Agricultural Engineering · Farm Machinery & Power · also Precision Farming, Agricultural Systems Management
Using positioning, sensing and variable-rate control to manage fields by the metre instead of by the hectare.
In the exams: ICAR AICE subject 43 Unit 8; elective in the ICAR 6th Deans' Committee B.Tech
Precision agriculture is the management of fields and animals according to the variation measured within them, rather than by a single average for the whole field. A wheat field of 40 hectares is not uniform: its soils, water supply, weed pressure and yield potential change over a few tens of metres. Conventional practice applies one seed rate, one fertiliser rate and one spray rate to all of it. Precision agriculture measures the variation, decides whether it is large and stable enough to act on, and then uses machines that can change what they do as they move across the field. The International Society of Precision Agriculture describes it as a management strategy that gathers, processes and analyses temporal, spatial and individual data and combines it with other information to support decisions according to estimated variability, for better resource-use efficiency, productivity, quality, profitability and sustainability.
The engineering core is a chain of technologies: satellite positioning (GNSS) to know where the machine is to a few centimetres; guidance and auto-steer to drive repeatable lines; sensors on the combine, the tractor, drones and satellites to measure yield, soil and crop; prescription maps and management zones that turn measurements into rates; variable-rate and section-control actuators that apply those rates; the ISOBUS standard (ISO 11783) that lets a tractor, a terminal and an implement from different makers talk to each other; and the data systems and farm management information systems that store, clean and use all of it. Each link has its own errors, and a good engineer knows where each error comes from and how big it is.
The subject is practised very differently around the world. On large mechanised farms in North America, Australia, Brazil, Argentina and Europe, RTK auto-steer, yield maps and section control are routine. On the small, fragmented holdings of South Asia and sub-Saharan Africa, the same ideas reach farmers through custom-hiring centres, service providers, smartphone decision tools, handheld crop sensors and drones, and through satellite imagery that is free at the field scale. This manual is written for both settings and says where practice or regulation differs.
In examinations, precision agriculture appears as Unit 8 of the ICAR AICE-JRF/SRF syllabus for Irrigation and Drainage Engineering (subject 43) and as an elective in the ICAR Sixth Deans' Committee B.Tech curriculum; it also appears inside farm-machinery papers, the machinery-systems part of the NCEES PE Agricultural and Biological Engineering exam, and in technician and chartered-engineer assessments in the UK. Expect recall of definitions and standards, short calculations (positioning error, yield from mass flow, sprayer rates, vegetation indices, break-even area), and case questions on choosing a system for a given farm.
Every field varies in space and in time. Spatial variability arises from soil-forming factors (parent material, texture, depth, slope position), from past management (old field boundaries, former manure heaps, lime applied unevenly, drainage lines, compaction on headlands), and from biological factors (weed patches, disease foci, pest hot spots). Temporal variability arises from weather: the same patch can be the best part of the field in a dry year, because it holds water, and the worst in a wet year, because it waterlogs.
A widely used classification (Zhang and co-workers, 2002) groups the sources into six kinds: yield variability (the historical and present yield distribution); field variability (elevation, slope, aspect, terrace and field boundaries); soil variability (fertility, texture, depth, moisture, electrical conductivity, organic matter, pH); crop variability (density, height, nutrient and water stress, leaf chlorophyll); variability in anomalous factors (weeds, insects, disease, wind and hail damage); and management variability (tillage, seeding rate, rotation, fertiliser and pesticide application, irrigation).
Not all variation deserves management. Three questions decide it. First, is the variation large compared with the error of measuring it? Second, is it spatially structured, so that neighbouring points are similar and a machine can respond to it, or is it random noise at a scale smaller than the machine? Third, is it stable enough over time, or predictable enough within the season, to act on? If yield differences are random from year to year, a fixed zone map adds little; in that case in-season sensing is the better tool.
Precision agriculture is usually taught as a cycle of four or five steps. Data collection: position-tagged measurements of yield, soil, crop and machine operations. Analysis and interpretation: cleaning the data, mapping it, looking for stable patterns and relating them to causes. Decision: choosing whether to treat the field uniformly or variably, and building a prescription map or a sensor algorithm. Implementation: variable-rate application, section control, guidance and controlled traffic. Evaluation: checking the result, ideally against strips or blocks left at the old uniform rate, and feeding the lesson into the next season.
Two strategies run through the cycle. Map-based management makes decisions before the operation from stored layers (soil samples, yield history, imagery) and loads a prescription map into the controller. Sensor-based management measures the crop or soil on the go and changes the rate within seconds, without a map.
The cycle rests on one enabling technology, positioning, and one enabling discipline, experimental design. Without accurate position, layers from different years and machines cannot be overlaid. Without unbiased on-farm trials, the farmer cannot tell whether the variable-rate prescription outperformed the uniform one or just coincided with a good season.
Adoption follows farm structure and the cost of labour. Guidance and auto-steer spread fastest because the benefit is immediate and easy to see: less overlap, less fatigue, longer working hours. Variable-rate fertiliser and seeding are adopted more slowly because their return depends on the field, the crop price and the quality of the prescription.
On smallholdings the unit of management is already small: a 0.4 ha plot is managed by one family who knows its good and poor corners. Here precision agriculture works through lower-cost routes: leaf colour charts and handheld chlorophyll or canopy sensors for nitrogen timing, soil health cards and grid sampling organised at village scale, satellite-based advisories delivered by phone, laser land levelling, and drones for spraying where aviation rules allow them. Drone spraying rules differ widely: some countries have specific approval schemes for agricultural drones and the pesticide products and rates used from them, others still restrict it; the operator must check the national aviation and pesticide regulators before work.
