Time Shift Evaluation to Improve Yield Map Quality

J. P. Beal, L. F. Tian

Published in Applied Engineering in Agriculture Vol. 17(3): 385-390 (© 2001 American Society of Agricultural Engineers ).

Article was submitted for publication in October 1999; reviewed and approved for publication by the Information & Electrical Technologies Division of ASAE in September 2000.

This research has been supported by the Illinois Council of Food and Agricultural Research (C-FAR), Project Number 97I-124-3. The use of trade names is only meant to provide specific information to the reader and does not constitute endorsement by the University of Illinois.

The authors are John P. Beal , Graduate Student, and Lei F. Tian , ASAE Member Engineer , Associate Professor, Dept. of Agricultural Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois. Corresponding author: L. Tian, Univ. of Illinois, 1304 W. Pennsylvania Ave., Urbana, IL 61801; phone: 217-333-7534; fax: 217-244-0323; e-mail: lei-tian@uiuc.edu.

Abstract. Quality data sets are essential for precision farming. Yield monitor data sets have to be processed to coordinate yield information with the proper geographic references because of time delays involved with threshing the crop and receiving the yield information. The time shift needed to account for these delays was shown to vary from 8 to 14 s. Yield data generated by three different combines in corn and soybeans were tested with a range of time shifts. Conventional yield maps were also processed to evaluate the results. A value termed surface area ratio (SAR) was developed to determine the best time shift for the data. SAR is the ratio of the upper surface area of a 3-D yield map to the projected area of that yield map. All of the data sets produced a localized minimum value of SAR at a time shift that also produced a conventional yield map with the best visual interpretation. The time shifts, as determined by the minimum SAR, varied by combine. SAR appears to be a valid method for determining the correct time shift for yield sensing systems.

Keywords. Global positioning systems, Precision farming technology, Site-specific crop management, Yield monitors, Yield maps.

Driven by economic and environmental concerns, precision agriculture technology has moved into the mainstream of production agriculture. The goal of precision agriculture is to manage intra-field variability. Since soils within a field may vary in inherent productivity, optimum use of resources cannot be achieved by uniform applications of seed, fertilizer, pesticides, etc., across an entire field. While the concept underlying precision agriculture dates to 1929 (Linsley and Bauer, 1929), implementation became feasible only with the development of modern electronic tools. These tools include sensors to measure the many variables affecting crop production, a global positioning system (GPS) for geo-referencing of the data, a geographic information system (GIS) for handling the geo-referenced data, and variable rate applicators able to adjust application rates as the applicator traverses a field. Successful implementation of precision agriculture may reduce both the financial cost and environmental impact of agricultural production by preventing excessive applications of chemicals and other agricultural inputs.

Intra-field variability is best visualized with a site-specific map. With current precision farming technologies, one can create maps depicting yield, elevation, soil type, input application, or chemical characteristics (e.g. OM, CEC, P, K, pH, etc.). It is critical to understand that quality data is essential for site-specific management (Bullock et al., 1998). Yield maps, in particular, need quality data because these maps are the proof of any comparison to be made in that field. Data for site-specific maps consist of two elements: geo-referenced location and information about that location. The site-specific information will have higher value when the sensing system is properly coordinated with the geo-referenced locations.

The current state of yield sensing technology has grain mass flow and moisture data, along with location references, recorded at specific time intervals using a combine yield monitor in coordination with a DGPS receiver. If only based on total weights over areas on the order of hectares, the accuracy of yield monitors can be as good as ? 1% (Birrell et al., 1996; Arslan and Colvin, 1999). Typically, the grain that enters the combine at the location marked by the DGPS receiver will not be measured by the grain flow and moisture sensors until approximately 10 to 12 s later. This time delay involves the separating, cleaning, and transport of the grain within the combine (threshing delay). The crop type, crop moisture, foreign material, separator design of the combine, and separator settings can affect the threshing delay. When the data logs are transferred to standard computer formats, the transfer program attempts to coordinate the yield data with the proper location references by shifting the yield data forward in the time sequence. This time shift is supposed to equal the threshing delay of the combine plus other delay factors. The time shift is also dependent on the fore and aft position of the GPS antenna on the combine as well as the processor speed of the yield monitor and DGPS receiver. Most transfer programs combine these different delay factors into a single time shift. Farm YES and Farm HMS (Red Hen Systems, Inc., Fort Collins, Colo.) have a transfer program that evaluates each delay factor separately with tests for determining the threshing and processor delays. The tests are not always successful.

