Agricultural forecasting calibrated to farm-specific decisions.
P1-Terra is a physics-based agricultural forecasting system. It produces calibrated environmental forecasts for individual farms, extending from operational windows (days) to seasonal planning horizons (months). The result is farm-specific environmental intelligence over which agricultural decisions can be made with measurable precision.
Continuous forecast.
Traditional agricultural forecasts are issued at a fixed cadence and held as the basis for decisions until the next issuance. Terra treats each issuance as the input to its next cycle. The forecast is recomputed continuously against the corrected environmental field as it updates, against the farm's latest observations, against the calibrators as they refresh.
Each cycle's correction is also the foundation of the next horizon. Terra reaches a forecast at the limit of its current accuracy, measures its error against what actually arrived, corrects for it, and uses the corrected state as the starting point for the next window. The horizon extends through chains of self-correction rather than through a single long-range model that loses accuracy with time.
A farm is not one forecasting problem. It is a horizon of forecasting problems, each grounded in the corrected state of the previous one. Terra is built around that chain. Where traditional forecasting ends, Terra begins.
Forward, with skill gating
Against on-farm observations
Against regional baseline
"A regional forecast is accurate at the resolution of the region. A farm's decisions are made at the resolution of the farm. The forecast that should govern those decisions is not the forecast accurate for the region, but the forecast accurate for the location where the decisions are made."
Per-farm calibration
The accuracy of a regional weather forecast is bounded by its resolution. A model that is reliable in aggregate can be inaccurate at a specific farm by margins that matter for operational decisions. Terra corrects for this. The system updates hourly against the latest forecast issuance and the latest observation data relevant to the farm's location. The forecast underneath a Terra output is the regional forecast as corrected against local data.
A farm is not one forecasting problem. It is a horizon of forecasting problems, each calibrated against the farm's own observation history. Terra is built around that chain.
Weather-conditioned environmental modelling
Rainfall, temperature, humidity, solar radiation, and wind forecasts are corrected against actual observations from the farm's own meteorological station, every hour. Open weather models are accurate at scale but consistently wrong in known, local ways at the farm level. Terra learns those local errors and removes them before the agricultural decision layer makes use of them. The quality of any downstream agricultural model is bounded by the quality of the environmental inputs. This is where the system spends most of its effort.
Agricultural variable modelling
Physics-based and observational models for the variables agricultural decisions depend on: soil moisture, evapotranspiration, growing degree days, sunlight exposure, erosion risk, and crop development indices. Each farm operates under conditions specific to its location, soil composition, crop mix, and management practice. Terra's models are configured per farm against those specifics. The variable underneath a decision is the variable as it applies to that farm, not a class average for the region or crop type.
Horizon-extending calibration
The forecast horizon is extended through chains of self-correction rather than through a single long-range model. Each forecast window is calibrated against the observations that arrived during it, and the corrected state becomes the starting point for the next window. Hard limits on horizon are imposed where the model has no demonstrated skill against historical baselines, and the system falls back to climatology at those leads rather than producing forecasts the system cannot defend.
The view below shows the corrected environmental conditions Terra is reasoning over in the current cycle for the selected farm. Each value is the post-calibration field, refreshed against the latest forecast and the latest observations from the farm's meteorological station. Adverse conditions surfaced here are the conditions the agricultural decision layer is reasoning over in the current cycle, at the resolution it sees them.
Current operating baseline
Terra operates today on regional NWP feeds calibrated against publicly available observation networks and on-farm meteorological stations where deployed. Forecast skill is documented against CHIRPS and station baselines. Current results outperform regional forecasts at the farm level on demonstrated metrics, before any deeper integration with on-farm sensing or proprietary observation networks.
Expansion with on-farm sensing
Integration with on-farm sensors expands the variables Terra can resolve directly and tightens the calibration of those it already produces. These integrations extend the agricultural decisions Terra can support without replacing the architecture that powers the baseline.
Terra is a Principia application, a domain-specific instantiation of the Intelligence OS architecture applied to agricultural operations. Calibrated environmental intelligence from P1-Tempo feeds the farm-level forecasting layer. Forecast precision at the Tempo layer determines the quality of the field over which Terra reasons, which determines the agricultural variables it can resolve and the operational consequences a farmer can act on. Forecasts are traceable to the calibration cycle, the observation set, and the agricultural model that produced them.