For the engineer, the practical lesson is to scale the tool to the decision. A 20 cm soil-sampling grid is useless on a farm that can only apply one fertiliser rate per plot; RTK is wasted on a broadcast spreader with a 1 m transition zone. Good design matches positioning accuracy, sensor resolution, actuator response and management unit to one another.
A global navigation satellite system (GNSS) is a constellation of satellites that broadcast timing signals. Four global systems operate: GPS (United States), GLONASS (Russia), Galileo (European Union) and BeiDou (China). Regional systems add coverage: QZSS over Japan and the Asia-Oceania region, and NavIC (IRNSS) over India. Agricultural receivers now usually track several constellations, which gives more satellites in view, better geometry and faster recovery under trees and near buildings.
A receiver measures the time a signal takes to arrive and multiplies by the speed of light to get a pseudorange, called 'pseudo' because the receiver's clock is not synchronised with the satellites' atomic clocks. With the receiver clock offset as a fourth unknown, a three-dimensional fix needs at least four satellites. Timing errors translate directly into distance: one microsecond is about 300 m, and 10 nanoseconds is about 3 m.
Two kinds of measurement are used. The code measurement uses the pseudo-random code; for the GPS C/A code the chip rate is 1.023 MHz, so one chip is about 293 m long, and code-based positions are good to a few metres. The carrier-phase measurement tracks the phase of the carrier wave itself; the GPS L1 carrier is 1575.42 MHz, a wavelength of about 19 cm, and phase can be measured to a small fraction of a cycle, giving millimetre-level precision once the whole number of cycles between satellite and receiver, the integer ambiguity, is resolved.
The main error sources are satellite clock and orbit (ephemeris) errors; ionospheric delay, the largest and most variable, which depends on the signal frequency and so can be largely removed by dual-frequency receivers; tropospheric delay, caused by the lower atmosphere and modelled from temperature, pressure and humidity; multipath, where signals reflect from buildings, grain bins, trees or the machine itself before reaching the antenna; and receiver noise.
The combined ranging error is called the user equivalent range error (UERE). Satellite geometry multiplies it. When satellites are clustered in one part of the sky the position is poorly determined; when they are spread out it is well determined. The multiplier is the dilution of precision (DOP): position error is approximately UERE multiplied by DOP. HDOP applies to horizontal position, VDOP to vertical, PDOP to three-dimensional position and GDOP adds time. Vertical accuracy is usually worse than horizontal because all satellites are above the horizon. Worked example: with a UERE of 4 m and an HDOP of 1.5, the expected horizontal error is about 1.5 × 4 = 6 m. Many operators set a mask of a maximum PDOP and a minimum elevation angle (often 10 to 15 degrees) to reject poor geometry and low, noisy satellites.
Stand-alone single-frequency positioning gives errors of a few metres, enough for field navigation and scouting but not for guidance. Corrections improve it in several ways.
Differential GNSS (DGNSS) uses a reference receiver at a known point to compute range corrections and transmit them; code-based differential gives sub-metre accuracy close to the station. Satellite-based augmentation systems (SBAS) broadcast wide-area corrections from geostationary satellites: WAAS (North America), EGNOS (Europe), MSAS (Japan), GAGAN (India), SDCM (Russia) and BDSBAS (China), with others being developed. SBAS gives roughly sub-metre to about one-metre accuracy, suitable for spreading and spraying with wide booms.
Precise point positioning (PPP) uses precise satellite orbit and clock products, delivered by satellite or internet, with a single receiver. It reaches decimetre to a few centimetres, but needs a convergence time, commonly tens of minutes, after start-up or loss of signal.
Real-time kinematic (RTK) positioning uses carrier-phase measurements from a base station at a known point and the rover on the machine. The base sends its observations by UHF radio or by internet over the cellular network, commonly using the NTRIP protocol. Once the rover resolves the integer ambiguities (a 'fixed' solution) horizontal accuracy is about 1 to 2.5 cm. Accuracy degrades with distance from the base; receiver specifications are often written as a constant plus a parts-per-million term, for example 8 mm + 1 ppm, which at a 15 km baseline gives 8 + 15 = 23 mm. Network RTK uses a network of permanent stations to model errors across a region and supplies a virtual reference station near the user, so one subscription covers a whole district. Where a 'float' solution appears the ambiguities are not fixed and accuracy falls to decimetres.
Accuracy for guidance is reported in two ways. Pass-to-pass accuracy is the error between adjacent passes made within about 15 minutes; it matters for avoiding overlap and gaps in one operation. Year-to-year (repeatable) accuracy is the ability to return to the same line months later; it matters for controlled traffic, strip tillage and planting beside last year's rows. SBAS and DGNSS can give good pass-to-pass figures but drift over hours, so their repeatability is poor. RTK from a fixed base at a surveyed point is repeatable.
Coordinates also need care. GNSS works in a global earth-centred frame (WGS 84 for GPS), which is then projected for mapping, commonly to UTM. National datums are fixed to the local tectonic plate, which moves by a few centimetres per year relative to the global frame, so a line recorded years ago in one frame can appear shifted in another. Mixing datums, or moving a base station without re-surveying it, is a common cause of lines that 'do not line up'.
Terrain matters too. The antenna sits on the cab roof, often about 3 m above the ground. When the tractor rolls on a side slope, the antenna moves sideways relative to the wheels by about h × sin θ: at 3 m height and 5 degrees of roll the error is about 0.26 m. Terrain compensation uses an inertial measurement unit (gyroscopes and accelerometers) to measure roll, pitch and yaw and to correct the position to ground level.
