Satellite Crop Yield Estimation via Statistical Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating crop yields using satellite imagery are inaccurate and costly, particularly when attempting to estimate yields for individual fields, due to high variation in estimates and the lack of reliable data and ground calibration.
Innovation Solution
A system and method that uses satellite image processing to generate crop yield estimates for areas as small as individual fields by applying a statistical model with environmental and crop information variables, trained using crop model simulations and observable quantities, to predict crop yields with high resolution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If satellite imagery is used to estimate crop yield for individual fields, then spatial resolution is improved, but measurement precision deteriorates due to high variation in estimates
Solution Approach 1:
The patent segments the crop yield estimation problem into multiple components: using multiple satellite images taken at different times throughout the growing season, dividing the field into manageable analysis units, and breaking down the estimation into multiple measurement variables (vegetation indices, environmental factors, crop parameters) that are combined to improve overall precision while maintaining fine spatial resolution
Solution Approach 2:
The patent applies preliminary action by using multiple satellite images captured at different times before the final yield measurement. These preliminary observations are used to build a statistical model that predicts final yield, allowing the system to capture crop development stages and environmental conditions throughout the growing season rather than relying on a single measurement
2Loss of information
If satellite image processing is applied to individual fields, then data granularity is improved, but device complexity increases due to processing requirements
Solution Approach 1:
The patent creates a universal statistical model that can be applied across multiple fields and regions. The same model structure and methodology are used whether estimating yield for one field or many, allowing the system to maintain high data granularity for individual fields while using a standardized processing approach that reduces overall system complexity
Solution Approach 2:
The patent changes parameters by using multiple satellite image dates and multiple environmental variables as inputs to the statistical model. Rather than attempting to process every possible parameter at full resolution, the system selects key parameters (vegetation indices, temperature, precipitation) that drive yield variations, reducing processing complexity while preserving essential information
3Measurement precision
If traditional satellite image methods are used for crop yield estimation, then cost is reduced, but measurement precision deteriorates due to lack of ground calibration data
Solution Approach 1:
The patent uses copying by creating statistical models that replicate the relationship between satellite observations and actual yield outcomes. These models are trained on historical data where both satellite imagery and ground truth yield measurements are available, then copied and applied to new fields without requiring expensive ground calibration for each individual field, thereby maintaining precision while reducing costs
Data Source
AI summary
Systems and methods for generating a crop yield estimate for an area as small as an individual field from images captured by a satellite are disclosed. The system generates simulations of crop yields in a region that includes the area by applying combinations of different parameters to a crop yield models. Observable quantities for simulated yields are determined from the simulations. The simulations and the observable properties are used to train a statistic model for the region that has two or more variables. Images captured by a satellite that include at least a portion of the area are obtained. Crop information is then determined from the images and weather information associated with the dates that the images where captured is obtained. The statistical model is then applied to the crop information and the weather information to determine a crop yield estimate.