Published online 7 November 2007
Published in Crop Sci 47:2547-2556 (2007)
© 2007 Crop Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA
FORAGE & GRAZINGLANDS
Relationships between Blue- and Red-based Vegetation Indices and Leaf Area and Yield of Alfalfa
Dennis W. Hancocka,* and
Charles T. Doughertyb
a Dep. of Biosystems and Agricultural Engineering, Univ. of Kentucky, Lexington, KY 40546
b Dep. of Plant and Soil Sciences, Univ. of Kentucky. Submitted with the approval of the Director, KY Agric. Exp. Stn. as publication 07-99-006
* Corresponding author (dhancock{at}uga.edu).
 |
ABSTRACT
|
|---|
The need for site-specific yield assessments of alfalfa (Medicago sativa L.) has spurred interest in developing methods to remotely sense biomass at harvest. Relationships between reflectance-based vegetation indices (VIs) and yield and yield-components of alfalfa have not been fully characterized. The objectives of this study were to evaluate the relationships between blue- and red-reflectance based VIs and canopy variables such as leaf area index (LAI), mass shoot–1, shoot length, and alfalfa yield . Canopy reflectance was obtained with two reflectance spectrometers 1 d before each of five harvests in 2005 within rainfed and subsurface drip-irrigated alfalfa. Blue- and red-based normalized difference vegetation indices (NDVIs) and wide dynamic range vegetation indices (WDRVIs) at three levels of a near-infrared (NIR) reflectance-scalar (‘
’ = 0.1, 0.05, and 0.01) were calculated and regressed on alfalfa canopy variables. A quadratic-plateau model was used to determine when VIs no longer detected yield increments. Both blue- and red-based NDVIs and WDRVIs exhibited significant (P < 0.0001) saturative responses to LAI, yield components, and dry matter (DM) yield. Decreasing widened the estimable yield range (0–1.82 vs. 0–2.76 Mg ha–1 and 0–2.60 vs. 0–3.74 Mg ha–1, respectively) of both blue- and red-based WDRVIs. Significant (P < 0.0001) yield regression models within the effective range of the VIs (<Yieldmax) were found within two harvests in 2005 and when data were pooled across all harvests. These results indicate that the use of a NIR reflectance-scalar can extend the range of herbage biomass (to 3.74 and 2.76 Mg ha–1, respectively) within which blue- and red-based indices may be used to estimate alfalfa yield.
Abbreviations: BNDVI, blue normalized difference vegetation index • BWDRVI, blue-wide dynamic range vegetation index • DM, dry matter • FS, Yara FieldScan • FWHM, full-width half-magnitude • GNDVI, green normalized difference vegetation index • GS, GreenSeeker • LAI, leaf area index • LAI', maximum leaf area index • NDVI, normalized difference vegetation index • NIR, near-infrared • RMSE, root mean square error • SDI, subsurface drip irrigation • SSM, site-specific management • VI, vegetation index • WDRVI, wide dynamic range vegetation index • Yieldmax, the point at which the quadratic function joins the plateau in the relationship between a given VI and yield
Relationships between Blue- and Red-based Vegetation Indices and Leaf Area and Yield of Alfalfa
Dennis W. Hancocka,* and
Charles T. Doughertyb
a Dep. of Biosystems and Agricultural Engineering, Univ. of Kentucky, Lexington, KY 40546
b Dep. of Plant and Soil Sciences, Univ. of Kentucky. Submitted with the approval of the Director, KY Agric. Exp. Stn. as publication 07-99-006
* Corresponding author (dhancock{at}uga.edu).
The need for site-specific yield assessments of alfalfa (Medicago sativa L.) has spurred interest in developing methods to remotely sense biomass at harvest. Relationships between reflectance-based vegetation indices (VIs) and yield and yield-components of alfalfa have not been fully characterized. The objectives of this study were to evaluate the relationships between blue- and red-reflectance based VIs and canopy variables such as leaf area index (LAI), mass shoot–1, shoot length, and alfalfa yield . Canopy reflectance was obtained with two reflectance spectrometers 1 d before each of five harvests in 2005 within rainfed and subsurface drip-irrigated alfalfa. Blue- and red-based normalized difference vegetation indices (NDVIs) and wide dynamic range vegetation indices (WDRVIs) at three levels of a near-infrared (NIR) reflectance-scalar (‘’ = 0.1, 0.05, and 0.01) were calculated and regressed on alfalfa canopy variables. A quadratic-plateau model was used to determine when VIs no longer detected yield increments. Both blue- and red-based NDVIs and WDRVIs exhibited significant (P < 0.0001) saturative responses to LAI, yield components, and dry matter (DM) yield. Decreasing widened the estimable yield range (0–1.82 vs. 0–2.76 Mg ha–1 and 0–2.60 vs. 0–3.74 Mg ha–1, respectively) of both blue- and red-based WDRVIs. Significant (P < 0.0001) yield regression models within the effective range of the VIs (<Yieldmax) were found within two harvests in 2005 and when data were pooled across all harvests. These results indicate that the use of a NIR reflectance-scalar can extend the range of herbage biomass (to 3.74 and 2.76 Mg ha–1, respectively) within which blue- and red-based indices may be used to estimate alfalfa yield.
