Executive Summary

During the 2026 cotton harvest, two John Deere CP770 cotton pickers fitted with Shepherd thermal monitoring worked side by side with an identical, unmonitored machine across the same fields for a full six-week season. The two protected machines, #20 and #21, completed the season without incident. The unprotected machine, #19, experienced a fire that developed undetected inside the round module accumulator—the exact zone Shepherd continuously monitors on the protected machines—and was only discovered when the burning module was ejected from the rear of the machine.

The fire produced a definitive, unplanned test of the technology. As a Shepherd-equipped harvester passed the burning module in the field, its thermal cameras—mounted inside its own accumulator—detected the heat anomaly from approximately 20 metres away, through the perforated metal walls of the accumulator, and triggered the threshold alarm. It did so on two separate passes. A system that alarms on an external fire at that distance, through its own housing, leaves little doubt about its ability to detect a fire developing centimetres from its lenses.

Over six weeks of continuous, every-second monitoring, the accumulator zone—the configuration specified for fleet deployment—produced zero false alarms. When the system did alert, it was for a real fire: the drive-past detections above triggered the threshold alarm on both occasions. The season demonstrated both halves of what a fire detection system must prove: it alarms when there is a fire, and stays silent when there is not.

Machine #19 escaped with minimal damage—the crew ejected and contained the fire quickly. That outcome was earned by fast response and favourable luck, not by early warning. With a CP770 replacement cost of approximately $2 million and average harvester fire losses of $1–2 million per event, the same fire discovered a few minutes later is a different story—and a single avoided loss repays a fleet-wide Shepherd deployment many times over.

The Problem: Harvester Fires in Australian Cotton

Cotton harvesting concentrates three ingredients of fire: highly combustible material, high-friction machinery, and hot, dry field conditions. Round module cotton pickers compound the risk—dense modules of compressed cotton are formed and carried on the machine, where a smouldering ember can develop unseen inside the accumulator or module handling system.

• High-Risk Catalyst: The high flammability of cotton lint makes pickers and balers exceptionally vulnerable. {source: AgHealth Australia}
• Contributing Factors: Excessively dry seasonal weather and "low-micronaire" cotton (which produces fine, floating fuzz) significantly increase fire occurrences. {source: Cotton Grower}
• Damage Source: Over 90% of non-ginning seed cotton fires take place prior to ginning, originating in the harvester basket, field cleaners, or subsequent module stacks. {source: The National Cotton Council}

For a 12-machine CP770 fleet, this represents roughly $24 million in rolling assets exposed for the duration of every harvest, insured at a premium of approximately $360,000 per year. Premiums respond to claims history and demonstrated risk controls; the industry-wide spike in harvester fire claims puts upward pressure on both.

The System: Shepherd Thermal Monitoring

Shepherd is a purpose-built thermal monitoring system for harvest machinery, using industrial FLIR thermal imaging cameras hardened for harvest conditions. The trial ran a wider multi-camera configuration across several machine zones; based on trial findings, the fleet deployment configuration targets the round module accumulator—the priority fire zone—with cameras mounted on either side of the accumulator interior.

Beyond fire detection, continuous thermal monitoring provides condition visibility—bearing temperatures, belt friction and component stress trends—supporting maintenance planning and reducing unplanned mid-season downtime.

The Trial: Method

The trial was not designed as a controlled experiment—#19 was simply the next machine in the fleet, working alongside the monitored pair. But the resulting conditions were as close to a natural side-by-side comparison as field operations allow: same model, same fields, same season, same crop, differing only in the presence of thermal monitoring.

Results

Monitored machines: a clean season in the zone that matters

Harvesters #20 and #21 completed the full six-week season with no fire events. Operators had continuous live visibility of accumulator-zone temperatures throughout harvest via the in-cab display, and the system logged thermal history and trends across the season.

In the accumulator zone—the configuration specified for fleet deployment—the system produced zero false alarms across six weeks of every-second monitoring in dust, chaff and vibration-heavy harvest conditions. False alarms are the primary reason operators lose trust in, and ultimately disable, monitoring systems; a season of nuisance-free operation in the priority zone means that when the alarm sounds, it will be believed and acted on.

The trial also monitored additional machine zones beyond the accumulator, and those zones did generate false alarms during the season—driven by very high ambient heat and alert thresholds initially set too low, with occasional camera dropouts in the harshest positions. These are calibration findings, and they did their job: they informed both threshold settings and the decision to focus the fleet configuration on the accumulator, where detection value is highest and the monitoring environment is proven. A trial that surfaces no calibration learnings hasn't tested anything.