Guidance systems help the operator drive parallel passes without markers or foam. They come in three levels. A lightbar or screen guidance system shows the cross-track error, the sideways distance from the intended line, and the operator steers. Assisted steering fits an electric motor to the steering wheel or column and turns it under the control of the guidance computer; it is easy to move between machines. Integrated (hydraulic) auto-steer places a proportional valve in the steering hydraulic circuit and reads a wheel-angle sensor; it gives the tightest control and is usually factory-fitted or dealer-fitted.
An auto-steer system has a GNSS receiver and antenna, a terrain-compensation module, a steering controller, a steering actuator and a display. The controller computes the cross-track error and heading error with respect to the guidance line and commands a steering angle to reduce both, much like a feedback controller that must be tuned: too aggressive and the tractor weaves, too gentle and it wanders and takes long to come back onto line after a turn.
Guidance patterns include straight A-B lines (defined by two points), A-plus-heading lines, curved or contour lines that follow a recorded edge, circle (pivot) patterns for centre-pivot fields, and adaptive curves that follow the last pass. Automated headland turning now executes the turn to the next line, and some systems also control implement functions in sequence at the headland.
Without guidance, an operator overlaps each pass to avoid missing strips. The overlap is wasted: seed, fertiliser and pesticide are applied twice, time and fuel are spent covering ground twice. With guidance the overlap can be reduced to a few centimetres.
Worked example. A 12 m sprayer is driven with an average 0.6 m overlap by eye. Effective width = 12 − 0.6 = 11.4 m, so the fraction of the field sprayed twice is 0.6 / 11.4 = 5.3 percent. With auto-steer and a 0.05 m overlap, the fraction falls to 0.05 / 11.95 = 0.4 percent. The difference, about 4.8 percent of chemical and time (5.26 − 0.42), is the saving from guidance alone for that operation.
Guidance also lets operators work at night, in dust and in the dark, and at a steadier speed. Because field capacity depends on effective width and field efficiency, the gain shows up as more hectares per hour and fewer operator hours per season. On wide implements a small percentage saving is worth a lot because input cost per hectare is high.
Steering the tractor precisely does not guarantee that the implement follows. On slopes and in curves, a trailed implement drifts downhill or cuts inside the tractor's path. Passive implement guidance moves the tractor off-line to pull the implement onto the line; active implement guidance places a second receiver on the implement and steers it directly, by a side-shift hitch, steerable coulters or disc steering, or steered axles.
Controlled traffic farming (CTF) uses repeatable guidance to confine all wheels to permanent traffic lanes year after year. The machine working widths are made multiples of one module (for example 3, 9 and 12 m) and wheel track widths are matched. The cropped beds between lanes are never driven on, so compaction is concentrated where it does the least harm, and the lanes give firm running for timely operations. The trafficked fraction depends on track width, tyre width and the module: with two wheel tracks each 0.5 m wide in a 9 m module, lanes occupy 1.0 / 9 = 11 percent of the field. In random traffic, by contrast, repeated operations spread wheel marks across a large share of the field each year.
Guidance lines should be designed before the season, laid out to minimise short runs and point rows, and stored with a clear name, datum and base station so they can be reused.
A combine yield monitor estimates grain yield every one or two seconds and tags it with position. Its main parts are: a mass-flow sensor, usually an impact (force) plate at the head of the clean-grain elevator that the grain strikes as it is thrown from the paddles, with other designs using optical volume measurement or radiometric attenuation; a grain moisture sensor, usually a capacitance cell through which a sample of grain flows; a ground-speed sensor (wheel, radar or GNSS); a header-position switch that stops recording when the header is raised; the cutting width, entered by the operator or reduced automatically by section data; and a GNSS receiver. A monitor also needs a temperature input for the moisture sensor and, for impact sensors, an elevator-speed sensor.
Yield per unit area is mass flow divided by the area harvested per second: yield (kg per m²) = mass flow (kg/s) ÷ [speed (m/s) × cutting width (m)]. Multiplying kg per m² by 10 gives tonnes per hectare.
Worked example. Wet grain flows at 10 kg/s at 5 km/h (1.389 m/s) with a 9 m header. Area rate = 1.389 × 9 = 12.5 m² per second. Wet yield = 10 / 12.5 = 0.8 kg/m², that is 8.0 t/ha. The grain is at 20 percent moisture (wet basis). Corrected to a standard moisture of 14 percent, dry-matter conservation gives 8.0 × (100 − 20) / (100 − 14) = 7.44 t/ha. The standard moisture for reporting differs by crop, by country and by buyer, so the monitor's setting must match local practice.
Impact sensors are not absolute; they must be calibrated against weighed loads. Good practice is to harvest several loads at different flow rates (by changing speed), weigh each on a certified scale or weigh wagon, and enter the weights so the monitor fits its calibration curve. The moisture sensor is checked against an oven or reference meter. Calibration is repeated for each crop and when conditions change markedly.
The main errors in raw yield data are: time lag, because grain takes some seconds to travel from the cutter bar through threshing and cleaning to the sensor, so yields must be shifted back along the path by the measured delay; fill-up and empty-out at the start and end of each pass, when the flow ramps up and down; narrow finishing passes and point rows where the actual cut width is less than the entered width, which makes yield look falsely high if the true width is smaller; sudden speed changes, when the smoothing of the grain flow no longer matches speed; GNSS errors and loss of fix; sensor drift and wear; and the mixing of grain within the combine, which smooths yield along the direction of travel so that the map cannot resolve features shorter than a few metres.