Abbreviations: BNDVI, blue normalized difference vegetation index • BWDRVI, blue-wide dynamic range vegetation index • DM, dry matter • FS, Yara FieldScan • FWHM, full-width half-magnitude • GNDVI, green normalized difference vegetation index • GS, GreenSeeker • LAI, leaf area index • LAI', maximum leaf area index • NDVI, normalized difference vegetation index • NIR, near-infrared • RMSE, root mean square error • SDI, subsurface drip irrigation • SSM, site-specific management • VI, vegetation index • WDRVI, wide dynamic range vegetation index • Yieldmax, the point at which the quadratic function joins the plateau in the relationship between a given VI and yield
|
INTRODUCTION
|
|---|
SPATIAL AND TEMPORAL variability of yield-limiting factors such as plant available soil moisture and nutrients, have spurred interest in developing site-specific management (SSM) strategies for alfalfa (Medicago sativa L.; Leep et al., 2000; Dolling et al., 2005). Several devices have been developed to measure and georeference forage dry matter (DM) yield (e.g., Michalk and Herbert, 1977; Martel and Savoie, 2000; Sanderson et al., 2001; Savoie et al., 2002; Shinners et al., 2003), but they are either not commercially available or too time-consuming to characterize field variation at a suitable resolution.
Advances in remote sensing and field-ready multispectral spectroradiometers offer promise in site-specific determination of alfalfa yield. Numerous vegetation indices (VIs) relate canopy reflectance to agronomically relevant variables (Bannari et al., 1995; Moran et al., 1997; Pinter et al., 2003; Gitelson, 2004). In general, these indices use the disparity between canopy reflectance in near-infrared (NIR), blue, green, or red wavebands to extract information about biomass (or cover) and/or nutrient status of the plant (Moran et al., 1997; Pinter et al., 2003). Because NIR reflectance increases and red and blue reflectance decrease with vegetative biomass, reflectance in these wavelength bands correlates with green phytomass (reviewed in Pinter et al., 2003). To account for environmental variation in the reflectance values, VIs are usually calculated as the difference, ratio, or other combination (linear or nonlinear) of reflected NIR light and one or more bands within the visible spectrum (Monteith and Unsworth, 1990; Pinter et al., 2003). The widely used normalized difference vegetation index (NDVI), is the normalized difference between NIR and red reflectance (Rouse et al., 1973, Table 6-1). It has been used to estimate alfalfa canopy height during regrowth (Payero et al., 2004), DM availability in variably stocked pastures (Mitchell et al., 1990), and yield of hayfields stressed by pests (Leep et al., 2000).
However, canopy reflectance and, therefore, VIs are affected by many canopy variables. For example, Monteith and Unsworth (1990) show how the fractional reflectance of light for a given wavelength from a plant canopy is a function of that canopy's properties (e.g., canopy height, leaf architecture, chlorophyll content, leaf thickness) and leaf area index (LAI). Thus, the relationships between VIs and canopy variables (such as LAI, shoot mass, and canopy height) provides the fundamental context for the relationships between the VIs and biomass (Pinter et al., 2003; Payero et al., 2004; Gitelson, 2004).
Furthermore, VIs demonstrate a saturative response to vegetative biomass (Moran et al., 1997; Pinter et al., 2003; Gitelson, 2004). This is because canopy reflectance asymptotically approaches a wavelength-specific limit as the LAI approaches a maximum (LAI'; Monteith and Unsworth, 1990). The saturative nature of canopy reflectance, therefore, confines the assessment of vegetation biomass to those conditions where LAI is substantially less than LAI'.
The NDVI and similar indices, such as the green- (GNDVI; Gitelson et al., 1996) and blue-based (BNDVI, Yang et al., 2004) indices (Table 1
), are very sensitive to "saturation" (Gitelson, 2004). The NIR reflectance is typically an order of magnitude greater than red reflectance (Gates et al., 1965; Gausman and Allen, 1973; Wiegand and Richardson, 1984; Slaton et al., 2001; Gitelson, 2004) and this difference widens as the canopy approaches LAI' (Gitelson, 2004). This led Gitelson (2004) to propose the wide dynamic range vegetation index (WDRVI), which incorporates a scalar ("") to de-emphasize the NIR reflectance contribution to NDVI (Table 1). As decreases, the WDRVI's rate of increase with biomass diminishes and causes the relationship to saturate at higher yields, effectively widening the yield range over which the index responds to phytomass. The success of this recent modification of NDVI in other important crops warrants an examination in alfalfa.
