K.L. Martin, W.R. Raun, K.W. Freeman, R.K. Teal, K. Desta1, D.B. Arnall, B. Tubaña, J.B. Solie, and M.L. Stone
Abstract
Improving crop management inputs with remote sensing devices is an emerging technology. This study documented the progression of NDVI (normalized difference vegetative index) over the life cycle of corn, evaluated the spatial variability of corn growth in terms of CV (coefficient of variation, calculated from NDVI readings) and documented relationships between NDVI, CV (calculated from NDVI), and grain, biomass yields and plant density. Four rows, 30 m in length from two locations over two years were randomly selected for this study. A GreenSeeker™ Handheld sensor was used to collect NDVI readings at multiple growth stages during the life cycle of corn. NDVI increased with progression of vegetative growth stages until V10, where a plateau was encountered, followed by a decline in NDVI after the VT growth stage. Coefficient of variation (CV) data from the NDVI readings revealed two dominant peaks during the life cycle of corn, one between the V6 and V8 growth stages and the second during the late reproductive growth stages. The CV data illustrated that the greatest variation expressed by corn during the vegetative growth stages was between the V6 and V8 growth stages. The highest correlation of NDVI with corn grain yield was found at the V7 to V9 growth stages; likewise CV and plant density were also more highly correlated from V7 to V9. The CV from NDVI readings was highly correlated with grain and biomass yields at all growth stages.

Introduction

Remote sensing technology is increasing in popularity in agricultural settings. Today, there is a wide variety of uses for remote sensors and nearly all applications continue to undergo updates and revisions to improve their effectiveness. Along with the many sensors, there are also many different indices produced by these various sensors. For this study, the focus is on better understanding NDVI in corn production to improve the application of yield based inputs (e.g. N fertilizers).

One key study for the background of NDVI (for use in production agriculture) was conducted in wheat by Sembiring et al. (1998). They used a PSD1000 Ocean Optics fiber optic spectrometer to evaluate the relationship of spectral radiance to wheat forage biomass, N, and phosphorus (P) uptake. This study used numerator/denominator indices and showed that numerator wavelengths between 705 and 735 nm and denominator wavelengths between 505 and 545 nm were good predictors of forage biomass, N, and P uptake at Feekes growth stages 4 to 6. Then, Lukina et al. (1999), estimated wheat vegetation coverage using binary pseudo-color images, which had a high correlation with NDVI measurements of the wheat canopy.

Using this background information, Raun et al. (2001) used the following equation to conduct further research in wheat and corn.

Where:

ρ_NIR – Fraction of emitted NIR radiation returned from the sensed area (reflectance)

ρ_Red – Fraction of emitted red radiation returned from the sensed area (reflectance)

Lukina et al. (2001) used NDVI to reliably predict plant N uptake and found that it was positively correlated with final grain yield in winter wheat.

Much of the past production agriculture research using NDVI has focused on wheat production, chiefly for yield prediction estimates. Since corn is the second largest crop produced in the world with 6.02 X 10⁸ MT produced globally and 2.29 X 10⁸ MT produced in the United States (2002 estimates, www.faostat.org), sensor based research is actively taking place in corn production. Having the ability to accurately predict yield in corn could dramatically impact corn production. If the farmer knew what the yield would be for each field or management zone within a field, every management decision that is regulated by yield could be made with much greater certainty.

To understand the impact yield estimation can have, we must understand what yield is related to and how to use that knowledge. Using our earlier example, N fertilization, Hergert et al. (1995), Schmitt et al. (1998), Franzen (2003) and others have shown that the corn yield goal multiplied by 1.2 pounds of N (subtracting residual N in the soil) results in an appropriate rate of N fertilization. Nitrogen is commonly considered the most limiting nutrient for many production systems and accounts for a significant portion of the production cost. Therefore, using the previously known information about yield relationships and yield potential estimates that have been demonstrated in wheat (Raun et al., 2001; Lukina et al., 1999; Lukina et al., 2001and Raun and Johnson 1999), equations can begin to be developed to calculate the input (constituents needed for crop growth and production) rate (assuming the input is yield related).

The appropriate growth stages for yield estimation are very important. It has been confirmed (e.g. Raun et al., 1999) that the appropriate growth stages for sensor technology utilization in wheat fall between Feekes physiological growth stages 4 and 6, but the specific growth stages for corn have not been established.

