Published in Agron J 99:650-656 (2007)
DOI: 10.2134/agronj2006.0155
© 2007 American Society of Agronomy
677 S. Segoe Rd., Madison, WI 53711 USA
Soybean
Development of Vegetation Indices for Identifying Insect Infestations in Soybean
James E. Boarda,*,
Vijay Makaa,
Randy Priceb,
Dina Knightc and
Matthew E. Baurd
a Dep. of Agronomy and Environmental Management, Rm. 104 Sturgis Hall, Louisiana State Univ. Agricultural Center, Baton Rouge, LA 70803
b Dep. of Biological and Agricultural Engineering, 148 Seaton Hall, Kansas State Univ., Manhattan, KS 66506
c Center for Geoinformatics, Louisiana State Univ., Baton Rouge, LA 70803
d Dep. of Entomology, Rm. 402 Life Sciences Bldg., Louisiana State Univ. Agricultural Center, Baton Rouge, LA 70803
* Corresponding author (jboard{at}agctr.lsu.edu)
Received for publication May 18, 2006.
 |
ABSTRACT
|
|---|
Because of greater efficiency relative to conventional methods, interest has developed for using vegetation indices in soybean [Glycine max (L.) Merr.] for identifying areas in a field experiencing injury by defoliating insects. Vegetation indices can indicate leaf area index (LAI) and light interception levels, canopy parameters affected by defoliating insects. Our objectives were to determine the relative accuracy of three vegetation indices for predicting LAI and light interception, and to outline a method for using vegetation indices for identifying areas in a field experiencing insect injury. Several commercial soybean cultivars were planted on a Commerce silt loam soil (fine-silty, mixed, nonacid, thermic Aeric Fluvaquent) near Baton Rouge, LA (USA) (30° N lat) in May 2004 and June 2005. In 2004, differences in LAI and light interception were created by manual defoliation, whereas in 2005, LAI/light interception differences occurred because of cultivars and planting dates. Results indicated that across canopies ranging from very low LAI to canopy closure (95% light interception), the normalized difference vegetation index (NDVI) most accurately predicted LAI and light interception (r2 = 0.93–0.97). Light interception and LAI were linked to NDVI by strong linear regression models, and did not show the quadratic response reported by others. A proposed method for adopting NDVI to identify insect-infested areas is presented.
Abbreviations: GNDVI, green normalized difference vegetation index • LAI, leaf area index • NDVI, normalized difference vegetation index • NIR, near infrared • SR, simple ratio
|
INTRODUCTION
|
|---|
DEFOLIATING insect pests are part of the largest and most diverse guild of insects affecting soybean production in the USA and are considered a major stress, especially in the southeastern USA (Higley, 1992). The three major defoliating insects of soybean are the soybean looper [Pseudoplusia includens (W.)], velvetbean caterpillar [Anticarsia gemmatalis (H.)], and green cloverworm [Plathyena scabra (F.)]. They usually invade soybean fields in the southeastern USA during the seed filling period (R5–R7 stages according to Fehr and Caviness, 1977) (Tynes and Boethel, 1993). Greatest frequency of these attacks occurs in September when soybean is in the mid- to late seed filling period. Previous defoliation studies have indicated that yield response is affected not only by the severity of insect infestation, but also the timing of the attacks. Board et al. (1994) demonstrated that 100% defoliation at the temporal midpoint and three-quarter point of seed filling resulted in yield reductions of 40 and 20%, respectively. Additional studies showed that 40 and 60% defoliation at the midpoint of seed filling resulted in yield losses of 8 and 17%, respectively; whereas the same defoliation at the three-quarter point of seed filling caused no yield reduction (Board et al., 1997).
Estimates of insect numbers are determined by "scouting" selected areas of a field with a sweep net that is brushed against the side of the canopy to collect insects. The collected insects are counted and population per sampling effort determined. Recommendations for spraying are usually made if pest populations (economic thresholds) are 13 to 26 larvae (1.3 cm) m–1 of row (Baldwin et al., 1997). Ideally, soybean fields should be sampled at weekly intervals starting at first flower (R1) and continuing into the seed filling period (Funderburk et al., 1998). Based on localized infestations, spraying of entire fields is recommended.
