precision agricultural system

A Precision AgriculturalI Systems

Chenghai Yang and Won Suk lee

CONTENTS

4.1 Introduction .............. """""""""644.2 Measurement of Spatial Variability """"""'65

4.2.r GPS......... """'65

4.2.2 Soil Sensors """"""' """'664.2-2-l NIR and MIR Spectroscopy """""""664.2.2.2 Raman Spectroscopy """""" """"""'684.2.2.3 Electrodes """"'684-1,.2.4 Microwave """"69

4.2.3 Crop Sensors"""""" """'69

4.2.4 Wireless Sensors """"""'704.2.5 Yield Monitors """""""" 71

4.2.6 Remote Sensing """""""734.2.6.1 Remote Sensors""' """"""734.Z.6.2AirborneMultispectralandHyperspectrallmaging

Systems """"""744.2.6.3 Sot"ltit" Imaging Systems"" """""""76+.2.A,.+ Image frocessirig and Analysis """"""' """"""'77

4.3 Data Analysis and Management""""""" """"""""""'784.3.1 GIS.......... """'78

4.3.2SpatialDataAnalysisandManagementZones'.,...........,..794.3.3 Site-Speciflc Appiication Maps """' 80

4.4 Variable Rate Application"""' """"""""""' 81

4.5 Economic and E]ivironmental Benefits """' 83

4.5.1 Profltabi1ity"""""""' """834.5.2 Environmental Beneflts """"""" ""'84

4.6 SummarY """"""""" 85

Disclaimer """"" 85

,........86References

63

Agricultural Automation64

4.1 INTRODUCTION

Precision agriculture is a new farming practice that has been developing since the

late 1980s. R"r"urfi activities in precision agricultur:e started with the development

of yield monitors, gria soit sampling, ,oii ,"r"rors, positioning systems, and variable-

rate technology at'universities in ttre united States and Europe in the iate 1980s'

By the early 1gg0r,;;; Gld monitors and variable rate controllers became com-

mercially available. witt udrunces in grouur positioning systems (GPs)' geographic

i,formation systems (GIS), remote ,"i'ing, and sensor technology' the agricultural

community has witnessed a ,upio gro*ti'ot u new body of precision agriculture

technologi", ,in." it e mid-1gggs. The flrst biannual international conference on

precision agriculture was held in l992.The flrst biannual European and Asian con-

ferences on precrsion agricultufe were held in 1991 atd}}os'respectively' The first

internationaljournal"rr-titl"dPrecisio'n'Agriculturewaslaunchedinlggg'andpreci-sion agriculture has been an important iopic in many agriculture-related journals'

Several books on this topic have U."n f'uiished' including The Precision-Farming

Guide for Agricuir.rirr, by Ess and ivlorgan (2010) and Handbook of Precision

Agriculture: principtle, ond Apptiroiio,nr,"*trich was edited by Srinivasan (2006)'

Theseconferences,proceedings,journals,andbooksprovideeffectiveforumsfordisseminating original and fundamental research and experiences in the fast grow-

ing area of precision agriculture'precision agricultuie has been variously referred to as precision farming' pr:-

scriptionru,*ing,,putiallyvariablefarming,site-speciflccropmanagement,varl-able rate t".t norogvi;;;;" bur a few. There are numerous deflnitions for precision

agriculture, but the central concept is to identify within-field variability and manage

that variability. More speciflcally, precision agriculture uses a suite of electronic

sensors a,d spatiai irt*rn*io, i..troiogies (.e., GPS, GIS, and remote sensing)

to map within-fleld soil and crop gr"*it, iariability and to optimize farming inputs

(fertilizers, pesticides, seeds, water,-etc.) to the splciflc "onditio,' for each area of

a field with the "i* oi increasi,g farm profits and reducing environmental impacts'

To automatically implement ,t, .oi..pi of precision agricult*re, the following

four main steps are generally involved:

1'Measuringspatialvariability.Ground-basedsensors,GPSreceivers,andremotesensingsystemsareneededtomapcropyield,soilattributes'p"rt

"onaitio"i "ra other importa.nt v11i1Uies affecting crop production'

co*p,rt"Jed data acquisition devices thlt:an-lntegrate field sensors and

GPSreceiversaren""o,o.yforeffectivedatacollection'2. Analyzing dara and making J""i.ionr. spatial data analysis tools' includ-

irrgGlsarrdimageprocessrng,areneededtomanagefield-collecteddataand information from other lources such as topographic maps and soil

maps.Statisticalandgeostatisticaltechniquesneedtobeusedtoana|yzedataarrdidentifythepatternsofspatialvarrabilityinmeasuredvariablesand the relationship, u*onglhe variables. Fields need to be divided into

eitherirregularmanagementzonesorregularlygriddedcellsformanage.ment based on fleld spatial ,o.iou*i y. rf," opiimal management plans for

Precision Agricultural Systems

farming inputs need to be determined for each management zone or cellbased on the specific conditions. This may require the use of an expert sys-tem that can integrate the knowledge of farmers and crop consultants withbiological, economic, and crop growth models to make better decisions.Implementing management decisions. Applicators with variable rate con-trollers are needed to apply various inputs at desired locations with correctrates based on site-specific application maps or data from real-time sensors.Evaluating economic and environmental beneflts. Precision agriculture hasthe potential to increase economic returns and minimize environmentalimpacts, but it requires investment in new equipment and time. To facilitatethe adoption of precision agriculture, it is important to document the eco-nomic and environmental benefits of precision agriculture operations.

This chapter provides an overview of the major technologies involved in pre-cision agriculture, including GPS, soil sensors, crop sensors, wireless technology,yield monitors, remote sensing, GIS, and variable rate technology. The emphasis isplaced on the principles and practice of these technologies for precision agricultureoperations.

4.2 MEASUREMENT OF SPATIAT VARIABITITY

Measurement of within-field variability is the first important step in precision agri-culture. Unless the level of variability is known, an appropriate management decisioncannot be made. GPS is the foundation of precision agriculture operations, includ-ing data collection and variable rate application. Various ground-based sensors inte-grated with GPS are used to measure spatial variability in soil attributes, crop yield,and crop pests. Airborne and high-resolution satellite imagery has become a majordata source for documenting soil and crop growth variability, because each imageprovides a continuous view of all fields in the imaging area.

4.2.1 GPS

The GPS was developed by the U.S. Department of Defense for military applica-tions, but it has been used for many other applications including precision agricul-ture. The system became fully operational on April 21, 1995.It provides service forpositioning, navigation, and timing.

There ale three segments in the GPS: space, control, and user. The space seg-ment consists of 24 active NAVSTAR (Navigation by Satellite Timing and Ranging)satellites with additional ones (seven more as of October 4, 20ll). The satellitesare positioned in six orbital paths, at least four in each path. One revolution takes12 hours. The space segment is designed to guarantee that users can see at leastfour satellites anytime and anywhere in the world. The satellites broadcast radiosignals in 1200-1500 MHz. It is equipped with atomic clocks, which are the mostcritical component, since the positioning is based on the exact timing. The coltrolsegment receives and transmits information to the satellites and ensures the accuracyof the satellite positions and clocks. There are several monitoring stations around the

65

--)-

4.