Time shift is a parameter that can be specified whenever data are read through the transfer program. For most systems, the selection of the proper time shift generally has been a trial and error process based on visual interpretation of the yield maps. To change the time shift, the original data has to be reprocessed through the transfer program and new yield maps developed and interpreted.

Yield data that are not properly coordinated with location can cause several problems. If the harvesting is done in a serpentine manner, coincident yield features in adjacent combine passes will be skewed. For example, a 2-s error in time shift with a combine travel speed of 8 km/h will cause data from adjacent passes to be skewed by 4.6 m in each direction from the proper location. GIS programs divide fields into management zones to correlate site-specific information of different data types. Typically these zones are 18 m squares. Time shift errors could place yield data that is supposed to be adjacent into different management zones. Another problem occurs when combine travel speed varies. The harvested area associated with each data location will vary as the travel speed changes. If the time shift is not correct, the harvested area will not be coordinated with the proper grain flow information, and the yield information derived will be erratic due to misalignment of area and grain flow data. This problem is often noticed at the ends of combine passes where extremely low or high yield values will be generated due to the misalignment. If these erroneous values are entered into a management zone, then that zone and possibly all the zones with end of passes will be inconsistent with the rest of the field.

We contend that a yield map can be viewed as a continuous 3-D surface, and that incorrect time shifts will result in abrupt yield changes due to skewed or misaligned data, which will alter the yield surface and increase the surface area. The correct time shift should produce a smoother surface with less area. The upper surface area of a 3-D plot of the transferred yield data can be used as the measure of the effect of changing the time shift. A ratio of this upper surface area to the projected area of the upper surface (test area) is calculated to standardize the measurement. This surface area ratio (SAR) is then plotted versus the various time shifts for each combine. The minimum SAR would determine the optimum time shift. A conventional yield map was processed for each transferred data set for visual comparison.

Objectives

The overall objective of this work was to improve data quality by optimizing yield map processing techniques. The following hypotheses were tested:

·         The time shift needed to coordinate location and grain information in combine-yield monitor systems is variable to the extent that it can reduce yield data quality.

·         Improved data processing algorithms can remove some spatial and temporal errors from yield maps.

Materials And Methods

Yield data logs of 1998 corn and soybean fields in Macon County, Illinois, from three different combines were obtained for this analysis. Two of the combines: a Case IH 1660 (Case IH, Racine, Wis.) and a 1996 Deere 9500 (Deere & Co., Moline, Ill.) were equipped with Yield Monitor 2000 yield monitors (Ag Leader Technology, Ames, Iowa) and CSI GBX-8 DGPS receivers (Communications Systems International, Calgary, Alberta). The other combine, a 1995 Deere 9500, had a Micro-Trak Grain-Trak monitor (Micro-Trak Systems, Inc., Eagle Lake, Minnesota) with an Ashtech BR2G DGPS receiver (Magellan Corp., Santa Clara, Calif.) installed in it. All of the combines were using the St. Louis Coast Guard radio beacon signal for differential correction. The GPS antenna location on all three combines was centered 3.7 m behind the initial gathering point of the crop. The grain flow sensor was in the top boot and moisture sensor was mounted on the fountain auger of the clean grain elevator on each combine. Specific information pertaining to the combine settings was not available. It is known that the two Deere 9500 combines had some differences in their separator settings. Grain moisture did not vary appreciably within the test areas because the areas were harvested in a short time period.

The AL2000 Ver. 1.1 (Applications Mapping, Inc., Roswell, Ga.) transfer program was used for the logs from the AgLeader monitors, and Micro-Trak Card Utilities Ver. 2.5 was the transfer program for the Grain Trak monitor logs.