The large amount of labor involved in collecting insect populations has stimulated research into alternative methods for identifying insect-infested areas. Because insect defoliation reduces yield through light interception effects on canopy photosynthetic activity and/or crop growth rate (Board and Harville, 1993; Ingram et al., 1981), light interception and LAI have been proposed as possible tools for identifying such areas (Browde et al., 1994; Higley, 1992; Haile et al., 1998a, 1998b). Higley (2001) has proposed an economic injury level based on a critical LAI level necessary to avoid yield loss.
Defoliation studies conducted during the first half of the seed filling period demonstrated that yield was reduced only when defoliation was severe enough to reduce LAI below about 3.5, a level at which light interception started falling below the optimal 95% level (Board and Harville, 1993; Board and Tan, 1995; Board et al., 1997). It was therefore concluded that either maintenance of an LAI of 3.5 to 4.0 and/or light interception of 95% could be used as criteria for identifying areas experiencing injury by defoliating pests. Corroborative research using manual and insect-induced defoliation (Malone et al., 2002) and a range of cultivars (Board, 2004) supported these results.
Remote sensing techniques that determine canopy reflectance ratios (vegetation indices) may be useful in determining critical levels of LAI and/or light interception for identifying areas in a field experiencing injury by defoliating insects. If verified, such methods would have much greater efficiency for identifying areas in a field experiencing injury by defoliating insects compared with current methods. Because green leaf surfaces reflect a much smaller amount of incident red light compared with infrared light, spectral reflectance ratios calculated from reflected red and infrared light can indicate leaf area indices between 0 and 100% canopy cover (Wiegand et al., 1991; Carlson and Ripley, 1997). A possible spectral reflectance ratio (spectral reflectance ratios from crop canopies are referred to as vegetation indices) to consider as a method for identifying areas experiencing injury is the normalized difference vegetation index (NDVI) defined as: (RefIR – RefR)/(RefIR + RefR) where RefIR = canopy reflectance of infrared radiation, and RefR = canopy reflectance of red radiation. Another possible criterion is the simple ratio (SR = RefIR/RefR) and the GNDVI (green NDVI) defined as (RefIR – Refg)/(RefIR + Refg) where Refg = canopy reflectance of green light. These vegetation indices have demonstrated useful agronomic applications in soybean and other crops. Leaf area index for wheat (Triticum aestivum L.) could be predicted by the SR and NDVI (Aparicio et al., 2000; Bellairs et al., 1996). They were also used in these studies to predict biomass accumulation, which in turn demonstrated efficacy as an indirect yield selection criterion in cultivar trials. Daughtry et al. (1992) demonstrated a highly significant correlation (r2 = 0.96) in corn between NDVI and light interception. Later research by Shanahan et al. (2001) showed that corn yield could be accurately predicted from the GNDVI (r = 0.70–0.92). Similar results have been found in soybean as in corn and wheat. Light interception and NDVI were shown to be highly correlated in soybean (Daughtry et al., 1992; Holben and Tucker, 1980; Ma et al., 2001). However, some researchers have reported poor relationships between LAI and NDVI when LAI levels approach canopy closure (LAI > 3.0) (Holshouser and Jones, 2002; Asrar et al., 1984; Ahlrichs and Bauer, 1983). Since soybean LAI frequently reaches levels greater than this during seed filling, these reports suggest that vegetation indices may not be useful for predicting canopy parameters under these conditions.
The purpose of this research was to determine the feasibility of using vegetation indices derived from digital photography to identify areas in a soybean field experiencing injury by defoliating insects. Vegetation indices tested were the NDVI, SR, and GNDVI. Specific objectives were to determine the relative accuracy of these three vegetation indices for predicting LAI and light interception across canopies ranging from very low LAI to canopy cover, and to develop a system for using vegetation indices to identify insect infestations based on the normal progression of LAI and light interception during the seed filling period.
|
MATERIALS AND METHODS
|
|---|
Culture
Field studies were done at the Ben Hur Research Farm near Baton Rouge, LA (USA) (30° N lat) in 2004 and 2005 on a Commerce silt loam soil. Row width was 97 cm and seed were sown to create a plant population of 220 000 plants ha–1. Experimental units were 6 m long by 11.6 m wide and consisted of 12 contiguous rows. Based on soil test recommendations, fertilizer was applied in both years of the study before planting at a rate of 0–0–112 kg ha–1 (N–P–K). Weeds, diseases, and insects were suppressed by recommended pesticides.