Agricu ltu ral Automation66

world, and the master control station is l0cated in the schreiber Air Force Base rn

Colorado springs, colorado. The user segment is basically GPS receivers that use

the satellite signals to calculate position' velocity' and time'

Differential Gps (DGPS) provides users beiter positioning accuracy (typically

1 to 3 m) by "ul"rluting"rro*

at a given location and transmitting it to the usets'

The following are the different ways to calculate the differential positioning errors:

(1)NationwideDGPD,(2)localbasestation,and(3)satellite-basedcorrectioninciud-ing wide Area A.ugmeniation Svstem'. omnisrA*' i11::1ii:"^S't:"?:3 Yllill;;frd;;"#]i*l?ir"matic GpS receiver provides centimeter positioning accuracy

in all three dimensions using carrier phase ryt$s; however, it is much more

*Til:tL?'lil*:."*T'f,ndu*"ntaltool"sNNNriculture,T.,:,:gl":i::l

referenced and site-sfecific data acquisition Uy)tivtOiilg coordinates of different crop-.^ G^trl

idiile coordi

;:il"rffiH;"*|ffi;ffi;;;;,.:ion ugii*t ore are numerous. Exampres are nerd

boundary mapping, yi"tA -upping, soii propJrties mapping' pli-"]T:t^t:ll,Tlln,il3lt"

;nffi ffi;J*?ppirg, *""a ona pest mapping, and auto,omous vehicle guida,ce'

Recently, Lightsquared Subsidiary l-ic ([eston, vA) developed a national 4G

LIE (fourth-generation long-term evolution) open wireless broadband network; how-

ever, it interferes with the hilh-precision GPS system' Because there is no engineering

solution to avoid the interference, currently discussions are being exchanged among

end users, LightSquared, the Department tf Ho*"lond Security' the Department of

Transportation, the National Telecommunications and Information Administration'

and the Federal Communications Commission (FCC). on September 13, 2o11, the

FCC (2011) announced that more tests are necessary to resolve potential harmful

interference of the LightSquared's network to the GPS'

4,2.2 Soll SrNsons

Soil properties afe very important for efficient crop management because they affect

crop growth. tmportant soii prop"rties include soif type, soil texture, soil pH' organic

matter content, fertility, moisture content, and soil compaction' Many different sens-

ing techniques have been implemented for measuring these soil properties' They

are near-infrared (NIR) and mid-infrared (MIR) spectroscoPY' Raman spectroscopy'

electrodes, and microwave'

4.2.2.1 NIR and MIR SPectroscoPy

Since Bowers and Hanks (1965) investigated the effect- of organic matter on reflec-

tance measulements, many studies have"be"n conducted to determine soil properties

using NIR and MIR spectroscopy toward the development of a real-time in-field

sensor (Sudduth and Humm el,li93a;Ben-Dor and Banin, 1995; Ehsani et al'' 1999;

Shibusawa et a1., 1999;Thomasson et al., 2001; walvoort and McBratney', 2001; Lee

at al', 2003;Mouazen et al., 2005a, 2005b; Brown et al., 2006; Maleki et al,, 2007,

2008).Soil properties that have been studied include soii organic matter' soil moisture'

and some prirnary nutrients such as nitrogen (N) and phosphorus (P)' Sudduth and


Hummel (1993a) developed a portable spectrophotometer to measure soil organicmatter, cation-exchange capacity, and moisture content, and tested it in the field(Sudduth and Humrnel, 1993b). Ehsani et al. (1999) developed calibration modelsusing partial least squares (PLS) and principal component regression to estimatesoil mineral-N content using soil NIR reflectance in 1100-2500 nm. They reportedthat the models were very robust, but suggested that site-specific calibration of themodels was necessary. For measuring underground soil reflectance, Shibusawa etal. (1999) developed a portable spectrophotometer in 400-1700 nm to measure soilmoisture, soil pH, electrical conductivity, soil organic matter, and NOr-N. Hummelet al. (2001) used an NIR soil sensor to predict soil moisture and organic mattercontent. Mouazen et al. (2005a, 2005b) developed a portable NIR spectrophotom-eter in 306-17ll nm to measure soil moisture content and to identify soil texture.Bogrekci and Lee (2005a) examined the spectral characteristics of four commonsoil phosphates (A1, Fe, Ca, and Mg phosphates) in Florida and reported that thosephosphates could be detected with a classification error of l.9Vo. Maleki et al. Q001)investigated a portable visible (VIS)-NIR P sensor for variable rate application ofelemental P. Maleki et al. (2008) implemented a real-time application of phosphate(PrOr) for maize planting using an on-the-go VIS and NIR soil sensor. Christy(2008) developed a shank-based spectrophotometer and reported that organic matterwas predicted best from field testing. However, as Ge et al. (2006) pointed out, oneof the major challenges for implementing a real-time soil property sensing is that soilproperties vary greatly from location to location since soil is a very complex mixtureof many different objects.

Soil moisture is another property to be estimated by NIR since there are verydistinct water absorption bands in the NIR region. Some of the studies include test-ing of a soil moisture rneter using NIR reflectance at 1800 and 1940 nm (Kanoet al., 1985), a global NIR calibration equation to determine soil moisture content(Slaughter et al., 2001), and an exponential prediction model (Kaleita et al., ZOO5).

Other than using NIR, a cornmercial device is available to measure soil moisture(EM38, Geonics Limited, Ontario, Canada) that uses electronic magnetic induction.A transmitting coil induces magnetic field in the soil, and a receiving coil measuresinduced current in the soil, which is used to measure soil conductivity, and then usedto estirnate soil moisture indirectly.

Using soil diffuse reflectance in the MIR range measured by a Fourier trans-form infrared (FTIR) spectrophotometer, Ehsani et al. (2001) estimated soil nitratecontent and found a strong nitrate absorption peak at 7194 nm. Linker et al. (2004)also used FTlR-attenuated total reflectance spectroscopy in the MIR region to esti-rnate soil nitrate content, and found that the best root mean square prediction errorsranged from 38 to 43 ppm N.

Another technique to detect soil properties is to use aerial and satellite images.Among the earlier studies, Landsat TM and SPOT images were used to detect differ-ent soil properties (Coleman et al., 1993; Agbu et al., 1990) and soil lines (Galvao andVitorello, 1998, Fox and Sabbagh, 2002). Some soil properties were detected usingaerial images, including soil P and organic matter (Varvel et a1., 1999), soil moisture(Muller and Decamps, 2000), and soil texture (Barnes and Baker, ZOOZ).

67

6B Agricu ltu ral Automation

For soil compaction detection, Glancey et al. (1989) tested a chisel device thatcould estimate soil cutting force distribution, Adamchuk et al. (2001) designed andtested a vertical blade for measuring soil mechanical impedance and resistance pres-sure. They reported highly correlated estimated values. Chung et al. (2003,2004)developed an on-the-go soil strength profile sensor and reported R2 values of 0.61and0.52 to estimate prismatic soil strength index for a claypan soil field and a flood-plain soil field, respectively. Mouazen and Ramon (2006) investigated an onlinesystem for measuring soil draft, cutting depth, and moisture content. Adamchukand Christenson Q007) used strain gauges to develop an instrumented blade to mapsoil mechanical resistance. Andrade-Sanchez et al. (2007) reported that the soil cut-ting force was influenced by soil bulk density, moisture content, and the location ofthe cutting element within the soilprofile. Andrade-Sanchez and Upadhyaya(2007)reported the development of the University of California-Davis soil compaction pro-flle sensor, and Andrade-Sanchez et al. (2007) reported that the device was ableto produce a soil cutting resistance variability map. Chung et al. (2006) developeda soil strength profile sensor using load cells. Then, Sudduth et al. (2008) tested thetwo previously developed on-the-go soil compaction sensors (soil compaction profilesensor and soil strength proflle sensor) and reported that the two sensors performedsimilarly. Hemmat and Adamchuk (2008) suggested that the fusion of different sen-sors could map spatially variable soil physical properties better.

4.2,2.2 Raman SpectroscopyRaman spectroscopy was also used to predict some soil properties. A portableRaman sensor was developed for measuring soil P content using a 785-nm laserprobe assembly and a detector array in 340-j460 cm-r (Bogrekci and Lee, 2005b).Its lowest root mean square error was reported to be 151 mg/kg by PLS regression.

4.2.2.3 Electrodes

Another type of sensor to measure soil properties is an electrode. Since Adsett andZoerb (1991) explored ion-selective electrode (ISE) technology to measure soil nitlatecontent, many researchers have tested this method. Adamchuk et aI. (1999) devel-oped an on-the-go soil pH sensing system and achieved good performance. Birrelland Hummel (2001) tested a multiple ion-selective field effect transistors (ISFETs)and reported that the ISFETs worked well for manually extracted soil nitrate contentin solutions, but not with samples from an automated soil solution extraction system.

