Australian Cotton Grower Article
Broughton Boydell, Brett Whelan, Alex McBratney and Murray Schoenfisch
Precision Agriculture
The term "precision agriculture", a bit of a buzz phrase around the traps these days, basically
refers to the utilization of technology in an attempt to achieve greater control over, and
efficiency with, the application of resources on your farm. Someone implementing PA
techniques should first characterize and understand within-field variability, and after that they
can alter management; fertilizer rates, water application , pesticides etc. so that they are applied
where they are needed, when they are needed and in the most safe and environmentaly correct
manner. If this is done properly then the increased application efficiency leads to a reduction in
the potential environmental jeopardy (which is a socio-political dividend) and should increase the
farms net profit (which is an economic dividend).
Engineering to deal with agronomic reality
Good tools are ones that help you to do a job that needs to be done. Great tools are ones that
allow you to do a job which otherwise could not have been achieved. Reduction in the variability
of a line has been a historic agenda in beef, wool, wheat and cotton breeding programs however
the commercial reality is that under field conditions, where environmentaly variability is added
to the equation, most lines exhibit a fairly high degree of variability. Subsequently, a variable
population is treated as an average. Some treatment rates will be over that required and some
under what they should be. How inefficient this is will be different on a field by field basis and is
related to the degree of spread between the higher and lower yielding regions. In production
agriculture the opportunity to treat individuals on an "as needed" basis rarely arises. An exception
to the rule may be the modern stud breeder who uses individual beast identifiers (ear or tail tags,
tattoo's or chips) and tools such as electronic scales to record and monitor an individuals lineage
and progress. The unique identifier allows the stud manager to recognize an individual's personal
attributes and thus to capitalize on these opportunities. Like the stud manager, a cotton property
manager may recognize variability between and within fields, however, due a number of factors
which include an inability to really tie down these locations and communicate them
unambiguously to others in anything more than vague "southwestern end" type terms, and the
difficulty in terms of operational efficiency of varying treatments so that they accurately match
demands of small areas all over the field using existing equipment.
The tools.
The global positioning system (GPS) and various continuous sensors and controllers have recently changed all of this. Differentially corrected GPS is able to define exact, re-visitable locations in a field (± 1m) which may be unambiguously relayed to anyone. Continuous sensors, including yield sensors, may be linked to the GPS location and used to create maps which record and illustrate the exact degree and location of yield variability. These locations may be revisited and investigated to help in the diagnosis of the cause of variability. Real-time continuous variable rate controls which meter fertilizers, and pesticides are now available, and, when linked to GPS locations and an electronic map controller enable a manager to identify areas which require differential management within a field. This information may be conveyed to the computer with confidence that an operator will perform the tasks assigned. Thus the ability to treat individuals within a field has evolved.
Like the scales which allow a stud manager to quantify the performance of a herd on a beast by
beast basis, the yield monitor may be used to quantify local yield performance, breaking a field
into potentially hundreds of individual management units. Once quantified, the yield information
may be used in management decisions aimed at exploiting variability and it is this use of
knowledge that makes accurate yield data valuable.
Yield mapping for cotton
Over the past few years there have been a number of alternative methods investigated which try to determine the yield of cotton at a point or in a region of a field. Conventional field trials use a labour intensive bagging combine or an even more intensive manual picking. While both are extremely accurate methods for quantifying yield, most farmers would chase you off the place with a big stick and plenty of colourful language if you suggested that either may be a suitable yield mapping option for them. Research has been conducted to assess the ability of remote sensing techniques to map yield however, the accuracy of predictions, and to a lesser extent, the cost of collection have halted its widespread adoption. Recent experience in the US grains industry indicates that the real adoption of yield mapping only occurs when there is a combine mounted continuous flow sensor available. Using a continuous sensor, data is collected accurately, easily and in a relatively cheap manner. Research into possible yield sensors for cotton began as far back as 1993 with the University of Tennessee looking into using a linear array of light beams shining across the flow of cotton to "count" locs as they passed. (see fig 1).
Sensors still under development in the US include options which may be located in the picking
head, in the basket delivery chute, actually in the basket and for behind the machine to detect and
quantify "tagged" cotton (similar to a grain-loss monitor). The perceived importance of this
sensor for PA in cotton is supported by Wayne Smith of John Deere's precision farming division
in the USA who has stated that "the most important sensor under development by Deere is the
cotton yield sensor". While Deere has yet to commercially release a monitor, two monitors were
successfully trialled in Australia this past season and both will be commercially released in the
USA this season. Micro-trak inc. (developed in Australia by Murray Schoenfisch at the
university of Southern Queensland) and Zycom inc. (USA) will release their monitors for the
1997 summer picking in the US (August-December). Both systems utilize an array of light
beams across the delivery chute which operate similar to a burglar alarm in that they are able to
detect each time a light beam is broken. The frequency that each beam is broken per second is
related to the yield of cotton that has passed through the sensor array. The fundamentals of the
process is the same as those for grain yield monitoring in that by knowing the distance that the
picker has traveled forward, and the width of pick (# of rows), the unit can calculate the area that
the recently picked cotton came from. It is then just a mathematical operation to take a yield
reading of kg/m2 to something more recognizable like bales or kg/ha.