The unmonitored machine: fire through the accumulator

During the same season, harvester #19—working the same fields without Shepherd fitted—experienced a fire that developed inside the accumulator and travelled with the module. The fire went undetected while inside the machine; it was discovered only when the module was ejected from the rear of the harvester already alight. 

The crew's fast response contained the incident, and damage to the machine was minimal. That outcome deserves emphasis for what it was: a near-miss. The fire had already passed through the highest-consequence zone of a $2 million machine before anyone knew it existed. The margin between a burning module on the ground and a burning harvester was response speed and circumstance—not detection. On a machine fitted with Shepherd, cameras mounted inside the accumulator would have registered the thermal anomaly as it developed, while the module was still in the machine and the event was still preventable.

The detection events: Shepherd found the fire—from another machine

The incident produced the trial's most compelling data point, unplanned and unprompted. As a Shepherd-equipped harvester drove past the burning module in the field, its thermal cameras—mounted inside its own accumulator—detected the heat anomaly from approximately 20 metres away, through the perforated metal walls of the accumulator housing, and triggered the temperature threshold alarm. The alarm fired on two separate passes.

The significance is hard to overstate. The cameras were not pointed at the fire; they alarmed on it through their own machine's perforated housing, at distance, while working. A thermal event developing inside a monitored accumulator—centimetres from the lenses, in direct line of sight—presents a detection challenge orders of magnitude easier than the one the system just passed twice in the field.

What the season demonstrates

This paper does not claim certainty about precisely when Shepherd would have alerted on the #19 fire had the machine been fitted—that timeline cannot be reconstructed. What the season established is direct and factual:

• A fire developed undetected in the accumulator of the one machine in the group without thermal monitoring
• The Shepherd system demonstrably alarmed on that same fire—externally, at ~20 metres, through perforated metal, on both passes
• The accumulator-zone configuration—the setup specified for fleet rollout—ran six weeks of continuous monitoring with zero false alarms

Together these establish the two facts that matter to an operator or an insurer: in the priority zone, the system alarms on real fires with sensitivity to spare, and it does not cry wolf.

The Business Case

Cost of the risk

The #19 incident resolved with minimal machine damage—which is exactly why it belongs in a business case. The same fire, discovered minutes later or with the module still in the machine, sits in the industry-average loss range below. Near-misses are the cheapest lessons available; the next event is not guaranteed to be one.

• Average harvester fire loss across industry: $1–2 million per event
• Replacement cost per CP770: approximately $2 million, before downtime and lost harvest window
• Fleet exposure: ~$24 million in CP770 assets operating simultaneously each harvest
• Lost harvest opportunity from a mid-season machine loss is typically not insured

The insurance dimension

At approximately $30,000 per machine per year, the 12-machine fleet carries around $360,000 in annual premiums. Documented, continuous fire detection changes the risk profile an insurer is pricing in three ways:

• Detection: thermal precursors identified before ignition reduce the probability of total-loss claims
• Evidence: continuous thermal history strengthens the insured's position in any claim, demonstrating both proactive risk management and the condition of the machine before an event
• Precedent: insurers are increasingly recognising fixed fire-detection technology in premium settings—operators should raise documented thermal monitoring directly with their broker 

Operator Perspective

“There’s no doubt in my mind—if Picker #19 had the Shepherd system installed, it would have detected the fire immediately.”

— BJ Smith

Harvest & Logistics Manager - Australian Food & Fibre (AFF)

Implications for Fleet Operators

• Fire risk on round module pickers concentrates in the accumulator and module handling zones—targeted monitoring of these zones delivers the majority of the protection at a fraction of whole-of-machine cost
• Targeted-zone monitoring is operationally mature: six weeks of harvest conditions in the accumulator zone produced zero false alarms and required no operator workflow changes. Wider-coverage configurations need threshold calibration for high ambient heat—a trial phase resolves this
• The same infrastructure doubles as condition monitoring, converting a risk-mitigation spend into a maintenance and uptime tool
• Documented monitoring creates an evidence trail with direct insurance value—in claims support and, increasingly, premium negotiation

For contract harvesters, where a machine loss also means contractual failure across multiple clients' crops, the downtime-avoidance case is stronger still.

Next Steps

Following the trial, a fleet-wide deployment across all 12 CP770 harvesters is under consideration, extending accumulator-zone thermal monitoring from 2 machines to the full fleet ahead of the 2027 season.

For trial data, technical specifications or a fleet assessment: Vince Carson, Business Development Manager, ESDH Technologies—vince_carson@esdh.com.au | 0400 500 651