Experimental Design and Data Obtained for the 2004 Study
Differences in LAI and light interception were created by manual defoliation. Experimental design was a randomized complete block with four replications in a split-plot arrangement. Main plots were two commercial soybean cultivars: Asgrow 5902 (Maturity Group V, determinate growth) and Hartz 4998 (Maturity Group IV, indeterminate growth). Split plots were five defoliation treatments administered during the 1st week of September 2004, 2 wk after the start of seed filling (R5). Previous studies indicated that for nonstressed soybean canopies, LAI reaches a maximum near R5 and no further leaf production occurs after this point (Board, unpublished data, 2006). This optimal LAI level remains constant during the first 2 to 3 wk of the seed filling period. Thus, defoliation occurred for both cultivars when seasonal LAI levels for both cultivars were optimal and regrowth could not occur after defoliation. Treatments were manual defoliation (random leaf removal) to create the following LAI levels: (i) 0% defoliation (control), (ii) 33% defoliation, (iii) 50% defoliation, (iv) 66% defoliation, and (v) 100% defoliation. Treatments were administered to six contiguous rows within each plot.
After completion of defoliation treatments, all plots were sampled (0.5 m2) for LAI and light interception on 8 Sept. 2004. Light interception was determined at randomly selected areas within the defoliated areas of each plot with a Li-Cor line quantum sensor (Li-Cor Co., Lincoln, NE; 1 m long) connected to a LI-1000 data logger (Li-Cor Co., Lincoln, NE). Photosynthetic irradiance was measured at ground level in micromoles per second per square meter as the average of three measurements made with the sensor placed diagonally between two contiguous rows. Irradiance was then measured at the top of the canopy and the light interception percentage determined. All recordings were made parallel to the rows between 1200 and 1300 h under full-sun conditions. Leaf area index was based on sampling a 0.5-m2 interior plot area and placing 50% (by fresh weight) of the leaf blades through a LI-3100 leaf-area meter.
Digital images of the plots were recorded on 15 Sept. 2004 using a camera system mounted on a pole truck. The system allowed viewing of the field from various perspectives around the border of the field easily during the growing season (airplane and good weather not required). However, caution was exercised to record images during either when the sun was out or constant cloud cover was present. Recordings were not taken during partial cloudiness.
The pole truck system consisted of a multispectral camera (Redlake Corp., San Diego, CA, Model MS4100) with a superwide angle lens (Sigma Corp., Tokyo, Japan, 14 mm, f/2.8). This camera was mounted on a 12.8-m telescoping mast (The Will-Burt Co., Orrville, OH, Model 7-42) that was mounted on a 1989, 0.9-Mg flatbed truck (Ford Motor Co., Model F-350). Approximate height of the camera (when the mast was fully extended) was 14.3 m from the ground (with a 0.3-m high camera mount and 1.2-m high truck bed). The mast was rated for up to a 120 km h–1 wind with a 0.3-m cross-sectional area, so the camera could be used while the truck was moving with no adverse effects to the mast or the machinery. The camera was mounted on a two-axis satellite dish rotator (Future Trax Int., Las Vegas, NV, Galaxis Space Scanner model) that allows movement of the camera in both pitch and compass direction and could be controlled inside the truck. A small form factor computer [Shuttle Corp., Montreal, QC, Canada, Model XPC; 2.4-GHz AMD (Sunnyvale, CA) Athlon Processor, 40-GB hard drive] was mounted along side the camera (on top of the mast) and contained a camera link frame grabber (National Instruments Corp., Austin, TX, Model PCI-1428) and software (DuncanTech Corp., Auburn, CA, DT Control-FG camera control software Version 1.04.63) to allow capture and recording of the images on the hard drive of the computer. A monitor, keyboard, and mouse were located in the truck cab (with 16.2-m cables extending to the camera and computer) to allow use of the system while the truck was driven around the field.
Previous experimentation with the camera truck indicated that a skew factor was needed in the calculations for the vegetation indices to account for distance to the plot. This factor was developed by looking at a single plot of both soybeans and grass to determine a curve of normalized vegetation index difference values versus distance from the truck.