Worked example on lag. If the grain delay is 12 s and the combine travels at 1.5 m/s, a yield reading must be moved 18 m back along the pass to sit over the crop that produced it.
Yield data cleaning removes points recorded with the header up, outlier speeds (very slow or very fast), start and end-of-pass ramps, overlapping points, points with the wrong width, and statistical outliers, commonly flagged by distance from the field mean in standard deviations or by comparison with neighbours. Published protocols and free tools exist; the principle is to document each filter so results are reproducible.
Cleaned points are interpolated to a grid by inverse distance weighting or kriging, or are averaged within grid cells or management zones. A single-year map shows that season's pattern; several years, each normalised to its own field mean (relative yield = point yield ÷ field mean), show stable high-yield, stable low-yield and unstable areas. That classification is a standard input for management zones.
Yield maps are used to calculate nutrient removal for replacement fertiliser programmes, to find drainage and compaction problems, to evaluate strip trials, to estimate profit by area, and to decide where not to farm: persistently unprofitable corners can be taken out of production for buffer strips or habitat. Yield monitoring also exists for other crops: cotton (optical flow sensors in the air ducts), sugarcane and root crops (load cells under conveyors), forage harvesters (throughput from feed-roller displacement or crop-flow sensors, with on-board near-infrared sensing of dry matter and constituents), and hand-harvested crops through position-tagged bin or crate weights.
Variable-rate technology (VRT) changes the application rate of an input as the machine moves. Two approaches exist. Map-based VRT reads a prescription map, knows its position from GNSS, looks up the target rate for that location and sends it to the rate controller. Sensor-based (real-time) VRT measures a property on the go, for example canopy reflectance, and converts it to a rate through an algorithm, with no prior map. Combined systems use a map for limits or zone-specific algorithms and a sensor for the in-season adjustment.
VRT is applied to seed (variable seeding rate and, in some crops, hybrid or variety), fertiliser (granular spreaders, liquid applicators, anhydrous ammonia), lime and gypsum, crop protection chemicals (spot spraying of weeds, patch spraying), plant growth regulators, irrigation (variable-rate centre pivots and zone-controlled drip), and even tillage depth.
A rate controller is a closed loop: the target rate and measured ground speed give a target flow; a flow meter, shaft encoder or other sensor measures the actual flow; and an actuator (control valve, hydraulic motor, electric drive, gate) is adjusted until actual equals target. Speed compensation is built in: if the tractor slows, the controller reduces flow so that the rate per hectare stays constant.
For a boom sprayer the required flow is q (L/min) = rate (L/ha) × speed (km/h) × width (m) ÷ 600. Worked example. At 150 L/ha, 8 km/h and a 24 m boom, total flow = 150 × 8 × 24 / 600 = 48 L/min; with nozzles at 0.5 m spacing, each nozzle must deliver 150 × 8 × 0.5 / 600 = 1.0 L/min.
Conventional rate controllers change flow by changing pressure. Nozzle flow is proportional to the square root of pressure, so to change flow by a factor k the pressure must change by k². Raising the rate from 100 to 150 L/ha at the same speed needs 1.5² = 2.25 times the pressure: from 2 bar to 4.5 bar. That changes droplet size and may take the nozzle out of its recommended range, so the practical turndown of pressure-based control is narrow. Wider ranges come from pulse-width modulation (PWM) of solenoid valves at each nozzle, which varies the fraction of time the nozzle is open at constant pressure; from twin-nozzle or variable-orifice bodies; and from direct injection of chemical concentrate into the carrier stream, which varies the product rate without changing the water rate, although with a transport delay from the injection point to the nozzles.
Granular spreaders vary rate by changing the metering gate opening or the belt or chain speed; spinner spreaders must also keep the spread pattern correct when the rate changes. Planters and drills vary seed rate by driving the metering units with hydraulic or electric motors whose speed is set by the controller.
Every VRT system has delays: the time to read the next rate, the controller response, the actuator travel and the product's travel from meter to ground. The distance travelled during the delay is speed × time. At 10 km/h (2.78 m/s), a 2 s delay moves the change 5.6 m past the zone boundary. Controllers therefore use a look-ahead: they request the new rate early by the known delay so that the change lands on the boundary.
The combination of delay and machine width sets the smallest management unit a machine can honour. A spreader with a 24 m throw width and a 10 m transition cannot treat a 5 m × 5 m cell; there is no point in writing a prescription at that resolution. Field and bench tests of a rate controller's dynamic response measure the time to reach a set percentage of a step change in rate.
VRT succeeds agronomically only when the prescription is right. A precise machine applying a poor prescription saves nothing. For this reason the engineering and the agronomy must be evaluated together, preferably with check strips at the uniform rate across the zones.
A prescription map is only as good as the data behind it. Soil fertility maps come from sampling. In grid sampling the field is divided into regular cells, commonly about 0.4 to 1 ha, and a composite sample is taken at a grid point or across each cell; the results are interpolated. Grid sampling finds patterns without prior knowledge but is expensive, and a coarse grid misses short-range variation. In zone sampling, samples are composited within management zones defined beforehand from yield, soil conductivity, elevation or imagery; it is cheaper and follows the causes of variation, but depends on the zones being right. Directed (targeted) sampling places samples where maps or sensors show contrasts.
Geostatistics gives the rule for grid spacing. The semivariogram plots half the mean squared difference between pairs of points against their separation distance (lag). Its nugget is the variance at zero distance, due to measurement error and variation shorter than the smallest lag; the sill is the plateau variance; and the range is the distance beyond which points are no longer correlated. A common rule of thumb is to sample at less than half the range, so that neighbouring samples are correlated and interpolation is meaningful. If the nugget is close to the sill, the property has little spatial structure at that scale and variable-rate management by map is not justified.
