Model Card: Regional Relative Yield Data Layer

Developer setup

• Authenticate with Cropwise – Learn how to authenticate your requests using your credentials.
• Model API Reference - WFS endpoint reference for the regional data layer (typeName=RelativeYield30km).

What it is

Overview

The Regional AgInsights Relative Yield Data Layer provides gridded, regional-scale in-season yield intelligence by comparing current-season crop yield predictions against a historical benchmark. It enables stakeholders across the organization — from sales teams to supply chain planners — to visualize and analyze yield performance patterns across territories, counties, and countries without requiring field-level API access.

This data layer is a regional, grid-based product within the Regional AgInsights product family. It transforms field-level Relative Yield model outputs into standardized 30 × 30 km geospatial layers, served via GRIS (WFS/WMS). The layer shows a range of likely yields: low (P25), middle (P50), and high (P75), based on how similar seasons performed historically to see whether this season is trending below, near, or above normal.

The data layer does not replace harvest measurements, field-level yield assessments, or professional agronomic expertise.

Inputs and Outputs

Inputs (generation-side)

At the batch generation level, the underlying Relative Yield model is executed per grid centroid with the following inputs:

Required:

• Grid centroid coordinates (latitude and longitude), derived from the 30 × 30 km standardized grid
• Crop (soybean, corn — depending on geographic scope)
• Maturity index and/or variety name (depends on crop) — configured via a planting date (PD) × maturity group (MG) matrix per state/region
• Planting date — representative planting dates per region (typically 5 PD × 2–3 MG per state)

Optional (applied at generation configuration):

• Soil profile: textural class, profile depth
• Planting parameters: depth, density, row spacing, percentage of available water at planting
• Water management: irrigated vs. rainfed
• Historical benchmark period (years)
• Maximum radial distance to fetch weather/soil data

Outputs

Each grid cell exposes the following attributes:

• Relative yield (decimal, 0 to ~2) as three percentiles:

  • P25 (25th percentile): Lower-bound scenario
  • P50 (50th percentile / median): Central estimate
  • P75 (75th percentile): Upper-bound scenario

  Interpretation:

  • 0.7 = expected yield is 70% of benchmark (approximately 30% below benchmark)
  • 1.0 = expected yield equals the benchmark
  • 1.2 = expected yield is 120% of benchmark (approximately 20% above benchmark)

  The P25–P75 band represents plausible variability and uncertainty driven by both past and future weather conditions. Median values below 0.2 or above 2.0 should be interpreted cautiously as potential outliers.

• Warning output: Indicates how many historical years were used to compute the benchmark yield. Fewer years generally means a less stable benchmark.

Definition:

Relative yield (Yr) = Absolute yield of current year (Yc) / Historical benchmark yield (Yh)

How it works

Algorithm principles

The underlying Relative Yield model is based on mechanistic equations that represent the soil–plant–atmosphere continuum. It models daily evolutions of:

• Soil water balance
• Biomass accumulation and partitioning
• Final yield and yield components

The model uses field-centric data (location, soil profile) coupled with historical local weather data to generate an in-season time series of yield predictions. The relative yield is computed as the ratio of the current-year absolute yield prediction to the historical benchmark yield.

Training and continuous improvement

• Annual model calibration: The model is updated every year by incorporating new growth stage observations for existing and newly launched varieties, through calibration and cross-validation.
• Performance monitoring: Scientists continuously monitor model predictions to detect and promptly address data quality issues and performance drift.

Generation (batch → grid)

Regional AgInsights generates predictions on a standardized 30 × 30 km grid by:

1. Constructing a coordinate set of grid centroids filtered to agricultural areas via cropland thresholds.
2. For each centroid, executing the Relative Yield API across the configured PD × MG matrix (typically 5 planting dates × 2–3 maturity groups per state).
3. Aggregating or selecting representative outputs per grid cell.
4. Managing execution metadata (timestamps, model versions, success rates) with retry logic and rate limiting.
5. Publishing results as geospatial layers in GRIS (PostGIS / GeoServer).

Serving (consumption)

The layer is served through the GRIS Proxy Service via OGC-compliant WFS endpoints under the typeName value:

• RelativeYield30km

Consumers can query features using CQL filters, request JSON or CSV output formats, and limit results with the maxFeatures parameter. Authentication is handled via OAuth2 client credentials.

How to use

Intended use

• Yield forecasting: In-season relative yield predictions inform stakeholders about expected seasonal yield compared to the benchmark field potential derived from historical weather.
• Scenario analysis: Compare different planting date and maturity group combinations to assess their impact on regional yield expectations.
• Sales / Commercial enablement: Territory-level yield context for customer conversations; identify areas with above- or below-trend expectations for targeted outreach.
• Marketing / Strategy: Regional opportunity analysis linking agronomic yield signals to commercial performance.
• Supply chain: Incorporate regional yield signals into demand planning and product allocation.
• Model integration: Serves as an upstream signal for other predictive workflows and Cropwise AI natural language queries.

How to interpret insights

Relative yield percentiles (P25 / P50 / P75)

Reading examples:

Warning output (benchmark robustness)

Limitations

• Predictions are sensitive to weather and soil-profile data quality. Grid-level soil assumptions may not reflect actual field conditions.
• The model does not account for certain abiotic stresses (beyond those captured by the weather-driven simulation) and pest or disease damage that could affect yield.
• The 30 × 30 km grid resolution aggregates multiple planting date and maturity group combinations; individual field-level accuracy is not guaranteed.
• The PD × MG matrix per state/region is subject to finalization; representative management assumptions may not reflect all local practices.
• Spatial masking rules (cropland thresholds and "main growing regions" definitions) are subject to finalization.
• This layer does not replace harvest measurements or expert agronomic yield assessments.

System information

Data sources

• ERA5T weather data (updated daily)
• Soil profile data (gridded defaults per centroid)
• Crop-specific calibration parameters (managed by agronomy experts)

Update frequency

• Weekly during the growing season

Spatial and temporal resolution

• Grid size: 30 × 30 km (Regional AgInsights product resolution)
• Temporal granularity: Weekly predictions during the growing season

Model availability note

The underlying Relative Yield API is unavailable between 5:30–6:30 AM UTC daily due to the data pipeline window.

Geographic Scope, Crops, and Availability

Last update: 30/04/2026