PDF version
for printing

© 2005 Plant Management Network.
Accepted for publication 7 December 2004. Published 4 February 2005.

Using Multispectral Aerial Imagery to Evaluate Crop Productivity

Clinton W. Jayroe, Research Specialist, William H. Baker, Associate Professor, and Amy B. Greenwalt, Graduate Assistant, Arkansas State University, Jonesboro 72401

Corresponding author: William H. Baker. wbaker@astate.edu

Jayroe, C. W., Baker, W. H., and Greenwalt, A. B. 2005. Using multispectral aerial imagery to evaluate crop productivity. Online. Crop Management doi:10.1094/CM-2005-0205-01-RS.

Abstract

Remote sensing offers the potential to provide quantitative and timely information on agricultural crops and could be utilized for creating variable rate prescriptions for fertilizer and chemical applications. This would be particularly useful for mid-season management of crop inputs by allowing producers to better allocate needed inputs, thereby increasing efficiency and hopefully overall profitability. The objectives of this study were: (i) to verify a correlation between data gathered by a yield monitor and multispectral imagery, (ii) to evaluate the relationship between soil electrical conductivity and vegetative growth patterns, and (iii) to establish a method of producing a vegetation map from images that could be used for directed scouting and mid-season chemical application decisions. Multispectral aerial imagery was used to observe crop canopy spectral reflectance in a study for rice (Oryza sativa), cotton (Gossypium hirsutum L.), and soybean (Glycine max) production fields in the Delta region of Arkansas. The images were acquired at various stages of crop growth and compared with ground reference observations consisting of soil electrical conductivity (ECa) and yield monitor data. Patterns were distinguishable with the use of the multispectral imagery along with some enhancements in a geographical information system (GIS). A classified, normalized difference vegetation index (NDVI) of the images was performed to further enhance the vegetative differences. The multispectral imagery proved to be a useful tool in assessing field variation through plant canopy reflectance. In the cotton and soybean production fields studied, yield variances due to soil characteristics were visible. Strong correlations, ranging from r² = 0.26 to 0.83, were seen in classified images, yield maps, and ECa maps. However in the rice production field, final yield had little correlation with the images acquired throughout the season (r² = 15), but the soil ECa map was related to the mid-season classified NDVI (r² = 0.48).

Introduction

Producers are interested in monitoring growth and development of crops and obtaining early estimates of final yield (8). Conventional methods of scouting are often labor intensive and variable conditions are only distinguishable to the very trained and experienced eye. With the use of infrared aerial photographs, plant physiological and morphological differences can be distinguished within fields as needed to identify smaller areas of possible plant disease (4). Crop Producers are aware of the productivity differences in their fields, and recognize the potential value of using variable rate technology instead of using uniform applications to manage crop production inputs (9). Site-specific, yield potential-based management of nitrogen and seed may improve profitability through both yield increases and material savings (12).

Remote sensing offers the potential to provide quantitative and timely information on agricultural crops over large areas (2). Analysis of remotely sensed and ground-truthed data could greatly improve management decisions in regards to crop inputs and irrigation efficiency. Multispectral and hyperspectral imaging sensors are able to view more than one particular band of energy. These bands are selected in various regions of the electromagnetic spectrum, based on the optimum range of energy being reflected by the objects observed. In-season canopy images have also been found useful in predicting yields in corn, soybean, and cotton plant canopy (6,8,11).

Soil variations can be a direct source of yield differences due to the diverse ratios of sand, silt, and clay which each conduct electricity at their own individual levels (5). These variances affect the water-holding capacity, `nutrient leaching, and plant root stability in soils. The electrical conductivity (ECa) of a soil is generally a good indication of soil texture and has been found to correlate well with biomass classifications in cotton (7).

The objectives of this study were: (i) to verify a correlation between data gathered by a yield monitor and multispectral imagery, (ii) to evaluate the relationship between soil electrical conductivity and vegetative growth patterns, and (iii) to establish a method of producing a vegetation map from images that could be used for directed scouting and mid-season chemical application decisions.

These images could potentially provide a map for variable-rate fertilizer and chemical applications in order to adjust inputs accordingly for site-specific management. By varying the rate of inputs, resources can be more efficiently managed and overall farming operations could be more profitable (10). This would also be a potential indirect benefit to the environment due to the reduction of fertilizer and chemical inputs.

Aerial Imagery of Rice, Cotton, and Soybean Fields

Aerial images were acquired periodically of the cotton, rice, and soybean canopies throughout the growing season of each crop with a multi-spectral camera mounted in a small aircraft. Aircraft altitudes and film format resulted in a resolution of approximately 2 m. Three filters were used to allow the camera to sense in wavelengths of green (~550 nm), red (~600 nm), and near-infrared (~800 nm).