The collected images were converted into NDVI, GNDVI, and SR using Leica Geosystems ERDAS software to process the digital imagery. Average Red (R), Green (G), and near infrared (NIR) values were determined as follows: R values were centered at a 670-nm wavelength with a 40-nm bandpass; NIR was centered at 800 nm with a 60-nm bandpass; and G was centered at a 500-nm wavelength having a 40-nm bandpass. Values for NDVI, GNDVI, and SR were calculated as follows: NDVI = (NIR – R)/(NIR + R); GNDVI = (NIR – G)/(NIR + G); and SR = NIR/R. After processing the images with each index, the spectral index values were extracted for rows in selected plots of interest.
Experimental Design for 2005 Study
More cultivars were included in the 2005 study for purposes of detecting possible interactions of Maturity Group (III, IV, and V) and growth habit (determinate and indeterminate) on relationships between LAI/light interception with vegetation indices. Differences in LAI and light interception were created by planting date and cultivar in the 2nd year of the study. Plant size generally declines as planting date is delayed from the normal period and cultivar maturity group declines. Our purpose was to use these two factors to create a wide range of canopy sizes to determine the predictive use of NDVI, GNDVI, and SR. Experimental design was a randomized complete block in a split plot arrangement with four replications. Main plots were three planting dates: an optimal planting on 5 May 2005, a moderately late planting on 14 June 2005, and a late planting on 25 July 2006. Splits plots were four cultivars: AG3905 (Maturity Group III), DP4331 (Maturity Group IV), P95M80 (Maturity Group V), and AG5903 (Maturity Group V).
Sampling for LAI and light interception occurred 17 to 18 Aug. 2005 for the May planting date, 18 to 22 Aug. 2005 for the June planting date, and 24 Aug. 2005 in the July planting date. Sampling dates corresponded to the mid- to late seed filling period for cultivars planted in May, the early seed filling period for those planted in June, and the early flowering period for the July planting date. Thus, treatment factors not only created wide differences in canopy size but also in developmental stage. Digital photography was accomplished on 26 Aug. 2005 near 1200 h using the same method as in 2004.
Data Analyses
Because some canopy reflectance values were missing from the May 2005 data, a Multiple Imputation method was used to calculate the missing data. This method predicts missing values by using existing values from other variables. These predicted values are called "imputes" and the procedure generates a series of imputed data sets for which data analyses are done to create one overall analysis. The imputed missing values were generated by PROC MI (SAS Institute, 1999) and the subsequent statistical analyses were done by PROC GLM (SAS Institute, 1999). Specific cultivar effects on relationships between vegetation indices with LAI and light interception were studied with multivariate analysis. This method employs "dummy variables" to account for the significance of an independent variable (cultivars) influencing dependent variables (LAI and light interception). One cultivar (AG5903) was removed from the analysis (dummy variable) to be a reference group for the other three cultivars for identifying differential cultivar effects on the relationship between vegetation indices with LAI and light interception. A regression model is generated that quantifies (using parameter estimates) significant effects of individual cultivars and vegetation indices on LAI and light interception.
Within both years, correlation and regression analyses between vegetation indices (NDVI, GNDVI, and SR) and canopy parameters (LAI and light interception) were conducted. Analyses were done for treatment combinations averaged across replications: defoliation x cultivar treatment combinations in 2004 and planting date x cultivar treatment combinations in 2005. Regression analyses were done using SAS PROC GLM in which linear, quadratic, and cubic components were successively tested for significance and included if the residual sum of squares was significantly reduced (p < 0.05). No procedure was employed to identify and remove outlier data points. Homogeneity of regression equations across years for specific vegetation index/canopy parameter relationships was accomplished with SAS PROC GLM. Homogenous regression equations were pooled across years.
|
RESULTS
|
|---|
Regression Relationships between Vegetation Indices and LAI/Light Interception
Regression relationships between LAI and light interception with NDVI, GNDVI, and SR varied from linear to quadratic to cubic (Fig. 1
–6
). Regression relationships between specific vegetation indices with either LAI or light interception were not homogenous across years, and therefore are presented separately by year. Simple linear relationships were shown for LAI and light interception regressed on NDVI in both years (Fig. 1 and 2) (r2 = 0.93–0.97). Green NDVI showed linear relationships with LAI and light interception in 2004, but not 2005 (relationships were quadratic) (Fig. 3 and 4). The SR showed the most complicated relationships with LAI and light interception, having quadratic and cubic relationships in 2004 and 2005, respectively (Fig. 5 and 6).
