Temporal Graph Convolutional Networks for Land Environmental Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for modeling ecological systems in agriculture face challenges due to the unstructured nature of catchments, which are difficult to analyze using classic deep learning approaches, and fail to naturally incorporate spatial information critical for hydrological and land-surface processes.
Innovation Solution
A computer-implemented method using similarity analysis and temporal graph convolutional neural networks to group sub-regions with similar environmental descriptors, encoding these into group-based graphs to improve model training and forecasting accuracy by incorporating domain knowledge and handling unstructured data effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If classic deep learning approaches such as convolutional neural networks are used to analyze catchments, then the analysis can be performed on structured pixel-based datasets, but the unstructured nature of catchments (interconnected irregular polygons) makes them difficult to analyze
Solution Approach 1:
The patent creates a graph-based representation (copy) of the catchment data that transforms the unstructured polygonal relationships into a structured format suitable for deep learning. The graph copies the spatial relationships and environmental descriptors into nodes and edges, making the data amenable to convolutional neural network analysis while preserving the original unstructured nature of the catchments.
2Adaptability or versatility
If alternative deep neural networks are used to analyze time series datasets, then time series analysis capability is improved, but spatial information which is critical for hydrological and land-surface processes is not naturally incorporated
Solution Approach 1:
The patent merges time series analysis capabilities with spatial information processing by integrating temporal convolutional networks with graph convolutional networks. This combination allows the model to simultaneously process time-series environmental data and spatial relationships between catchments, preserving both temporal dynamics and spatial dependencies in the hydrological forecasting.
3Reliability
If connectivity between neighboring catchments is determined by geographical relationships, then spatial proximity is considered, but catchments with similar environmental descriptors may be missed despite their geographical distance
Solution Approach 1:
The patent applies local quality by creating graph-based neighborhoods around each catchment that are defined by environmental descriptor similarity rather than purely geographical proximity. This allows each catchment to have a customized neighborhood structure that reflects its specific environmental characteristics, improving the accuracy of connectivity assessment by considering local environmental conditions rather than applying a uniform geographical criterion.
Data Source
AI summary
Embodiments are directed to a computer-implemented method of analyzing a land region that has been decomposed into a plurality of regular or irregular sub-regions. The computer-implemented method includes applying, using a processor system, a feature extraction process that extracts a set of sub-region environmental descriptors for each of the plurality of sub-regions. The processor system applies a similarity analysis to the set of sub-region environmental descriptors to generate groups of the plurality of sub-regions. The processor system creates a plurality of group-based graphs by encoding each of the groups into a corresponding group-based graph. A spatio-temporal neural network is used to train a model based at least in part on the plurality of group-based graphs.