Model Card: Relative Yield Prediction Model

What it is

Overview

The relative yield model provides in-season, field-specific relative yield predictions for various crops grown in the main agricultural regions of the world.

The model generates predictions by comparing forecasted current-season yields against a benchmark derived from multiple simulations using long term historical weather data. Users can customize benchmark parameters through the API, including the historical weather data timeframe and crop management specifications.

The predictions do not attempt to replace actual harvest yield measurements in fields.

Inputs and Outputs

Required field inputs

• field location (latitude and longitude, point)
• crop
• maturity index and/or variety name (depends on crop)
• planting date

Optional inputs

The optional inputs can be modified by users for more custom model tuning, otherwise default values will be used.

Advanced crop practice inputs can be provided in order to increase prediction accuracy. These parameters can be defined for the current season scenario as well as for the historical benchmark. They include:

• soil profile characteristics: soil textural class and soil profile depth
• planting parameters: depth, density, raw spacing, percentage of available water at planting
• water management: irrigated vs. rainfed

Model configuration:

• Historical Benchmark Period: Time window defining the historical years used to establish baseline yield predictions
• Maximum radial distance to fetch weather and soil data

Response configuration:

• Desired prediction period during the season
• Prediction Frequency: User-defined temporal resolution (e.g., weekly, 10 days) defined as the time interval between 2 consecutive yield predictions during the growing season.

Outputs

• Relative yield predictions expressed as decimal values ranging from 0 to 2, representing the following statistical distributions: 25th percentile (lower quartile), 50th percentile (median), 75th percentile (upper quartile). Interpretation: Values represent the proportion of expected yield compared to the benchmark yield. For example, a value = 0.5 = 50% of benchmark yield. For detailed interpretation guidelines, refer to the 'How to interpret results' section.

• Warning outputs indicate how many historical years were used to calculate the yield benchmark value.

How it works

Algorithm principles

The Relative Yield algorithm is based on mechanistic equations which consider the soil-plant-atmosphere continuum and model daily evolutions of soil water balance, biomass accumulation and partitioning, final yield, and yield components.

The model uses field-centric data such as field location and soil profile characteristics, coupled with historical local weather data, to return a timeseries of relative yield predictions during the season.

Finally, the relative yield API returns the ratio between the current year yield absolute prediction with respect to the historical yield benchmark according to the following formula:

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

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. This ongoing process enhances prediction accuracy through calibration and cross-validation, ensuring reliable results for both current and new commercial varieties.
• Performance monitoring: our scientists continuously monitor model predictions to detect and promptly address data quality issues and model performance drift.

How to use

How to interpret insights

Explanation of predicted values

The yield prediction model provides a relative yield estimate, presented as a percentage of the expected average yield for a given field, aka. benchmark yield. This estimate is given as a median value (50th percentile) along with a range of likely variability (25th and 75th percentiles). The range provided by the 25th and 75th percentiles accounts for uncertainty due to both past and future weather conditions, giving users a likely band within which the final yield will fall.

For example, a median relative yield of 0.7 (or 70%) means the model predicts a final yield 30% lower than the expected benchmark. A median of 1.2 (or 120%) suggests a final yield 20% higher than benchmark.

Detailed use-case examples

Intended use

The relative yield predictions can serve multiple purposes such as:

• Yield forecasting: the relative yield predictions are primarily used to inform growers about the expected yield of the season compared to a benchmark yield field potential derived from historical predictions. analysis.

• Scenario analysis: enables comparison of different crop management practices, including: planting date, variety, and other management practices, to assess their impact on yield.

• Model integration: serves as an enabler component for other agricultural predictive models.

Limitations

Insights accuracy

Relative yield predictions may be affected by poor weather data quality, as well as by the quality of the soil profile information used for inference.

Some abiotic stresses as well as pest damages that could affect the crop's yields are not accounted for by the model.

Note that model predictions are intended to inform growers the expected relative yield level for each field during the current season but cannot, under any circumstances, replace yield measurements and expert agronomist assessments.

System availability

The model uses ERA5T weather data for simulation and prediction. ERA5T data is updated every day in the inference data store.

Due to the execution of the inference data pipeline, the model is not available from 5:30 AM to 6:30 AM UTC.