View this table:
[in this window]
[in a new window]
|
Table 1. Equations and the reflectance (R) bands used for calculating the normalized difference vegetation indices (NDVI) and wide dynamic range vegetation indices (WDRVI) used in this study.
|
|
Therefore, the objectives of this work were (i) to evaluate how NIR reflectance of alfalfa canopies influences blue- and red-based NDVIs and WDRVIs, (ii) to characterize the relationship between these indices and canopy variables, (iii) to determine if these VIs exhibit different ranges in yield within which these indices respond to alfalfa phytomass, and (iv) to evaluate the relationship between these VIs and alfalfa yield within those effective yield ranges.
|
MATERIALS AND METHODS
|
|---|
Stands of alfalfa cultivar Garst 631 were established on Maury silt loam (fine, mixed, semiactive, mesic Typic Paleudalf, 2–6% slope) on 1 May 2003 at the Univ. of Kentucky Animal Research Center (84° 44' W, 38° 4' N) in a 4.5-ha site. Five blocks of two main plots (18.3 by 39.6 m) were laid out according to slope and depth to bedrock. Main plots were randomly assigned to be irrigated by subsurface drip irrigation (SDI) or to be rainfed. On 16 to 17 Apr. 2003, the SDI lines were installed in irrigated main plots at a depth of 0.38 cm and on 150 cm centers.
On 5 Aug. 2004, 10 soil cores (2-cm diam.) were taken to 10-cm depth from each of the SDI and rainfed plots. These showed water pH (6.7 ± 0.3) and P (192 mg kg–1, Mehlich III P) to be nonlimiting, but indicated Mehlich III soil K (114 mg kg–1) was suboptimal (Thom and Dollarhide, 1994). Four split-plots (2.4 by 6.1 m, with 0.6 m borders) were randomly located in a cluster within each main plot. These split-plots received 0, 112, 336, or 448 kg K2O ha–1 (as KCl) on 1 Oct. 2004. Observations from these plots were designated as 2005K. Since the split-plots occupied only a portion of each main-plot, the remaining area within each main plot received a uniform application of 336 kg K2O ha–1 at the same time. This enabled an additional set of yield and reflectance observations, 2005o, to be obtained from an additional four, predetermined locations (randomized for each regrowth cycle) within the remaining area of the main plots.
Yield Measurements
Favorable growing conditions supported five harvests in 2005 (2005K and 2005o): 5 May (H1), 15 June (H2), 22 July (H3), 23 August (H4), and 30 September (H5). All harvests were made at 10% bloom, with the exception of H4 which was taken at 25% bloom. Herbage samples were cut from a 1.5- by 5.5-m area at 4 cm with a Hege Model 212 Forage Plot Harvester (Wintersteiger Ag, Niederlassung, Germany) and weighed to within ± 0.1 kg. The length of the harvested area was restricted to 0.5 m from the ends of the plots and measured to within ± 3 cm. Forage mass was corrected for dry weight after drying samples from each plot to a constant weight at 60°C in a forced air dryer.
Leaf Area Index and Yield Component Measurements
Immediately before each of the final four harvests (H2–H5), two herbage subsamples (0.3 m2) were clipped and the number of viable crowns were counted in each plot for the 2005K observation set. All herbage samples were taken within a 3 h period and within 1 d of plot harvest. The subsamples were taken at 2 cm above the soil surface using a Model HS 80 Stihl (Stihl, Inc. Virginia Beach, VA) hedge trimmer. The mass from each clipping was weighed, placed in labeled plastic bags and covered in ice for transport to a 2°C refrigerator. Within 1 wk of sampling, the number of shoots was counted and a subset of 10 shoots was randomly selected and stem length, stem diameter above the first node, and number of fully expanded trifoliate leaves were recorded for each shoot (i.e., leaves shoot–1).
Fully expanded trifoliate leaves and petioles were excised from 10 shoot subsamples acquired from the 0 and 448 kg K2O ha–1 treatments. Leaf area was measured to the nearest 0.01 cm2 on a LICOR, LI-3100 area meter (Li-Cor, Lincoln, NE), consistent with the procedures of Powell and Bork (2005). The LAI was determined from the leaf area per 10 stems and stems m–2. The dry mass and the mean mass shoot–1 from the 10 shoots (leaves and stems were pooled for 0 and 448 kg K2O ha–1 treatments) was obtained following 3 d at 60°C in a forced air dryer.
