Comprehensive information on Nitrogen Use Efficiency for cereal crop production

REAL-TIME SENSING AND N FERTILIZATION WITH A FIELD SCALE GREENSEEKERTM  APPLICATOR

J.B. Solie, M.L. Stone
Biosystems and Agricultural Engineering Dept.Oklahoma State University, Stillwater, Oklahoma

W.R. Raun, G.V. Johnson, K. Freeman, R. Mullen
Plant and Soil Science Dept., Oklahoma State University, Stillwater, Oklahoma

D.E. Needham, S. Reed, C.N. Washmon
Biosystems and Agricultural Engineering Dept., Oklahoma State University, Stillwater, Oklahoma

INTRODUCTION

     Researchers have long recognized that vegetative indices calculated from optically sensed measurements can correlate with plant properties such as biomass and factors inducing plant stress.  Researchers have also recognized the utility of on-vehicle, real-time sensing of plant properties and variable rate application of nutrients and herbicides.  Machines with these capabilities would enable:  on-the-go sensing and diagnoses of nutrient deficiencies, variably applying materials to correct those deficiencies, and accurately and precisely treating each area sensed without preprocessing data or determining location within a field.  However, as late as 1994, authors (Sawyer, 1994) believed these machines would not be economically feasible. 

By 1994, optical sensor sprayers had been introduced to sense and spot spray weeds.  These sensors utilized passive sunlight illuminated sensors (McCloy and Felton, 1992) or active sensors equipped with their internal lighting (Beck and Vyse, 1995).   The Beck sensor was commercially successful and is sold as the Patchen PHD650 WeedSeekerTM.  The active lighting sensor was particularly attractive, because it could operate under all lighting conditions from total darkness to bright sunlight.

The technology for real-time optical sensing and variable rate nitrogen fertilizer application was introduced in the United States by Stone et al. (1995 and 1996) and in Germany by Heege and Reusch (1996).  By 1996, at least one on-the-go passive optical sensor, variable rate applicator had been demonstrated to the public (Crummet, 1996 and Kimberlin 1996).  This machine, developed at Oklahoma State University, sensed and variably applied liquid UAN fertilizer at 0.75 by 0.75 m resolution.  The Heege sensors have been incorporated into a variable rate, granular fertilizer applicator that senses and treats swath width areas.  This passive lighting sensor applicator is being manufactured and marketed by Norsk-Hydro under the name Hydro N-Sensor.

By 1997, Oklahoma State University engineers had determined they needed a computer based, active lighting, optical sensor to produce a variable rate applicator that could implement variable rate, N-fertilizer strategies, developed by OSU soil fertility scientists.  The resulting sensor used the patented Patchen pulsed lighting and optics.  Oklahoma State University and NTech Industries, Ukiah, CA., the successor company to Patchen, Inc, entered into an agreement to produce and market the new sensor, GreenSeekerTM, along with supporting hardware and software to variably apply N fertilizer.  This paper describes the field scale variable rate applicator, reviews the agronomic research supporting the algorithm for fertilizer rates, and discusses results of field trials with the applicator.

SENSOR APPLICATOR

The field scale GreenSeekerTM variable rate sensor/applicator (U.S. Patent No. 5389781, other Patents, and Patents pending) consisted of 30 GreenSeekerTM sensors mounted on an 18.3 m cantilevered boom of a self-propelled sprayer (Fig. 1).  Sensors were spaced 0.61 m apart.  Each sensor controlled three solenoid valves equipped with agricultural spray nozzles, located directly behind the sensor.  Application rate was varied by turning on various combinations of these valves.  Each sensor scanned a 0.61 by 0.61 m area while applying fertilizer to the adjacent previously sensed area.  All sensors were linked to a user interface (UI), located in the applicator cab, through a controller area network (CAN) bus.  The UI monitored sensor applicator status, transferred application algorithms to the sensors, and collected and recorded sensor measurements and valves settings for post-application analysis.  A computerized pressure controller and GPS receiver were connected to the network.  Vehicle velocity was measured by radar.  The following discussion describes specific components of the sensor/applicator.

