Simulated Application Errors for Granular Materials for Fixed and Variable-Rate Application

 J.P. Fulton, S.A. Shearer, M. E. Anderson, T.F. Burks and S.F. Higgins


Fulton, J.P., S.A. Shearer, M.E. Anderson, T.F. Burks and S,G. Higgins. 2000.  Simulated Application Errors for Granular Materials for Fixed and Variable-Rate Application.  ASAE Paper No. 001153.  Annual International Meeting, Midwest Express Center, Milwaukee, Wisconsin, July 9-12.


Abstract

This paper formulates a method to create "As Applied" surfaces for fixed and variable-rate application of granular fertilizers and lime using a spinner spreader.  A field test was conducted where variable-rate application of murate of potash was required.  Grid soil sampling was performed on a 0.4 hectare (1 acre) grid.  Based on the results of the fertility analysis, a prescription was developed for the field and imported into the sofware package AgView which is used to control the variable-rate control system on the spinner spreader and also record material application across the field.  ArcView was used to create an "As Applied" surface.  A script was written to take the  application transverse created by AgView and apply rectangular polygons at each point to represent the transverse spread pattern for that point.  Transverse spread patterns were determined and selected from previous research performed using the same spinner spreader.  Three patterns (168.1, 112.3 and 56.0 kg/ha) were selected to represent high, medium and low application rates based on the results of this previous research.   The script was able to take this information create a surface of polygons and then overlay a matrix of points to calculate the rate applied at each point.  The result was an "As Applied" layer of points that can be used to represent the application of  potash to the field and estimate variable-rate application errors.  The "As Applied" surface was then compared to the desired prescription surface.  A difference surface was then created and a correlation coefficient of .97 was calculated for comparison between these two surfaces.

Introduction

             Precision farming provides a new means for farmers to manage variability existing within and among agricultural fields.  The underlying idea of precision farming is to better manage variability, optimize crop yields, and lesson the environmental impact by only using the required inputs to produce the crop.  While some areas of a field might be suited for high yields, others might be incapable of generating any economic return and therefore should not be cropped.  The use of grid sampling has become a norm for those utilizing precision farming practices by providing a method to determine fertility levels across fields.  Grid soil sampling can be implemented in a variety of ways but it entails subdividing a larger field into smaller areas typically called management zones.  Within each zone soil samples are collected and analyzed to determine the necessary fertility requirements for each particular management zone.  The existence of variable-rate (VR) controllers on application equipment allows farmers to then vary the application rate of fertilizers and lime to apply the necessary product to these zones.  In return, product application should be improved along with minimizing over- and under- application of fertilizers, unlike the traditional application method of using a single fixed rate over the entire field. 

            While VR application equipment offers a method to improve application of materials to fields, limitations exist on this equipment that can create application errors where the actual application rate and pattern might differ from the desired rate and pattern, causing inaccuracies.  This might be most evident on spinner spreaders used to apply granular fertilizers and lime.  It is known that these types of spreaders do not provide uniform application of products due to the nature of spinner discs and nonuniform density of the product.  However, they provide an adequate and acceptable means of applying granular products.  Retrofitting these spreaders with VR controllers might actually increase application errors by altering the spread pattern during rate changes.  A method is needed to assess application errors of VR spinner spreaders and evaluate the effectiveness of VR application of granular fertilizers.  There exist software packages that record application rates along with the position at which the material was applied providing an application map.  Combining this information with spread patterns at different rates for a particular spreader can be used to generate a representative surface of applied granular fertilizer or lime.  In return, this "As Applied" surface can be used to estimate VR application errors when compared to the desired application surface.

 Objectives

1)      To generate an “As Applied” surface for granular fertilizer using logged data from the application of granular fertilizer using a variable-rate spinner spreader.

2)      To compare the “As Applied” surface to the desired application surface.

 

 Background

 Precision agriculture has permitted farmers to start managing their fields on a much smaller resolution rather than on a whole field basis. This approach to managing nutrients has three advantages over traditional approaches; agronomic, economic and environmental.  While variable rate technology is one of the most distinctive tools used by those implementing precision agriculture, the question always arises about how well are nutrients being really applied in the field?  With VR equipment more complexity is introduced due to rate changes when covering a field.  Therefore, methods need to be designed to test the accuracy of VR equipment and determine whether they are able to improve application accuracy over traditional fixed rate methods.

