam (MiC); Field 2 was composed of 17% Dob, 48% MiB, 2% MiC, and 33% McAfee silt loam (MnC) (fine, mixed, active mesic Mollic Hapludalfs); Field 3 consisted of 1% Dob, 4% Huntington silt loam (Hu) (fine-silty, mixed, active mesic Fluventic Hapludolls), 48% Lowell silt loam (LwB) (fine, mixed, active, mesic Typic Hapludalfs), 37% LwC, and 10% MiB (http://websoilsurvey.nrcs.usda.gov/app [verified 27 Oct. 2007]).

Canopy Reflectance
Canopy reflectance data were recorded by a Greenseeker RT500 variable rate application and mapping system. The scanner has eight active, on-the-go, ground-based sensors spaced 0.76 m apart on a 6.1-m boom. The sensors are independent of ambient light and can be used either day or night regardless of light conditions. High intensity light emitting diodes pulse the canopy with red (660 nm) and near infrared (780 nm) radiation (25-nm band widths) at high frequencies while a photodiode detector measures the reflected light across a 61-cm swath. From the reflectance data, the NDVI was calculated and averaged approximately every linear 0.76 m to create individual data points that represented 0.46 m2 of canopy. The Greenseeker system calculates NDVI as follows:

Formula

Data were processed using a Greenseeker specific post-processing program that georeferences and stores data in a format suitable for global information system (GIS) analysis. For a more detailed description of this system and the sensor specifications, see www.ntechindustries.com/greenseeker-home.html (verified 27 Oct. 2007).

Spindletop Hayfield
At Spindletop, NDVI was correlated with biomass determined with a calibrated RPM or a weighing forage harvester. To prevent direct and indirect measurements from interfering with each other, methods were evaluated during two different periods of growth, with indirect measurements taken on 14, 21, and 29 July, and direct measurements taken on 7 and 22 October, and 7 November.

The RPM was calibrated against herbage mass by taking five measurements along a linear transect (2 by 0.41 m [0.82 m2]) and then cutting to the soil surface with a Stihl HS 80 hedge trimmer (Stihl, Virginia Beach, VA). Herbage samples were oven-dried to a constant weight at 80°C and weighed. The Hege 212 Forage plot harvester (Wintersteiger Inc., Salt Lake City, UT) was used to cut and weigh biomass from 36 transects (6 m x 6 m). Subsamples from each transect were dried to a constant weight at 80°C.

The Greenseeker was used to collect NDVI data each week for 3 wk throughout the month of July (14th, 21st, and 29th). A sampling grid (22.86 x 22.86 m) was used to locate 50 sampling points to relate biomass and NDVI. Ten simple random samples of compressed sward surface height measurements (CSSH) were taken within 9 m of each point to estimate the average CSSH. Biomass at each sampling point was estimated from an algorithm developed during the calibration of the RPM.

Data were processed in ArcMap 9.0. (Environmental Systems Research Institute, 2004. Spatial buffers, 11.43 m in radius, were created and centered on each sampling point using the Buffering Wizard tool of ArcMap. Each buffer was assigned the mean CSSH, the global positioning system (GPS) coordinate of sampling point, and mean NDVI. Average biomass calculated from CSSH was regressed against the average NDVI for each buffer.

For three consecutive weeks in October 2004, NDVI was also determined for 36 random 6-m x 6-m areas (12 per week), which were then cut to 5-cm stubble height with the forage harvester to determine biomass. Biomass was regressed against the mean NDVI for each area.

ARC Stocking Rate Pastures
The second grassland site comprised of three endophyte-infected tall fescue pastures grazed with beef stockers at SRs of 4.3, 6.3, and 8.3 head ha–1 (1226, 1780, or 2344 kg BW ha–1, respectively). The Greenseeker was used to determine NDVI on Day 56 (18 May 2005) of continuous grazing since spring greenup. Variograms were used to determine if NDVI data were spatially structured (Nielsen and Wendroth, 2003).