Description of Multispectral Sensors
Canopy reflectance measurements were made with two field-ready multispectral sensors: the Yara FieldScan (FS; Yara International ASA, Oslo, Norway) and the GreenSeeker Model 505 (GS; NTech Industries, Inc., Ukiah, CA). Since the FS and GS are fundamentally different and examples of the passive and active sensors currently available to producers, it is important to analyze these relationships using both types of devices.
The Yara FieldScan is a passive device that uses two, factory-calibrated, diode-array spectrophotometers to quantify light reflected from the target as a proportion of the incident solar radiation at each of up to 20 wavelength bands (± 10 nm full-width half-magnitude [FWHM]; [tec5USA, 2005]). To accommodate the plot width, the sensor toolbar was mounted parallel on a four-wheeled cart and oriented so that only reflectance from the plot was recorded without disturbing the crop canopy (Hancock, 2006). Data were georeferenced ( ± 2 m) using a Holux GM-210 (HOLUX Technology, Inc., Taipei, Taiwan) GPS receiver that matched the output frequency (1 Hz) of the FS. The cart carrying the sensor was pushed at a walking pace recording 7 to 10 observations per plot. Analysis of the canopy reflectance data revealed that blue, red, and NIR wavelengths were strongly correlated with alfalfa yield, yield components, and LAI (Hancock, 2006). Two blue- and red-based normalized difference vegetation indices (NDVIFS and BNDVI, respectively) and two wide dynamic range vegetation indices (WDRVI and BWDRVI, respectively) were calculated at each of three levels of "" (0.1, 0.05, and 0.01) using the blue (450 nm) or red (660 nm) wavelengths and the NIR (770 nm) wavelengths. The specific red and NIR wavelengths were chosen because they are used by the Greenseeker to estimate NDVIGS.
In contrast to the FS, the GS is an "active" device in that it illuminates the target with red (660 nm) and NIR light (770 nm) in a linear 0.6 by 0.01 m strip using two rows of light-emitting diodes. It measures reflected light in only the red (660 nm ± 10 nm FWHM) and NIR (770 ± 15 nm FWHM) bands with a single, factory-calibrated photoelectric diode (NTech Industries, 2005). Reflectance data is recorded at 1000 measurements s–1 but NDVI was estimated, georeferenced ( ± 2 m) and recorded at 1 Hz using GPSCapture software (NTech Industries, Inc., Ukiah, CA) and a Holux GM-270 (HOLUX Technology, Inc., Taipei, Taiwan).
Canopy Reflectance Measurements
Canopy reflectance was recorded in the 2005K and 2005o plots 1 to 2 d before each of the five harvests on days and times when the weather was "mostly sunny" to "partly cloudy" (Table 2
). On partly cloudy days (4 May, 14 June, and 22 Aug), data were only taken when plots were in full sun. Canopy reflectance data was taken ± 1 h of solar noon and recorded in opposite (NE and SW) directions along the plot length to minimize effects of azimuth, as recommended by Guan and Nutter (2001). Data were converted to ASCII text files and processed using ArcGIS 9.0 (ESRI, Inc., Redlands, CA). Measurements taken 0.5 m or less from inside the plot edge were discarded and reflectance values were averaged for each plot.
View this table:
[in this window]
[in a new window]
|
Table 2. Radiant flux characteristics while reflectance was measured from alfalfa canopies on the day before harvest.
|
|
Data Analysis
The repeated measures option of MIXED models procedure of SAS 9.1 (SAS Institute, 2004) was used to analyze for treatment effects on the VIs across multiple harvests. Regression equations were derived with the MODEL and REG procedures of SAS 9.1 (SAS Institute, 2004). A quadratic-plateau analysis was performed using the NLIN procedure in SAS 9.1 (SAS Institute, 2004). This analysis is performed by fitting a quadratic model to the dataset, setting the first derivative of the equation equal to zero to determine the peak of the curve (joint point), then setting all predicted values that correspond with independent variable values greater than the value of the joint point's independent variable value equal to the value of Y predicted at the joint point (plateau). Then, additional iterations of quadratic model fitting and joint point determinations are made to optimize the fit of the quadratic and plateau splices. Standard errors for the joint points were calculated using an IML procedure script in SAS 9.1 (SAS Institute, 2004; P.L. Cornelius, personal communication, 2006).
|
RESULTS AND DISCUSSION
|
|---|
Repeated measures analysis of observation set 2005K indicated that the effects of K fertilization was not dependent (P > 0.05) on harvest date (Hancock, 2006; Hancock and Dougherty, 2006). Furthermore, yield within a harvest and vegetation indices were not affected (P > 0.05) by K application (Hancock, 2006; Hancock and Dougherty, 2006). Repeated measures analysis also indicated that although there was a significant (P < 0.001) interaction between harvest date and irrigation treatment, it was only seen in data from the droughted H3 and H4 regrowth periods (P < 0.05). Therefore, we pooled canopy reflectance and yield data from both observation sets (2005o and 2005K) across K and irrigation treatments to facilitate interpretation and to allow for an evaluation of the relationships between the response variables within a wide range.