All the yield data and maps were processed with SSToolbox Ver. 3.1.2 (SST Development Group, Inc., Stillwater, Okla.). Surfer for Windows Ver. 6 (Golden Software, Inc., Golden, Colo.) was used to produce the 3-D yield plots and to calculate surface area.

A yield monitor logs yield data and location references at 1-, 2-, or 3-s intervals at the combine operator?s discretion. Thus, only time shifts that are multiples of the recording intervals need be considered so that the shifted data can be matched with a referenced location. The recording interval selected for both of the Deere combines was 2 s, while for the Case IH combine it was 1 s. After selecting a set of time shifts to test, the original data logs were processed through the transfer program for each time shift. Then the data sets were entered into SSToolbox to process the conventional yield maps, and to prepare the data for Surfer. A rectangular test area that did not include the ends of each combine pass was selected from the data set for the Surfer grid. The test area is only a geographic reference for Surfer to select data points. At the end of each combine pass extreme yield values may be caused by the misalignment of grain flow and area values. These inappropriate yield values then would produce errors in the surface area calculations. The test areas were from 1.5 ha to 8.1 ha in size. Each of the fields had been harvested within one day?s time. Each of the transferred data sets was mapped for the test area with Surfer to prepare the 3-D yield data plot, measure the upper surface area and the projected area. Figure 1 shows 3-D yield plots of a test area at two time shifts. During the mapping procedure, the scattered yield data points are transformed to a uniform grid pattern by linear Kriging interpolation in the Surfer program. The grid node spacing was determined to have an effect on the surface area measurement. A ratio of grid nodes to the data points in the test area (grid node ratio) was used to set the grid node spacing. A grid node ratio > of 2 provided the most consistent surface area results.

IET450_files/image1.gif

(a)

IET450_files/image2.gif

(b) Figure 1. Yield surface plots for a soybean test area harvested with a 1995 Deere 9500 combine. (a) 3-D yield plot at 14-s time shift. (b) 3-D yield plot at 8-s time shift.

Results and Discussion

Conventional yield maps were processed for each time shift in a test range. The maps for one of the data sets are included in this article (figs. 2a-e). The combine direction of travel is left and right with blocks of several passes in each direction, and the end rows have been removed. This data set has known yield features labeled in figure 2a: an underground petroleum pipeline and an undrained depression area adjacent to the pipeline where rainwater can pool and reduce grain yield. On the yield maps at either end of the time shift test range, the dot patterns that represent the physical features have jagged outlines or are indistinguishable because the dots are skewed from their adjacent positions. In figure 2a, the dot pattern for the depression is skewed one way and in figure 2e, it is skewed the opposite way. The dot pattern for the depression and the pipeline come into focus with the 8-s time shift (fig. 2c). The ends of each combine pass are another indication of improper time shift. Irregular ends of adjacent passes, with extreme yield values, are evident in figures 2a and e. Misalignment of area and grain flow information causes the extreme yield values as the combine speeds up or slows down at the end of each pass. The irregularity of the ends is caused by some referenced locations not having any grain flow information due to misalignment. The ends of adjacent passes are more even with the edge of the field, and the yield values are more consistent with the rest of the field when the time shift is correct (fig. 2c). Also, there are fewer erratic yield points throughout the field due to misalignment of area and grain flow data.

IET450_files/image3.gif

(a)

IET450_files/image4.gif

(b)

IET450_files/image5.gif

(c)

IET450_files/image6.gif

(d)

IET450_files/image7.gif

(e)

IET450_files/image8.gif

Figure 2. These yield maps were produced from data collected on a 1995 Deere 9500 combine harvesting soybeans using the indicated time shifts when transferring the data. Combine direction of travel is left and right. (a) Yield map produced from data using a 14-s time shift. (b)Yield map produced from data using a 10-s time shift. (c) Yield map produced from data using an 8-s time shift. (d) Yield map produced from data using a 6-s time shift. (e) Yield map produced from data using a 4-second time shift.