Twenty CSSH were taken with a GPS-enabled RPM (Flynn et al., 2006) along three transects within each field and downloaded to ArcMap 9.0 (Environmental Systems Research Institute, 2004). Georeferenced buffers were created around each sampling point and assigned CSSH and mean NDVI. Optimum buffer size was determined from correlation coefficients and mean absolute error of linear regression models of buffers ranging from 0.5 to 4.0 m. Spatial coregionalization models were also used to confirm optimum buffer size (Nielsen and Wendroth, 2003).

Herbage biomass data for fields were derived from relationships between herbage biomass and NDVI. Frequency distributions of dry matter (DM) in fields are presented as histograms and defined by skewness and kurtosis. Guidelines for classes of herbage DM availability (kg DM ha–1) for grazing cattle were at levels specified by Dougherty and Collins (2003).

Statistical Analysis
Regression models for NDVI and CSSH were made with SAS Proc Reg (SAS Institute, 2003). Spatial analysis of NDVI and CSSH was conducted with GS+ software (Gamma Design Software, 2005). When NDVI and CSSH were used in cokriging, CSSH was the primary variate with NDVI being the covariate. Semivariograms and crossvariograms created for cokriging were fitted with exponential or Gaussian models. Models are defined as (Deutsch and Journel, 1997):

Exponential:

Formula

Gaussian:

Formula
where h = lag distance; Co = nugget variance ≥ 0; C = structural variance ≥ 0; Ao = range parameter (Gamma Design Software, 2005). When regression and cokriging were compared, the mean absolute error was used as a measure of precision:

Formula
where xi = the measured value of CSSH;xi=the predicted value; n = the number of samples (Nielsen and Wendroth, 2003). Histograms were presented and analyzed using ArcMap 9.0 (Environmental Systems Research Institute, 2004), with class intervals set at increments of 100 (e.g., 100–199 kg). The moment of order k (µk) is defined as:

Formula
where n = sample size, zi = the sample element, z=sample mean, i = the moment, and k = the order (Nielsen and Wendroth, 2003).

Skewness and kurtosis were calculated as follows:

Formula

Formula
where µ2, µ3, and µ4 = the second, third, and fourth moments.


   RESULTS
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Spindletop Field
Estimation of Herbage DM Biomass from NDVI and CSSH
The CSSH was correlated with herbage DM biomass over the three sampling dates in July 2004 (wet year) (DM = 233.6 CSSH- 382.5; P = 0.02; n = 24; r2 = 0.89). Ten CSSH measurements were taken and averaged and then converted into DM estimates by the previously described equation for each buffer (11.43 m radius) for the 50 predetermined sampling grid points. DM estimates were then regressed against mean NDVI within each buffer. Coefficients of determination for linear relations increased with the date of sampling (14 July, 21 July, and 29 July) (r2 = 0.10, 0.31, and 0.43, respectively). Regression of pooled data from all sampling dates (DM = 12,637 NDVI – 8346.5; n = 150; r2 = 0.54; P = 0.0001) (Fig. 1 ) supported the use of NDVI to estimate herbage DM biomass. Cokriging was used in an effort to improve estimates of herbage DM biomass by taking into account spatial correlations between sampling sites. When the mean absolute error of cokriged estimates was compared to that of the linear regression models, only the first sampling date favored cokriging as the estimation method (errorabs = 0.54 compared with 0.58).


Figure 1
View larger version (21K):
[in this window]
[in a new window]

  		Fig. 1. Regression of herbage dry matter (DM) biomass determined with the rising plate meter (RPM) and the normalized difference vegetation index (NDVI) for the Spindletop hayfield in 2004 (wet year).

 
Estimation of Herbage DM Biomass from NDVI by Destructive Harvests
The relationship between NDVI and herbage DM biomass collected with the forage plot harvester (pooled over 3 wk) in autumn 2004 (wet year) (DM = 743.58 NDVI – 369.53; P = 0.0001; n = 36; r2 = 0.68) (Fig. 2 ) was more precise than the regression for NDVI and biomass estimated by the RPM method. When analyzed by sample date (7 and 22 October and 7 November), the correlation coefficients of the regression models were inferior to those determined with the pooled data. In contrast to the DM-NDVI relationships, the linear correlation coefficients decreased with the date of sampling (r2 = 0.62, 0.47, and 0.27, respectively). Cokriging of herbage DM biomass and NDVI was deemed invalid because of the limited number of observations.