Relationships between the Canopy Properties of Alfalfa and the Red- and Blue-based Vegetation Indices
All of the VIs exhibited a significant (P < 0.0001) saturative exponential response with LAI and mass shoot–1 (Fig. 1
and 2
). Each VI also exhibited a strong (r2 = 0.66- 0.82) saturative exponential response to shoot length (Fig. 3
). This is consistent with Payero et al. (2004) who found a strong (r2 > 0.92) exponential relationship between the height of an alfalfa canopy during regrowth and 11 red-based vegetation indices. The relationship between NDVIGS and LAI, yield components, and shoot length was slightly, but consistently stronger than NDVIFS (Fig. 1). However, this may be partly attributable to the larger number of NDVIGS observations per plot (i.e., recording rate of the GS is greater than FS). Mitchell et al. (1990) also found NDVI was correlated with leaf, stem, and total yield of grazed pastures.
View larger version (33K):
[in this window]
[in a new window]
|
Figure 1. Relationship of the blue- and red-based normalized difference vegetation indices (NDVIs) to leaf area index (LAI), mass shoot–1, and shoot length. At each of the final four harvests in 2005, 20 LAI measurements and 40 mass shoot–1 and shoot length observations were taken from each harvested plot.
|
|
View larger version (34K):
[in this window]
[in a new window]
|
Figure 2. Relationship of red-based wide dynamic range vegetation indices (WDRVIs) at ‘’ levels of 0.1, 0.05, and 0.01 to leaf area index (LAI), mass shoot–1, and shoot length. At each of the final four harvests in 2005, 20 LAI measurements and 40 mass shoot–1 and shoot length observations were taken from each harvested plot.
|
|
View larger version (35K):
[in this window]
[in a new window]
|
Figure 3. Relationship of blue-based wide dynamic range vegetation indices (WDRVIs) at ‘’ levels of 0.1, 0.05, and 0.01 to leaf area index (LAI), mass shoot–1, and shoot length. At each of the final four harvests in 2005, 20 LAI measurements and 40 mass shoot–1 and shoot length observations were taken from each harvested plot.
|
|
Understanding how these VIs are affected by canopy variables provides the fundamental context for the relationships between the VIs and biomass (Pinter et al., 2003; Payero et al., 2004; Gitelson, 2004). Monteith and Unsworth (1990) show how the fractional reflectance of light for a given wavelength from a plant canopy is a function of that canopy's properties (e.g., canopy height, leaf architecture, chlorophyll content, leaf thickness, etc.) and LAI.
Implicit in the calculation of NDVI is the dominance of NIR reflectance (Table 1). Moreover, as LAI increased, the resulting increases in NIR reflectance were at least an order of magnitude greater than the corresponding decrease in red reflectance (data not shown). This is consistent with the findings of Gitelson (2004). Decreasing in the red- and blue-based WDRVIs did not substantively alter the response of these indices to LAI, mass shoot–1, or shoot length (Fig. 1, 2, and 3), however, the coefficients of determination (r2) did decrease, especially for the blue-based WDRVIs. We observed that red reflectance became increasingly variable as LAI and yield increased, but NIR reflectance remained stable (coefficient of variation [CV] < 18%) across all LAI and yield values (Hancock, 2006). Thus, one reason for the decreased coefficients of determination may have been the result of the increased emphasis on the red reflectance component when NIR reflectance was scaled down.
Relationships between Alfalfa Yield and Red- and Blue-based Vegetation Indices
The saturative responses of each VI to increasing LAI and alfalfa yield (Fig. 4
, 5
, and 6
) were similar. The spliced quadratic-plateau model closely approximates a saturative exponential function, but identifies the yield (Yieldmax) above which the VI does not respond to increases in yield. Thus, Yieldmax (i.e., the point at which the quadratic function joins the plateau) estimates the upper limit of the range in which a given VI predicts alfalfa yield. To determine Yieldmax over the widest range, a spliced quadratic-plateau model was fitted to a dataset pooled across all harvest dates. This model explained much of the observed variation (r2 > 0.65) in each of the blue- and red-based NDVIs and WDRVIs (‘’ = 0.1, 0.5, and 0.01). Further, these VIs demonstrated a range in Yieldmax values (Table 3
).