             Research has shown that VR application of fertilizers can save farmers money (Smith et al., 1990).  This study reported that one farmer saved over $18,000 on 400 hectares (1,000 acres) of corn using VR application over the traditional method used the previous year.  While many farmers are using VR technology, many do not recognize the possible application errors associated with this equipment.  One of the more popular VR applicators are spinner spreaders.  These spreaders are used widely used to apply granular fertilizers and lime.  However, there are inherent problems with this technology.

 Fulton et al. (1999) collected potash distributed by a spinner spreader by performing multiple pass tests as would occur under field conditions.  Their results show variability in the deposition of potash across their test area.  This variability in application compounded by those associated when moving to VR application can create undesirable application situations.  Problems with VR spinner spreader application are pattern shifts during rate changes (Fulton et al., 1999; Olieslagers et al., 1997) and delayed rate changes due to system latency.  System latency can be improved with the look ahead feature provided in most software packages.  The improvement in pattern needs to be investigated further but the divider on at the rear of spinner spreaders needs to be simultaneously adjusted during rate changes to maintain a desired Gaussian spread pattern.  To comprehend and quantify the effect of fixed and VR granular fertilizer application, a model or program needs to be developed to assess application errors.

             Fulton et al. (1999) modeled fixed and VR application of murate of potash.  They established a two-dimensional array of pans that could be used to collect the spread of potash over the test area.  Single and multiple tests were performed in order to model uniform and VR application.  Through this testing, they were able to model and describe both fixed and VR transverse spread patterns at a low rate (56.0 kg/ha), high rate (168.1 kg/ha), and rate change from low to high.   This information can then be used to estimate application errors when used in conjunction with an "As Applied" file for a particular field. 

 Methods

             A field was selected in Shelby County, Kentucky to apply murate of potash.  The field is subdivided into 0.4 hectare (1 acre) management zones with soil samples collected in each zone and then analyzed at the University of Kentucky's Regulatory Services.  The University of Kentucky's Lime and Fertilizer Recommendations (AGR-1, 1999) was used to determine the amount of required murate of potash to apply in each zone.  An application prescription was generated in SSToolbox (SST, 1999) and is presented in Figure 1.  The software package AgView by GIS Solutions (AgView, 1999) has the capability of recording "As Applied" accumulative fertilizer quantities and was used to pass application rates from the prescription to the MidTech controller.  The MidTech controller then echoes back to AgView the amount of material being applied.  AgView accumulates the amount of applied material and saves this information as a 'shapefile'.  Since the quantity applied is cumulative within the file, the difference must be taken between consecutive points to compute the amount applied per point.  Figure 2 provides the recorded field transverse recorded by AgView during field application.

             Previous work using the same VR spinner spreader truck in modeling VR application of potash permits the selection of transverse spread patterns to describe VR application (Fulton et al., 1999) which can then be used to generate an "As Applied" surface.  For simplicity and initial program development, three transverse spread patterns were chosen to represent high, medium and low application rates.  Figure 3 presents the selected transverse spread patterns to represent these three ranges.  The actual application rate ranges for these three categories are presented in Table 1.  The modeled uniform 168.1 kg/ha and 56.0 kg/ha application rates determined by Fulton et al. (1999) denotes the high and low ranges, respectively.  The medium rate was selected as 112.3 kg/ha and the associated transverse spread pattern was determined by choosing the longitudinal position along the modeled rate change (56.0 to 168.1 kg/ha) application surface created by Fulton et al. (1999).   Once determining where this rate occurred on the modeled surface, the cross section provided the representative pattern for 112.3 kg/ha. 

 The 'shapefile' generated by AgView containing the geographic points and amount of murate of potash applied can be used in combination with the selected transverse spread patterns to create an "As Applied" surface for this field.  To merge this data and create such a surface, ArcView by Esri (ArcView, 1999) will be used.  Avenue Script provides a programming language which can be used to write programs or Scripts to perform various functions and analysis within ArcView.  To accomplish the task of generating an "As Applied" surface, one Script was written to read the shape file created by AgView, determine the amount applied, and then assign application polygons at each point representing the appropriate transverse spread pattern applied for each point.  To properly perform these operations to create the "As Applied" surface, the 'shape file' was transformed using ArcViews Projection Utility Wizard from geographic coordinates in WGS 84 to State Plane, Kentucky North.  Either State Plane or Universal Transverse Mercator (UTM) can be used but are required in order to create the polygons representing the distribution of potash across a field.