Figure 2
View larger version (14K):
[in this window]
[in a new window]

  		Fig. 2. Regression of the normalized difference vegetation index (NDVI) against herbage mass (dry matter [DM], kg ha–1) determined by direct harvesting in the Spindletop hayfield in autumn 2004 (wet year).

 
ARC Stocking Rate Pastures
Spatial Structure of NDVI in Grazed Pastures
Isotropic variograms, derived from {approx}56,000 NDVI observations per field, were used to determine the spatial structure of NDVI within the SR pastures. Exponential semivariogram models were the best fit for all fields, and all were similar with respect to the nugget, sill, and range parameters (Table 1 ). However, these parameters still highlighted key differences in spatial variability within swards attributable to differential SRs (Fig. 3 ).


View this table:
[in this window]
[in a new window]

  		Table 1. Parameters of the exponential models used to describe the semivariograms of the normalized difference vegetation index data for each stocking rate (Animal Research Center stocking rate pastures).

 

Figure 3
View larger version (16K):
[in this window]
[in a new window]

  		Fig. 3. Semivariograms of the normalized difference vegetation index (NDVI) data from the Animal Research Center pastures grazed by beef stockers at the following stocking rates (SRs): circle denotes light SR; triangle denotes intermediate SR; square denotes heavy SR.

 
Determination of Buffer Radius for CSSH and NDVI
Approximately 50 to 60 CSSH observations were recorded along three transects within each field and then overlaid on NDVI maps using ESRI ArcMap Software (Environmental Systems Research Institute, 2004). Each CSSH observation was logged and georeferenced individually and assigned a series of buffers ranging from 0.5 to 4.0 m in 0.5-m increments to help identify an optimum buffer size for averaging NDVI data. Mean absolute errors of cokriged buffer data were similar and could not be used to identify optimum buffer radius (Table 2 ). Regression models, however, identified the 2.0-m buffer radius as optimum (Table 2, and Fig. 4 ). Herbage biomass within 2-m buffers was estimated from CSSH using the regression equation derived previously (Herbage biomass = 231.68 CSSH- 57.67; P = 0.0001; n = 19, r2 = 0.69). When cokriging was applied to the same 2-m buffer data set it was determined that cokriging (errorabs = 2.23) only had a slight advantage over linear regression estimates (errorabs = 2.28).


View this table:
[in this window]
[in a new window]

  		Table 2. Comparison of buffers using regression and cokriging of the normalized difference vegetation index and herbage dry matter to identify an optimum buffering size.

 

Figure 4
View larger version (24K):
[in this window]
[in a new window]

  		Fig. 4. Relationship between the normalized difference vegetation index (NDVI) and herbage mass of all Animal Research Center stocking rate pastures using the optimal 2.0-m buffer radius (dry year).

 
Optimum Buffer Size for Individual Fields
Analysis of DM estimates on a per-field basis revealed differing optimum buffer sizes (Table 3 ) with Field 1 (light SR) showing the strongest correlation at 3.0 m (DM = 19,066 NDVI – 11,536; P = 0.0001; n = 69; r2 = 0.23), Field 2 (intermediate SR) at 0.5 m (DM = 2916 NDVI + 561; P = 0.11; n = 58; r2 = 0.04), and Field 3 (heavy SR) at 2.0 m (DM = 5065.8 NDVI –1880; P = 0.0003; n = 53; r2 = 0.23). When cokriging was applied to these buffer data sets from individual fields, the mean absolute errors of cokriging models were higher than those of regression models with the exception of Field 1 (light SR) (Table 4 ). Although cokriging of buffer data showed a slight improvement in the mean absolute error in some situations it could not be used to interpolate herbage DM biomass from raw NDVI data sets due to the lack of identical range parameters for variograms and crossvariogram. Due to this inconsistency, coregionalization models failed to meet the conditions needed to ensure "positive definiteness" (Nielsen and Wendroth, 2003). Thus DM was estimated from NDVI with a linear regression model.