An additional benefit sensor technology adds to production agriculture is the ability to quantitatively identify variability within a field. When considering the spatial resolution of photographic images, Simonett (1983) stated that spatial resolution may be defined as the minimum distance between two objects that a sensor can record distinctly. Raun and Johnson (1999) recognized that the natural variability within a field causes single rate applications of nutrients (one rate over the entire field) to be excessive or inadequate depending on the variability and location in the field. Furthermore, Solie et al. (1999) showed that significant variability in nutrient levels in the soil exists at areas < 1m². Independent of resolution or amount of variability, many studies have shown that variability has a negative impact on yield (Krall et al., 1977; Nielsen, 2001; Nafziger et al., 1991; Raun et al., 1986 and Raun and Johnson, 1999) and many state that quantifying the variability could spatially improve crop inputs (as stated before, yield based inputs) (Raun et al., 2001; Washmon et al., 2002).

Coefficient of variation is a statistical parameter that can easily be used to calculate the variability in NDVI measurements taken over a given area. Freund and Wilson (2003) define CV as the ratio of the standard deviation to the mean, expressed in percentage terms, or simply the standard deviation as a proportion of the mean. Washmon et al. (2002) used CV (calculated from Landsat satellite images) to evaluate variability in wheat and reports that if within-field CVs could be predicted, the potential response to added nutrients could be established and in-season nutrient applications could be adjusted.

The objectives of this study were to 1) document the progression of NDVI over the life cycle of corn, 2) evaluate the spatial variability of corn growth in terms of CV over the life cycle of corn and 3) determine if relationships exist between NDVI, CV (calculated using NDVI), and grain, biomass yields and plant density.

MATERIALS AND METHODS

Two experimental sites were established in the spring of 2003, one at the Stillwater (EFAW) Research Station in Stillwater, Oklahoma on a Easpur loam (fine-loamy, mixed superactive thermic Fluventic Haplustoll) and one at the Lake Carl Blackwell Research Station West of Stillwater, Oklahoma on a Pulaski fine sandy loam soil (coarse-loamy, mixed, nonacid, thermic, Typic Ustifluvent). The production system utilized conventional tillage in a rainfed environment. Two Bacillus thuringiensis (bt) gene enhanced corn hybrids identified by their maturity (108 or 111 day) were used in this study. At Efaw hybrid ‘111’ was used in 2003 while hybrid ‘108’ was used in 2004. At Lake Carl Blackwell, hybrid ‘108’ was used in both years (planting dates, harvest dates, plant populations, maturities, and grain moisture are reported in Table 1).

Four rows, each 30 m in length, were randomly identified at each location (from an area >2 ha). Two rows at each location in 2004 were thinned to the population reported in Table 1 to evaluate differences due to plant density. Within each row, the cumulative distance of each plant from the beginning of the row was recorded at the V3 to V5 growth stage using a tape measure extended to the length of the row. Plant locations were documented to evaluate relationships between collected data and plant density (computed using plant location measurements).

A GreenSeeker™ Hand Held Optical Sensor[1] was used to record NDVI measurements at each growth stage. The NDVI measurements were taken starting from V3 growth stage at all locations and years except at Lake Carl Blackwell in 2003 (which began at the V6 growth stage) (growth stages at which data was accumulated are reported in Table 2). Each growth stage was identified using the classification terms developed at Iowa State University (1993). This sensor unit was manually carried with the data collection region of the sensor (sensor head) in the nadir position 0.8 m directly above the corn canopy. The sensor was carried at a relatively constant pace the length of each (30 m) row to collect approximately 600 readings per row.

The mean NDVI and CV of NDVI were calculated for each row at each respective growth stage. At physiological maturity, the above ground biomass was hand harvested and calculated for each plant, then averaged for the entire row (harvest dates are reported in Table 1). The ear(s) were removed from each plant and dried in a forced air oven at 75°C for 4 days at which time, the weights were recorded. The kernels were removed from the ears and reported as grain yield. The corn grain and plant biomass yields reported in Table 3 were calculated for each row and averaged over each location. Statistical analysis was preformed using the Regression Procedure in SAS (SAS, 2002)

RESULTS and discussion

NDVI Over Time

Although the progression of NDVI over the life cycle of corn has not been well documented in the US, Lukina et al. (2001) noted that NDVI can be used to estimate vegetation coverage. The NDVI readings that were collected throughout the life cycle of corn follow the expected trend of vegetation coverage over time. As corn plants emerged, biomass per unit land area was small and NDVI values were low (Figures 1-4). Average NDVI of all rows at all sites during vegetative growth was lowest at the V3 growth stage because soil reflectance dominated the NDVI measurement. However, as plants grew and developed, NDVI rapidly increased (between growth stages V3 and V10) as the canopy covered the soil with overlapping leaves (Figures 1-4). During these stages, it was apparent that NDVI was proportional to the level of vegetation coverage.