Color infrared digital orthoquarterquads (DOQQ) were obtained from the website of the University of Arkansas’s Center for Advanced Spatial Technologies (CAST). These images had been obtained during the winter to record tonal patterns in fields that are attributed to the changes in bare soil reflectance. The DOQQs were used in the geographical correction of the multispectral aerial images.

The images were visually and statistically compared to various data layers such as yield and ECa maps, which have been noted as being good indicators of soil variability and texture changes (7). They were also enhanced through a classified, normalized difference vegetative index (NDVI) so that the best view of variability in crop canopy could be viewed. The NDVI is a standard image calculation expressed as the difference of two bands divided by their sums. In this study, the red band (B2) and the near-infrared band (B3) were used to calculate NDVI = [(B3-B2) / (B3+B2)]. This provided an enhanced view of the fields with respect to the extent of overall variability. The ECa data was gathered with a Veris 2000XA (Veris Technologies, Inc., Salina, KS). The device consists of four coulters that were lowered 2 inches into the ground and pulled through the field on a parallel path in 60-ft intervals. A DGPS receiver, differential global position systems, was used to gathered information simultaneously. The yield and Veris ECa data was interpolated using inverse distance weighted algorithm. All calculations and interpolations were performed in the commercial GIS software package ARC GIS version 8.3 (ESRI, Redlands, California).

The statistical analysis was performed by converting each of the interpolated surfaces into 10-m grid files. The large grids were used to eliminate any influence of local scale variability. Regression analysis was used to evaluate the relationship between each of the data layers.

Cotton

Cotton (Gossypium hirsutum L.) spectral reflectance had similar patterns as the yield maps and ECa data (Figs. 1 to 5). The (NDVI) contrasted the differences in vegetation as observed through weekly field visits. The maps created from the biomass reflectance changed as the weeks progressed. However, some patterns remained distinguishable throughout the season.

Fig. 1. Classified NDVI image on June 21, 2003 cotton vegetation canopy.		Fig. 2. Classified NDVI image on July 28, 2003 cotton vegetation canopy.

Fig. 3. Classified NDVI image on August 18, 2003 of cotton vegetation canopy.

Fig. 4. Cotton yield map of a field grown near Jonesboro, AR, during the 2003 growing season.		Fig. 5. Soil ECa interpolation from a cotton field grown near Jonesboro, AR, during the 2003 growing season.

Crop biomass patterns are distinguishable in the image acquired on June 21 (Fig. 1). These patterns continue to develop throughout the season and are distinct in the images on July 28 (Fig. 2) and on August 18 (Fig. 3). The three images possess a strong correlation with each other as the strongest correlation was r² = 0.80 between the June 28 and August 18 images and the weakest correlation found was r² = 0.40, June 21 and June 28 (Table 1). The June 21 and August 18 images were particularly interesting since the earliest crop patterns strongly correlated with the late season patterns (r² = 0.60) (Table 1), which agrees with the findings of Vellidis et al. (11).

Table 1. Correlation of cotton data layers measured in a field near Jonesboro, AR during the 2003 growing season

Spatial patterns in vegetative growth are similar to the spatial variability in yield data, (Fig. 4) the correlations between the imagery and yield data range between r² = 0.44 and r² = 0.66 (Table 1). Further, the crop patterns also correlate well with the ECa data (Fig. 5 and Table 1). Early crop biomass appears to be strongly influenced by soil variations as a correlation of r² = 0.72 exists between the June 21st image and the Veris electrical conductivity (Table 1). Following the growth patterns of cotton throughout the season appears to be an adequate method of predicting yield levels as indicated by the strong correlation between the images and yield. Utilizing classified images of crop growth patterns throughout the season would be an adequate method of scouting in cotton. This information could potentially be used as a source of directing site-specific application during the season.

Rice

Spatial variation in the NDVI classification from June 21 (Fig. 6) did not match that of the spatial patterns in the final rice (Oryza sativa) yield (Fig 7), as the correlation between them was r² = 0.11 (Table 2). However, the contours seen in (Fig 6) are patterns seen in both the image and in the interpolated ECa surface (Fig 8). These contours depict areas of the field where land leveling had taken place. Land leveling is a common practice in the mid-South to facilitate irrigation in rice production and is often related to declines in soil fertility as well as reduced crop productivity (1). Even though a correlation did exist between the mid season image and the ECa interpolation (r² = 0.48), neither were related to the final yield patterns as seen in (Table 2).

	Fig. 6. Classified NDVI image on June 21, 2002 of rice canopy.		Fig. 7. Rice yield map of a field grown near Walnut Ridge, AR, during the 2002 growing season.

Fig. 8. Soil ECa interpolation from a rice field grown near Walnut Ridge, AR, during the 2002 growing season.

Table 2. Correlation of rice data layers measured in a field near Walnut Ridge, AR during the 2002 growing season.

