Satellite Plume Source Estimation Using Segmented Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating emission source location from satellite data face challenges due to low temporal and spatial resolution of satellite concentration data and lack of granular local wind data, making direct inversion of concentration and wind data impossible.
Innovation Solution
A method involving the creation of a dataset from satellite plume concentration data, downsampling, partitioning into training and validation sets, and training machine learning models to identify the presence and position of emission sources with subpixel resolution, using synthetic and actual data to account for variable wind conditions and source magnitudes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If direct inversion of concentration data and wind data is attempted, then emission source location can be estimated, but the method fails due to insufficient data resolution
Solution Approach 1:
The patent segments the problem into two distinct machine learning models: one for detecting plume presence and another for estimating source location and magnitude. This segmentation allows each model to specialize in specific aspects of the data, overcoming the limitation of low-resolution satellite concentration data by focusing computational effort on extracting meaningful signals from limited observations.
Solution Approach 2:
The patent introduces machine learning models as intermediary components between the raw satellite concentration data and the final emission source estimation. These models act as mediators that can infer additional information (such as source location and magnitude) from the limited concentration data, effectively bridging the information gap caused by low data resolution.
2Reliability
If constant remote satellite observations are implemented to capture intermittent emission patterns, then detection accuracy improves, but observation frequency and resource consumption increase
Solution Approach 1:
The patent performs preliminary action by training machine learning models on synthetic plume data that represents various emission scenarios before actual satellite observations are processed. This pre-training enables the system to make accurate predictions from sparse, intermittent observations without requiring constant satellite passes, as the models have already learned to recognize emission patterns from the synthetic training data.
Solution Approach 2:
The patent creates synthetic copies of plume concentration data through computational models to train the machine learning systems. These synthetic data copies allow the models to learn from a wide variety of emission scenarios without requiring actual constant satellite observations, thereby maintaining high detection reliability while reducing the need for frequent real observations.
Data Source
AI summary
In an approach for estimating emission source location from satellite plume data, a processor creates a dataset of plume concentration data. A processor down samples the dataset to an array at satellite resolution. A processor partitions the array into two separate datasets according to a preset proportion. A processor trains two machine learning models on at least one of the two separate datasets, wherein a first machine learning model of the two machine learning models is for identifying a presence of a plume and a second machine learning model of the two machine learning models is for identifying a source position and magnitude of the plume. A processor applies the two machine learning models to new concentration data.