 The first function performed by the Script is to compute the amount of potash applied per point by taking the difference in the accumulative amount between consecutive points.  This amount is stored in pounds.  A new column is generated within the database file to represent the pounds applied per point.  The next process was to create polygons at each point that represent the assigned transverse spread pattern.  A rectangular polygon was selected to represent the transverse spread pattern of this truck.  This polygon was then subdivided into 13 equal width polygons; each assigned an application rate based upon the amount of material applied per point.

 In reality, the longitudinal spread pattern of a spinner spreader is in a semicircle shape rather than rectangular in nature as used here.  Since the longitudinal spread pattern is unknown for this spinner spreader, the user is prompted for an offset to help improve the positioning of the application polygons.  The offset provides a means to move the polygon forward or backwards depending upon the user's estimation of where thrown material actually lands relative to where it was dispersed by the discs.  Additionally, the Global Positioning System (GPS) receiver on most spreaders are mounted on the cab or near front of the spreader allowing the offset to compensate for this offset distance from the spinner discs.  Therefore, the user is asked to enter an offset before the Script begins any operations.

 Spread width was set at 34.67 m (113.75 ft) based upon the single pass test used by Fulton et al. (1999).  Thirty two meters was used by Fulton et al. (1999) as the spread width but with the outside pans positioned 32 m apart an additional 2.67 m was added to the total width so equal width rectangular polygons (2.67 m wide) centered on the transverse pan location could be used.  The length of the polygon was calculated by using data points on either side of a particular point.  The midpoint is determined between the point of interest and either point on each side (Figure 4).   These distances set the length of the polygon for a particular point.  Based on these dimensions, a rectangular polygon can then be assigned to each point but oriented North-South on the point.  Initially orienting the polygon North-South provides the easiest manner to affix the polygon at the point.

 To properly orient the polygon, the heading of the spreader truck must be determined for at each point.  The heading is calculated by looking ahead to the next point and calculating the azimuth (see Figure 4).  Once calculated, the polygon is rotated so perpendicular to the heading as shown in Figure 5.  The polygon is then divided into 13 equal polygons to represent the transverse spread pattern at that point (Figure 6).  Figure 7 presents the resulting polygons for this particular field.  The next step is to then apply the amount of material received in each polygon using the three selected spread patterns.

 It should be noted that starting and ending points require a slightly altered method to apply a polygon since there are not three points (one on each side) to determine a midpoint and heading.   These points occur at the start and end of a field plus when turning around on the ends of a field.  In this case, a ghost point is created to complete the three point series based upon the other two points.  The ghost point is placed at the same distance from the point of interest and the other existing point.  This provides a method to calculate a heading and assign a polygon at these distinct points.

 The next step performed within the Script is assigning the amount of material placed in each polygon.  Each of the transverse spread patterns representing the three ranges was normalized by the total sum of each pattern (Table 1).  By performing the normalization, the amount of potash per point can be multiplied by the ratio representing each of the 13 polygons across the transverse spread pattern of that point to compute how much potash each polygon receives.  The amount of potash in each polygon is then converted to an application rate by dividing the amount by the area to get the rate in pounds per acre.  This procedure leaves an application rate for each individual polygon.

 A matrix of points was constructed as shown in Figure 8 to develop the 'As Applied" surface.  The user selects the spacing for the matrix before being generated.  For this particular field, a 3.05 m spacing was chosen (10 ft).  The matrix is overlaid to cover the entire field.  Each point is selected individually with all polygons intersecting the point summed to obtain the total application rate for each individual point.  Upon completion of this, all points outside the field boundary are deleted to generate a matrix of points representing the "As Applied" layer (Figure 9).   These points can then be use to compare what was actually applied to the desired application surface in Figure 1.

 

 Results and Discussion

             Figure 10 presents the overlay of the "As Applied" surface and the desired prescription map.  The "As Applied" layer does provide insight of what occurred during field application.  For the most part, an acceptable job was performed on this field in applying potash.  There exist some discrepancies with overlap into zones requiring zero murate of potash.   While application errors occur, zones requiring murate of potash received it.

             A difference surface was created to compare the "As Applied" surface to the desired prescription surface (Figure 11).  This surface was created by averaging the "As Applied" points for each management zone polygon.  The difference was then computed between these two layers.  Nine of the 31 zones received between 15 to 30 lb/ac over- or under- application of potash.  Under application occurred on a high percentage of the zones but most of the zones where between -15 to 15 lb/ac.  Eight zones received between -5 and 5 lb/ac; a very acceptable range for granular application.