Inverse distance weighting (IDW) estimates a value at an unsampled point as a weighted average of nearby samples, with weights proportional to 1 / dᵖ, where d is distance and p is the power, commonly 2. Worked example. Two samples, 20 mg/kg at 10 m and 40 mg/kg at 20 m, give weights 1/100 and 1/400. Estimate = (20 × 0.01 + 40 × 0.0025) / (0.01 + 0.0025) = 0.3 / 0.0125 = 24 mg/kg. The closer sample dominates.
Kriging uses the fitted semivariogram to choose weights that give an unbiased estimate with minimum variance, and it also gives an estimation variance for each point, which IDW does not. Ordinary kriging is the usual choice for soil properties; block kriging estimates averages over cells, which suits machine-scale management; co-kriging and regression kriging use a densely measured secondary variable such as soil electrical conductivity to improve the map of a sparsely sampled property.
A management zone is a sub-region of a field with a relatively homogeneous combination of yield-limiting factors, for which a single rate of a given input is appropriate. Zones are delineated from stable layers: multi-year normalised yield, soil apparent electrical conductivity, elevation and terrain attributes (slope, wetness index), bare-soil imagery, and farmer knowledge.
The usual workflow is to standardise the layers, reduce redundancy (for example by principal component analysis), and group the cells by unsupervised clustering, most often k-means or fuzzy c-means. The number of zones is chosen by indices such as the fuzziness performance index and normalised classification entropy, which are minimised when the classes are most distinct, and by practical judgement: two to five zones is typical, because more cannot be managed or interpreted. Small fragments are then merged into neighbours so that no zone is narrower than the machine can treat. Zones should be checked against independent data: if yields in the zones do not differ, the zones are not useful.
A prescription assigns a rate to each zone or cell. Rates come from agronomic rules. For nutrients, the main approaches are sufficiency (apply only where soil test is below a critical level, at a rate from a response curve), build-and-maintain (raise soil test to a target over some years and then replace crop removal) and removal-based replacement using yield maps. Worked example. If a zone yielded 8 t/ha and the grain removes 3.5 kg P per tonne, replacement is 8 × 3.5 = 28 kg P/ha, which is 28 × 2.29 = 64 kg P₂O₅/ha (P₂O₅ contains 43.6 percent P). Removal coefficients vary by crop, variety and region and must be taken from local data.
For nitrogen, zone targets are often combined with in-season sensing. For seed, the rate per zone follows the yield potential: in many crops higher-potential zones receive more seed, though for some crops and environments the optimum runs the other way; local trials decide.
The prescription is exported in a format the controller reads: a shapefile of polygons or a grid, or the ISO 11783-10 task data file (ISO-XML) for ISOBUS task controllers. It must carry the product, the units (kg/ha, seeds/ha, L/ha), a default rate for areas outside the map or when GNSS is lost, and the coordinate reference. A last check against machine limits, such as minimum and maximum flow, prevents a controller from being asked for a rate it cannot deliver.
Healthy green leaves absorb strongly in the blue and red parts of the spectrum, where chlorophyll absorbs, with the red absorption centred near 670 nm, and reflect more in the green, which is why they look green. In the near-infrared (NIR), roughly 700 to 1300 nm, reflectance is high, often 40 to 50 percent or more, because light is scattered at the air-cell boundaries in the leaf's spongy mesophyll. The steep rise between red and NIR, around 690 to 740 nm, is the red edge; its position shifts to shorter wavelengths when chlorophyll falls. In the short-wave infrared (about 1300 to 2500 nm) reflectance is controlled by leaf water, with water absorption near 1450 and 1940 nm. Bare soil has a gently rising spectrum without the red-NIR contrast, and water absorbs strongly in the NIR.
Stress changes the signature: less chlorophyll raises red reflectance and moves the red edge; loss of leaf structure and canopy cover lowers NIR; water stress raises canopy temperature because transpiration cools the leaf less.
Indices combine bands to emphasise vegetation and suppress background. The normalised difference vegetation index is NDVI = (NIR − Red) / (NIR + Red). It ranges from −1 to +1; water is negative, bare soil is near 0.1 to 0.2 and dense green canopy is 0.8 or above. Worked example: NIR reflectance 0.45 and red 0.05 give NDVI = 0.40 / 0.50 = 0.80. NDVI saturates in dense canopies (roughly beyond a leaf area index of about 3), so it loses sensitivity just when nitrogen decisions are made in high-yielding crops.
The soil-adjusted vegetation index SAVI = (1 + L)(NIR − Red) / (NIR + Red + L), with L commonly 0.5, reduces the effect of soil brightness in sparse canopies; with the same reflectances, SAVI = 1.5 × 0.40 / 1.00 = 0.60. The normalised difference red-edge index NDRE replaces red with a red-edge band and saturates later, so it is preferred for mid- and late-season nitrogen. The green NDVI uses the green band. The enhanced vegetation index (EVI) adds a blue-band correction for atmosphere and a canopy background term.
In-season nitrogen algorithms compare the crop with a well-fertilised reference, an N-rich strip, to remove the effect of variety, growth stage and weather. The sufficiency index is the target-area index divided by the reference-strip index; when it falls below a threshold the crop is considered N-deficient and a rate is calculated. Thresholds and the equations that convert index to rate are crop- and region-specific and must come from local calibration.