Kim et al. (2006) investigated nitrate and potassium ion-selective membranes andreported that the membranes showed linear response with higher nitrate and potas-sium concentrations than 10-a mol/L. Kim et al. Q007a) found that cobalt rod-basedelectrodes showed sensitive response over a typical phosphorus concentration rangein agricultural fields. Kim et al. (2007b) studied the applications of ISE to simul-taneous measurement of soil primary nutrients (N, P, and K) and reported that theNO, ISEs worked well; however, K and P ISEs showed lower detection accuracies.Sethuramasarryraja et al. (2007) investigated the ISEs to detect soil pH, residualnitrate (NOe-), and soluble potassium (K*) contents, and reported that tlie soil typeand the soil/water ratio affected sensor performance. Adamchuk et al. (2007) sug-gested field-specilic calibration for more accurate pH mapping. Sethuramasamyraja


et al. (2008) used ISEs to measure soil pH, soluble potassium, and residual nitratecontents, and achieved stable calibration for pH and K electrodes, but not for nitrate,demonstrating the potential on-the-go soil property sensing. Lee et al. (2010)reported that commercial electrodes are currently available for measuring differentsoil properties including moisture, pH, nitrate, potassium, bromide, and chloride,manufactured by London-Phoenix Company (Houston, TX), Cole-Parmer (VernonHills, IL), and Zhejiang Top Instrument Co. Ltd. (Hangzhou, Zhejiang, China).

4.2.2.4 MicrowaveMicrowave is also used to measure soil moisture content. Since Schmugge (1918)investigated thermal IR approach, passive and active microwave sensing methodshave been studied to detect soil moisture including large-scale measurements of soilmoisture (Jackson and Schmugge, 1989; Vinnikov et al., 1999: Tien et al., 2007).Judge Q007) presented a brief review of different techniques and models to measuresoil moisture using microwave remote sensing, and reported that major chalienges ofthe microwave sensing would include lack of long-wavelength satellite-borne radi-ometefs, seasonal components in theoretical models, and integration of hydrologicand microwave measurements.

4.2.3 Cnop SnNsons

Crop sensors include sensing systems for yield, nutrients, water, weed detection, cropbiomass, and health. Many different sensing techniques have been developed andtested, and some of them have become commercially available.

NIR spectral reflectance and thermal imaging are used to monitor crop health andnutrient/water contents. NIR reflectance has been used extensively by many research-ers. Thornas and Oerther (1972) reported a strong relationship between reflectanceat 550 nm and sweet pepper leaf N content. Blackmer et al. (1996) reported signifi-cant wavelengths (450, 630, 690, 710, and 760 nm) to estimate nitrogen contents ofcorn canopies. Min and Lee (2005) developed prediction models for citrus nitrogenconcentrations using multivariate statistical analyses and reported0.LZVo predictionerror in the validation set. They also reported several important wavelengths (448,669,719, 1377,1773, and223l nm) for citrus N detection.

Thermal imaging and multispectral/hyperspectral imaging are used to identifycrop status. For example, Alchanatis et al. (2006) investigated mapping of waterstatus in a vineyard using thermal and VIS images, and reported that stomatal con-ductance and stem water potential were highly correlated with the crop water stressindex. Cui et al. (2010) investigated automatic soybean rust detection using theratio of infected area and rust color index extracted from multispectral images, anddemonstrated the feasibility of detecting the disease under laboratory conditions.Moshou et al. (2011) developed a multisensor decision system using hyperspectralreflectauce and multispectral imaging along with neural networks, and demonstratedthe functionality of automatic disease (yellow rust disease in winter wheat) detectionthrough field tests.

Lee et al. (2010) reviewed different methods for crop canopy and biomass detec-tion, including laser scanning, ultrasonic sensing, light penetration of the canopy,

69

70 Agricultu ral Automation

synthetic aperture radar satellites, and Landsat TM imagery. These techniques can

be used for agricultural chemical applications, irrigation, and plant health assessment.

They described that the following different remote sensing techniques are used todetect plant diseases: (1) reflection based sensing inciuding spectral reflectance andmonochromatic and multispectral imaging, (2) emission based sensing such as ther-mography, and (3) fluorescence. Their advantages, disadvantages, and potential uses

are summarized. Lee et al. also reported an electronic nose to identify fruit ripening,disease, or physical damages by detecting simple or complex volatile (odors) changes.

Many studies were conducted for testing commercial sensing systems. Wendrothet al. (2011) compared the Hydro N-sensor (Oslo, Norway) with the GreenSeeker(Trimble Navigation Limited, Sunnyvale, CA; formerly NTech Industries, Ukiah,CA), and reported that the GreenSeeker had better sensitivity on measuring theNormalized Difference Vegetation Index and better identifled the need of nitro-gen than the Hydro N-sensor; however, its data processing was a complex process.GreenSeeker is used to determine the amount of nitrogen for crops on-the-go andto apply variable amounts of nitlogen at different locations. The device measuresreflectance of crops at656 and770 nm to determine nitrogen concentration. A com-mercial optical sensor was developed by Force-A Scientific, which can be used todetect nitrogen deflciency, wheat protein, thermal stress, and polyphenol antioxi-dant of leaf and fruit epidermis by measuring absorbance of samples using differentlight-emitting diodes as illumination sources. An internal GPS receiver is used forgeoreferenced data acquisition. Cerovic et al. (2008) used three commercial devices(Dualex FLAY Dualex ANTH, and Multiplex; Force-A, Orsay, France) to measureflavonol and anthocyanin contents of grapes, and reported that the Multiplex showedconsiderable potential to be used for precision viticulture in the vineyard. Louis etal. (2009) measured chlorophyll and epidermal phenolic compounds using a portableMinolta SPAD meter and a Dualex meter (Force-A, Orsay, France), and proposedoptical signatures of immature and mature leaf phenological stages with contrastingnitrogen and carbon economy. Another example of commercial units is Crop Circle(Holland Scientific Inc., Lincoln, NE). This system can be used for mapping plantbiomass or soil color, and also for variable rate fertilization application. Solali et al.(2008) used a Crop Circle sensor to measure corn N status, and reported that thechlorophyll index (Clrpo = (NIR880/VIS.e0) - 1) showed the greatest potential for Nestimation. However, they noted that further studies would be needed for differentsoils, climate, and locations.

4.2.4 Wrnrlrss SrNsons

Wireless sensors are typically used toward automation for farm management such as

continuous soil moisture monitoring for efficient irrigation. A wireless sensor net-work (WSN) consists of different sensors and communication devices. Wireless dig-ital communication devices include ZigBee IEEE802.15.4, Bluetooth IEEE802.L5.l,and Wireless LAN IEEE802.1lblgln, Their frequencies are mostly 2.4 GHz, butZigBee uses also 915 MHz.

Wang et al. (2006) provided an overview of wireless technology developmentalong with its applications and discussed the advantages and some obstacles in

Precision Agricultu ral Systems

adopting the technology. They reported that future application of the wireless tech-nology could include precise farm management, food safety, traceability of agricul-tural products along with radio frequency identification (RFID) tags, and ubiquitouscomputing. Lee et al. (2010) described basic information of wireless technologyand different applications related to specialty crops and presented the following as

typical applications in agriculture: management of farming, precision agriculture,optimization of piant growth, surveillance in farms, advertisement for consumers,education and training for efflcient farming, and research. They suggested the fol-lowing factors to increase adoption of the technology: Iow cost, easiness, rugged-ness, long-range communication, and scalability to a high number of sensor nodes.

The following are some examples of the wireless sensor technology. Hamrita andHoffacker (2005) implemented a wireless system to monitor soil moisture contentusing a microcontroller and passive RFID tags. Vellidis et al. (2007) developed a

WSN for smart irrigation in cotton using moisture sensors, a circuit board, and an

active RFID tag to provide wireless sensor interface. Darr and Zbao (2008) describeda model that can predict losses of wireless transmission signal due to structural inter-ference and quantify them in a poultry layer facility. Zhang et al. (2011) developed a

four-layer (sensor node layer, gateway, central platform layer, and application layer)wireless network system, and reported good performance of the system.

4.2.5 Yrrlo MoNrroRs

Commercial yield monitors are being adopted steadily in the United States, Europe,and other parts of the world in recent years. Yield monitoring is widely used ingrain harvesting, but yield monitors have been developed and used for non-graincrops such as cotton, potatoes, sugar beets, sugarcane, forage, and tomatoes. Someof the commonly used yield monitors include monitoring systems from Ag LeaderTechnology (Ames, IA), John Deere (Moline, IL), and Case IH (Racine, WI). Theintegration of yield monitors with GPS enables yield raeasurernents to be associatedwith their geographic positions for creating yield maps. Yield maps are critical toprecision agriculture because they can be used for determining management strate-gies and for evaluating the results of these strategies.

