Remote Sensing Labs
Browse all docs

Agriculture

Agriculture is one of the domains where remote sensing has the longest track record. Crops are spatially extensive, they change week to week, and nearly every decision about them is made under uncertainty about a field nobody has walked recently. The mechanics are covered elsewhere on this site — Agricultural Analysis is the seasonal workflow, and NDVI Monitoring is the index-through-time engine underneath it. This page is about who uses the answers, what they decide with them, and what changes when the answer arrives.

The decisions imagery is actually feeding

A grower or agronomist decides where to scout this week, whether to vary seed, nitrogen, or irrigation across a field, and when a block is ready to harvest. A cooperative or agribusiness decides how to allocate scarce agronomists across hundreds of fields, and how much volume to expect at the elevator. An insurer or agricultural lender decides whether a claimed loss is consistent with what the season looked like, and how to price an area-based index product. A ministry of agriculture or a food-security early warning program decides whether a region is heading for a shortfall, and whether to move procurement, imports, or aid before prices do.

Each of those users reports on a different unit — a management zone, a farm portfolio, a policy district — and a wrong scouting trip costs an afternoon while a wrong national forecast moves markets. Above all, agriculture’s clock is set by the crop, not by the analyst: an answer that arrives after the decision window has closed is indistinguishable from a wrong one.

Field scale: managing variability you cannot see from the gate

The premise of precision agriculture is that a field is not uniform, and that the variation inside it is large enough and persistent enough to be worth treating differently. Soil depth, drainage, compaction, planting quality, and pest pressure all vary at scales far below the parcel, and a grower standing at the gate sees almost none of it. Repeated overhead observation is one of the cheapest ways to cover the whole field every few days when skies allow, which is why it anchors variable-rate input plans, irrigation zoning, replant decisions, and trial evaluation.

The value is differential treatment. If nitrogen, water, or seed can be varied across a field, then knowing which parts of the field respond differently converts directly into either lower input cost or higher yield on the same ground. If they cannot be varied — a single-rate applicator, a flood-irrigated block — then the same map is interesting but not yet actionable, and it is worth saying so before the season starts.

A vegetation index tells you where the crop is developing differently, not why; the scout, the soil probe, or the yield monitor supplies the cause. Growers quickly lose confidence in a system that asserts a diagnosis their own eyes contradict, and the return comes from the scouting trips avoided and the problems caught early, not from the map itself.

Regional scale: yield outlook and food security

Change the unit to a district, a province, or a growing region and the same observations serve a different purpose. Here the question is not which corner of a field is struggling but whether an entire production area is tracking above or below normal, and how early that can be said with enough confidence to act on. Consumers of that answer include national statistical and agriculture agencies, famine early-warning systems such as FEWS NET and the GEOGLAM Crop Monitor, humanitarian logistics planners, and commodity analysts.

Lead time is the whole product. A credible signal that a harvest will fall short is worth far more in the middle of a growing season, when import contracts, strategic reserves, and pre-positioned aid can still be moved, than it is at harvest when the shortfall is a fact. The claim is always an anomaly — relative to what this region normally looks like at this point in its calendar — so a long, consistent archive is a domain requirement rather than a preference: a baseline reaching back only three seasons cannot separate a bad year from a new normal. At this scale that baseline usually comes from coarse, near-daily records rather than fine ones: AVHRR vegetation-index records reaching back to 1981, continued by MODIS and VIIRS, paired with long satellite-based rainfall estimates. Their frequent revisit survives cloud in a way a 16-day sensor cannot, and the archive is long enough to define “normal”. Finer archives such as Landsat earn their place at the field-to-landscape scale, where parcels must be resolved. The difference between a long record and a dense one is the subject of Time Series. Imagery is also rarely used alone here — it is one input beside weather data, crop calendars, market prices, and survey statistics, and its contribution is spatial completeness and timeliness rather than precision on any single field.

What determines whether the value is real

Three things separate agricultural remote sensing that changes a decision from work that merely describes a season. The first is timing against the decision window, which is set by phenology: an answer that arrives after the top-dress window, the irrigation set, or the planting date has passed cannot be acted on at any accuracy. The second is whether the reporting unit matches the decision unit. Field boundaries, management zones, and administrative districts are the units people act on, and where fields are smaller than a few pixels of the sensor in use — as many smallholder fields are at 30–250 m — the mismatch described by the Resolution concept is not a technical footnote but a question of who the analysis can serve at all. The third is the asymmetry in the cost of being wrong: a false alarm costs a scouting trip, a missed stress event costs yield, and an insurance or subsidy decision made on a bad signal costs money and trust — which is why the error tolerance belongs to the program, not to a default threshold.

Where the domain stays hard

Yield monitor records, field boundaries, and input applications are commercially sensitive, so the reference data needed to validate agricultural products is often private. Many of the world’s most food-insecure regions have growing seasons that coincide with persistent cloud, which pushes serious work toward radar and the SAR reasoning that comes with it. Mixed cropping, intercropping, and irregular smallholder parcels break assumptions that hold comfortably on large uniform fields. And adoption is a real constraint: an output that contradicts a grower’s direct experience without explaining itself gets ignored, however good the underlying analysis.

Where to go next

The workflow mechanics live in Agricultural Analysis, which covers the crop calendar, seasonal metrics, and validation, and in NDVI Monitoring for the index and baseline reasoning it builds on.

Sources