Changes in NDVI values were small between the V10 growth stage and the VT growth stage, reaching a maximum value at or just before tasseling. At the point of canopy closure, the sensor was almost exclusively measuring plant material. Therefore, data acquired after canopy closure involved almost no red reflectance, so changes in NDVI thereafter relied on NIR differences. When tassels were fully emerged, NDVI decreased due to the yellowing of mature tassels. As plants entered the reproductive stages and senescence occurred on the lower portions of the plant, NDVI decreased more rapidly. As senescence moved to the top of the plant, NDVI was depressed as low as 0.30.

Differences in NDVI across locations were encountered in both years of this study. These differences are due to natural differences in soil type, texture, water holding capacities, and differences in timely rainfall that ultimately influenced the growth of the corn. During the tassel period, abnormal deviations in NDVI (usually lower values) are likely due to the low quantity of chlorophyll present in the tassel.

CV Over Time

Figures 1-4 show the data for each specific site for both years. During the early growth stages (V3 to V4), the NDVI data were generated from the soil surface and a very small portion of plant material. Therefore, CVs (calculated from NDVI) for each row were low. Coefficient of variation values reached a maximum between the V6 and V8 growth stages (Figures 1, 3, and 4). At this peak, the plants did not accumulate enough vegetation to cover the within-row soil surface, and expressed the greatest amount of variability. After one to two additional growth stages, a rapid decline in variability followed at all sites (Figures 1-4).

The CVs reached a minimum value at or just before the VT growth stage, followed by another rapid increase in CV (Figures 1-4). These growth stages (V12 to VT) correspond to the time at which the NDVI values reached their highest point as illustrated in Figures 1-4. As a visual determination, the corn at the V10 to V12 growth stage appeared to be at the most uniform stage, as differences were very difficult to recognize. Likewise, the CV data generated over time reached its lowest value just prior to tasseling when complete canopy coverage and leaf overlap created the most heterogeneity encountered in the growing season. Immediately after tasseling, the small peak in CV values expressed in some years (EFAW 2003) was likely due to the full expression of the tassels and the uneven tassel emergence (Figure 1). This small peak carried over into the early reproductive growth stages (R1 to R2) at the EFAW location.

After the reproductive stages (R2 to R5), the expressed CV of the NDVI values increased and as a result, the highest CVs were found at these growth stages. As senescence occurred, the sensor measured NDVI from leaves containing a wide range of green and non-green color intensities. The highest CVs were found as the plants approached full maturity, when the senescence neared the top of the plant and the lighter color had a greater impact on NDVI values.

Some differences can be noted in the 2004 growing season due to differences in plant population. The variation in plant population was implemented to more fully understand how the population affects the CV over time. In Figures 3 and 4, there is a trend for the lower population to display elevated CVs. Although the CVs increased in the lower population, the trend for the CV over time remained the same, more specifically, the vegetative growth stage at which a peak was observed did not change. Other differences in CV observed at over locations (similar to NDVI) were likely due to soil differences or water availability changes.

Corn Grain and Plant Biomass Yields

At the early growth stages (V3 to V5), both grain and biomass yields were poorly correlated with NDVI (R²<0.12, P>0.26), but increased between the V6 and V7 growth stages (R²<0.29, P>0.03) and was the highest from the V8 growth stage to the V12 growth stage. Between the V8 and V12 growth stage, the correlation of NDVI with grain and biomass yields were at P<0.008 and R² = 0.56 to 0.66 for grain yields and P<0.006 at R² = 0.59 to 0.66 for biomass yields. The highest relationship for any one growth stage was found at V8 for both grain and biomass yields (R² = 0.66) (Table 2).

Since NDVI was highly correlated with grain and biomass yield at the V8 growth stage (at the locations and in the years in this study), the V8 growth stage is likely the most appropriate stage to evaluate corn for potential grain yield and biomass estimation. At all locations, the expressed variability was greatest at the V6 to V8 growth stage (high CV) compared to the later growth stages. Also, rapid uptake of nutrients (e.g. N, P, and potassium (K)) and rapid growth begin just after the V6 growth stage (V8 to V10), (Iowa State University, 1993). Therefore, the decision to apply some inputs will depend on the processes of the uptake and utilization of the input.

