Custom datasets and calibrated models for research institutes
Custom geospatial datasets, calibrated hydrological models, and remote sensing time series built to the documentation and reproducibility standards research publications require.
Research institutes often need a dataset that does not exist yet: a land cover classification for a specific region and period, a calibrated model for an under-studied basin, a consistent satellite time series across a boundary that no published product respects. Assembling it is technical work rather than research, but the research cannot start without it.
We build those inputs to the standard the downstream research requires — documented provenance, stated accuracy, and a processing chain another team can reproduce and extend.
What makes this sector different
- Required datasets frequently do not exist in published form at the needed extent or period.
- Published global products are often too coarse or too generalised for regional research.
- Reproducibility requires documenting the whole processing chain, not just the output.
- Accuracy has to be quantified rather than asserted for downstream use.
What we deliver
- Custom land cover and land use classifications with quantified accuracy assessment
- Consistent multi-year satellite time series across a defined study area
- Calibrated hydrological models with documented parameterisation
- Processing code and documentation enabling independent reproduction
Common questions from research institutes
How do you quantify classification accuracy?
Through a confusion matrix built against independent reference samples that were withheld from training, reported as overall accuracy plus per-class producer and user accuracy. A single headline accuracy figure hides the classes that perform badly, which are often exactly the ones a study depends on.
Who owns the resulting dataset?
You do, under the terms we agree at the outset, subject to the licensing of the underlying imagery. Sentinel and Landsat archives are open, so derived products from those carry no upstream restriction; commercial imagery does, and we flag that before it becomes a problem at publication.