A yield monitoring system consists of a display console and a set of sensorsinstalled on a harvester for measuring crop flow rate, moisture content, groundspeed, and cutting width that are mathematically related to yield. A GPS receiveris usually used with a yieid monitor for yield mapping. Some yield monitors rely ona header position sensor to accurately calculate harvested acreage. Several types offlow sensors are used for measuring grain flow, but impact-based mass flow sensorsare commonly used in many yield monitoring systems. Grain flow can be sensed byplacing an impact plate in the path of clean grain flow to measure either the forceapplied by the grain impacting the plate or the amount of plate displacement thatoccurs wheu grain strikes the spring loaded plate. The force or the displacementmeasured is proportional to the grain flow. A cotton flow sensor uses light emittersand light detectors mounted on opposite sides of a cotton picker's delivery ducts suchthat cotton passing between the emitters and detectors reduces transmitted light. Themeasured reduction in light is converted to flow rate. A moisture sensor allows grain

72 Agricu ltu ral Automation

yield to be converted to standard moisture content. A capacitance-based sensor is

often used for measuring instantaneous moisture and the sensor measures the con-ductivity of the grain as it moves past the sensor plates. This conductivity is directlyrelated to the moisture in the grain. Ground speed can be measured by a magneticwheel/shaft sensor, a radar speed sensor, or a GPS receiver. The cutting width can be

easily determined for row crops, but a cutting width sensor may be needed for non-row crops for accurate yield measurement. Instantaneous yield can be calculatedusing the following formula:

\, 36fl-

SW(4.t)

where Ydenotes yield (t/ha),/is the flow rate (kg/s), s is the ground speed (km/h),and w is the effective cutting width (m). To convert grain yield to standard moisturecontent, yield can be multiplied by a factor of (l - actual moisture)/(l - standardmoisture). Generally, when actual moisture is less than the standard moisture, no

conversion is necessary.

Yield mapping is easy in principle, but presents a challenge if accurate and reli-able maps are to be obtained, because so many sensors are involved in a yield moni-toring system. Therefore, it is important to understand the errors associated with a

yield data set and to keep them to a minimum. Blackmore and Marshall (1996) andBlackmore and Moore (1999) identified a number of errors associated with yieldmonitor data. Some of the signif,cant errors include unknown cutting width, timelag of grain through the threshing mechanism, sensor accuracy and calibration, andGPS errors. Numerous techniques have been developed to address these problems.Time lag, also known as time delay and throughput lag, refers to the time it takes forthe crop to pass through the mechanisms of a harvester before reaching the point ofmeasurement by a yield monitor. Time lag will result in mismatches between yieldmeasurements and their positions. Obviously, time lag varies with harvester equip-ment, sensor location, crop being harvested, and other factors.

Different methods have been used to determine accurate estimates of time lag.Searcy et al. (1989) used a first-order time delay function with a step input to modelgrain flow for the combine. Wagner and Schrock (1989) obtained crop entry and exittime lags based on the times the combine header entered and left the plot as well as

the times grain flow measurement began and ended. Stott et al. (1993) determinedtime lag by comparing grain flow rates collected in alternating directions acrossa known zero-yielding portion of the field. Birrell et al. (1996) used both a simpletime delay rnodel and a first-order model to calculate instantaneous yield response.Several other studies have been conducted to estimate time lag and examine its sta-bility (Murphy et al., 1995; Nolan et al., 1996; Whelan and McBratney, 1997; Chosaet al., 2001).

Altlrough different models and transfer functions have been used to characterrzeflow dynamics in a harvester, they are technically difflcult for adoption by practition-ers. Therefore, applying a constant time lag to match flow data to positions is themost widely used method. Most commercial yield mapping software packages such


as Ag Leader's SMS Basic software use a constant time lag to compensate for theeffect of the flow delay. However, it is difficult to objectively determine a correct oroptimum time lag for a whole field or for each area of the field. Chung et a1. (2001)used geostatistical and data segmentation methods for determining yield monitoringtime lag with objective criteria. Beal and Tian (2001) used the ratio of the surfacearea of a three-dimensional yield map to its projected area for determining yieldmonitoring time lag. Correct time delays were determined based on minimum arearatio values. Yang et aI. (2002) developed a method for determining the optimumtime lag for yield monitoring based on remotely sensed imagery taken during thegrowing season. The underlying assumption for the method is that there exist statis-tically significant correlations between crop yield and remotely sensed imagery andthat incorrect time lags will cause a reduction in the correlations. Therefore, a timelag that maximizes the correlation can be considered the correct or optimum timelag.

4.2.6 Rrrraorr SrNsrNc

Remote sensing is the science and technology of acquiring information about theearth's surface without physically touching it. It uses sensors to measure and recordthe reflected and emitted electromagnetic radiation from the target area in the fieldof view of the sensor instrument. The detecting and recording instruments are gen-erally referred to as remote sensors. Remote sensors are typically carried on air-craft and Earth-orbiting satellites, but some sensors can be handheld or mounted onground-based vehicles. Remote sensing applications in precision agriculture havebeen steadily increasing in recent years because of improvements in spatial, spectral,and temporal resolutions of both airborne and satellite remote sensors. Airborneor satellite imagery allows a farmer to have a bird's-eye view of the crops grow-ing on the entire field or entire farm. This section will provide a brief overview ofthe remote sensing systems that have been used for precision agriculture, includingground-based spectroradiorneters, airborne digital multispectral and hyperspectralirnaging systems, and high-resolution satellite imaging systems.

4.2.6.1 Remote Sensors

Remote sensors include all the instruments that detect and measure reflected andemitted electromagnetic radiation from a distance. These instruments fall into twobroad categories: non-imaging (i.e., spectroradiometers) and imaging (i.e., cameras).According to the types of sensor-carrying platforms, remote sensors can be ground-based, airborne, and spaceborne. Both non-imaging and imaging sensors can be car-ried in all three types of platforrns, although non-imaging sensors are primarily usedfor ground-based applications.

Portable non-imaging remote sensing instruments include radiometers and spec-troradiometers. The types of radiometers can be single-band radiometers, whichmeasure radiation intensity integrated through one broad waveband, and multispec-tral radiometers, which measure radiation intensity in more than one broad wave-band. Spectroradiometers Ineasure radiation intensity over a continuous range ofwavelengths by simultaneously sampling a large number of narrow spectral bands.

73

74 AgriculturalAutomation

Forexample,aFieldSpecHandHeldportablespectroradiometerg:ii*:lSpectralDevices,Inc.,Denver,Co)acquiresacontinuousspectrumbymeasuringradiationintensity in Stzband; u"i*""t 325 and 1075 nm' The FieldSpec 3 portable spec-

trometerfromtheSamecompanycantakemeasurementsfrom350to2500nmwithsampling int".uut, ii i.+ n* at iSO-tOO6 nm and 2 nm at 1000-2500 nm' Ground-

based spectroradiometers have been *;;;y used in precision agriculture for estimat-

ingsoilproperties(sooo,,tt.andHumm.t,tggla;Thomassonetal',2001),assessingcrop nitrog"r, ,tutrJiBor"ugtiu and varco , 2OO2; Zhao et aL., 2005), and detecting

"t"'p p*,<Mirik et a1'' ZOOT;Liu et aI" 2010)'

Imaging ,"nrorc are design"o to proulJ" ,i"*, of a target area from vertical

(nadir) perspectives. Aerial photogra;hy is the.oldest and simplest form of remote

sensing and provides film-baitd ph#;""ph: ;1ry,"^tf fine spatial resoiution' but it

hasbeengraduallyreplacedbycontinuinginnovationsinaigitatimagingtechnol-ogy. Electro-opticat ,"rro., are the main imaging sensors being used today' These

sensors use detectors to convert the ,"n""r"d ulalot emitted radiation from a ground

scene to proportionot "t""t irur ,ignoir, which are theu recorded on magnetic' opti-

cal,and/orsolid-statemediaandcanbeviewedastwo-dimensionalimagesonacomputer or television monitor. El"";;;;tical imaging systems are capable of oper-

atinginnumefousbarrdsf,o,n*o,",p".'or'"gion's,yftheelectromagneticspec.trum,includingn"u'-ttft'u'io1et'VIS'NtR'tuttR'andthermalinfrared'