Visual interpretation of yield maps can be subjective. With the series of maps shown, the data changes from a haphazard mixture of dots in several yield ranges with a 14-s time shift to recognizable patterns representing known features and then on to another incoherent jumble of multi-range dots as the time shift decreases to 4 s. On the best-appearing maps, yield features were distinct, the ends of adjacent combine passes were even and of consistent yield, and the amount of yield data fluctuation was reduced. Similar results were achieved with the yield maps from the other combines. The time shift for the best appearing maps ranged from 8 to 14 s.

The 3-D plots (figs. 1a and b) show the texture of the yield surface in the test area. The 14-s surface has a rougher texture with several yield spikes due to data misalignment. The SAR of this surface is 1.629E+05. The 8-s surface is smoother and has the minimum SAR of 1.274E+05.

All of the data sets produced a local minimum value for the surface area ratio when the SAR was plotted versus various time shifts (figs. 3-5). The minimum SAR values varied by combine at time shifts from 8 to 14 s. For each combine, the minimum value time shift did not vary noticeably by crop. The time shifts for these minimum SAR values corresponded with the conventional yield maps that produced the best visual impression.

IET450_files/image9.gif

Figure 3. Surface area ratios for 1996 Deere 9500 combine. The minimum surface area ratio value indicates the optimum time shift value.

IET450_files/image10.gif

Figure 4. Surface area ratios for 1995 Deere 9500 combine.

IET450_files/image11.gif

Figure 5. Surface area ratios for Case IH 1660 combine.

Time shift variations were shown among the different combines. These variations may be due to combine settings and/or design, or crop conditions. The difference in yield monitors and DGPS receivers may also be a factor in these variations. The crop type did not cause a time shift variation. For this study, the ability to recognize that variations are present and determine the optimum time shift is more important than to determine the causes of the variations.

Surface area ratios of normalized data could also serve as indexes of yield data variability. Higher SAR values would indicate yield data that is more volatile.

Conclusions

The subjective analysis of conventional yield maps and the corresponding minimum SAR values indicate variations in the correct time shift of yield data among the combines tested. On some maps, where the time shift error was 4 s and combine travel speed of 9.7 km/h, yield information was mapped at a location that is 10.7 m from the appropriate location. Various factors such as combine separator design and settings and monitoring systems can affect the data gathering process so that the time shift should be adjusted. Without this adjustment, the grain flow and moisture values cannot be properly coordinated with location and area information to deliver data that accurately represents that location.

The 3-D yield plots illustrate the variability of the yield surface at different time shifts. Within its own group of time shift tests, SAR produced a minimum value at the optimum time shift. Surface area ratios are a valid analytical method for determining the correct time shift for a particular combine-yield monitor system.

This article is a preliminary study on the initial processing of yield monitor data. It is meant to bring to light one of the intricacies involved in getting the best quality information from the mountains of data being generated by precision agriculture today. Data downloaded from yield monitor logs are not ready to use. The initial processing step of running the data through the transfer program is where grain information and location references are reconciled. If the time shift is handled properly, subsequent analysis of the yield data and development of application maps or other derivatives can be done with the knowledge that the grain values and location references are properly coordinated.

REFERENCES

Arslan, S., and T. S. Colvin. 1999. Laboratory performance of a yield monitor. Applied Engineering in Agriculture 15(3): 189-195.

Birrell, S. J., K. A. Sudduth, and S. C. Borgelt. 1996. Comparison of sensors and techniques for crop yield mapping. Computers and Electronics in Agric. 14(2-3): 215-233.

Bullock, D. G., D. S. Bullock, E. D. Nafziger, T. A. Peterson, P. Carter, T. Doerge, and S. Paszkiewicz. 1998. Does variable rate seeding of corn pay? Agron. J. 90(6): 830-836.

Hair, J., R. Anderson, R. Tatham, and W. Black. 1995. Multivariate Data Analysis , 4th Ed. Upper Saddle River, N.J.: Prentice Hall.

Linsley, C. M., and F. C. Bauer. 1929. Test your soil for acidity. Agricultural Experiment Station Circular No. 346. Urbana, Ill.: University of Illinois.

About, Search, Contact Us, E-Mail Alert, Subscribe/Order,
Join ASAE, ASAE, Table of Contents, Home, Get Acrobat Reader

Copyright ? 2003. American Society of Agricultural Engineers. All rights reserved.

error