View this table:
[in this window]
[in a new window]

  		Table 3. Correlation coefficients for the normalized difference vegetation index and compressed sward surface heights for 0.5- to 4-m buffer zones for pastures grazed at three stocking rates.

 

View this table:
[in this window]
[in a new window]

  		Table 4. Mean absolute errors of regression and cokriging for the estimation of CSSH from NDVI at optimum buffer radius for stocking rate pastures.

 
Estimating Biomass from NDVI
Herbage biomass (kg DM ha–1) was estimated for each NDVI point (0.46 m2) from the pooled 2-m buffer equation previously described. Herbage biomass was set to 0 kg DM ha–1 when NDVI values were <0.48. This was determined by the point of interception between the 2-m buffer equation and the x-axis. Guidelines for herbage DM availability were set as follows: inaccessible <840 kg DM ha–1, restricted 840 – 1680 kg DM ha–1, and nonrestricted >1680 kg DM ha–1, as suggested by Dougherty and Collins (2003).

Spatial Analysis of Herbage Mass
Histograms of herbage mass of fields grazed at the light and intermediate SRs indicate that data were negatively skewed (–0.97 and –0.62, respectively), and with an above normal (3.0) coefficient of kurtosis (4.75 and 3.5, respectively) (Fig. 5 ). In contrast, data from the heavy SR was positively skewed (0.27) and exhibited below-normal kurtosis (2.75). These distributions may reflect differences in herbage utilization and grazing behavior. The field grazed at the light SR had a mean herbage biomass of 2693 ± 532 kg DM ha–1 with sampled areas (.46 m2) ranging from 0 to 4218 kg DM ha–1. About 95% of these areas were above 1680 kg DM ha–1 and, thus, may not limit intake. At the intermediate SR, which had a mean herbage mass of 2370 ± 618 kg DM ha–1 and similar range of DM, 86% of areas should not have limited intake per day. The field grazed at the heaviest SR had a much lower mean herbage biomass (1576 ± 627 kg DM ha–1) but only 41% of the areas were considered to be limiting of intake. However, this field did have a similar range of herbage biomass (0–3923 kg DM ha–1).


Figure 5
View larger version (22K):
[in this window]
[in a new window]

  		Fig. 5. Histograms of herbage mass observations for Animal Research Center pastures grazed at three stocking rates (SRs). Each observation represents herbage mass per 0.46 m2 ({approx}56,000 observations per 3 ha).

 
When only pasture areas above the level of limiting intake were analyzed, a mean herbage biomass of 2763 ± 430 kg DM ha–1 and 2539 ± 449 kg DM ha–1 were observed for the light and intermediate SRs, respectively. The field grazed at the heaviest SR, however, had a mean biomass of 2182 ± 383 kg DM ha–1. It was determined from these observations that the pastures had 3087, 2216, and 617 kg of herbage DM available for grazing before intake may become limiting at light, intermediate, and heavy SRs, respectively,


   DISCUSSION
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
CONCLUSIONS
 REFERENCES
 
Spindletop Hayfield
Calibration Methods
The nondestructive RPM procedure was better suited than direct harvesting for the definition of spatial variability of grasslands because it allows more observations and minimizes perturbations. The higher coefficient of determination for the regression of NDVI against biomass determined with the direct harvesting method was expected due to the accuracy of direct measurements (Gourley and McGowan, 1991; Harmoney et al., 1997; Sanderson et al., 2001; Tarr et al., 2005; Martin et al., 2005). However, harvesting of 12 sampling sites took 3 h compared with 1.5 to 2 h needed to record 50 CSSHs with a RPM calibrated for the estimation of herbage mass, making it less desirable for large areas.