 

 

FIGURE 1.  GreenSeekerTM optical sensor, variable rate N fertilizer applicator

GreenSeekerTM Sensor

The GreenSeekerTM active lighting optical sensor uses high intensity light emitting diodes (LED’s) that emit light at 660 nm (red) and 780 nm (NIR) as light sources.  These LED’s are pulsed at high frequency.  The magnitude of the light reflected off the target is measured by a photodiode detector.  Electronic filters remove all background illumination.  Magnitude of the filtered signal is measured by a multiplexed A/D converter.  Measurements are accumulated and averaged over the 0.61-m sensing and treatment distance.  The computer calculates reflectance values for red and NIR and calculates the normalized difference vegetative index (NDVI).  The sensor is temperature stable.

The sensor utilizes the structured lighting optics of the WeedSeekerTM sensor modified for sensor spaced 0.61 m apart.  Intensity of light from this sensor varies inversely as the square of the distance from the source, as do all light sources.  Consequently, reflectance varies inversely with the square of the distance.  However, NDVI, the simple NIR/Red ratio and other normalized and ratiometric vegetative indices do not change with changes in reflectance.  Within the range of the sensor, these indices are insensitive to height.  The field-of-view of the GreenSeekerTM sensor does not change between height above the target between 0.81 m and 1.2 m.  The applicator is equipped with a boom height controller to help keep the sensors within this range.

The sensor calculates application rates and determines the combination of the three valves needed to apply that rate.  The computer sends this information to a second computer located in the valve control module attached to the sensor.  That computer controls the valves.

Valves and Nozzles

Three solenoid valves (Fig. 2) are used to discretely change application rates.  These valves are designed to accept all standard agricultural nozzles, screens, and nozzle caps.  They operate at high enough volumes to treat 0.61 by 0.61 m areas at applicator speeds as great as 24 km/h.  These valves are designed to apply UAN fertilizer.

Nozzle selection criteria for this applicator are different from standard sprayers.  Nozzles should uniformly treat a fixed width (0.61 m) over a range of boom heights.  Because it is possible that adjacent sensors/valves may not be applying fertilizer, nozzle overlap cannot be counted on to improve distribution uniformity. Sixty-five degree (65 deg.) regular or even fan nozzles are relatively insensitive to changes in boom height and can be used to apply material if uniform distribution is required.  Single jet nozzles can be used to stream fertilizers in row crops.  Oklahoma State University created 3-jet nozzles by drilling blank nozzle caps to stream fertilizer at 0.20 m spacing for wheat and other small grain crops.  These nozzles were relatively insensitive to changes in boom height and to spray drift from wind, but were subject to plugging and restrictions in flow because of the small size of the orifices.

Nozzles are binary rated to apply at 1X, 2X, and 4X rates.  By turning on different combinations of these nozzles 8 discrete rates can be applied:  0X, 1X, 2X, 3X, 4X, 5X, 6X, and 7X.  Nozzles are sized to apply the maximum 7X rate at the nominal applicator velocity and sprayer pressure.  For testing in 2002, the applicator was equipped with the custom drilled 3-jet caps and applied UAN at the rate of 13.4, 26.8, 40.2, 53.4, 70.0, 80.4, and 93.6 kg N/ha

 

FIGURE 2.  Tri-valve configuration for variable rate application of 7 discrete rates and off.

 

Other System Components

The sprayer is equipped with a user interface designed and constructed by Oklahoma State University engineers.  The UI is equipped with a back-lit display that can be easily read by the operator.  The UI uses a flash card, which contains input data files for each field and stores output data from the sensors.  The input file contains information of field ownership, location, fertilizer rate algorithm for the crop and conditions, sprayer set point operating pressure, set point operating velocity, and nozzle type and size.  The UI transfers these data to the sensors and pressure controller.  The UI monitors applicator velocity, transmits commands to the sensors to average NDVI data, and transmits new nozzle status to the valve control module.  The UI monitors the CAN network and records CAN messages from each sensor for NDVI and valve status, GPS latitude and longitude, vehicle velocity, spray system pressure and application rate, and other pertinent messages in binary format on the flash card.  Two or more sensor readings may be averaged in the sensor, because the high rates at which data are generated can overwhelm the CAN bus.  Although these data may be averaged, the sensor continues to sense and treat each 0.61 m interval.

