two numbers for the same field

Problem statement

Reducing post-harvest loss is a Sustainable Development Goal target (SDG 12.3, tracked by FAO's Food Loss Index), yet for fruits and vegetables the loss numbers themselves are unreliable. National statistics offices have two ways to get them: ask farmers and traders (enquiry) or send enumerators to weigh what is discarded at each operation (actual measurement). FAO's field test of its own measurement guidelines in Mexico showed the two methods can disagree by a factor of three or more on the same farms — and each is wrong in a different direction. Enquiry systematically under-reports; actual measurement is a "one-shot picture" that captures whichever season, weather and operation the enumerator happened to witness. Without a measurement approach that is cheap enough to run routinely and robust to timing, loss-reduction investments are aimed using numbers no one trusts.

Why this matters

FAO's statisticians write that "knowledge and reporting about food loss measurements along the supply chain is still scarce, and only a few countries measure and report food losses on a limited number of products," so the SDG 12.3.1a Food Loss Index is largely modelled rather than measured. In the Mexico test, broccoli losses "by enquiry show similar results for both municipalities of about 6 to 11 percent in harvesting and grading, but results by actual measurement diverge considerably and range from 10 to 43 percent," with actual measurement in Valle Santiago estimating "35 percent of losses in harvesting and 40 percent of losses in grading." Off-farm, broccoli enquiry estimates reached about 50 percent at wholesale and 34 percent at retail. If a country's loss baseline can swing from 6% to 43% depending on method, then a cold-room, packhouse or road investment cannot be justified, prioritized, or evaluated, and SDG reporting rests on the cheaper, lower-biased number.

What’s been tried and why it hasn’t worked

FAO's Global Strategy guidelines prescribe sample surveys combining declarations with physical measurement, and the Mexico test applied them to banana and broccoli. Enquiry worked for banana (on-farm loss 3.67 percent, both methods "relatively consistent") because bananas are cut pre-mature in bunches with little field damage; it failed for broccoli because producers "declared average losses of both seasons, while actual measurement could only capture losses occurring during the rainy season," and because rain made it "difficult to set the boundaries between pre-harvest and harvest losses." Actual measurement failed on logistics: data collection "was restricted to only 15 days, wherefore harvesting could not be timed appropriately and several plots needed to be replaced," operations at each stage "are distributed throughout the day or along various days," so capturing every operation "implies more than one visit by the enumerator," and sample sizes for transport and storage collapsed. Definitions did not transfer either: "the commonly used definition based on cereals and pulses is not sufficient" for fruits and vegetables, harvest and grading were "difficult to separate, as grading is often conducted while harvesting," and discarded produce that flows to freezers or processors is not straightforwardly a loss. FAO's conclusion is that "combining both methods through statistical pooling can help to improve food loss estimates" — but pooling two biased instruments does not remove the bias, and the report's authors suggest that larger samples would likely reveal a "tendency to underestimate by enquiry and overestimate food losses by actual measurement."

What would unlock progress

What is missing is a third instrument: a low-cost, continuous or repeat-visit measurement of discards at grading and packing points that does not depend on a single enumerator visit or on farmer recall — the perishable-crop equivalent of a flow meter. Candidate ingredients exist in adjacent fields: smartphone image-based grading and volume estimation from precision agriculture, low-cost load cells and event logging from industrial IoT, and diary/recall-calibration methods from consumption surveys, which measure and correct the systematic gap between recall and observation. A crop- and stage-specific correction model (enquiry × season × operation → expected measured loss) built from paired data would let statistics offices keep using cheap enquiry while removing its known bias.

Entry points for student teams

A student team could design and pilot a repeat-visit, low-burden discard-logging tool for a single perishable at grading (e.g., a scale-plus-camera station or a farmer-operated tally with photo verification) and quantify how its estimate compares with enquiry and one-shot enumeration on the same plots. A statistics team could use FAO's Mexico dataset structure to build and test a pooling estimator that models the direction and magnitude of enquiry bias by crop and season rather than simply averaging methods. A design team could rebuild the enquiry questionnaire around the boundary problems FAO documents (pre-harvest vs. harvest, harvest vs. grading, discards diverted to processing) and test whether re-worded instruments narrow the gap. Relevant skills: survey statistics, agricultural economics, embedded sensing/computer vision, field methods.