Soybean

The yield data patterns, imagery, and ECa measurements were very comparable in the soybean (Glycine max) field studied. The classified image (Fig. 9) displays a very distinct horse-shoe pattern that can be seen. This same pattern was present in the yield data (Fig. 10). The lower yielding areas (red, orange, and yellow) were locations where land leveling took place. The ECa data also displays this same pattern (Fig. 11). The strongest correlation exists between the image and the yield data (r² = 0.35), however a relationship between the ECa interpolation and the other data layers did exist as shown in (Table 3).

Fig. 9. Classified NDVI image on July 28, 2002 of soybean canopy.		Fig. 10. Soybean yield map of a field grown near Jonesboro, AR, during the 2002 growing season.

Fig. 11. Soil ECa interpolation from a soybean field grown near Jonesboro, AR, during the 2002 growing season.

Table 3. Correlation of soybean data layers measured in a field near Jonesboro, AR during the 2002 growing season.

Conclusions

With the use of multiple images strategically scheduled throughout the growing season, better management decisions can be formulated to control yield-limiting factors. This is a reliable method of scouting crops due to its ability to extend the view of production field beyond the human eye’s limitations. Producers in the future will likely implement these processes in to their practices, especially as environmental concerns arise and cost savings are realized from site-specific management and application.

Cotton is a highly intensive crop to manage, with several chemicals needed throughout the season. A real economic advantage is possible by utilizing remotely sensed data for a guide in evaluating plant health and for producing variable rate prescription maps for site-specific application of chemicals and fertilizer. Soybeans similarly could benefit in the same aspects as cotton. However, soybeans are not typically as costly to produce and may not be economically justifiable. The use of multispectral imagery in rice production could be used as a guide for early-season management. Providing that aerial variable rate technology is available, chemicals and fertilizers could be applied site-specifically. High cost inputs, such as fungicide treatments, could be varied to reduce chemical costs if identified by images to be isolated to specific management zones. In general, remotely sensed data has the capability to offer very timely information that can be used as a tool in many management decisions. These methods offer the chance to evaluate lower production areas, irrigation efficiency, diseased areas, and insect infestation damage during development phases (3).

Literature Cited

1. Brye, K. R., Slaton, N. A., and Chen, P. 2004. Can crop productivity be predicted by soil characterization after land leveling? Online. Crop Management doi:10.1094/CM-2004-0908-01-RS.

2. Clevers, J. G. 1997. A simplified approach for yield prediction of sugar beet based on optical remote sensing data. Remote Sens. Environ. 61:221-228.

3. Jayroe, C. W., Baker, W. H., and Robertson, W. H. 2003. An analysis of aerial imagery and yield data collection as management tools in rice production. Arkansas Soil Fertilities Studies, Univ. Ark. Agric. Exp. Sta. Res. Ser. 502:270-279.

4. Johannsen, C. J., and Sanders, J. L. 1982. Remote Sensing for Resource Management. Soil Conserv. Soc. Am., Ankeny, IA.

5. Lund, E. D., Christy, C. D., and Drummond, P. E. 1999. Practical applications of soil electrical conductivity mapping. Proceedings of the 2nd European Conference on Precision Agriculture. July 1999. Denmark.

6. Ma, B., Dwyer, L., Costa, C., Cober, E., and Morrison, E. 2001. Early prediction of soybean yield from canopy reflectance measurements. Agron J. 93:1227-1234

7. Robertson, W., Jayroe, C., Baker, W., Plunkett, D., and Kirkpatrick, T. 2002. Remote sensing technologies used as management tools in cotton production. Arkansas Soil Fertilities Studies, Univ. Ark. Agric. Exp. Sta. Res. Ser. 502:70-74.

8. Senay, G. B., Ward, A. D., Lyon, J. G., Fausey, N. R., and Nokes, S. E. 1998. Manipulation of high spatial resolution aircraft remote sensing data for use in site-specific farming. T. ASAE 41:489-495

9. Shanahan, J. F., Doerge, T. A., Johnson, J. J., and Vigil, M. F. 2004. Feasibility of site-specific management of corn hybrids and plant densities in the great plains. Precis. Agric. 5:207-225.

10. Sisk, K. S., Baker, W. H., Jayroe, C. W., Stiles, H. S., Kirkpatrick, T. L., Plunkett, D. E., and Popp, J. S. 2002. Using multispectral imagery to identify stressed cotton as a means to improve cotton profitability. Arkansas Soil Fertilities Studies, Univ. Ark. Agric. Exp. Sta. Res. Ser. 507:316-323.

11. Vellidis, G., Thomas, D., Wells, T., and Kvien, C. 1999. Cotton yield maps created from aerial yield maps. 1999 ASAE Annual International Meeting. Toronto, Ontario, Canada. ASAE, St. Joseph, MI.

12. Wilkerson, J., and Moody, H. 2004. Variable rate plant population and nitrogen. 2004 Southern Plant Nutri. Manag. Conf. Proc., Ardmore, OK.