View larger version (14K):
[in this window]
[in a new window]
|
Fig. 1. Relationships for leaf area index (LAI) regressed on normalized difference vegetation index (NDVI), 2004 and 2005.
|
|
View larger version (20K):
[in this window]
[in a new window]
|
Fig. 2. Relationships for light interception regressed on normalized difference vegetation index (NDVI), 2004 and 2005.
|
|
View larger version (16K):
[in this window]
[in a new window]
|
Fig. 3. Relationships for leaf area index (LAI) regressed on green normalized difference vegetation index (GNDVI), 2004 and 2005.
|
|
View larger version (22K):
[in this window]
[in a new window]
|
Fig. 4. Relationships for light interception regressed on the green normalized difference vegetation index (GNDVI), 2004 and 2005.
|
|
Leaf area indices ranged from 0 to 3.5 in 2004 and 0.5 to 4.5 in 2005, producing a wide LAI spectrum for testing the predictability of NDVI (Fig. 1). Regression of LAI on NDVI showed similar linear patterns in both years, although the regression coefficient was greater in 2004 compared with 2005. Thus, specific NDVI levels did not predict similar LAI levels across years. An NDVI of 0.1 predicted an LAI of 2.70 in 2004 compared with an LAI of 2.30 in 2005. A similar result occurred for NDVI/light interception relationships (Fig. 2). In both years, light interception ranged from a low of 10 to 20% to a high of 80 to 90%, slightly below the optimal 95% level. The greater regression coefficient in 2004 vs. 2005 reflected that given NDVI levels predicted much higher levels of light interception in the 1st year compared with 2nd year. For example, in 2004, an NDVI of 0.2 was associated with a light interception of about 80%, whereas in 2005, the same NDVI predicted light interception of only 60%.
Data for regression of GNDVI with LAI and light interception were less consistent across years compared with NDVI regressed against these two canopy parameters (Fig. 3 and 4). Although LAI and light interception showed strong linear relationships with GNDVI in 2004, such was not the case in 2005. When LAI rose above 3.0 and light interception above 70%, data points tended to cluster rather than keeping to a linear continuum (Fig. 3 and 4). Relationships of LAI and light interception with SR were quadratic and cubic in 2004 and 2005, respectively (Fig. 5 and 6). In both years, SR reached a plateau level at LAI > 3.0, and so was effective at identifying LAI only with the 0 to 3.0 range. A similar pattern occurred for light interception. Values for SR topped out at about 70% light interception, and were not effective at distinguishing light interception differences above this level.
Multivariate Analyses for Relationships between Cultivars and Vegetation Indices on Light Interception and LAI
The multivariate analyses revealed significant differences between how the four cultivars in 2005 affected the regression models of LAI and light interception with the vegetation indices (Table 1). As evidenced by parameter estimates, NDVI showed the greatest impact on LAI and light interception. Both Maturity Group V cultivars P95M80 and AG5903 (determinate growth) showed consistent relationships between NDVI and LAI. However, NDVI/LAI relationships for Maturity Group III AG3905 and Maturity Group IV DP4331 (both indeterminate) were significantly different from reference cultivar AG5903. The relationship between NDVI and light interception was less affected by cultivar differences (Table 1). Only Maturity Group III cultivar AG3905 showed a slightly significant (P < 0.05) difference from reference cultivar AG5903, while the other two cultivars behaved similarly to AG5903.