4.2.6.2AirborneMultispectralandHyperspectrallmagingSystemsThegrowinginterestinairborneremotesensingwasstimulatedbyresearchanddeveiopment on muttispectral i*.gi;;r,"*, u"na ttt"it applications in the 1980s

and 1gg0s (Meisner and Lindstro*, t"98i; Pearson- et 11"

tgg+;p'v.,'itt et a1" 1995)'

The increased use of this technology *ut uu'iuutedto its iow cost' high spatial reso-

lution, immeoiate availability of irrlog"ry for visual assessment, compatibility with

computer processing system-s,-a"a "&riiv to obtain data in narrow spectral bands

in the vIS to MIR region of the spectrum (Mausel et a1'' L992; King' 1995)' Most

airborne digital imaging systems t"r, p*"iae multispectral imag^e data at spatiai

resolutions ranging from less than 1 m to a few meters and at l to lznarrow spectral

bands in the vIS to NIR regions of the electromagnetic spectrum (Pearson et a1''

1994;Escobareta1.,1998;Yang,2010)'Air.bornemultispectralimageryhasbeenwidely used in precision agriculture u""'sing soil variability 6arnes et a1'' 2003)'

mapping crop gr.owth a,d yield. vJabiilrv ciong and Anderso,, 1999; Pinter et al"

2003;Inman ei ai., 2008), and detecting ".op inse-ct and disease infestations (Moran

et al., lgg1; yang et al., 2005; Frank" ird tti"nr, }}}7;Backoulou et a1'' 2011)'

Most airborne multispectral imaging systems. use muitiple charge-coupled device

(ccD) "u*"rur,

;; "i which ir?q,iipp"d with.a different bandpass fllter' This

approach has the advantage that each "o*"'o can be individually adjusted for opti

mum focus and aperture settlngs, iut has the disadvantage that the images from al1

bandshavetobeproperlyaligned'onesuchsystemisafour-camefarnultispec-tral imaging system assembled ^i

,t " USDA-ARS kika de la Garza Subtropical

Agricultural Rlsearch Center in Weslaco, TX' The multispectral system consists

of four high-resolution ccD digital cameras and a ruggedized PC equipped with

aframegrabberandimageu.q,,i.,i,ionsoftware.Thecamerasaresensitiveinthe


400-1000 nm spectral range and provide 2048 x 2048 active pixels with 12-bit datadepth. The four cameras are equipped with blue (430-470 nm), green (530-570 nm),red (630-670 nm), and NIR (810-850 nm) bandpass interference filters, respectively.Another approach to multispectral imaging is to use a beam splitting prism andmultiple CCD sensors built in one single camera to achieve multispectral imagery.One such system is the MS4100 multispectral 3CCD camera (Geospatial Systems,Inc., West Henrietta, NY), which uses a beam splitting prism and three CCD sensorsto acquire images in three to five spectrai bands within the 400-1100 nm spectralrange.

Hyperspectral imaging sensors or imaging spectrometers are a new generation ofelectro-optical sensors that can collect image data in tens to hundreds of vely nar-row, continuous spectral bands throughout the VIS, NIR, MIR, and thermal infraredportions of the spoctrum. These systems offer new opportunities for better differen-tiation and estimation of biophysical attributes for a variety of remote sensing appli-cations. Airborne hyperspectral imagery has been evaluated for characterizing soilfertility (Bajwa and Tian, 2005), mapping crop yield variability (Goel et al., 2003;Yang et al., 2004, 2007, 2010a: Zarco-Tejada et al., 2005), and detecting crop pests(Fitzgerald et a1., 2004; Yang et al., 2010b).

Many commercial airborne hyperspectral sensors such as AVIRIS, CASI,HYDICE, and HyMap have been developed and used for various remote sensingapplications. Advances in CCD cameras, frame grabber boards, and modular opticalcomponents have also led to developments of low-cost airborne hyperspectral imag-ing systems from off-the-shelf products (Mao, 1999). A hyperspectral imaging sys-tem assembled at the ARS Weslaco Research Center is an example of such a system(Yang et al., 2003). The system consists of a digital CCD camera, an imaging spec-trograph, an optional focal plane scanner, and a PC computer equipped with a framegrabbing board and camera utility software. The CCD camera provides LZ80 (h) xl0Z4 (v) pixel resolution and true 12-bit dynamic range. The imaging spectrographis attached to the camera via an adapter to disperse radiation into a range of spectralbands. The effective spectral range resulting from this integration is from 467 to932 nm. The optional focal plane scanner can be attached to the front of the spectro-graph via another adapter for stationary image acquisition. The horizontal and verti-cal binning capability of the camera makes it possible to obtain images with variousspatial and spectral resolutions. For most applications, the hyperspectral sensor isconfigured to capture images with a swath of 640 pixels in 128 bands.

As hyperspectral imagery is attracting more interest, more commercial airbornehyperspectral imaging sensors have become available in recent years with improvedspatial and spectral resolutions and high performance inertial navigation systems forincreased position accuracy. For example, the AISA family of airborne hyperspec-tral sensors from Spectral Imaging Ltd. (Oulu, Finland) includes two sensors in the0.4- to 0.97-pm range (AisaEAGLE and AisaEAGLET), one sensor in the 0.91- to2.5-prm range (AisaHAWK), one sensor in the 0.4- to 2.5-pm range (AisaDUAL),and a thermal sensor in the 8- to 12-pm range (AisaOWL). The AisaEAGLE sensorcan capture images with a swath of I0Z4 pixels and in up to 488 bands, whereas theAisaOWL can get a 384-pixel swath in up to 84 bands. All sensors are equipped witha high-performance, three-axial inertial navigation sensor for monitoring the aircraft

75


position, and altitude. The sensor integrates solid state gyros and GPS with a real-time Kalman filter for increased accuracy.

4.2.6.3 Satellite lmaging Systems

Satellite remote sensing systems not only cover large sulface areas on the earth,but also view the same target area repeatedly. Traditional satellite systems such as

Landsat and SPOT have been widely used for agricultural applications over largegeographic areas, but this type of imagery has limited use for precision agricul-ture because of its coarse spatial resolution. Remote sensing from space is rapidlychanging with many countries and commercial flrms developing and launching newsystems on a regular basis. Commercial availability of high-resolution satellite sen-sors (i.e., IKONOS, QuickBird, GeoEye-I, and WorldView-2) has opened up newopportunities for rnapping within-field variability for precision agriculture. Thesesatellite sensors have significantly narrowed the gap in spatial resolution betweensatellite and airborne imagery. IKONOS and QuickBird imagery has been evaluatedfor assessing soil properties (Sutlivan et a1., 2005), estimating crop leaf nitrogencontent (Bausch et al., 2008), and mapping crop yield variability (Chang et al., 2003;Dobermann and Ping, 2004; Yang et al., 2A0O.

When GeoEye, Inc. (Herndon, VA) successfully launched the IKONOS satellite in1999, rt made history with the world's first high-resolution commercial remote sensingsatellite for civilian uses. IKONOS provides 1-m panchromatic images in the 0.45- to0.90-pm spectral range and 4-m multispectral imagery in the blue (0.45-0.52 pm),green (0.51-0.60 pm), red (0.63-0.70 pm), and NIR (0.76-0.85 pm) bands. The pan-chromatic and multispectral imagery can be merged to create l-m coior imagery (pan-sharpened). The radiometric resolution is 11 bits, or 2048 gray levels. The image swathis 11.3 km at nadir, and the revisit time is less than 3 days. Shortly after the successfullaunch and operation of IKoNos, DigitalGlobe, Inc. (Longmont, Co) launched theQuickBird satellite in 2001. QuickBird provides panchromatic and multispectral datain essentially the same spectral ranges as those of IKONOS, but at a higher spatialresolution. QuickBird acquires panchromatic data with 0.60-m resolution and fourmultispectral bands with 2.4-m resolution. Similarly, pixel depth is 11 bits. The imageswath at nadir is 16.4 km, and the sensor can tilt up to 45' off nadir.

