- I am seeing and hearing about the benefits of combining LiDAR with spectral imaging.
- There is often a wide gap between the promise of hyperspectral imaging and reality.
- If there is a synergy between LiDAR and spectral imaging, beyond a DTM I am sure it would be of interest.
I have been hearing and reading about the combination of airborne LiDAR and multi and hyperspectral imaging. In most cases I don’t understand the value of the combination beyond a 3D terrain model. Having worked for an applied remote sensing consulting firm for a couple of years I can assure you there is often a significant gap between the promise of spectral imaging technology and its actual application.
The promise is that one can develop a library of unique spectral signatures that can identify materials and species. The reality is the world is a mixed pixel, and the signatures are influenced by a number of factors. This is particularly true in the case of hyperspectral, where the number of variables can boggle the mind.
The traditional method for attempting to visualize hyperspectral data is to use a 3D cube. The one shown is from the University of Texas. The leader in terms of software for analyzing this data is ENVI. I have not been keeping up with the latest research in this field. 5 years ago it was difficult, and expensive to even schedule a hyperspectral sensor mission.
If anyone is working in this area I would appreciate an update, particularly with regards to the value of combining LiDAR with spectral imaging.
The cost to collect HSI data have gone down DRAMATICALLY in the last five years, and the data quality has gone up by an order of magnitude. One of the best improvements has been the increase in the accuracy of HSI geo-coords when adding a LiDAR system.
The other reason to incorporate HSI and LiDAR together is if you are collecting one, the cost to add the other is insignificant since they have similar swath widths. The real cost is getting the aircraft on location and air time. The better HSI sensor vendors have automated their calibration and georefrencing routines taking a large portion of the labor cost out of the HSI data collection.
Since you have some expereince with HSI and LiDAR you know how long it can take to apply spectral algorithms over areas that are 100s of square miles (I get paid by the hour!). During a recent project to identy a certain tree species we used the LiDAR data to remove any object less than seven feet tall, then applied a vegetation mask to remove soils and manmade objects resulting in the ability to process the remaining pixels two orders of magnitude quicker than if we tried to spectrally match evey pixel. (Note: this process can have very negative consequences when employing a matched filter.)
“often a wide gap between the promise of hyperspectral imaging and reality”
I could not agree with this statement more. Many “analysts” think one can just convert the data to reflectance, grab a library spectrum, apply an algorithm and find all of that material in the scene. This could not be further from the truth. The company I work for will only deliver radiance calibrated data or the answer (material map). If you are good enough to do the data analysis, you should be good enough to perform your own atmospheric compensation. If you deliver reflectance data it will never exactly match the ground measurements and you become the patsy for an analyst poor results.
There will be good example of combining LIDAR and HSI in the upcoming August 2009 edition of PG&RS Magazine.
Great information – thanks Dan.
There are many advantages combining lidar and multi or hyperspectral data in vegetation studies. The main one is that lidar provides the height of the canopy that otherwise would not be available. It is also easy to detect trees with lidar and to assign them a height. This height, a constant for the whole tree, can be used together with the spectral channels in the classification. The segmentation of the image can be based on the lidar detected trees and allows to perform an object-based classification. You should also rectify the image with a DSM (I mean taking into account the height of the trees) if you want to combine high resolution images from different epochs.
Best regards,
Toni