Digital Elevation Model Error Correction Using Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current global digital elevation models (DEMs) suffer from large vertical errors, especially in densely populated areas, where satellite radar sensors misinterpret building tops as hills, leading to inaccurate vulnerability assessments for coastal communities facing sea level rise and flooding.
Innovation Solution
A system utilizing a convolution neural network (CNN) is developed to improve DEM accuracy by receiving input data including vegetation, architecture, and population density information, trained with high-quality data from the NASA ICESat-2 mission to predict error corrections and generate more accurate elevation models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If satellite radar sensors are used to create global DEMs, then coverage area is improved, but measurement precision deteriorates due to misinterpreting building tops as hills
Solution Approach 1:
The patent uses machine learning models as an intermediary to process satellite radar data and correct elevation errors. The models learn from high-quality reference data (lidar) and apply correction algorithms to satellite-derived DEMs, effectively mediating between the coarse satellite data and accurate elevation measurements needed for coastal flood risk assessment.
Solution Approach 2:
The patent transforms the satellite radar data by applying machine learning-based parameter corrections. The models learn optimal correction parameters from training data and apply these transformations to adjust the elevation values, changing the parameters from raw satellite measurements to corrected elevation estimates that account for urban structures.
2Manufacturing precision
If traditional DEM correction methods are used, then processing time is reduced, but manufacturing precision deteriorates because corrections are limited to small areas or specific vegetation types
Solution Approach 1:
The patent develops universal machine learning models that can correct elevation errors across diverse environments including urban areas, forests, and mixed landscapes. The models are trained on varied data and designed to handle multiple land cover types simultaneously, making the correction system universally applicable rather than requiring separate solutions for different terrain types.
Solution Approach 2:
The patent replaces traditional mechanical correction methods (manual processing, area-by-area adjustment) with machine learning-based automated correction. The ML models automatically learn correction patterns from data and apply them systematically across the entire study area, substituting manual or rule-based mechanical processes with intelligent automated systems.
3Measurement precision
If CoastalDEM v1.1 trained on US ground truth data is applied globally, then vertical bias is reduced in trained regions, but reliability deteriorates in areas with dissimilar vegetation, architecture, and population density
Solution Approach 1:
The patent applies local quality by training or fine-tuning machine learning models with region-specific ground truth data. Different regions (urban centers, rural areas, different climate zones) use models adapted to their local characteristics, ensuring that the correction algorithms are optimized for local vegetation types, building styles, and population densities rather than applying a single global model.
Solution Approach 2:
The patent performs preliminary action by pre-training models on extensive ground truth data from multiple regions before deployment. The models are prepared in advance with region-specific training data to account for local characteristics, so when applied to new areas, they already have the contextual knowledge needed for accurate corrections rather than being applied blindly without local adaptation.
Data Source
AI summary
A system and method for creating a digital elevation model, and for reducing vertical bias and/or root mean square error (RMSE) of an elevation dataset may be provided. The system may include one or more processors configured to receive input data, provide the input data to a neural network (NN), and generate a digital elevation model based on the predicted elevations output by the NN. The NN may be configured to include an input layer; a plurality of hidden layers connected to the input layer, the plurality of hidden layers configured to iteratively analyze the input data and learn nonlinear relationships between the input data and actual elevation; and an output layer connected to the plurality of hidden layers, the output layer configured to output a predicted elevation based on the analysis of the input data.