View this table:
[in this window]
[in a new window]
|
Table 1. Multivariate analysis of the leaf area index (LAI) and light interception as influenced by cultivars and vegetation indices (NDVI = normalized difference vegetation index, GNDVI = green normalized vegetation index, SR = simple ratio), 2005.
|
|
|
DISCUSSION
|
|---|
Regression Analyses for LAI and Light Interception vs. Vegetation Indices
Data in the current study demonstrate the feasibility of using vegetation indices for making management decisions about defoliating insects. Such methods have potential use for models using LAI levels as economic injury levels for defoliating pests (Higley, 2001). Important advantages for this are improved prediction of defoliation levels requiring insecticide application, greater accuracy for insecticide application, and reduced sampling costs for defoliating insects. For purposes of identifying canopies where light interception falls below 95% (indicating the possible presence of defoliating pests), the NDVI was the most appropriate vegetation index to use. The strong linear regressions of NDVI with LAI and light interception levels ranging from near-total defoliation to canopy closure (95% light interception, LAI > 4.3; Fig. 1 and 2) demonstrate that this reflectance ratio accurately predicts LAI and light interception as these parameters fall from optimal to suboptimal levels. The GNDVI and SR did not maintain linear relationships up to canopy closure as shown by NDVI (Fig. 3, 4, 5, 6). For both vegetation indices in 2004 and 2005, relationships with LAI and light interception were linear up to an LAI of 3.0 and 70% light interception. However, above this level, both canopy parameters showed plateau responses to further increases in either GNDVI or SR. Consequently, neither vegetation index could distinguish between optimal LAI/light interception levels (3.5–4.0, 95%), and the suboptimal levels below this which may indicate infestations of defoliating insects.
The level of precision shown by NDVI for predicting LAI/light interception makes it an ideal criterion for identifying insect-infested areas of a soybean field during the seed filling period. Although linear relationships have been reported in previous research between NDVI and LAI at LAI levels below 3.0 (Asrar et al., 1984; Ahlrichs and Bauer, 1983), data points typically cluster at LAI levels above this resulting in either quadratic or exponential relationships (Holshouser and Jones, 2002). In contrast, the linear relationships between LAI/light interception and NDVI in our study were maintained up to an LAI of 4.5 and light interception of 95% (Fig. 1 and 2). Based on previous work, it is expected that if LAI in the current study had risen above 4.5 to the 5.0 to 6.0 range (levels supraoptimal for 95% light interception) a similar clustering of points would have occurred. Although NDVI would not be expected to accurately predict LAI and light interception in this range, its ability to identify these parameters up to canopy closure demonstrates its usefulness for identifying areas experiencing defoliating pests.
Multivariate Analyses for LAI and Light Interception
A major consideration in development of vegetation indices for use as insect-infested areas is the consistency of vegetation index/canopy parameter regression models across cultivars. Differences in canopy architecture exhibited by determinate and indeterminate cultivars are an area of particular concern. Results of the multivariate analysis demonstrated that cultivars significantly affected regression relationships between LAI vs. NDVI and light interception vs. NDVI; although the interaction was greater in the former compared with the latter (Table 1). Use of a given LAI level to identify potential insect infestations as predicted by NDVI would require regression models tailored to specific Maturity Groups and growth habits. In contrast, regression models between light interception and NDVI would be more broadly applicable across cultivars. In either case, the inconsistency of LAI and light interception relationships with NDVI across cultivars presents a major barrier to their adoption.
NDVI as an Identifier of Insect Infestations
Use of NDVI for identifying insect infestations would also be compromised by year effects. Regression relationships of NDVI with LAI and light interception were not homogenous across years (Fig. 1 and 2). For example, NDVI indicating 95% light interception differed from 0.33 in 2004 to 0.68 in 2005. A number of factors could have accounted for this: percentage canopy cover, soil color, crop developmental stage, crop condition, and atmospheric characteristics (Ma et al., 2001). Use of NDVI for identifying insect infestations would be more consistent if regression models were site/cultivar specific and not extrapolated to other locations, years, and cultivars. Greater consistency could also be achieved with newer technology in which NDVI is based on light emitted from the instrument and reflected from the canopy back to the sensor.