GeoEye again made history with the launch of GeoEye-l in 2008.It offers unprece-dented spatial resolution by simultaneously acquiling 0.41-m panchromatic and 1.65-mfour-band multispectral imagery. The spectral ranges are similar to those of IKONOS.The pixel dynamic range is also ll bits. The image swath is increased to 15.2 km.On October 8, 2009, DigitalGlobe launched WortdView-Z, the first high-resolutioneight-band multispectral satellite, to acquire panchromatic data at 0.46-m resolutionand multispectral irnagery at 1.84-m resolution. WorldView-2's unique combination ofhigh spatial and spectral resolution provides new opportunities and potential for a vari-ety of practical remote sensing applications. The imagery is distributed at either 0.5- or0.6-m resolution for the panchromatic band and at either 2.0- or Z.4-mresolution forthe multispectral bands, depending on the sensor's viewing angle. The image swath atnadir remains at 16.4 krn, and the average revisit time is about 1.1 days. Table 4.1 givesthe spectral characteristics for the four high-resolution satellite sensors.

Precision Agricultural Systems 77

TABTE 4.1

spectral characteristics for worldView-2, euickBird, ceoEye-l, andIKONOS

Spectral Band (pm)

Band Name

Panchromatic

Coastal

Blue

Green

Yellow

Red

Red Edge

NIRINIR2

IKONOS

0.450-0.900

0.445-0.516

0.505-0.595

0.6324.698

0.757-0.8s3

GeoEye-1

0.450-o.800

0.450-o.510

0.510-0.580

0.6s5-o.690

0.7804.920

QuickBird0.450-0.900

0.45M.5200.520-0.600

0.630-0.690

0.760-0.900

WorldView-2

0.4s0-0.800

0.400--0.450

0.450-0.510

0.510-0.580

0.58s-0.625

0.630-0.690

0.1as{.7450.770-0.895

0.860-1.040

In addition to their high spatial resolution, these satellite sensors offer image dataat 8 times as many gray levels as the 8-bit traditional satellite sensors. Moreover, thehigh revisit frequency and fast turnaround time of these high-resolution satellites arecertainly advantages over traditional satellites. These advantages, combined withtheir relatively large area coverage and ability to take imagery over any geographicarea, make high-resolution satellite imagery attractive for many applications, includ-ing precision agriculture.

4.2.6.4 lmage Processing and AnalysisImage processing and analysis is an important component of remote sensilg technol-ogy' Different imaging systems provide different types of imagery, and therefore avariety of techniques need to be used to process and, analyze the image data. Thesetechniques are diverse, tanging from simple visual interpretation to sophisticatedcomputer processing methods. Image processing and analysis generally involvesimage display and enhancements, image registration and rectificition, image clas-sification, accuracy assessment, and more advanced spectral analysis techniques.Because of limited space, the reader can refer to other textbooks for this special topic(campbell,2002; Richards and Jia, 2005; Lillesand et al,, 2007; ERDAS, 2010).

Two other spatial information technologies closely related to remote sensing areGPS and GIS. GPS data are often required to determine the geographic locationsof airborne imagery and to geometrically correct and georefer=r". the imagery. AGIS provides a platform for GPS data and remote sensing imagery to be diJplayed,analyzed, and integrated with other spatial data. At the same time, remote ielsingimagery and GPS data have become primary data sources for GIS analysis. Indeed,these technologies have been interrelated to one another, especially ior precisionagriculture applications.

7B

4.3 DATA ANALYSIS AND MANAGEMENT

Agricu ltu ral Automation

Data analysis and management is probably the most difficult area and the area wherethe greatest amount of work has been done in precision agriculture. It involvesmanipulating and analyzing rneasured data and determining proper contiol actionsor at least presenting useful information to the farmer. One of the most importantaspects of these data is their spatial or geographic nature. Therefore, spatial infor-mation technologies such as GIS and geostatistics are important for processing thistype of data and for developing management zones or site-specific application maps.

4.3."t CIS

By the U.S. Geological Survey Q}Il} a GIS is defined as "a computer system capa-

ble of capturing, storing, analyzing, and displaying geographically referenced infor-mation; that is, data identified according to location. Practitioners also regard thetotal GIS as including the procedures, operating personnel, and spatial data that gointo the system." GIS has found many applications in precision agriculture as wellas in other areas.

There are two different types of GIS data: raster and vector. Raster data are acell-based data format and each cell has a value. Images and grids are examples ofraster data. Vector data are based on coordinates of different map features. A pointis stored as a single x, y coordinate, and a line is stored as a pair of x, y coordinates.Similarly, a polygon is stored as a set of x, y coordinates.

Precision farming data often need to be interpolated to fill in gaps between datapoints. Common interpolation methods include nearest neighbor, local averaging,inverse distance weighting (IDW), contouring, and kriging. For the nearest neighbormethod, an unknown point is set equal to its nearest neighbor. Local averaging is toestimate unknown values by a simple average of a selected number of points aroundthe desired location. IDW is based on the fact that points closer to an unknown pointare more likely to have similar properties than those farther away, and thus weightsare determined inversely proportional to the distance between data points whenestimating the unknown point. Contouring is to connect points of the same va1ue,

and unknown value can be estimated between the known points. Kriging is knownto be an optimal interpolation method. It first estimates the variability of the knowndata set. Then to estimate unknown value, IDW is conducted for points closer, andequal weights are used for points farther away. This method is slower than othersbecause of intensive computations.

Like GPS, GIS can be considered a major tool for implementing precision agri-culture. In this regard, Earl et al. (2000) provided an overview of the role of GIS inautonomous field operations, emphasizing that GIS could play an important role insimultaneously interpreting rnultiple spatial and temporal field attributes for efficientfarm management. Pierce and Clay (2007) also described various GIS applicationsin agriculture including nitrogen management of sugar beet using GIS and remotesensing, development of productivity zones from multiple years of yield data, site-specific weed management, soil salinity mapping, variable depth tillage assessmentusing GIS, and on-the-go soil strength sensing. However, Nemenyi et al. (2003)


reported potential problenis (e.g., incompatibility of GIS and precision agriculturesoftware) related to commercial GIS and GPS systems in the market.

GIS can be applied to precision farming in many different ways. It can be used to

select locations or areas based on certain characteristics (e.g., high yielding areas and

low yielding areas in a field). Another application is data manipulation and analysis

including mathematical or logical operations. An example of this application wouldbe to create a dry yield map using moisture content and yield maps. GIS is veryeffective for handling multiple data layers in precision agriculture. It can integrateyield maps, nutrient maps, soil type maps, and other data layers for making manage-

ment decisions.

4.3.2 Spmnl Dnrn AruRrvsrs AND MANAGEMENT ZoNrs

Precision agriculture involves a large number of data layers. A data layer can be asoil organic map, a soil nitrogen map, a crop disease incidence map, a yield map, orany other map characterizing the spatial variability of a variable within fields. Each

data layer can be stored in either vector format (i.e., soil type and sampling points) orraster format (i.e., remote sensing imagery). When discrete samples are taken from a

fleld as in the case of grid soil sampling, data interpolation methods such as IDW and

kriging are commonly used to estimate the values of a variable at unsampled loca-tions. Discrete data are generally intelpolated into regularly spaced raster format,which can then be used for generating contour maps and for performing spatial GISanalysis. On the other hand, an airborne image or satellite image can be convertedto a polygon map by using image classiflcation techniques to statistically clusteringimage pixels into categories of similar spectral response.

The real impetus for site-specific crop managelnent is within-field spatial vari-ability. Understanding the magnitude and patterns of spatial variability in measuredvariables provides an important basis for dividing a field into appropriate man-agement units for site-specific crop managernent. Geostatistics is a useful tool fordescribing the spatial dependence of a variable such as crop yield or a soil attribute inprecision agriculture. Spatial dependence implies that samples collected at smallerseparation distances are more likely to have similar values than those collected at

larger separation distances. The semivariogram, or simply variogram, in geostatis-tics describes the spatial dependence of a variable (Isaaks and Srivastava, 1989). Thevariogram shows how strongly and extensively the samples are related in space. Theinfluence range of the variogram can be used to determine appropliate pixel cell sizeinto which the field should be divided for variable rate application. Another impor-tant use of the variogram is for kriging to generate unbiased estimates of a variableat unsampled locations or at a regular grid.