The computerized pressure controller measures applicator velocity with a radar unit connected directly to the controller.  The controller maintains constant sprayer pressure independent of the number and combination of nozzles actuated by the sensors.  The controller varies pressure to compensate for changes in speed.

The applicator is equipped with an agricultural GPS receiver and sensor data are tagged with latitude and longitude.  The system accepts meassages in either NMEA 0183 format for RS-232 serial communication or as CAN messages compatible with ISO 11783 standards.

The sensor/applicator system is mounted on a Cherokee 560 self-propelled sprayer (Equipment Technologies, Inc., Mooresville, IN).  The sprayer is equipped with an 18.3-m boom and operates in the field at speeds up to 24 km/h.  Boom control switches located on the sprayer joy stick are interfaced with the UI.  The UI senses the status of those switches and actuates boom sections by turning valves on and off with CAN messages.  The sprayer operator can elect to operate the sprayer manually using joy stick control of spray boom sections or automatically with the sensors controlling the valves. 

Post processing software has been written to read the binary format CAN data from the UI flash card, extract GPS latitude and longitude, vehicle velocity, sprayer system pressure, and the average NDVI and valve states for each sensor.  The software calculates application rates for each valve state.  Each sensor NDVI reading and application rate is georeferenced and the data exported as an ASCII file that can be read by most GIS programs.  Future versions of this program can support shape files.

The GreenSeekerTM sensor applicator can apply any nutrient or plant protectant if application rate is a function of any combination of bands for which LED’s are available.  Indices must be formulated as normalized difference vegetative indices, but in the future may accept ratios or other vegetative indices that are insensitive to changes in reflectance values from changes in boom/sensor height.  Customized algorithms can be downloaded to the sensors by the UI.  In addition, the GreenSeekerTM sensor will have the same capabilities to detect and spot spray weeds as the Patchen WeedSeekerTM sensor sprayers.

AGRONOMIC CONSIDERATIONS

Although the relationships among NDVI biomass, chlorophyll and total plant N have been investigated for some time, the capability of using this information to increase N use efficiency and economic returns to farmers has proved elusive.  Commencing in 1993, Oklahoma State University soil scientists developed a methodology to use NDVI to determine N requirements of wheat (Triticum aestivum L.).  As the result of this research, these scientists have identified three concepts that control this process:  sensing and treating at the optimum resolution, determining crop yield potential, and assessing the response index for additional N fertilizer.

Text Box:   To optimize the agronomic benefits of applying N fertilizers based on plants needs, sensing and treatment should occur a the finest resolution at which variation occurs.  Lengnick (1997) observed that a grid size of 30 m was not adequate for accurate estimation of nugget semivariance of total soil C and N and corn N concentration.  Raun et al. (1998) showed that significant differences in soil and plant variables, including soil N, occurred within sampling distances as short as 0.3 m.  Solie et al. (1999) demonstrated using geostatistical analysis, that to precisely measure soil and plant variables (measurements with high relatedness), measurements should be made at the meter or submeter interval.  Solie et al. (1996 and 1999) showed that indirect, spectral, measurements should be made at that same resolution.  Resolutions for by-plant fertilizer application and weed detection must be even finer (Criner et al. 1999). 

 

 

 

 

 

 

 

 

 

FIGURE. 3.  Grain yield of 24 experiments at different location conducted over a four-year period as predicted by an in-season (Feekes 5) measurement of NDVI adjusted by days from planting when the crop is actively growing (temperature
> 4.4 0C).

Crop yield potential and the expected response to additional fertilizer must be assessed in order to determine N fertilizer application rates.  Yield potential (YP0) is the predicted or expected yield with no additional inputs and no subsequent changes in the crop’s environment.  NDVI measurements at the time hard red winter wheat is normally top dressed in Oklahoma (Feekes growth stage 5) are correlated to yield using an exponential model.  A single exponential curve can predict wheat grain yields for 24 locations collected over a 4-year period.  Adjusting NDVI (Raun et al. 2001) by dividing by the number of days after planting when wheat is actively growing (temperature 4.4 0C or greater) improved correlation, R2=0.591 (Fig. 3).  Prediction of actual yield varied from location to location and year to year, because of intervening factors such as additional fertilizer, rainfall, and disease that affected yields after the date of sensing.  However, the model proved to be a good predictor of grain yield without additional fertilizer over sites and years where planting dates, varieties, and rainfall were variable.

