Enhanced georeferencing of VHR satellite imagery using multiple-resolution LiDAR data
Background
In 2009 the Swedish Land Survey initiated a nationwide project aiming at establishing a new ground elevation model (Ny nationell höjdmodell, NNH). The source data will be acquired using medium-resolution LiDAR scanning (0.5 – 1 point/sqm) and this whole dataset will be made available along with the new ground elevation model. Although sparser than the LiDAR data used in the previous project, if the developed method can be extended to use this new and widely available data to accurately georeference satellite images, there is significant potential for extended use of satellite imagery in many application areas such as:
• Forestry
• Agriculture
• Mapping
• Land coverage classification
• Nature preservation
• Environmental monitoring
In previous work it has been shown that high-resolution LiDAR data (8-10 points/sqm) can be used to improve the accuracy of the rectification process for VHR satellite imagery, making the rectification RMSE as low as 0.5m (from an image pixel size of 2.4m). This allowed the spectral information in the satellite imagery to be well associated with the corresponding tree canopies identified using LiDAR data processing (FORAN SingleTree™). In the previous work, although the satellite imagery was accurately georeferenced, the spectral information within each single tree crown proved to be insufficient for accurate tree classification. Thus, methods for extending the spectral information are needed. By using image data from multiple registrations, different sensors or wavebands, this could to enhance the tree species classification accuracy for single trees.
On the other hand, since the developed methods allowed for very accurate referencing of images, there is
great potential to use the results for area-based forest inventory methods such as FORAN ForestGrid, in order to determine the tree composition. Area based methods work with sparse data, so if accurate referencing of images can be obtain using sparse LiDAR data such as NNH, this opens up for a multitude of applications involving imagery and LiDAR data combined.
Objective
The purpose of the project is to build upon the previous investigation into the potential synergies between VHR satellite imagery and LiDAR data. It is proposed to extend these methods and investigate the applicability of image georeferencing using the new medium-resolution National Land Survey LiDAR data. The result will be used to evaluate synergistic use of VHR (Very High Resolution) satellite imagery and airborne laser (LiDAR) for area-based forest inventory methods such as FORAN ForestGrid, where the imagery is used for accurate tree composition determination within 15mx15m area cells. Furthermore, since experience from other projects shows insufficient accuracies in separating tree classes using only one source of satellite imagery, the project will investigate techniques to improve the classification at the level of single tree individuals by using multiple image sources.
Expected results
The expected results of the projects are:
• Knowledge as to with what accuracy VHR satellite imagery can be georeferenced using the new data
from the National Land Survey
• Knowledge as to with what accuracy tree composition can be determined for small areas, typical
15mx15m, using satellite imagery and new LiDAR data.
• Knowledge as to how the spectral information for single tree crowns can be increased by multiple
image sources that are georeferenced using high resolution LiDAR data.
• Knowledge as to with what accuracy a single tree can be classified using more spectral information.