Sensors are either passive, using sunlight, or active, carrying their own light source. Satellite, aircraft and most drone cameras are passive, so they need clear skies and radiometric calibration for changing illumination, commonly with a reflectance panel or an irradiance sensor. Tractor-mounted and handheld canopy sensors are usually active: they emit modulated light at red or red-edge and NIR wavelengths and read the reflected part, so they work in any light, even at night, and give readings independent of sun angle.
Four resolutions describe an imaging system: spatial (ground sampling distance, the size of one pixel on the ground), spectral (number and width of bands), temporal (revisit interval) and radiometric (bits per pixel). Free multispectral satellite imagery at 10 to 30 m resolution with revisits of a few days is widely used for field-scale crop monitoring; commercial constellations offer metre-scale imagery; synthetic aperture radar (SAR) sees through cloud, which matters in monsoon and tropical seasons, and responds to canopy structure and soil moisture.
For drone cameras, ground sampling distance (GSD) = pixel pitch × flying height ÷ focal length. Worked example: a 3 µm pixel, an 8.8 mm lens and 100 m height give GSD = 3×10⁻⁶ × 100 / 0.0088 = 0.034 m, about 3.4 cm. Halving the height halves the GSD but quadruples the number of images needed for a field. Photogrammetric mapping requires forward and side overlap between images, commonly 70 to 80 percent forward and 60 to 70 percent side, and ground control points or RTK/PPK positioning for accurate location.
Thermal sensing gives canopy temperature. The crop water stress index CWSI = [(Tc − Ta) − (Tc − Ta)lower] / [(Tc − Ta)upper − (Tc − Ta)lower], where the lower baseline is a well-watered, fully transpiring canopy and the upper is a non-transpiring canopy. CWSI runs from 0 (no stress) to 1 (maximum stress).
Proximal soil sensors work at or near the soil surface, often on the go. The most widely used measures apparent electrical conductivity (ECa), either by electromagnetic induction (EMI), which needs no contact and induces currents in the soil from a transmitter coil, or by galvanic contact, where coulters inject current and measure voltage in a Wenner-type electrode array. ECa rises with salinity, clay content, water content and temperature, so it maps soil texture and water-holding capacity in non-saline soils and salinity where salts dominate. Readings are standardised to 25 °C, and the effective depth depends on coil spacing and orientation or electrode spacing. ECa maps are a primary layer for management zones and a covariate for kriging sparse soil samples.
Other proximal sensors include gamma-ray spectrometers (passive, measuring natural potassium, thorium and uranium emissions linked to mineralogy and texture), visible and near-infrared (VNIR) spectrometers for organic carbon and clay (they require local calibration), on-the-go pH systems using ion-selective electrodes on soil cores, mechanical sensors of penetration resistance (cone index) for compaction, and capacitance or time-domain reflectometry probes for soil moisture. Crop proximal sensors include active canopy reflectance sensors, handheld leaf chlorophyll meters that measure leaf transmittance at red and NIR, ultrasonic and lidar sensors of plant height and biomass, and machine-vision cameras that detect weeds for spot spraying.
Section control switches parts of an implement off where the ground has already been treated or should not be treated, and on again where it has not. The implement is divided into sections: a sprayer boom into groups of nozzles (or, at the limit, individual nozzles), a planter into row units with row clutches or electric drives, a spreader into spread-width segments by changing the spinner or the drop point, and a seed drill into metering sections.
The controller keeps a coverage map, the as-applied area, in memory. As the machine moves, it projects the position of each section from the GNSS antenna using the machine geometry (antenna offsets, hitch point, boom distance) and checks whether that section is about to enter covered area, a headland already worked, an exclusion zone (waterway, buffer strip near a watercourse, field boundary) or an area outside the prescription. It then switches the section off, and back on when it leaves.
Because valves, clutches and product flow take time to respond, each section has a turn-on delay and a turn-off delay (sometimes called look-ahead times). At 12 km/h (3.33 m/s), a 0.8 s turn-off delay equals 2.7 m of travel; without look-ahead the section would keep applying for that distance into covered ground. The delays are measured in the field by driving over a marked line and observing where application starts and stops, then entering the offsets.
Overlap happens mostly at headlands, at point rows (where passes meet a boundary at an angle), around obstacles and in irregular fields. In a rectangular field with passes parallel to the long side and a single headland, overlap is small. In fields with angled boundaries, waterways and many short runs, it can be a substantial share of the area, and savings are highest there. Narrower sections save more, since each switch leaves less unwanted area: a 36 m boom with 6 sections of 6 m leaves up to 6 m-wide wedges of overlap at an angled edge, whereas individual-nozzle control reduces the wedge to the nozzle spacing.
Worked example on a wedge. A boom enters covered ground at an angle so that the overlap region per pass is a triangle 6 m wide at the base and 20 m long. Area = ½ × 6 × 20 = 60 m². Over 50 such entries the overlap is 3000 m², or 0.3 ha. At 250 currency units of input per hectare, the waste is 75 units per operation, before counting yield loss from double doses.
Section control on planters also reduces double planting in point rows and headlands, where excessive plant density lowers yield and wastes seed.
Section control depends on an accurate field boundary and headland definition. Boundaries are recorded by driving the edge with the implement's outer section, by importing them from earlier operations or from mapping sources, and they should be checked against the land registry or survey where they matter for compliance. Errors in the boundary or in the machine offsets cause either skips (untreated strips) or overlaps.
The coverage map also becomes the as-applied map: which product, what rate and where. Together with the prescription it proves what was done, a record required for pesticide use in many jurisdictions and increasingly by buyers and certification schemes. In ISOBUS systems section control runs through the task controller with the section-control functionality (TC-SEC), which lets any compliant terminal switch the sections of any compliant implement.