Different trends in correlation coefficients observed for each calibration method may have been the result of several different factors. For instance in July, DM of the hayfield became increasingly spatially heterogeneous over time, probably as a result of spatial variations in soil fertility and availability of soil water (Heckrath et al., 2005). This was reflected in a wide range of biomass (727 – 3144 kg DM ha–1) over a narrow range of NDVI (0.73 – 0.84). In contrast, fall growth was poor, as indicated by herbage biomass (34 – 293 kg DM ha–1) and NDVI ranged from 0.53 to 0.78. This apparent anomaly may be explained by growth of C3 grasses in autumn. In fall, these grasses exhibit high rates of tillering and low rates of leaf elongation (Nelson et al., 1977; Zarrough et al., 1983). This growth pattern may have increased herbage mass below 5 cm (the sample cutting height) and altered the relationship between NDVI and herbage DM. Weiser et al. (1986) and Serrano et al. (2000) indicated that canopy reflectance and NDVI are more closely related to cover than biomass.

ARC Stocking Rate Pastures
Rising Plate Calibration
Regression between CSSH and NDVI was inferior to the regression attained in the Spindletop hayfield. This may have been an artifact of pasture management. Mowing of tall fescue swards at 20 cm to suppress flowering culms resulted in spatially variable populations of senescent pseudostems that may have impeded sward compression and RPM function (Arias et al., 1990).

Spatial Dependency of NDVI
The exponential variogram model type chosen for fitting the experimental variogram provides a range parameter only and, therefore, does not allow a precise definition of the spatial range of dependence, as would a spherical model (Nielsen and Wendroth, 2003). Nevertheless, we determined that a 0.76-m distance between sampling locations to be appropriate for identifying the spatial continuity of NDVI.

Expression of Patch Grazing Behavior
The spatial distribution of biomass in the grazed tall fescue pastures may indicate the occurrence of patch grazing. Patch grazing occurs when herbivores seek out feeding stations (patches) that maximize rate of herbage DM intake and minimize grazing time (Launchbaugh and Dougherty, 2007) in accordance with foraging theory (Stephens and Krebs, 1986). Patches within tall fescue pastures were likely selected for grazing because of high herbage mass, perhaps perceived as sward surface height and dark color. Patch selection may also involve avoidance of swards contaminated by excreta (Aiken et al., 1997) or the presence of "grazing barriers" that physically deter prehension (Arias et al., 1990), such as sharp stubble formed from remnants of rigid pseudostem caused by mowing or grazing. Grazed patches accounted for more of the pasture area at the heaviest SR as reported by Aiken et al. (1997) and Cid and Brizuela (1998) (Supplemental Fig. 1).

Effect of Patch Grazing on Spatial Variability and Regression
Comparison of field variograms (NDVI) reveals that small-scale variability was the most pronounced at the intermediate SR (Fig. 3). This was a key observation in explaining why a relatively weak correlation was observed between NDVI and DM (r2 = 0.04) for the intermediate SR, in which the variogram indicated the sharpest increase in the semivariance and the yield map exhibited the most patch grazing. Although patch grazing had occurred at some level in the light SR and heavy SR, these SRs may have reduced the amount of small-scale variability. Most likely, the light SR was not heavy enough to cause extensive patch grazing, leaving large uniform areas of ungrazed forage interspersed with small areas of patch grazing, while the heavy SR was heavy enough to cause large coalesced grazed patches interspersed with small patches of ungrazed forage. These relatively large areas of uniform biomass apparently desensitized the error associated with GPS inaccuracies (Trimble Navigation Limited, 1999) and varying buffer sizes and thus improved regression models. The patchiness of Field 2 (intermediate SR) exacerbated these sources of error making the regression weaker. Graphic representations can be seen in Supplemental Fig. 1.