View larger version (13K):
[in this window]
[in a new window]
|
Figure 4. Quadratic-plateau functions describing the relationship between alfalfa yield and normalized difference vegetation index (NDVI) as measured by the GreenSeeker (NDVIGS) and FieldScan (NDVIFS). Canopy reflectance measurements were made 1 d before each of the five harvests during 2005. Observations for a given harvest consist of the mean VI and total plot yield measured from each plot (n = 80).
|
|
View larger version (22K):
[in this window]
[in a new window]
|
Figure 5. Quadratic-plateau functions describing the relationship between alfalfa yield and wide dynamic range vegetation indices (WDRVIs) calculated using one of three near-infrared (NIR) reflectance-scalars (‘’ = 0.1, 0.05, and 0.01). Canopy reflectance measurements from which the indices are calculated were made 1 d before each of the five harvests during 2005 using the FieldScan. Observations for a given harvest consist of the mean VI and total plot yield measured from each plot (n = 80).
|
|
View larger version (26K):
[in this window]
[in a new window]
|
Figure 6. Quadratic-plateau functions describing the relationship between alfalfa yield and blue-based vegetative indices (blue normalized difference vegetation index [BNDVI] and blue-wide dynamic range vegetation indices [BWDRVIs] calculated using one of three near-infrared [NIR] reflectance-scalars [‘’ = 0.1, 0.05, and 0.01]). Canopy reflectance measurements from which the indices are calculated were made 1 d before each of the five harvests during 2005 using the FieldScan. Observations for a given harvest consist of the mean VI and total plot yield measured from each plot (n = 80).
|
|
Yieldmax values indicate that NDVIGS and NDVIFS would only be useful in differentiating alfalfa yields in the range of 0 to 1.83 ( ± 0.118) and 1.82 ( ± 0.122) Mg ha–1, respectively. Decreasing increased Yieldmax in both the red- and blue-based WDRVIs (Table 3). These more gradual changes in VI in response to alfalfa yield are consistent with the results of Gitelson (2004) with corn (Zea mays L.), soybean [Glycine max (L.) Merr.], and wheat (Triticum aestivum L.). Interestingly, the blue-based VIs exhibited larger Yieldmax values than the red-based counterparts. The canopy floor (i.e., soil, plant residue, etc.) reflects much more red than blue light and the crop canopy absorbs more red than blue light (Monteith and Unsworth, 1990). So as the LAI increases, blue reflectance is expected to decrease less than red reflectance (Monteith and Unsworth, 1990). We observed this trend and found that blue reflectance was linearly related to yield, while red reflectance was curvilinearly related (data not shown). This likely caused the blue-based VIs to respond more gradually to increases in yield.
Evaluation of Red- and Blue-based Vegetation Indices for Predicting Alfalfa Yield within Their Effective Range
After establishing Yieldmax and the effective range for a specific VI, alfalfa yield was regressed on each VI within its effective yield range for each harvest (i.e., only using observations were yield was less than Yieldmax) and with data pooled across all harvests (Table 4
). Significant quadratic relationships were established between each VI and alfalfa yield in H3 (P < 0.05), H4 (P < 0.0001), and when data were pooled across all harvests. Dry conditions are typical in this location during the H3 and H4 regrowth cycles and were observed in 2005 (Hancock, 2006; Hancock and Dougherty, 2006). The drought during these periods effectively widened the range of yields observed in both H3 and H4 (Fig. 4, 5, and 6). Though the quadratic relationship between NDVIGS and alfalfa yield was the strongest (r2 = 0.68) of all VIs when the data were pooled across all harvests, only 164 (41%) of 400 possible observations fell within the effective yield range of the NDVIGS (Table 4). The relationship between NDVIFS and alfalfa yield explained less of the variation (r2 = 0.58).
View this table:
[in this window]
[in a new window]
|
Table 4. Best-fit regression equations, F ratios, fit statistics, and the number and mean value of yield observations included in the analysis of the relationship between blue- and red-based vegetative indices and alfalfa yield. Analysis included only those observations for which the yield value fell within the effective range of the respective indices. The analysis was performed within each of five harvests in 2005 (H1, H2, ... H5) and on data pooled across all harvests (All).
|
|
The wider effective yield range of the WDRVIs encompassed nearly 60% more data points for red-based VIs and more than 25% more observations for blue-based VIs. Further, the blue-based indices maintained significant (P < 0.0001) relationships with alfalfa yield within H3 (r23 0.18), H4 (r23 0.81), and across all harvests (r23 0.55). In addition, the use of a scalar generally enhanced the fit of the quadratic models within harvests. The relative error [(RMSE of the model/mean of yield values included) x 100] of the significant quadratic models was <30% for each VI. The level of error for the models based on these VIs, both within and across harvests, was at or slightly less than the error (25–40%) reported for conventional in situ forage biomass measurement devices, such as the pasture ruler, capacitance meter, and rising plate meter (Michalk and Herbert, 1977; Sanderson et al., 2001).