Before a common standard, each implement came with its own control box and cable, and a tractor cab could hold several screens that did not share data. ISOBUS is the common name for the international standard ISO 11783, 'Tractors and machinery for agriculture and forestry — Serial control and communications data network'. It lets a tractor, one terminal and implements from different manufacturers exchange control and data over one network, and it defines the file format for moving tasks and records between the farm office and the machine.
ISOBUS is built on the controller area network (CAN), the same serial bus used in cars and trucks, and borrows much from SAE J1939, the truck and bus standard. It uses the CAN 2.0B extended frame with a 29-bit identifier, at 250 kbit/s. Each frame carries up to 8 bytes of data; longer messages use a transport protocol that splits them into packets of 7 data bytes, up to 255 packets, or 1785 bytes, with an extended transport protocol for larger transfers such as uploading an implement's user interface.
The standard has many parts. The most asked are: Part 1, general standard for mobile data communication; Part 2, physical layer (cable, connectors, termination, including the implement bus breakaway connector at the hitch and the in-cab connector); Part 3, data link layer; Part 4, network layer; Part 5, network management, including address claiming, where each electronic control unit (ECU) claims a source address using its unique 64-bit NAME; Part 6, virtual terminal (VT); Part 7, implement messages application layer; Part 8, power-train messages; Part 9, tractor ECU (TECU), which is the gateway between the tractor's internal bus and the implement bus and provides speed, hitch position, PTO speed and other signals; Part 10, task controller and management information system data interchange; Part 11, mobile data element dictionary (the list of defined data items such as application rate, with their units and resolution); Part 12, diagnostics services; Part 13, file server; and Part 14, sequence control, for automated headland sequences.
The virtual terminal is the key idea for the operator. An implement ECU stores an 'object pool', a description of its screens, buttons and graphics, and uploads it to the tractor's VT, so the single terminal in the cab can display and operate any compliant implement. The task controller (Part 10) runs jobs: it receives a task containing products, prescriptions and field boundaries in an XML file (TASKDATA.XML, commonly called ISO-XML) from the farm management information system, controls the implement during work and records totals and position-based logged data, which it returns to the office.
Because the standard is large and optional in places, the industry body AEF (Agricultural Industry Electronics Foundation) defined functionalities that a product can be certified for and that buyers can compare. The main ones are: UT (universal terminal, the VT function); AUX (auxiliary control, such as a joystick controlling implement functions); TC-BAS (task controller basic, recording totals); TC-GEO (task controller geo-based, for prescriptions and position-logged data, hence variable rate); TC-SEC (task controller section control); TECU (tractor ECU); ISB (ISOBUS shortcut button, a common stop for implement functions); and TIM (tractor implement management, which lets an implement control certain tractor functions such as speed, hitch or hydraulic valves under defined safety rules). A tractor and an implement work together for a function only if both are certified for it, and certification is to a version, so compatibility must be checked in the AEF database or equivalent before buying.
Common faults are physical: missing termination, damaged connectors, low supply voltage; and mismatched software versions.
A precision farm generates several kinds of data. Spatial layers: field boundaries, soil sample points, soil ECa, elevation, imagery. Operation records: as-planted (variety, seed rate), as-applied (product, rate, timing), as-harvested (yield, moisture). Machine data: fuel use, engine load, idle time, speed, location tracks and fault codes from the CAN bus, often sent by telematics (cellular modems) to the cloud. Agronomic observations: scouting notes, photographs, lab analyses. Business data: costs, contracts, stock.
Each record needs metadata to be usable: what was measured, when, by which machine and sensor, in which units, in which coordinate reference system, and what filters or calibrations were applied. Loss of metadata is the most common reason that archives of yield and application data cannot be used years later.
Spatial data are stored as vectors (points, lines, polygons, commonly shapefiles or GeoJSON or GeoPackage) or rasters (grids such as GeoTIFF). Machine task data in ISOBUS systems use ISO 11783-10 ISO-XML, with binary files for logged time-series. Many terminals also write proprietary formats; open conversion frameworks and industry data standards exist to translate them, but conversions can lose attributes, so the original files should be archived.
Data quality has several dimensions: positional accuracy, attribute accuracy (was the monitor calibrated?), completeness (missing passes), consistency (the same product named the same way across seasons) and timeliness. Cleaning is part of the job, not an afterthought: an uncalibrated yield map or a prescription in the wrong units can cost more than having no data. A simple discipline helps: consistent naming of farms, fields, crops and products; one master field boundary per season; and a log of calibrations.
Farm data raise questions of ownership, control, privacy and value. Who may see a farm's yields and input use: the equipment maker, the agronomist, the input supplier, the lender, the government? Legal protection differs by country. Several regions have voluntary codes or principles, prepared by farm organisations and industry, that say the farmer should own or control the data generated on the farm, that collection and use need clear consent, that terms must be transparent, and that data should be portable. In some jurisdictions data about identifiable people are protected by general data-protection law, which covers farm records that identify the farmer.
The engineer's practical duties are to read the terms of service before connecting machines to a cloud platform; to know where data are stored and whether they can be exported and deleted; to aggregate or anonymise data before sharing for benchmarking or research; and to secure telematics, since connected machines are also cyber-security targets.
