Field-Specific Nutrient Modeling Using Spatial Residual Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer-implemented predictive models for agricultural crop fertility struggle to be agronomically field-specific, often overestimating or underestimating nutrient requirements due to unaccounted field characteristics, and require massive data sets that are rarely available for non-experimental fields.
Innovation Solution
An agricultural intelligence computer system that uses historical input data for a specific agronomic field to compute required nutrient application values, generates residual values for locations where actual nutrient application falls short, and creates model correction data by correlating residual values with spatial characteristics of the field to calibrate the nutrient model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a digital nutrient model uses aggregated historical data from multiple fields to compute nutrient requirements, then the model can be applied to any field with available data, but the model fails to account for unique field characteristics causing consistent overestimation or underestimation of nutrient requirements
Solution Approach 1:
The patent segments the nutrient modeling process into two distinct components: a general digital nutrient model that provides baseline nutrient requirements, and a field-specific residual model that captures unique field characteristics. This segmentation allows the system to maintain model applicability across different fields while improving accuracy for each specific field through the residual component learned from historical data.
Solution Approach 2:
The patent changes the parameters of the nutrient model by introducing field-specific residual values that are learned from historical nutrient application and yield data. These residual parameters are specific to each field and adjust the general model's predictions to account for unique field characteristics such as soil properties, topography, and management practices that the general model cannot capture.
2Measurement precision
If a digital nutrient model is calibrated to be highly field-specific using massive amounts of historical data from each field, then the model accuracy for that field improves, but the approach becomes infeasible for non-experimental fields where such data is rarely available
Solution Approach 1:
The patent applies partial action by using only the historical data that is actually available for each field rather than requiring complete datasets. The system learns field-specific residuals from whatever historical nutrient application and yield data exists, even if incomplete, and combines this partial field-specific information with the general digital nutrient model to provide accurate recommendations without requiring extensive data collection.
3Ease of operation
If a digital nutrient model uses current year inputs only without historical field data, then the model can be applied immediately to any field, but the model is biased by unmeasured or poorly measured field properties leading to unreliable outcomes
Solution Approach 1:
The patent implements feedback by using historical nutrient application data and actual yield outcomes to learn field-specific residual patterns. This feedback loop allows the model to continuously improve its accuracy for each field by comparing predicted nutrient requirements with actual field performance, thereby reducing bias from unmeasured field properties while maintaining ease of implementation.
Data Source
Figure 1
Figure 2(a)~2(b)
Figure 3
AI summary
In an embodiment, an agricultural intelligence receives agronomic field data for an agronomic field, comprising one or more input parameters, nutrient application values, and measured yield values. The system uses a digital model of crop growth to compute, for a plurality of locations on the field, a required nutrient value indicating a required amount of nutrient to produce the measured yield values. The system identifies a subset of the plurality of locations where the computed required nutrient value is greater than the nutrient application value and computes, for each location, a residual value comprising a difference between the required nutrient value and the nutrient application value. The system generates a residual map comprising the residual values. Using the residual map and the one or more input parameters for each of the plurality of locations, the system generates and stores model correction data for the agronomic field.