What satellite, weather and soil data can (and can’t) tell a farmer

A decade ago, detailed information about a farm’s weather history, soil and crop health was mostly out of reach for smallholder farmers. Today, a remarkable amount of it is freely available from space agencies and research institutions. The harder question is what that information can reliably tell someone deciding what to plant, when to irrigate or how much fertiliser to apply.

Three layers of public data

Satellite imagery: how the crop is growing

The European Space Agency’s Sentinel-2 satellites photograph the Earth’s land surface every few days, with detail down to 10 metres in key wavelengths, and the data is free and open¹. By comparing how plants reflect red and near-infrared light, analysts calculate vegetation indices such as NDVI, which track how green and vigorous a crop canopy is through the season.

These indices are good at showing change: a field greening up after planting, a patch falling behind its neighbours, a crop drying down toward harvest. They are much less good at explaining why. A weak signal could mean water stress, nutrient deficiency, pests, a late planting date or simply a different crop.

Weather and climate data: the conditions the crop faces

Services such as NASA’s POWER project provide decades of daily solar radiation, temperature, rainfall and humidity estimates for any point on Earth, built from satellite observations and climate models². Combined with short-range forecasts, this makes it possible to estimate heat stress, water balance and the likely timing of crop growth stages.

The trade-off is resolution. Global weather datasets typically describe grid cells tens of kilometres across, which can blur the difference between a coastal field and one further inland, or between a valley and a hillside.

Soil maps: what lies beneath

ISRIC’s SoilGrids uses machine learning and hundreds of thousands of soil profile observations to predict properties such as texture, pH, organic carbon and nitrogen worldwide, on a 250-metre grid³. For many locations, it is the best soil information available without taking a sample.

But it is a prediction, not a measurement. Its uncertainty is highest where few samples exist, and it cannot capture recent changes from liming, fertiliser use or flooding. It also cannot detect contaminants such as heavy metals.

Where public data falls short on smallholder farms

• Scale. Many smallholder plots are smaller than a single soil-map cell and cover only a handful of satellite pixels, so field edges, paths and neighbouring crops mix into the signal.
• Clouds. Optical satellites cannot see through cloud, and in monsoon climates clouds can hide fields for weeks at the moments that matter most. Radar satellites help, but measure different things.
• Management. None of these datasets records what the farmer actually did: the variety, the planting date, the fertiliser or the irrigation. That context often explains more of the outcome than the environment does.

Turning data into a decision

The value of these datasets comes from combining them and being honest about uncertainty. A crop recommendation grounded in soil, weather history, seasonal forecasts and satellite-observed vegetation trends is more robust than one based on any single layer. It becomes far stronger when it is calibrated against local ground truth: field observations, yield records and laboratory tests.

That principle runs through Culvera’s approach. Public global data provides a starting point for any location; local partners and field measurements make it trustworthy; and the output should be a clear recommendation, not another dashboard of raw numbers. See how the Culvera platform works, or read about how better field data supports low-emission rice.