 The application of potash along the boundaries of zero application zones occurs because of the lag time and the manner the operator applied material.  The control system on the spreader is unable to perform instantaneous rate changes creating a situation where the desired rate at zone borders might not receive the desired rate.  AgView does have a look ahead time that can be used to look ahead for rate changes.  3 seconds is used for this spreader system based upon the results of Fulton et al. (1999).  This feature helps shift the rate change so it occurs more at zone boundaries rather than entirely when entering the zone.   

The path chosen by the driver also influences material application.  Many times, the driver is unable to drive parallel to zone bounders as shown in Figure 2.  For this field, the operator drove diagonal due to the shape and lay of this field.  This can cause problems in corners where the spreader is in a zone for a very short time.  Another known problem is how well the driver maintains the correct parallel distances on adjacent passes.   For this particular applicator, 16.0 m (52.5 ft) is the effective spread width and needs to be maintained for adjacent passes to minimize over- and under- lap of material spread.  While this paper does not investigate the average distance maintained on this field, this can cause application errors.

 Figure 12 presents one of the problems associated with the current methodology.  When traveling in a curved path, the rectangular polygons are unable to cover the outside area and represent the actual application of material.  The outside path is longer than the inside and since the midpoint is calculated between consecutive points, open areas will occur.  Therefore, points over these areas on the "As Applied" surface have a zero application rate whereas they actually receive potash.

 

 Summary

 This paper demonstrates that a recorded application transverse for murate of potash can be used to generate a VR "As Applied" surface which can be used to assess application errors and view how potash was applied to a field.  To generate such a surface, the transverse spread pattern for one or multiple rates must be determined using a similar procedure used by Fulton et al. (1999).  For this paper, three spread patterns were chosen and incorporated into the Script used by ArcView to create the "As Applied" surface.  However, more or less can be incorporated into the Script as desired.  In the same manner, this process can be used to look at the application of any granular fertilizers or lime in a fixed or variable-rate manner.  The benefit of using Avenue to develop the Script is that it can be easily used and transferred to anyone using ArcView.

 Variability exists in those areas requiring potash but knowing the nature of spinner spreaders and their drawbacks, the spreader does a decent job of VR application of potash.  The results were better than expected compared to the results discovered by Fulton et al. (1999).  While there exists some shortcomings of this method, improvements can be incorporated to improve the generation of the "As Applied" surface and more accurately represent the VR application of granular materials.  Future research consists of determining the longitudinal spread pattern for this spreader.  This will be accomplished by keeping the spreader stationary and setting up a matrix of pans behind the spinner discs to catch the dispersed potash.  The longitudinal spread pattern can then be modeled based upon this test and used to better describe the spread of material over the rectangular polygons.  Knowing this and the velocity of the spreader will eliminate the problem shown in Figure 12.  Other granular materials such as lime and phosphate will also be investigated and incorporated into the Script.  Another aspect that will be added into the Script will be grabbing the desired prescription rate for each of the "As Applied" points.  These two rates can then be compared to quantify application accuracy rather than averaging the "As Applied" points for each management zone.

 Overall, this spreader performs rather well for VR application of potash.  The resulting "As Applied" surface tends to show errors exist and need to be addressed to help improve VR application of granular products.  However, the results provide encouragement that VR application with spinner spreaders is possible with a correlation coefficient of .97 found between the "As Applied" and prescription surfaces.  With further improvements to the Script to generate the "As Applied" surfaces, fixed and VR application errors can be closely estimated to help provide insight to the quality job spinner spreaders and operators perform.

  

References

 ArcView.  1999.  Version 3.2.  Environmental Systems Research Institute Inc., Redlands, CA.

AGR-1.  1998.  Fertilizer and lime recommendations.  University of Kentucky, Lexington, KY.

AgView. 1999.  GIS Solutions Inc., Springfield, IL.

Fulton, J.P.,  S.A. Shearer, G. Chabra, and S.F. Higgins.  1999.  Field Evaluation of a Spinner Disc Variable-Rate Fertilizer Applicator.  Presented at the 1999 Annual
            International Meeting, July 18-21, Sheraton Centre, Toronto, Canada.  Paper No. 991101.

Olieslagers, R., H. Ramon, and J. De Baerdemaeker.  1997.  Performance of a continuously controlled spinning disc spreader for precision application of fertilizer.
            
Precision Agriculture 1997.  BIOS Publishers Ltd.  pp. 661-668.
 Smith, D., D. Seim and C. Finck.  1990.  Micro-managed fields save fertilizer.
            
http://www.sri.bbsrc.ac.uk/webdocs/research/strat_res/preagric/svfo.html.

 SST.  1999.  SSToolbox. Site-Specific Technologies Development Group, Stillwater, OK.