Regression models relating NDVI to light interception, rather than to LAI, would be more effective for identifying insect infestations. Based on the multivariate analysis, the effect of NDVI on light interception was less affected by cultivar differences than was the NDVI effect on LAI. Also, reduced light interception directly affects yield, while LAI indirectly influences yield through its effect on light interception (Ingram et al., 1981). Although the two parameters are closely linked, LAI effects on light interception can be confounded by factors such as row spacing (Board, unpublished data, 2006). A farmer or agricultural consultant could calculate the NDVI used to identify insect infestations by first determining the NDVI associated with 95% light interception (canopy closure) for a particular field. Canopy closure usually occurs by R5, but can occur earlier at narrow row spacings or with larger plants (Board, unpublished data, 2006). Recordings should be taken as soon as canopy closure is recognized. Since 95% light interception is maintained throughout early and midseed filling (Carpenter and Board, 1997), any decrease in NDVI below the level associated with optimal light interception would indicate the possible presence of insect damage. The rate of NDVI reduction from the initial optimal level would indicate the severity of suspected insect attacks. Once a problem area is identified, the cause of the reduced light interception would require on-site investigation. Insect sampling would still be required in these areas to ensure that insecticide application was warranted. Insects causing the original damage may not be present. Also, other agents (e.g., diseases and animal defoliation) may have been responsible for the damage. In crop situations where canopy closure was not achieved, maximal light interception would occur near R5 (Egli and Leggett, 1973). The NDVI associated with this level would then be used as a benchmark to assess canopy damage at subsequent periods.
|
CONCLUSIONS
|
|---|
Because of its strong linear relationships with LAI and light interception across a broad range of canopy cover, NDVI demonstrated potential use for identification of areas experiencing insect-induced defoliation. Thus, NDVI could be used to increase the accuracy and efficiency for detecting defoliating pests of soybean and determination of pesticide application. Regression models relating light interception to NDVI appeared more useful for this purpose than those between LAI and NDVI; mainly because of greater applicability across cultivars and more direct effect on limiting yield. Based on remote sensing methods used in this study, regression models between NDVI and light interception would need to be time and site specific. However, recent technological advances may result in more robust models having wider applicability across location and time.
|
ACKNOWLEDGMENTS
|
|---|
We wish to express appreciation to the Louisiana State University Board of Regents and the Louisiana Soybean Promotion Board for providing funding and other support for this research.
|
REFERENCES
|
|---|
- Ahlrichs, J.A., and M.E. Bauer. 1983. Relation of agronomic and multispectral reflectance characteristics of spring wheat canopies. Agron. J. 75:987–993.[Abstract/Free Full Text]
- Aparicio, N., D. Villegas, J. Casadesus, J.L. Araus, and C. Royo. 2000. Spectral vegetation indices as nondestructive tools for determining durum wheat yield. Agron. J. 92:83–91.[Abstract/Free Full Text]
- Asrar, G., M. Fuchs, E.T. Kanemasu, and J.L. Hatfield. 1984. Estimating absorbed photosynthetic radiation and leaf area index from spectral reflectance in wheat. Agron. J. 76:300–306.[Abstract/Free Full Text]
- Baldwin, J.L., D.J. Boethel, and R. Leonard. 1997. Control soybean insects. Louisiana Coop. Ext. Serv. Publ. 2211. Revised ed. Louisiana State Univ., Baton Rouge.
- Bellairs, S.M., N.C. Turner, P.T. Hicks, and R.C.G. Smith. 1996. Plant and soil influences on estimating biomass of wheat in plant breeding plots using field spectral radiometers. Aust. J. Agric. Res. 47:1017–1034.[CrossRef][Web of Science]
- Board, J.E. 2004. Soybean cultivar differences on light interception and leaf area index during seed filling. Agron. J. 96:305–310.[Abstract/Free Full Text]
- Board, J.E., and B.G. Harville. 1993. Soybean yield component response to a light interception gradient during the reproductive period. Crop Sci. 33:772–773.[Abstract/Free Full Text]
- Board, J.E., and Q. Tan. 1995. Assimilatory capacity effects on soybean yield components and pod number. Crop Sci. 35:846–851.[Abstract/Free Full Text]
- Board, J.E., A.T. Wier, and D.J. Boethel. 1994. Soybean yield reductions caused by defoliation during mid to late seed filling. Agron. J. 86:1074–1079.[Abstract/Free Full Text]
- Board, J.E., A.T. Wier, and D.J. Boethel. 1997. Critical light interception during seed filling for insecticide application and optimum grain yield. Agron. J. 89:369–374.[Abstract/Free Full Text]
- Browde, J.A., L.P. Pedigo, M.D.K. Owen, and G.L. Tylka. 1994. Soybean yield and pest management as influenced by nematodes, herbicides, and defoliating insects. Agron. J. 86:601–608.[Abstract/Free Full Text]
- Carlson, T.N., and D.A. Ripley. 1997. On the relation between NDVI, fractional vegetation cover, and leaf area index. Remote Sens. Environ. 62:241–252.