Cokriging
Precision and applicability were considered in deciding whether regression or cokriging were to be used to determine the best model to estimate biomass from buffer data sets. Spatial parameters of the cokriging model would have favored cokriging over regression regardless of the small difference between the mean absolute errors of the CSSH estimates from cokriged buffer data (errorabs = 2.23) and the linear regression model of buffer data (errorabs = 2.28). However, cokriging was not a valid application for the raw CSSH and NDVI data sets because of a failure to achieve positive definiteness (Nielsen and Wendroth, 2003). To achieve positive definiteness, two semivariograms, one for each variable, and their crossvariogram must be fitted with variogram models that are similar with respect to model type and range. While a common model type (exponential) could be determined with CSSH and NDVI variograms, a common range could not be defined. Therefore, it was concluded that linear regression would be the better model for estimating biomass from NDVI.

Frequency Distribution of Herbage Biomass
All pastures had areas where herbage would not limit DM intake, as indicated by the diagrams of herbage DM biomass above 1680 kg DM ha–1 (Supplemental Fig. 2). While herbage DM above the level of limiting intake was least (617 kg DM) on the pasture grazed at the heaviest SR, there was 3087 and 2216 kg DM, respectively, at the light and intermediate SRs. Skewness and kurtosis statistics of herbage DM distribution (Fig. 5) contribute some information about the spatial structure of grazed pastures. Kurtosis of the low (4.75) and intermediate (3.5) SR pastures indicates a high concentration of herbage in areas about the mean (2693 and 2370 kg DM ha–1, respectively), which may indicate that SRs were too light for optimal utilization. Negative skewness observed for these pastures (Fig. 5) could be the result of patch grazing and selective grazing, by creating more and larger areas of low biomass that resulted in more pronounced small-scale variability (Fig. 3).

The frequency distribution of herbage DM of the pasture grazed at the heaviest SR had kurtosis below normal (2.75), a slightly positive skewness (0.27), and a mean yield (1575 kg DM ha–1) that was lower than that of pastures grazed at lighter SRs (Fig. 5). These statistics may indicate that animals on this pasture utilized more of the available herbage and had fewer opportunities for diet selection than cattle on the other pastures. Although some areas of herbage in all pastures were above the suggested intake-limiting level, more areas may have been rejected near dung patches at the heavier SR. Only the pasture grazed at the heaviest SR may have limited intake per day; the others may be considered to be understocked. Therefore, as grazable patches and feeding stations become further apart, one might anticipate that herbage availability declines and that herbage allowances must be increased to maintain herbage intake per day (Dougherty et al., 1989). Spatial and temporal information may significantly contribute to our understanding of grasslands systems and help to improve the determination of SRs in grazing systems.


   CONCLUSIONS
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
REFERENCES
 
Spatial properties of herbage mass of pasture and hayfields may be readily determined from NDVI data taken at submeter resolution and with CSSH data taken with an RPM calibrated for the estimation of herbage mass.

Frequency distribution parameters of herbage mass, such as skewness, kurtosis, and mean yield, may provide useful information about the spatial properties of pastures and, indirectly, about grazing behavior. Spatial descriptors of continuously grazed swards may indicate patch grazing over a range of SRs.

Spatial and temporal statistics applied to functional grazing systems and to large-scale grazing experiments in combination with remote, nondestructive measurements of soil, plant, and animal components will surely play a primary role in future grassland research, development, and management.


   ACKNOWLEDGMENTS
 
Dr. Eric Vanzant of Dep. of Animal and Food Sciences, Univ. of Kentucky granted access to the stocking rate study at Animal Research Center, Woodford County, KY. Dr. Greg Schwab, Dep. of Plant and Soil Sciences offered use of the Ntech Greenseeker.


   NOTES
TOP
 NOTES
ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher.


   REFERENCES
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 





This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Supplemental Data
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Flynn, E. S.
Right arrow Articles by Wendroth, O.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Flynn, E. S.
Right arrow Articles by Wendroth, O.
Agricola
Right arrow Articles by Flynn, E. S.
Right arrow Articles by Wendroth, O.
Related Collections
Right arrow Grazing Management
Right arrow Geostatistics
Right arrow Remote Sensing

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Crop Science Vadose Zone Journal
Journal of Natural Resources
and Life Sciences Education		 Soil Science Society of America Journal
Journal of Plant Registrations Journal of
Environmental Quality		 The Plant Genome