Potential accuracy: Grain vs. Cotton
It takes approximately 12 seconds for a grain to travel from the cutter bar to the yield sensor in a grain combine. In this time the crop is carried from its original location in the field to another further down the row. To compensate for this, yield estimates are tagged not to the combines position in the field when sensed but where it was 12 seconds ago. While this is a necessary correction, the question has been asked whether it is enough. Research conducted by Brett Whelan of the Australian center for Precision Agriculture indicates that as a result of the dynamics of combine separation, individual (1 second) yield estimates for grain crops are significantly less accurate than the manufacturer claimed 2-3% error which is determined by full load calibration tests. This reduced local accuracy is explained by mixing in the machine. Not all of the grains from a single ear of wheat will hit the sensor at the same time. Some grains are threshed and separated immediately, while others are walked to the back of the machine before entering returns to be re-thrashed. The result is that the grain sensed at a point in time may have been cut anything from 6-20 seconds ago. Brett says that "this temporal spread manifests itself as a spatial error, and really acts on yield data like a smoothing filter". In yield maps, low yielding areas (like a road or weed patch) is smoothed out by higher yielding neighbors and vica-versa. The dynamics of the cotton picker are far less convoluted. Locs of cotton are picked and pass the sensor within a couple of seconds. Product is not carried for great periods of time before being sensed and subsequently, for the cotton picker, yield measurements are far more accurate on the local scale. When it comes to interpreting cotton yield data, you can have far greater confidence in the local accuracy. More confidence will allow stronger questions to be asked. This filtering effect is evident in figure 2 where simulated estimated yields from a time corrected grain combine sensor and estimates from a cotton picker sensor are compared to the true yield at that point in a field.
.
Results for yield monitoring in 1997
Both of the now commercially available yield monitors for cotton were run in Australia this year
and preliminary results in the form of yield maps and calibration results indicate that the sensors
are indeed superior to those available for grain. Claimed accuracy based on load calibrations for
these monitors is <2-3% error (approximately the same as for grain sensors). Figure 3 is a yield
map of a cotton field near Wee Waa NSW from this past season.
With yield expressed as un-ginned seed-cotton mass (kg/ha) there is considerable evidence of
within-field variability and a number of interesting features. (north = top, east = right etc.). The
southern end of the field shows a significant reduction in yield (this area was re-planted due to
poor establishment). The best yields are generally in the eastern third of the field near the head-ditch. Another interesting feature is the linear region of higher yield running north south and
marked at either end (A). This feature is apparently the expression of an old gravel road that used
to pass through the field. This ability to capture and accurately represent a narrow feature is
further (albeit anecdotal) evidence to support the accuracy of these sensors.
Research into the viability of PA techniques for Australian Cotton
The CRC for Sustainable Cotton Production, the Cotton Research and Development Corporation (CRDC) along with the Australian Center for Precision Agriculture (ACPA) initiated a research project in the 1996/7 picking season aimed at investigating the usefulness of cotton yield maps for the Australian cotton industry. Initially the aims of this research is to characterize the variability of yield and attempt to understand the true reasons behind yield variability as seen in an Australian field environment. Yield data, collected with both of the commercially available yield monitoring systems, will be investigated and an attempt made to understand the reasons for variability based on associated data from intensive field scouting, remote sensing and soil sampling throughout the season. Ultimately the objective of the research is to
1. Evaluate the relative potential of PA techniques for Australian cotton.
2. To prioritize future research into these applications.
ACPA, the CRC for Sustainable Cotton Production and the CRDC have anticipated grower
interest in this PA technology and are seeking to generate a knowledge base that will be useful
for growers when it comes to making managerially beneficial decisions based on yield maps and
in determining the best method for achieving an increase in farm productivity, profit, and
environmental sustainability using Precision Agricultural techniques.
Contact: Broughton Boydell
ACPA- CRC for Sustainable Cotton Production
Agricultural Chemistry and Soil Science
The University of Sydney
(02) 9351-2947 or (018) 435-088
bboydell@agec.usyd.edu.au
THE UNIVERSITY OF SYDNEY
© 2008 - Australian Centre for Precision Agriculture