Because of the limitations associated with using intensive grid soil sampling todevelop prescription maps, the concept of management zones has received consider-able attention. Researchers have understood the value of dividing whole fields intosmaller, homogeneous regions for fertility management. Ear'lier studies proposedthe division of fields by soil type (Carr et al., l99l) and landscape position (Fiez et

al., 1994). Other methods of management zone delineation have used remote sens-ing technologies to charactqrize within-field spatial variation. Remote sensing-based

79

BO Agricu ltu ral Automation

management zones, which are obtained by classifying image pixels into categories

of similar spectral response, should reduce both the variance within each zone and

the number of soil samples required to characterize each zone. Yang and Anderson(1999) u$ed airborne multisp-ectral imagery and unsupervised classification tech-niques to detelmine within-field management zones for two grain sorghum fields

with multiple stresses. Two of the zones identifled were soil related. One representedareas with insufflcient soil moisture and the second depicted areas where plants

suffered severe chiorosis due to iron deficiency. The remaining zones represented

areas with different production levels due to a combination of soil and environmentalfactors.

Airborne images taken at early stages of crop growth can reveal plant growthpatterns that could be observed in images acquired later in the season. However,

imagery obtained at the time or shortly after the crop has reached its maximumcanopy cover shows more stable patterns that remain during the rest of the growingseason. The important implications of these observations are that plant growth varia-tions and stress conditions can be detected within the growing season so that propermeasures may be taken to correct some of the problems such as nutrient deficien-cies. Significant correlations existed between yield and image data, and yield was

more strongly related to images taken around peak growth, indicating that imagerytaken at this particuiar stage could be a better indicator of yield (Yang and Everitt,2002). Significant differences in grain yield among the spectlally determined zones

indicate that aerial digital imagery can adequately capture within-field yield vari-ability. Although the spatial plant growth patterns identified by within-season digitalimages may not always perfectly match those revealed on yield maps from yieldmonitor data, aerial digital imagery does provide important information for bothwithin-season and after-season management in precision agriculture.

Data from yield monitors have also been investigated as a means of generating

management zones. Spatial and temporal yield patterns can be variable and incon-sistent between growing seasons (Colvin et al., 1997; Stafford et a1., 1998). Althoughyield monitor data alone might be unsuitable for the delineation of managementzones, they are a valuable source of ancillary information, especially when compiledover several growing seasons (Stafford et al., 1998). Generaily, the techniques fordelineating management zones involves the use of multiple sources of data, includ-ing yield monitor data, soil properties, remotely sensed imagery, and topography(Yang et al., 1998; Fleming et aL,2004; Hornung et al., 2006; Khosla et al., 2008;Franzen et a1., 20LI).

4.3.3 Srrr-Sprcrnc AppLrcATroN MAps

Although identifying spatial variability of soil and crop growth with flelds is the firstirnportant step toward site-specific management, using that variability to formulatevariable rate application plans of farming inputs is another essential step in precisionagriculture. The major types of crop production inputs for variable rate applicationinclude fertilizers, limestone, pesticides, and seeds. There are two basic methodsfor implementing variable rate application: map-based and sensor-based. Map-basedvariable rate application systems adjust the application rate of a product based on


the information contained in a digital map for each cell or each managemerrt zone,

whereas sensor-based systems use data from real-time sensors to directly control

variable rate operations. Both methods require the application rate to be determined

based on the site-specific conditions, although only map-based variable rate applica-

tion methods require a site-specific application map. Although traditional guidelines

for uniform applications of these inputs can be used to determine the appropriate

input rate, a decision support system implemented in a GIS environment is gener-

ally necessary to integrate measured soil and crop variables with the knowledge ofexperts and farmers.

Site-specific recommendations for fertilizers are mainly based on the same fertil-izer guides developed for whole-field management. These guides were developed

by combining results from a number of fertilizer response studies over a wide range

of physiographic areas and soil types into simplifled recommendation equations

(Kitchen et al., 1995). Most recommendation equations require that yield expecta-

tions and existing soil nutrient levels be determined. Yield maps made from yieldmonitor data over several years can be used to determine yield potential maps forvariable rate applications. However, because crop yields are greatly affected by fac-

tors such as annual weather variations, crop rotations, and infestations of weeds,

insects, and diseases, it may be difficult to obtain reliable yield potential maps. For

this reason, many research studies of variable rate applications use a uniform yieldgoal (Ferguson et al., L996). Attempts have been made to make variable rate fertil-izer application using expert systems, GIS and crop models, and fertilizer guides (He

and Peterson,lggl: Havlin and Heiniger, 2009). Variable-rate recommendations forlime are based on soil pH level and agronomic guides (Bongiovanni and Lowenberg-Deboer, 2000; Johnson and Richard, 2010).

Managing crop pests (diseases, insects, and weeds) requires the use of pesticides

(fungicides, insecticides, and herbicides). Some crop pests occur randomly withina field, whereas others tend to occur in similar patterns spatially and temporally.Certain diseases such as cotton root rot tend to occur in the same areas of the field inrecurring years. Some weeds often occur in aggregated patches of varying size or instripes along the direction of cultivation. If pest occurrences are consistent in densityand locations over years, maps from previous years can be used to regulate pest con-trol methods in subsequent years. Otherwise, sensor-based variable rate applicationis more effective for real-time control of the pest. Remote sensing can be an effectivetool for detecting and mapping crop pests within and across seasons for variable ratepesticide application (Yang et a1., 2010b).

Variable rate seeding has the potential for reducing seed costs for fields with largevariations in topographic attributes and yield potential. Optirnum crop variety and

plant density vary with variation in yield potential across different landscape posi-tions (Shanahan et a1.,2004). Variable rate technology can also be used for variablerate water application through center-pivot irrigation systems.

4.4 VARIABLE RATE APPLICATION

Variable rate technology enables farmers to improve crop production efficiency and

reduce environmental impacts by adjusting rates of fertilizer, pesticide, and seed to

B1


specific conditions within discrete areas of a Iield. Efforts devoted to identifyingspatial variability and developing variable rate maps will have limited value unlessequipment can automatically regulate application rates as it travels across a field.Variable rate application equipment is available for a variety of substances includinggranular and liquid fertilizers, pesticides, seeds, and irrigation water. Several com-panies are marketing variable rate application equipment.

The major components of a typical map-based variable rate control system includean in-cab computer (or controller) Ioaded with application software and variable rateapplication maps, a DGPS receiver that provides vehicle position information to thecomputer, and an actuator that regulates material rates under direction of the com-puter. When the equipment is operating in the field, the computer receives positioninformation, matches the required application rate as a function of vehicle position,and then sends a set-point signal to the controller that adjusts the application to thedesired rate. A variable rate system may also record actual application rates alongwith GPS position. This information serves as a record of what was applied to thefield and allows for review of application for future recommendation considerations.

Among the earlier studies, Schueller and Wang (1994) described the conceptof variable rate fertilizer and pesticide application, described commercial applica-tors using GPS, and suggested more research to understand various error sources.Al-Gaadi and Ayers (1999) developed a site-specific variable rate herbicide applica-tion system, and reported that the system produced an actual application rate mapand that its highest application rate error was ZVo- Carrara et al. (2004) implementeda variable rate herbicide application system for durum wheat and reported a savingsof 29Vo of herbicides compared to the conventional spray operation. Tumbo et al.(2007) evaluated a commercial variable rate controller for properly applying aldi-carb outside the buffer zones around water wells in citrus groves and deterrnined thedynamic performance of the system using two common drive mechanisms.

Variable rate application will not affect the basic functions of most applicationmachinery. The required changes will be necessary to accommodate the addi-tion of sensors and controllers. Most of the current liquid systems can be modifiedfor variable rate application. Yang (2001) adapted a FALCON variable rate con-trol system (Ag-Chem, AGCO Corp., Duluth, GA) to an existing liquid fertilizerapplicator to vary rates of nitrogen and phosphorus simultaneously. Ess and Morgan(2010) described ACCU-RATE (Rawson Control Systems, Inc., Oelwein, IA) andSOILECTION (Ag-Chem, AGCO Corp.) as examples of commercial variable rateapplication systems. The ACCU-RATE is a variable rate seed metering drive system,which can be added to planters to implement variable rate seeding. It can vary theapplication rate of seeds and fertilizers. The SOILECTION system is used for vari-able rate dry and liquid fertilizer application. It can variably apply up to l0 differentproducts in a single pass and can record actual application rates for future use.