|
CONCLUSIONS
|
|---|
Increases in LAI, and related canopy variables, such as mass shoot–1 and shoot height, caused the VIs to exhibit a saturative exponential response. The relationships between the canopy variables and NDVIGS, NDVIFS, and WDRVIs at each level of were stronger when red reflectance was used than when based on blue reflectance. Decreasing slowed the exponential increase in the red- and blue-based WDRVIs in response to increases in LAI, mass shoot–1, and shoot height.
Decreasing also increased the effective yield range for both red- and blue-based WDRVIs. Further, blue-based VIs exhibited a larger effective range of yields than red-based counterparts. The relative error for all models that estimated yield from VIs was at or slightly less than the error (25–40%) reported for other forage biomass measurement devices. We conclude that using an NIR reflectance scalar of 0.01 in calculating red- and blue-based WDRVIs enabled yield variations of alfalfa to be accurately quantified up to 2.76 and 3.74 Mg ha–1, respectively.
|
ACKNOWLEDGMENTS
|
|---|
The contribution of equipment and technical advice by Drs. Thomas Mueller, Dennis Egli, Scott Shearer, and Timothy Stombaugh is gratefully acknowledged. The authors also wish to thank Dr. Paul L. Cornelius, Professor of Statistics, jointly in the Plant and Soil Sciences and Statistics departments, at the University of Kentucky, for creating the NLIN scripts in PROC IML that estimated and compared the joint points in the spliced quadratic-plateau analyses. This material is based on work supported by the Cooperative State Research, Education and Extension Service, U.S. Department of Agriculture, under Agreement No. 2004-34408-15000. Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the authors and do not necessarily reflect the view of the U. S. Department of Agriculture. This is a publication of the Kentucky Agricultural Experiment Station (07-99-006).
|
NOTES
|
|---|
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. Permission for printing and for reprinting the material contained herein has been obtained by the publisher.
Received for publication January 20, 2007.
|
REFERENCES
|
|---|
- Agricultural Weather Center. 2005. Climatology data. University of Kentucky Agricultural Weather Center. Available at http://wwwagwx.ca.uky.edu/climdata.html (verified 10 Sept. 2007).
- Bannari, A., D. Morin, F. Bonn, and A.R. Huete. 1995. A review of vegetation indices. Remote Sens. Rev. 13:95–120.
- Dolling, P.J., R.A. Latta, P.R. Ward, M.J. Robertson, and S. Asseng. 2005. Soil water extraction and biomass production by lucerne in the south of Western Australia. Aust. J. Agric. Res. 56:389–404.[CrossRef][Web of Science]
- Gates, D.M., H.J. Keegan, J.C. Schleter, and V.R. Wiedner. 1965. Spectral properties of plants. Appl. Opt. 4:11–20.
- Gausman, H.W., and W.A. Allen. 1973. Optical parameters of leaves of 30 plant species. Plant Physiol. 52:57–62.[Abstract/Free Full Text]
- Gitelson, A.A. 2004. Wide dynamic range vegetation index for remote quantification of biophysical characteristics of vegetation. J. Plant Physiol. 161:165–173.[CrossRef][Web of Science][Medline]
- Gitelson, A.A., Y. Kaufman, and M.N. Merzlyak. 1996. Use of green channel in remote sensing of global vegetation from EOS-MODIS. Remote Sens. Environ. 58:289–298.[CrossRef]
- Guan, J., and F.W. Nutter, Jr. 2001. Factors that affect the quality and quantity of sunlight reflected from alfalfa canopies. Plant Dis. 85:865–874.[CrossRef]
- Hancock, D.W. 2006. Spectral reflectance of canopies of rainfed and subsurface irrigated alfalfa. Ph.D. diss. Univ. of Kentucky, Lexington.
- Hancock, D.W., and C.T. Dougherty. 2006. Response of alfalfa to potassium nutrition and subsurface drip irrigation during a drought year. p. 35. In M.W. Alison, Jr. (ed.) Proc. Am. Forage Grassland Conf., San Antonio, TX. 10–14 Mar. 2006. AFGC, Georgetown, TX.