Plant Density

Arnall (2004) used an optical sensor to evaluate the relationship of CV of NDVI readings and the plant density in wheat. He found that CV increased as plant density decreased and established a critical CV using a linear plateau model. He concludes that adding an estimate of plant density to yield prediction models can improve the model. This study showed that the CV of NDVI measurements and NDVI alone was related to plant density (Table 2). The stage with the highest correlation of CV to plant density was between V7 to V9 (R²>0.85, P<0.0001, negative relationship) (Table 2). At the earlier growth stages (V3 to V6), the correlation was between 0.59 and 0.77 (inversely related). After the V9 growth stage, the R² decreased from 0.85 to 0.56 (negative relationship) at V10 and continued to decrease thereafter. At the V6 to V12 growth stages, NDVI was related to plant density (R² = 0.30-0.72, P<0.05), but no statistical correlation was found before the V6 growth stage or at the VT growth stage.

Based on these results, plant density can be estimated in corn via CV generated from NDVI readings. The relationship decreased dramatically as canopy closure occurred (V10 growth stage), thus suggesting that sensor technology application for assessment of plant density should occur before the V10 growth stage.

Determining the spatial variability within each fixed area via CV generated from remote sensors could be very important. In wheat, integrating the CV component into the fertilizer application system allowed Arnall (2004) to recognize plant stands that would not reach the yield potential determined by NDVI alone. In corn, CV (calculated from NDVI) could allow for the estimation of plant density, thus revealing the areas where plants are too sparse to reach the yield potential of other areas with a greater plant density. To illustrate this, Table 3 clearly shows that the low populations in 2004 did not achieve the yield as the “higher” populations. Here, the only difference between rows 1-2 and 3-4 was the population, creating a reduction in mean grain yield of 2536 and 1341 kg ha^-1 at EFAW and LCB respectively.

Plant density was related to grain yield (R² = 0.64, P<0.0002) and biomass yield (R² = 0.71, P<0.0001). Again, this illustrates the positive relationship between plant density and yield (both grain and biomass). It could also be inferred from the relationships described here that the CV of NDVI readings and the yield of both grain and biomass would be related, which was an additional result of this study. Grain and biomass yields were related with CV at R²>0.52 and P<0.05 (containing a negative slope) over a wide range of vegetative growth stages (V3 to VT) (Table 2), thus showing that the yield decreases as the variability in the corn increases.

CONCLUSIONS

This study documents the progression of NDVI and CV over time, which is similar to that found by Raun et al. (2005). The study conducted by Raun et al. (2005) took place in Texcoco, Mexico. The difference in climate, crop growth period (April to August vs. July to December) and that the study was conducted over only one growing season raised doubt as to whether a phenomenon was observed or if a common trend was observed. However, the data collected in Mexico does follow the same trend as the data collected in this study (in central US). This paper does address several other factors that have not been previously researched. These include correlations of NDVI with grain and biomass yields, CV (calculated from NDVI) relationships with grain and biomass yields, and correlations concerning the estimation of plant density over the vegetative life cycle of corn. It is common for remote sensing studies to develop relationships at growth stages that are convenient for their purpose (yield estimation, biomass estimation, plant density estimation). Few investigate these relationships over the entire life cycle of the crop.

The NDVI trend can be used to estimate vegetation coverage throughout the life cycle of corn. The CV data revealed that the growth stage at which the sensor used in this study can identify the greatest variations in plant characteristics is centered around the V6 to V8 growth stage, but is still relatively good at recognizing the variation from the V5 to V8 growth stage. If remote sensing devices are used for management decisions, it is likely that they should recognize differences between plants and plant spacing to make the best decisions.

Corn grain and plant biomass yields were found to have the highest correlation with NDVI from the V8 to V12 growth stages. Since remote sensing devices are commonly used to evaluate expressed plant characteristics and in some applications, estimate yield potential, the time at which sensors have the greatest correlation with yield is the time at which these sensors should be used. This time frame should correspond to the period at which the plant can utilize an input most effectively and that the need for this input could be recognized using sensors and treated accordingly.

Plant density was highly correlated to the CV from NDVI readings at the V7 to V9 growth stages and decreased at V10. Also, corn grain and biomass yields were correlated with both plant density and negatively correlated with CV. Therefore, CV of NDVI measurements should be able to improve yield potential estimation above that of NDVI alone. Coefficient of variation could be used to assess the influence of plant spacing on yield or could be used to identify a threshold CV in corn similar to that of Arnall (2004).

By combining the results found from NDVI generated over time, CV over time, yield, and plant spacing, the optimum growth stage at which remote sensors could be used can be deciphered for the varying uses of remote sensors. For yield potential estimation, many of these factors will be important depending on the development of a yield potential prediction equation.