Sensor-based variable rate application can be used to apply fertilizers as a sidedressing or foliar spl'ay during the growing season. Biermacher et al. (2009) devel-oped a site-specific nitrogen fertilizer application system that uses optical reflec-tance measurements of glowing wheat plants to estimate N requirements. Anotherexample of sensor-based liquid feltilizer application is the GreenSeeker selectivespraying system (Trimble Navigation Limited). The system uses an optical sensor to


measure red and NIR light reflectance from a crop canopy. The canopy reflectance

can be related to crop status and the need for nitrogen fertilizer. The sensing system

is integrated with a spraying system that can produce a variable rate application of

liquid nitrogen fertilizer. Although sensor-based variable rate systems do not require

a positioning system, the data from the sensors can be used for creating variable rate

maps for other map-based field operations if a GPS receiver is equipped.

4.5 ECONOMIC AND ENVIRONMENTAL BENEFITS

Variable rate application of production inputs has the potential to increase farm prof-

its and reduce adverse envilonmental impacts. Economic and environmental beneflts

of variable rate chemical application are important for both producers and the general

public and will affect the pace of adoption of precision agriculture. Numerous short-

term studies of economic and environmental impacts of variable rate application

have been reported, but more research is needed to document its long-term impacts.

This information will facilitate the adoption of precision agriculture technologies.

4.5.1 Pnorlranlurv

For a new practice to be widely adopted in production agriculture, the practice must

yield an economic profit except for regulatory requirements. Variable rate applica-

tion requires additional costs associated with soil sampling, data analysis, and new

equipment. Economic beneflts from variable rate application can only be derived

from increased yieids and/or savings in reduced inputs or a combination of both. Ifthe initial investment for equipment is high, actual economic returns of variable rate

application will be low ol even negative for the flrst few years. Some costs associated

with soil sampling for a field can be accurately determined, whereas other costs for

new equipment and data analysis are difficult to estimate for each field. Nevertheless,

it is certain that these costs will go down if the same equipment and data analysis

software are used for more fields over more years.

Maly studies have shown that variable rate applications of fertilizers are supe-

rior to uniform rate application in terms of economic benefits, but evidence of prof-

itabitity has been mixed based on nine published fleld research studies (Swinton

and Lowenberg-DeBoer, 1998). Numerous recent studies present more evidence

of profitability for variable rate application of fertilizers, pesticides, and seeds.

Mohammadzamani et al. (2009) conducted a study to develop aprecision application

of a preemergence herbicide and reported a savings of l3%o compared to a uniformherbicide application. Robertson et al. (201I) described the extent of variable rate

adoption in Australia using a survey, and reported that the adoption rate increased to

20Vo of grain growers from less than 5Vo six years earlier, and this figure is expected

to increase. However, they also described technical issues with equiptnent, software

access, and the incompatibility of equipment remains as constraints for adoption.

Bullock et al. (1998) conducted a study to estimate the economic value of variable

rate seeding (VRS) with data consisting of more than 42,000 individual experimen-

tal units from 170 fields in the Midwestern U.S. Corn Belt, and reported that VRS

will not be economically feasible unless detailed information for field characteristics,

B3

B4 Agricu ltu ral Automation

production inputs, and stochastic factors are properly obtained. Lowenberg-Deboer(1998) also reported that VRS is potentially profitable only when farmers have somelow yield potential land (<100 bu/ac), and when the proportion of low yield land issmall. Bullock and Lowenberg-DeBoer (2407) reviewed different studies regardingthe economics of variable rate technology, showed that mixed results were reportedfor the profltability of the technology, and suggested more sufficient use of spatialanalysis, longer-term data, and the need for ex ante analysis to yield consistent results.

4.5.2 EttvtRoNrrrENTAL BrNrrrrs

Precision agticulture technologies are known to be environmentally friendly, sinceonly the needed amount of agricultural chemicals (pesticides, fertilizers, etc.) areapplied to areas as needed (site-specific treatment). Thus, the technologies helpmaintain the quality of the environment and the sustainability. One of the goals ofprecision agriculture is to optirnize crop inputs so that the environmental impactcould be minimized.By optimizing the input, growers can increase the efficiency ofchemical applications, reduce time and labor, and thereby increase profit. Variablerate technology or precision application of agricultural chernicals can help increaseenvironmental benefits as well as reduce waste.

The National Research Council (1997) reported that the environmental benefits ofprecision agriculture may not be a primary reason for the technology adoption, sincegrowers are depending on economic savings or more proflt. Precise calibration of inputmaterials will increase plant uptake of those inputs, and precise fertihzq applicationwill not leave any excess amount in the soil, thus reducing runoff and contaminationto the groundwater. The council recommended that synergy between biotechnologyand valiable rate technology could increase environmental benefits. It described poten-tial environmental problems, including adverse effects from variable rate applicationto steeper slopes, more fertiiization to potentially incorrectly interpreted high yieldingareas, and improper use of environmental data by the regulating agencies. The councilfurther stated that "the committee found no credible research that contains consistentevidence of environmental benefits from precision agriculture."

However, Thrikawala et al. (1999) reported that groundwater quality was bet-ter when variable rate technology was used than when a constant rate was applied.Hatfleld (2000) described that nonpoint-source pollution issues shouid be consid-ered. Bongiovanni and Lowenberg-Deboer (2004) reviewed studies regarding sus-tainability and environmental benefits with precision agriculture and reported thatprecision agriculture contributes to long-term sustainability by reducing excessiveuse of fertilizers and pesticides. Aeurnhammer (2001) argued that precision farm-ing would become more important when environmental beneflts and increased flowof information are achieved. Harmel et al. (2004) found that their study results didnot show any evidence that variable rate nutrient application improved water qual-ity, but indicated that in the foilowillg year the reduced rates of nitrogen applicationmay potentially improve water quality. They suggested that an intensive field-scaleresearch would be necessary. McBratney et al. (2005) mentioned that it would be dif-flcult to estimate the economic value of environmental damage from agriculture, andthey also described the proo-lems and challenges in auditing by regulating agencies.

Precision Agricu Itu ral Systems

Clearly, there is a need for more comprehensive field-scale research to investigate the

environmental impacts of variable rate technology.

4.6 SUMMARY

Precision agriculture as a new farming strategy is gradually changing the way farm-ers manage their fields. Some technologies developed for precision agriculture have

become standard practice in agriculture. Yield monitoring is a good example and is

probably the single most widely used precision agriculture technology. Yield mon-

itors have become standard or optional equipment in new harvesters from manymanufacturers. Farmers use yield monitor data not only for variable rate applica-tion, but also for evaluating different management practices and identifying prob-

lem areas. Although real-time soil sensors have evolved drastically in the past twodecades, continuing breakthroughs are the key to cheaper and more reliable soildata collection. Because of the time and cost involved in ground-based soil and crop

sensors, airborne and high-resolution satellite imaging systems have gained interest

for mapping crop growth variability. Remote sensing imagery obtained during thegrowing season has potential not only for after-season management, but also forwithin-season management.

Developing accurate site-specific recommendation input maps remains a great

challenge. Variable rate fertilizer recommendations for each cell or managementzone are based on the same guide developed for whole field management. Moreresearch is needed to develop improved site-specific recommendation algorithms.Variable rate technology has been developed mainly for seed, fertiltzer, and pesticide applications. This technology is also being developed for valiable-depth plant-ing, tillage, manure, and water application. Variable rate application can be bothprofitable and environmentally beneficial for fields with large variability in cropyield and soil nutrients and for fields with patched weeds and isolated disease infec-tions. However, variable rate application may not be suitable or necessary for fieldswith little soil and crop growth variability.

Precision agriculture continues to evolve, and it has reached a level that allowsa farmer to measure, analyze, and manage within-field variability that could notbe automatically managed before. Although precision agriculture involves a great

deal of technologies and requires additional investments of money and time, it can

be practiced at various levels depending on the resources and technology servicesavailable to the farmer. If practiced properly, precision agriculture can increase farmprofitability and minimize adverse environmental impacts, thus improving the long-term sustainability of production agriculture.

DISCLAIMER

Mention of trade names or commercial products in this article is solely for the pur-poses of providing specific information and does not irnply recommendation orendorsement by the U.S. Department of Agriculture. USDA is an equal opportunityprovider and employer.

B5

precision agricultural system

Documents

remote sensors

precision farming pr

pi of precision agricult

crop sensors

soil sensors

variable rate application

variable rate controllers

formation systems gis