Skip to content
Shyenova

Area 04

How much of what you are currently estimating could simply be measured?

A satellite pass is not a finding. The work is turning reflectance into a number somebody will act on, and proving it is right.

85–93%

Classification accuracy

Typical overall accuracy for major classes with adequate ground truth and a full-season series.

4–8 weeks

Earlier signal

How far ahead of conventional field reporting a reliable estimate becomes available.

60–80%

Survey cost

Reduction once field effort is redirected from measurement to model validation.

These are the improvement ranges these levers are known to produce in industry where the underlying practice has not been examined recently. They are targets to test against your own baseline, not claims about named past clients.

Earth observation has quietly become cheap. Sentinel-1 and Sentinel-2 are free, revisit every five to twelve days, and cover the whole country. Commercial constellations offer sub-metre imagery at daily cadence. The constraint stopped being data access some years ago.

The constraint now is that an index map is not a decision. Somebody has to establish what a value of 0.62 means for this crop at this growth stage in this district, validate it against ground observation, and put the result in front of a person with the authority to act on it. That gap between imagery and decision is where most earth observation projects quietly stall, and it is the whole of what this practice does.

What this is not

We do not sell imagery, operate satellites or resell a platform licence. We are frequently the people who tell a client that free Sentinel data answers their question and the sub-metre subscription they were quoted does not.

Artificial intelligence

Where machine learning earns its place in earth observation

This is the discipline where it has the longest and most credible record. Image classification, change detection and segmentation are exactly what these methods were built for, and the accuracy assessment culture in remote sensing is unusually rigorous, confusion matrices, per-class producer and user accuracy, held-out validation. What it does not solve is ground truth. A model trained on two hundred labelled plots will not become reliable because the architecture improved.

How we approach applied AI

What an engagement looks like

A representative shape.

Illustrative. This is the shape of work of this type, not an account of a specific client. We do not publish named results, and we would be wary of a firm that publishes them without the client in the room.

16 weeks

  1. 01Situation

    An agri-processor was committing to procurement volumes months before it had any reliable view of planted acreage, relying on dealer and field-officer reporting whose historical error nobody had measured.

  2. 02What we did

    Built a Sentinel-1 and Sentinel-2 time series across the sourcing districts, trained crop classification on collected ground truth, and ran the satellite estimate in parallel with the incumbent reporting for a full season.

  3. 03Where it landed

    The satellite estimate was available several weeks earlier and diverged from field reporting by a consistent margin in two districts, which ground checks confirmed. The parallel season is what earned the commercial team's trust; the technology alone would not have.

Questions

About remote sensing.

Do we need to buy commercial satellite imagery?

Frequently not, and this is often the first useful thing we tell a client. Free Sentinel data at ten metres answers a great many questions well. Commercial sub-metre imagery earns its cost for small plots, individual asset inspection and anything needing daily revisit, and we would rather establish which category you are in than have you subscribe first and find out later.

How much ground truth do we actually need?

For classification, a few hundred well-distributed labelled points per class per agro-climatic zone, split into training and validation. For yield modelling, measurements spanning a genuine range of outcomes, models trained only on good plots predict badly on poor ones. If you have none, the first season is a collection season and we will say so up front.

Can this work through the monsoon?

With radar, yes; with optical imagery, no. Sentinel-1 sees through cloud, which is why kharif monitoring in most of India runs on it. It is harder to work with and genuinely less accurate for some discriminations. We state which of your seasons are strong and which are compromised rather than averaging the two into a single headline figure.

We have a GIS team already. What would you add?

Usually the classification and validation methodology rather than the mapping. Many capable teams have never had to defend an accuracy assessment or propagate error through a change product. We often build one epoch or one season with the team and hand the rest over.

Start a conversation

Talk to us about remote sensing.

Most useful first conversations are thirty minutes and involve no proposal. Describe the problem you are looking at and we will tell you honestly whether it is one we can help with.

Or write directly to hello@shyenova.com