OVERVIEW OF SITE-SPECIFIC CONTROL TACTICS FOR TPB IN COTTON USING REMOTE SENSING
Since 1997, it has been found that remote sensing (RS) imagery improves the assessment of tarnished plant bug (TPB) populations by field scouts. Since 1999, spatially variable prescriptions have been built from imagery to instruct differential global positioning system (DGPS) equipped, variable-rate (VR) ground sprayers where to apply pesticides for TPB on a spatial scale. We have found (beginning in 2000) that when the concepts are modified they can also apply to the control of Heliothines on cotton.
The processing of information is similar for most of the season; however, for Heliothine spray decisions made in July, August, and September of the Mid-South, the prescriptions have to be built using temporally separated images. If the application is needed during June, then only a single image needs to be used. The steps are:
1. The process first begins with the acquisition of an image, its ortho- and geo-rectification (Fig. 1), and if necessary the building of a mosaic from two or more frames.
Fig. 1. Geo-referenced, false-color composite for 06 July 2001 of Arkansas Field Group, Perthshire, MS. Backwaters of the Mississippi River are just visible to the North.
2. The next step is to enhance the image in order to sharpen the contrast in the growth and development of the cotton plants in the field (Fig. 2). Numerous methods exist to create the enhanced map. Experience to date is suggesting that different methods are needed at different times of the season. One useful method is a RS technique called density slicing. This method is applied to the data obtained after transforming the 'by band' information of the image into a vegetative index. There are many forms of these indices.
Fig. 2. The color-coded normalized difference vegetation index (NDVI) of target fields (ca. 105 ha) for 06 July 2001. Red, yellow, orange, green, and blue colors represent cotton in increasing states of vigor, respectively.
3. The next step is to use the proper sampling method to minimize the sample variance of pest density by using the map built in Step 2 to define sampling strata. An assumption (which is a robust one) is that pest densities differ in intensity by habitat strata and the dispersion pattern within the strata is random. The sample variance can also be minimized by avoiding the use of sample unit sizes that are too small. Cumulative distribution functions of the Negative Binomial Distribution (NBD) (forced to converge to the Poisson) to illustrate the influence of sample unit size upon the probability of observing different numbers of insects per sample unit of various sizes. Lloyd's mean crowding index and Lloyd's index of patchiness (Lloyd, 1967) for several simulated infestation rates and identical sample unit size is near unity when the dispersion pattern is random.
4. The next step is to question if whether the map used to scout the field for insect pests agrees with scouting data from the past (other years) and recent (last week) experiences of the field consultant. Sometimes field conditions change causing effects that make the map that was built incorrect for describing the conditions of the crop (Contrast Figs. 2 and 3 as shown in Fig. 4).
Fig. 3. The color-coded normalized difference vegetation index (NDVI) of target fields for 19 August 2001. Red, yellow, orange, green, and blue colors represent cotton in increasing states of vigor, respectively, at this time of the season.
5. When the map is corroborated it is possible to next create spray 'on' or 'off' zones in correspondence to those areas where the insect pests are deemed too high and it is necessary to implement control measures (Fig. 4). These different zones are spatially located because the imagery is georeferenced. Note, at times, a blanket application may be the best choice.
Fig. 4. Color-coded scouting or spray decision map based upon a change analysis between the NDVI maps of 06 July and 19 August 2001. If used for a spraying operation, this map would be the basis of the spatial prescription. For a mid- to late-August spray decision, the colors red, yellow, orange and green are the 'no-spray' zones, whilst the blue color represents the 'spray-on' zone for Heliothine control. Use of temporal images earlier than that of 19 August (Fig. 3) would have established other maps where the spray zones assignments differ from that shown here.
6. After the application, it is necessary to evaluate the results. Sometimes, even though the application was applied at the correct place and rate, the pest population still is not controlled. The 'as applied' map built by the VR controller can also help generalize areas where the post-spray scouting information indicates that the pest still remains at a density of concern.
7. However, the RS information is most valuable in understanding these 're-treat' questions if more than one image over several weeks is available. Using imagery from two separate times at least several weeks apart and by completing a processing step known as change analysis (Richards and Jia, 1999) it is possible to determine relevant edges in cotton fields that are not discernible within an image acquired at one time (Fig. 4).
8. Once the refined map is built, apply, or re-apply the pesticide as required by the recommendation of the field scout. Thus, through an iterative decision making process working with the support of RS and GIS, it is possible to match the experience and judgement of the field consultant with VRT to conduct site-specific pesticide applications.
EXAMPLE OF REMOTE SENSING TECHNOLOGY
The continual improvement of this type of technology will increase the useability and usefulness of it in the future.
For more information contact:
Jeff L. Willers
PO Box 5367
Mississippi State, MS 39762-5367
Lloyd, M. 1967. Mean crowding. J. Anim. Ecol. 36: 1-30.
Richards, J. A., and X. Jia. 1999. Remote sensing digital image analysis. An introduction (3rd ed.). Springer-Verlag, Berlin.
Willers, J. L., D. L. Boykin, J. N. Jenkins, P. D. Gerard, S. A. Samson, P. L. Mckibben, K. B. Hood, W. L. Ladner, and M. M. Bethel. In revision. Site-specific approaches to cotton insect control. Part 1. Statistical concepts, sampling and analysis techniques. Prec. Agric.
Willers, J. L., J. M. Mckinion, P. D. Gerard, and J. N. Jenkins. In review. Site-specific approaches to cotton insect control. Part II. The relevance of non-aggregated dispersion patterns. J. Econ. Entomol.
J. Willers and J. N. Jenkins- Genetics & Precision Agriculture Research Unit, Missi. State, MS
Andy Zusmanis- Leica Geosystems (ERDAS, Inc.), Atlanta, GA
S. A. Samson- Extension GIS, Geospatial Resources Institute, Mississippi State University
K. B. Hood, J. H. Freeman, J. R. Bassie Sr., M. D. Cauthen- Perthshire Farms, Gunnison, MS
P. L. McKibben- Mckibben Ag-Services, Mathiston, MS