A farm management information system (FMIS) is a system for collecting, processing, storing and disseminating the data needed to carry out a farm's operations and management. A widely cited analysis (Sørensen and co-workers, 2010) lists its functions: field operations management (planning, recording and documenting operations), best-practice and agronomic decision support, finance (costs, budgets, profitability per field), inventory (inputs and produce), traceability (linking a lot of produce to the fields and treatments behind it), reporting (to regulators, buyers and certification schemes), site-specific management (handling maps and prescriptions), sales, machinery management (maintenance, costs, utilisation), human resource management and quality assurance.
Modern FMIS are commonly cloud-based with a web interface and phone apps, exchange tasks and records with machines by ISO-XML or proprietary links, pull weather and satellite imagery automatically, and connect to accounting and to buyers. Decision-support modules include crop growth and nitrogen models, disease-risk forecasts from weather, irrigation scheduling from soil-water balance, and machinery scheduling. The weakness of many FMIS is not software but input: records must be entered or imported reliably, or the outputs are wrong.
Precision tools change machine-management numbers. Effective field capacity is C = S × W × E / 10 in ha/h, with speed S in km/h, width W in m and field efficiency E as a decimal. Guidance raises the effective width (less overlap) and can raise E by allowing work at night and fewer stops. Telematics give actual hours, fuel and idle time per machine, so cost-per-hectare figures and fleet logistics (such as grain cart to combine and truck) can be managed from data rather than estimates.
Worked example. A 12 m drill works at 10 km/h with field efficiency 0.70: capacity = 10 × 12 × 0.70 / 10 = 8.4 ha/h. If auto-steer reduces overlap so the effective width rises from 11.4 to 11.95 m, the capacity at the same speed and efficiency becomes 10 × 11.95 × 0.70 / 10 = 8.37 ha/h, compared with 7.98 ha/h at 11.4 m, about 4.8 percent more.
Most precision investments are judged with a partial budget, which counts only what changes: added returns plus reduced costs, minus added costs and reduced returns. If the result is positive the change pays.
Ownership cost is annualised with the capital recovery factor CRF = i(1 + i)ⁿ / [(1 + i)ⁿ − 1], adjusted for salvage value where relevant. Worked example. An auto-steer system costs 12,000 currency units, lasts 5 years with negligible salvage, and money costs 8 percent. CRF = 0.08 × 1.08⁵ / (1.08⁵ − 1) = 0.2505, so the annual cost is 12,000 × 0.2505 = 3006 units. If guidance saves 12 units per hectare (reduced overlap on seed, fertiliser, chemicals, fuel and time), the break-even area is 3006 / 12 = 250 ha per year. A farm covering fewer hectares, or a machine shared by a contractor over more hectares, changes the answer; that is why custom hiring and contractors are the route to adoption on smaller farms.
Variable-rate returns are harder to predict because they depend on how much the optimal rate differs between zones, how much yield responds, and the price ratio of input to crop. Where the whole field responds alike, VRT gains little; where zones differ strongly in optimum, gains can be large. Environmental benefits are real but often not paid for in the farm budget.
A student uses this subject to understand how a modern farm machine knows where it is and what to apply. In practicals you will survey a field with a GNSS receiver, compute NDVI from a satellite or drone image, interpolate soil-test points into a map, and write a simple prescription. In exams, expect to compute positioning error from UERE and DOP, yield from mass flow, sprayer flow and pressure changes, and break-even area. Learn the ISO 11783 parts, the difference between SBAS, PPP and RTK, and the meaning of nugget, sill and range. Practise explaining why a precise machine cannot rescue a poor prescription.
Five years into practice you are setting up and supporting systems: choosing the correction service for a farm (SBAS for spreading, RTK for planting and controlled traffic), entering antenna and implement offsets, measuring turn-on and turn-off delays for section control, calibrating yield monitors with weighed loads, cleaning yield data, and building zone maps from yield, ECa and imagery. You check ISOBUS compatibility before a customer buys an implement, diagnose bus termination and address-claim faults, and export ISO-XML tasks. You also run strip trials so the farm can see whether variable-rate nitrogen or seed actually paid.
With 25 years of experience you design whole systems and decide investments. You judge whether a region's farms should build a network RTK service, how contractors and custom-hiring centres can spread the cost over enough hectares, and which data agreements protect farmers. You set company or agency policy on data ownership, interoperability and cyber-security, lead teams that write prescriptions for thousands of hectares, and advise on regulation of drone spraying and buffer-zone compliance. You translate between agronomists, engineers, finance and farmers, and you know from experience which promised benefits appear in practice and which do not.
A researcher with 25 years' experience works on the unsolved problems: how to make management-zone and sensor-based nitrogen algorithms transfer across soils and climates; how to design on-farm experiments and analyse them with spatial statistics so that site-specific response functions can be estimated honestly; how to fuse satellite, drone, proximal soil and machine data; how to quantify the uncertainty of prescriptions and the economic value of information; and how precision tools can serve smallholders without large machines. Such a researcher publishes, reviews standards, trains students and links engineering with agronomy, economics and policy.
This manual does not yet cover precision livestock farming, robotics and autonomous field machines beyond guidance, variable-rate irrigation hardware in detail (centre-pivot zone control and drip automation), machine-learning methods for image analysis, or country-by-country drone and data-protection law. It gives no crop-specific nitrogen-sensor algorithm coefficients, soil-test critical values or nutrient-removal coefficients, because these vary by crop, soil and region and must come from local calibration. Adoption statistics are deliberately left out because published figures vary widely by survey and year.
Original practice questions written to the syllabus and pattern of the exams named in each question's examTags. They are not past papers and do not reproduce any board's questions.
These notes are free to read here, with no account.
What the Laisha app adds:
Free for ten days, then a subscription. We would rather write that here than have you find it later.