Evaluation of grain yield response to N fertilization in 15-yr maize and 30-yr wheat experiments has shown that check plots where no N has been applied exhibit wide variation in the supply of soil N from year to year (Johnson and Raun, 2002).  This temporal dependence of N availability reinforces the need for mid-season measurements that account for N supplied from non-fertilizer sources.  Although potential grain yield could be estimated, it was necessary to determine the potential yield increase that could be achieved from in-season applications of fertilizer N.  The concept of the fertilizer response index (RI) was developed to meet that need.  RI was calculated by dividing average NDVI from a non-N limiting strip (created in each field by fertilizing a strip at a rate where N would not be limiting throughout the season) by the average NDVI in a parallel strip that is representative of the N availability across the field as affected by N fertilizer applied by the farmer (Johnson and Raun, 2002). Farmer N- fertilization practices could result in 0-N availability to non-N limiting conditions, thus, the initial preplant non-N-limiting strip would likely range anywhere from 20 to 100 kg N ha-1. Thus, computing RI will require the addition of a non-N limiting strip in each field where NDVI from that strip will be divided by NDVI from any strip in the rest of the field receiving the fixed farmer preplant N rate.  The response index computed with spectral measurements of NDVI at Feekes growth stage 5 are well correlated with the response index at harvest, grain yield from the nitrogen rich strip divided by the grain yield of the farmer’s fertilizer rate (Fig. 4).  The model fit improved as the wheat matured with R2=0.77 at Feekes growth stage 9 (Mullen et al. 2002).

Figure 4..  Relationship between RINDVI (response index determined spectrally at Feekes 5) and RIHarvest (grain yield) over 22 locations, 1998,1999, 2000, and 2001.

To determine N fertilizer application rate, an algorithm (Patent pending) was developed to calculate predicted yield with additional N (YPN) adjusted for potential yield with no additional N (YP0), determined while sensing and treating by the expected response to additional N fertilizer.  The amount of additional N needed to obtain the predicted yield was calculated and adjusted by the efficiency of which it is taken up by the grain.  The algorithm accounts for maximum N application rate, RI, active growing days. The algorithm, which is tailored to each field is loaded into each sensor.  Application rate for each 0.61 by 0.61 m area is calculated by the sensor that scans and treats that area.

EXPERIENCE – SPRING 2002

The GreenSeekerTM variable rate sensor/applicator was operated at ten locations and sensed and treated approximately 120 ha.  Areas treated ranged from 0.7 ha strips to greater than 15 ha fields.  Sensors, valves, pressure controllers, and the CAN communications system performed well.

NDVI varied greatly from field to field with CV’s ranging from 12 to 55 from whole plot sizes all greater than 1 ha (Fig 5).  Similar to the results reported by Solie et al., 1999, large differences in NDVI were noted over very short distances (<0.6 m).  Variable rate application of N should therefore be made on a scale at least 0.6 x 0.6 m in order to adhere to our understanding of ‘precise.’  Variable N application on a scale larger than 0.6.x 0.6 m would not conform to what we would expect for ‘precision’ agricultural application.

Figure 5.  NDVI map of one pass by GreenSeekerTM self-propelled field sprayer.

 

At the Amorita, OK site (plots 1 km x 20m), applied N rates (using the variable N rate applicator) ranged from 30 to 105 kg N ha-1, but averaged 41 kg N ha-1.  Wheat grain yield from these same plots was 3.34 Mg ha-1.  When N was applied at a constant rate of 45 kg N ha-1, wheat grain yields averaged 3.04 Mg ha-1.  At a wheat grain value of $44.1 per Mg, and $0.55 kg N fertilizer, the increased yield and decreased N applied would have increased gross revenue by $15.43/ha.  When N was applied uniformly at rates greater than that used for variable N application, no increase in wheat grain yields were observed.  These results are consistent with results reported by Raun et al., 2002 where increased yields at decreased N rates were observed in smaller-scale research plots employing these same sensors and the nitrogen fertilization optimization algorithm (NFOA) developed at OSU.

