Satellite and Drone Imagery Analysis for Agriculture and Property

From a Land Rover to a map on a laptop

A farm manager with 1,200 hectares of arable land drives the fields to spot problems: a patch of poor emergence, waterlogging in a corner, disease starting on the north side. It takes days, and parts of the farm get seen less than others.

A property manager with 90 commercial units sends a surveyor up ladders to check roofs after storms. It is slow, expensive and a working-at-height risk.

Both are problems of looking at a lot of ground or a lot of buildings regularly. Aerial and satellite imagery combined with machine learning does the looking at scale, and sends people only where something has changed.

Satellite or drone: resolution decides

A useful test: how big is the smallest thing you need to see? A field-level drop in crop vigour shows up at 10 metres. A slipped roof tile does not.

What machine learning measures in agriculture

• Crop health indices from multispectral bands, tracked across the season to spot areas falling behind
• Plant and tree counts in orchards, vineyards and forestry, from drone imagery
• Field boundary and crop type mapping across large areas, useful for agronomists and insurers
• Weed patches and gaps in establishment, feeding spot-spraying or re-drilling plans
• Waterlogging and drought stress patterns across a season
• Livestock counts from drone footage on large grazing land

Indices such as vegetation health scores are not new and do not need machine learning. The machine learning adds interpretation: separating a disease pattern from a drainage one, predicting yield zones, or flagging which of 300 fields has changed unusually since last week. That last one is often the most useful. A model that tells an agronomist where to walk first saves more time than one that tries to diagnose from space.

What it measures for property and estates

• Roof condition and damage after storms, from drone surveys
• Solar panel suitability and panel faults via thermal imaging
• Building footprint and extension changes over time, useful for portfolio and planning checks
• Vegetation encroachment near buildings, rail lines and power infrastructure
• Construction progress on development sites, compared against plans

For insurers and property managers, the same detection models also feed damage assessment workflows. We cover the photo-based side in our post on damage assessment from photos.

The practical limits

• Cloud cover. Optical satellites cannot see through it. In a wet British spring you may get few clear images for weeks. Radar satellites can, but interpreting them is a specialist job.
• Ground truth. A model that says a field zone is stressed is only useful if someone checks what is really there, at least while you calibrate it.
• Drone regulations. Commercial drone flights in the UK and EU need appropriate operator registration and flying within the rules for the category, particularly near people and buildings.
• Data volume. A single drone survey can produce tens of gigabytes of imagery. Plan storage and processing before the first flight.
• Seasonality. Models trained on one crop stage or one season often perform worse on another.

Imagery tells you where to look. It rarely tells you, on its own, what to do when you get there.

Buy a platform or build

Established agronomy and drone survey platforms already provide crop indices, plant counts and roof inspection reports, and for most farms and property firms they are the sensible first step. A custom build makes sense when you need to combine imagery with your own data, such as yield maps, tenancy records, maintenance history or insurance claims, or when you want a detection model for something platforms do not offer.

When we take this on at SpiderHunts, the imagery pipeline is usually the smaller part. Joining results to your land parcels or property records and sending the right alert to the right person is most of the machine learning build.

A first step that costs little

Free satellite imagery for your fields or sites over the last two seasons can be pulled and analysed before anyone buys a drone. If field-level change detection already highlights the problems you care about, keep going with satellite. If the problems are too small to see at that resolution, you have a clear, evidence-based case for drone surveys on specific sites.

Either way, agree up front who acts on what the imagery shows and how fast. A stress map delivered on Friday that the agronomist opens a fortnight later has missed its window. The same goes for a storm damage report that sits in an inbox while water gets into a unit. The value is in the loop from image to action, and that loop is ordinary operational design as much as it is machine learning. Our broader machine learning guide covers how to scope that kind of project.

Frequently asked questions