- Leep, R., R. Brook, D. Warncke, D. Pennington, J. Bronson, B. Steenwyck, P. Jeranyama, J. Lindquist, and J. DeYoung. 2000. Spatial and temporal dynamics of alfalfa production in relation to precision agriculture and integrated crop management. Available at http://www.kbs.msu.edu/farm/reports/2000/Spatial%20and%20Temporal%20Dynamics.pdf (assessed 7 Sept. 2007).
- Martel, H., and P. Savoie. 2000. Sensors to measure mass-flow-rate through a forage harvester. Can. Agric. Eng. 42:123–129.
- Michalk, D.L., and P.K. Herbert. 1977. Assessment of four techniques for estimating yield on dryland pastures. Agron. J. 69:864–868.[Abstract/Free Full Text]
- Mitchell, A.R., P.J. Pinter, Jr., J.N. Guerrero, C.B. Hernandez, and V.L. Marble. 1990. Spectral reflectance measurements of alfalfa under sheep grazing. Agron. J. 82:1098–1103.[Abstract/Free Full Text]
- Monteith, J.L., and M. Unsworth. 1990. Principles of environmental physics. 2nd ed. Edward Arnold, New York.
- Moran, M.S., Y. Inoue, and E.M. Barnes. 1997. Opportunities and limitations for image-based remote sensing in precision crop management. Remote Sens. Environ. 61:319–346.[CrossRef]
- NTech Industries. 2005. Model 505 GreenSeeker Hand Held Optical Sensor Unit Operating Manual. Available at http://www.ntechindustries.com/pdf/Model-505-Hand-Held-Manual-rev-G.pdf (verified 28 Aug. 2007). NTech Industries, Ukiah, CA.
- Payero, J.O., C.M.U. Neale, and J.L. Wright. 2004. Comparison of eleven vegetation indices for estimating plant height of alfalfa and grass. Appl. Eng. Agric. 20:385–393.[Web of Science]
- Pinter, P.J., Jr., J.L. Hatfield, J.S. Schepers, E.M. Barnes, M.S. Moran, C.S.T. Daughtry, and D.R. Upchurch. 2003. Remote sensing for crop management. Photogramm. Eng. Remote Sens. 69:647–664.
- Powell, G.W., and E.W. Bork. 2005. Simulated aspen understory microclimate effects on alfalfa growth. Agron. J. 97:1361–1366.[Abstract/Free Full Text]
- Rouse, J.W., R.H. Haas, Jr., J.A. Schell, and D.W. Deering. 1973. Monitoring vegetation systems in the Great Plains with ERTS. p. 309–317. In Third ERTS Symp., Washington, DC. 10–14 Dec. 1973. NASA SP-351. Vol. 1. U.S. Gov. Print. Office, Washington, DC.
- Sanderson, M.A., C.A. Rotz, S.W. Fultz, and E.B. Rayburn. 2001. Estimating forage mass with a commercial capacitance meter, rising plate meter, and pasture ruler. Agron. J. 93:1281–1286.[Abstract/Free Full Text]
- SAS Institute. 2004. The SAS system for Windows, release 9.1.3. SAS Inst., Cary, NC.
- Savoie, P., P. Lemire, and R. Theriault. 2002. Evaluation of five sensors to estimate mass-flow rate and moisture of grass in a forage harvester. Appl. Eng. Agric. 18:389–397.[Web of Science]
- Shinners, K.J., B.M. Huenink, and C.B. Behringer. 2003. Precision agriculture as applied to North American hay and forage production. In G. Quick (ed.) Proc. of the Int. Conf. on Crop Harvesting and Processing, Louisville, KY. 9–11 Feb. 2003. ASAE No. 701P1103e. ASAE, St. Joseph, MI.
- Slaton, M.R., E.R. Hunt, and W.K. Smith. 2001. Estimating near-infrared leaf reflectance from leaf structural characteristics. Am. J. Bot. 88:278–284.[Abstract/Free Full Text]
- tec5USA. 2005. YARA FieldScan: Multi-spectral system for remote sensing. Available at http://www.tec5hellma.com/Products/FieldScan/index.html (verified 28 Aug. 2007). tec5USA, Plainview, NY.
- Thom, W.O., and J. Dollarhide. 1994. Long-term alfalfa study with K on a Maury silt loam. Bull. 728. Univ. of Kentucky, College of Agriculture, Agricultural Exp. Stn., Lexington.
- Wiegand, C.L., and A.J. Richardson. 1984. Leaf area, light interception, and yield estimates from spectral components analysis. Agron. J. 76:543–548.[Abstract/Free Full Text]
- Yang, C., J.H. Everitt, J.M. Bradford, and D. Murden. 2004. Airborne hyperspectral imagery and yield monitor data for mapping cotton yield variability. Precis. Agric. 5:445–461.[CrossRef]
TOP
ABSTRACT
INTRODUCTION