ACKNOWLEDGEMENT

The authors wish to thank Kent Dieball for his contributions in software and hardware development and the over 30 graduate and nearly 100 undergradueat students who contributed to the project.  The commercial version of the GreenSeekerTM sensors were developed jointly by Oklahoma State University and NTtech Industries, Inc.  The authors acknowledge the contributions of Mr. John Mayfield and Mr. David Smith of Ntech.  The authors also thank the OSU Agricultural Experiment Station, the Samuel Roberts Noble Foundation, USDA-NRI program, USDA/NASA-IFAFS program, and the Oklahoma Soil Fertility Research and Education Advisory Board for their financial support of this project.

REFERENCES

Beck, J. and T. Vyse. Structure and method usable for differentiating a plant form soil in a field. United States Patent No. 5389781. Feb. 14, 1995.

Criner, B.R., J.B. Solie, M.L. Stone, R.W. Whitney. 1999. Field-of-view determination for a bindweed detection sensor. Trans. ASAE. 42(5): 1485-1491.

Crummet, D. 1996. On-the-fly live plant analysis. Oklahoma Farmer Stockman. Farm Progress Publ. Co., Stillwater, OK. 109(2):8-9.

Heege, J.J. and S. Reusch. 1996. Sensor for on the go control of site specific nitrogen top dressing. ASAE Paper No. 961018. ASAE, St. Joseph, MI. 49085-9659.

Kimberlin, K. 1996. OSU demonstrates sensor-based farming. Rancher Farmer New. Maverick Publ. & Design, Inc. Norman, OK. 2(3)6-7.

Johnson, G.V. and W.R Raun. 2002. Nitrogen response index as a guide to fertilizer management. J. Plant Nutr. (In Press)

McCloy, K. and W. Felton. 1992. Controller for agricultural sprays. U.S. Patent No. 5144767

Mullen, R.W., Kyle W. Freeman, William R. Raun, G.V. Johnson, M.L. Stone, and J.B. Solie. 2002.  Use of an in-season response index to predict potential yield increases from applied nitrogen.  Agron. J. (in press).

Raun, W.R., J.B. Solie, G.V Johnson, M.L. Stone, R.W. Whitney, H.L. Lees, H. Sembiring and S.B. Phillips. 1998. Micro-variability in soil test, plant nutrient, and yield parameters in bermudagrass. Soil Sci. Soc. Am. J. 62:683-690.

Raun, W.R., G.V. Johnson, M.L. Stone, J.B. Solie, E.V. Lukina, W.E. Thomason and J.S. Schepers. 2001. In-season prediction of potential grain yield in winter wheat using canopy reflectance. Agron. J. 93:131-138.

Raun, W.R., J.B. Solie, G.V. Johnson, M.L. Stone, R.W. Mullen, K.W. Freeman, W.E. Thomason, and E.V. Lukina. 2002. Improving nitrogen use efficiency in cereal grain production with optical sensing and variable rate application. Agron. J. (in press)

Sawyer, J.D. 1994. Concepts of variable rate technology with considerations for fertilizer application. J. Prod. Agr. 7:195-201.

Stone, M.L., J.B. Solie, W.R. Raun, S.L. Taylor, J.D. Ringer, and R.W. Whitney. 1995. Use of spectral radiance for correcting in-season fertilizer nitrogen deficiencies in winter wheat. ASAE Paper AETC 95133. Am. Soc.Agr. Engr., St. Joseph, MI 49085-9659.

Stone, Marvin L., John B. Solie, Richard W. Whitney, William R. Raun and Heather L. Lees. 1996. Sensors for detection of nitrogen in winter wheat. SAE Technical paper series.  SAE Paper No. 961757.  SAE, Warrendale PA.

Solie, J.B., W.R. Raun, R.W. Whitney, M.L. Stone and J.D. Ringer. 1996. Optical sensor based field element size and sensing strategy for nitrogen application. Trans. ASAE 39(6):1983-1992.

Solie, J.B., W.R. Raun and M.L. Stone. 1999. Submeter spatial variability of selected soil and bermudagrass production variables. Soil Sci. Soc. Am. J. 63:1724-1733.