BIBLIOGRAPHY

Arnall, D.B. 2004. Relationship between coefficient of variation measured by spectral reflectance and plant density at early growth stages in winter wheat. M.S. Thesis. Oklahoma State University, Stillwater, OK.

Franzen, D.W. 2003. North Dakota fertilizer recommendation tables and equations based on soil test levels and yield goals. (Available online at http://www.ext.nodak.edu/extpubs/plantsci/soilfert/sf882-8.htm.) North Dakota State Univ. Ext. Serv., Fargo, (accessed 03/2005).

Freund, R.J., and W.J. Wilson. 2003. Statistical methods. 2^nd ed. Academic Press, San Diego, CA.

Hergert, G.W., R.B. Ferguson, and C.A. Shapiro. 1995. Fertilizer suggestions for corn. NebGuide G74-174-A. Univ. of Nebraska, Lincoln.

Iowa State University. 1993. How a Corn Plant Develops, Special Report No. 48. Cooperative Extension Service, Ames, IA. (available online at http://maize.agron.iastate.edu/corngrows.html#v9mg), (accessed 03/2005).

Krall, J.M., H.A. Esechie, R.J. Raney, S. Clark, G. TenEyck, M. Lundquest, N.E. Humburg, L.S. Axthelm, A.D. Dayton, and R.L. Vanderlip. 1977. Influence of within-row variability in plant spacing on corn grain yield. Agron. J. 69:797-799.

Lukina, E.V., M.L. Stone, and W.R. Raun. 1999, Estimating vegetation coverage in wheat using digital images. J. Plant Nutr. 22(2):341-350.

Lukina, E.V., K.W. Freeman, K.J. Wynn, W.E. Thomason, R.W. Mullen, M.L. Stone, J.B. Solie, A.R. Klatt, G.V. Johnson, R.L. Elliott, and W.R. Raun. 2001. Nitrogen fertilization optimization algorithm based on in-season estimates of yield and plant nitrogen uptake. J. Plant Nutr. 24(6):855-898.

Nafziger, E.D., P.R. Carter, and E.E. Grahm. 1991. Response of corn to uneven emergence. Crop Sci. 31:811-815.

Nielsen, R.L. 2001. Stand establishment variability in corn. Dept. of Agronomy publication # AGRY-91-01. Purdue University, West Lafayette, IN.

Raun, W.R., D.H. Sander, and R.A. Olson. 1986. Emergence of corn as affected by source and rate of fertilizers applied with the seed. J. Fert. Issues 3:18-24.

Raun, W.R., and G.V. Johnson. 1999. Improving nitrogen use efficiency for cereal production. Agron. J. 91:357-363.

Raun, W.R., J.B. Solie, K.L. Martin, K.W. Freeman, M.L. Stone, G.V. Johnson, and R.W. Mullen. 2005. Growth stage, development, and spatial variability in corn evaluated using optical sensor readings. J. Plant Nutr. 28:173-182.

Raun, W.R., J.B. Solie, M.L. Stone, E.V. Lukina, W.E. Thomason, and J.S. Schepers. 2001. In-season prediction of potential grain yield in winter wheat using canopy reflectance. Agron. J. 93:131-138.

SAS Institute. 2002. The SAS system for windows version 8.02. SAS Institute Inc., Cary, NC.

Sembiring, H., W.R. Raun, G.V. Johnson, M.L. Stone, J.B. Solie, and S.B. Phillips. 1998. Detection of nitrogen and phosphorus nutrient status in winter wheat using spectral radiance. J. Plant Nutr. 21(6): 1207-1233.

Schmitt, M.A., G.W. Randall, and G.W. Rehm. 1998. A soil nitrogen test option for N recommendations with corn. Publ. FO-6514-GO. Univ. of Minnesota Ext., St. Paul.

Simonett, D. S., 1983, The development and principles of remote sensing. pp. 1–35. oIn Manual of Remote

Sensing, 2nd edn, Vol. I, Theory, Instruments and Techniques, edited by D. S. Simonett

Solie, J.B., W.R. Raun, and M.L. Stone. 1999. Submeter spatial variability of selected soil and bermudagrass production variables. Soil Sci. Soc. Am. J. 63: 1724-1733.

Washmon, C.N., J.B. Solie, W.R. Raun, D.D. Itenfisu. 2002. Within field variability in wheat grain yields over nine years in Oklahoma. J. Plant Nutr. 25:2655-2662.

[1] Mention of trademarked instruments does not imply endorsement by authors.