Learning Model for Altitude-Resolved Water Vapor Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing rainfall prediction systems face challenges in improving temporal resolution for calculating precipitable water amount and are unable to accurately estimate altitude-by-altitude water vapor densities using GNSS data.
Innovation Solution
A microwave radiometer system that utilizes machine learning to generate models for estimating altitude-by-altitude water vapor densities and precipitable water amount by integrating radio field intensities with climatological data, such as temperature, humidity, and air pressure, and sonde data from external servers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If GNSS data is used to calculate precipitable water amount, then measurement capability is provided, but temporal resolution cannot be improved and altitude-by-altitude water vapor densities cannot be calculated
Solution Approach 1:
The patent combines microwave radiometer data with GNSS data to create a hybrid measurement system. The microwave radiometer provides high temporal resolution observations while GNSS provides absolute calibration reference, achieving both precision and high temporal resolution simultaneously through data fusion
Solution Approach 2:
The patent introduces machine learning models as an intermediary that processes microwave radiometer measurements and converts them into precipitable water amount estimates calibrated against GNSS data. This intermediary enables the system to achieve GNSS-level accuracy with microwave radiometer temporal resolution
2Measurement precision
If GNSS data is used for water vapor observation, then measurement capability is provided, but altitude-by-altitude water vapor densities cannot be estimated
Solution Approach 1:
The patent segments the atmosphere into multiple altitude layers and uses microwave radiometer measurements at different frequencies to probe each layer. The machine learning model processes these segmented measurements to retrieve vertical profiles of water vapor density, preserving altitude-resolved information that would be lost in integrated measurements
Solution Approach 2:
The patent transitions from horizontal integration (GNSS provides path-integrated delay) to vertical dimensionality by using microwave radiometer frequency channels to sense different atmospheric layers. This dimensional transformation enables retrieval of altitude-by-altitude water vapor densities through the machine learning model
3Measurement precision
If machine learning models are trained with large datasets, then estimation accuracy is improved, but processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary action by pre-training machine learning models offline using large historical datasets containing microwave radiometer measurements, sonde data, and GNSS data. This offline pre-training phase, which requires significant computational resources and time, is performed once or periodically, allowing rapid online deployment with minimal real-time processing requirements
Solution Approach 2:
The system implements self-service through automated data collection, preprocessing, and model retraining pipelines. The machine learning model continuously learns from incoming data streams and automatically updates itself, reducing the need for manual intervention and optimizing performance over time without additional operational overhead
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Provided is a generation method for a learning model that can efficiently estimate altitude-by-altitude water vapor densities or precipitable water amounts. According to the present invention, a generation method for a learning model involves: acquiring training data that includes first measurement data measured by a microwave radiometer and altitude-by-altitude water vapor densities or precipitable water amounts obtained from second measurement data measured by a radiosonde: and, on the basis of the acquired training data, generating a learning model that, when first measurement data has been inputted, outputs altitude-by-altitude water vapor densities or precipitable water amounts.