- Carpenter, A.C., and J.E. Board. 1997. Growth dynamic factors controlling soybean yield stability across plant populations. Crop Sci. 37:1520–1526.[Abstract/Free Full Text]
- Daughtry, C.S.T., K.P. Gallo, S.N. Goward, S.D. Prince, and W.P. Kustas. 1992. Spectral estimates of absorbed radiation and phytomass production in corn and soybean canopies. Remote Sens. Environ. 39:141–152.[CrossRef]
- Egli, D.B., and J.E. Leggett. 1973. Dry matter accumulation patterns in determinate and indeterminate soybeans. Crop Sci. 13:220–222.[Abstract/Free Full Text]
- Fehr, W.R., and C.E. Caviness. 1977. Stages of soybean development. Agric. Exp. Stn. Spec. Rep. 80. Iowa State Univ., Ames.
- Funderburk, J., R. McPherson, and D. Burtin. 1998. Soybean insect management. p. 273–290. In L.G. Heatherly and H.F. Hodges (ed.) Soybean production in the Midsouth. CRC Press, Boca Raton, FL.
- Haile, F.J., L.G. Higley, and J.E. Specht. 1998a. Soybean cultivars and insect defoliation: Yield loss and economic injury levels. Agron. J. 90:344–352.[Abstract/Free Full Text]
- Haile, F.J., L.G. Higley, J.E. Specht, and S.M. Spooner. 1998b. Soybean leaf morphology and defoliation tolerance. Agron. J. 90:353–362.[Abstract/Free Full Text]
- Higley, L.G. 1992. New understandings of soybean defoliation and their implications for pest management. p. 56–65. In L.G. Copping (ed.) Pest management in soybean. Elsevier Sci. Publ., London.
- Higley, L.G. 2001. Yield loss and pest management. p. 13–22. In R.K.D. Peterson and L.G. Higley (ed.) Biotic stress and yield loss. CRC Press, Boca Raton, FL.
- Holben, B.N., and C.J. Tucker. 1980. Spectral assessment of soybean leaf area and leaf biomass. Photogramm. Eng. Remote Sens. 46:651–656.
- Holshouser, D.L., and B.P. Jones. 2002. Prediction of double-crop soybean leaf area index and yield variability utilizing the normalized vegetation index. In Annual meetings abstracts [CD]. ASA, CSSA, and SSSA, Madison, WI.
- Ingram, K.T., D.C. Herzog, K.J. Boote, J.W. Jones, and C.S. Barfield. 1981. Effects of defoliating pests on soybean canopy CO2 exchange and reproductive growth. Crop Sci. 21:961–968.[Abstract/Free Full Text]
- Ma, B.L., L.M. Dwyer, C. Costa, E.R. Cober, and M.J. Morrison. 2001. Early prediction of soybean yield from canopy reflectance measurements. Agron. J. 93:1227–1234.[Abstract/Free Full Text]
- Malone, S., D.A. Herbert, Jr., and D.L. Holshouser. 2002. Relationship between leaf area index and yield in double-crop and full-season soybean systems. J. Econ. Entomol. 95:945–951.[Web of Science][Medline]
- SAS Institute. 1999. SAS/STATE user's guide. Version 8. SAS Inst., Cary, NC.
- Shanahan, J.F., J.S. Schepers, D.D. Francis, G.E. Varvel, W.W. Wilhelm, J.M. Tringe, M.R. Schlemmer, and D.J. Major. 2001. Use of remote sensing imagery to estimate grain yield. Agron. J. 93:583–589.[Abstract/Free Full Text]
- Tynes, J.S., and D.J. Boethel. 1993. Control soybean insects. Louisiana Coop. Ext. Serv. Publ. 2211. Louisiana State Univ., Baton Rouge.
- Wiegand, C.L., A.J. Richardson, D.E. Escobar, and A.H. Gebermann. 1991. Vegetation indices in crop assessments. Remote Sens. Environ. 35:105–119.[CrossRef]
TOP
ABSTRACT
INTRODUCTION
