Top 3 Applications for Data Fusion
I've been doing quite a bit of work lately in the realm of data fusion, and with every unique project I learn a new and unique way to interpret what “data fusion” can mean. Here are the top 3 recent applications for data fusion that I have had the opportunity to explore.
Imagery + LiDAR
Fusion of imagery with LiDAR is probably the number one application for data fusion in terms of popularity. In fact, combining a 3-D point cloud with 2-D imagery provides a long list of analysis options that are not possible with just one modality alone. One example of this richness is illustrated in a recent webinar to use image information to perform DEM hydro-flattening from a LiDAR-derived surface. While LiDAR is one of the best data sources to derive topographic information, there are times when a return over shallow water or even algae or animals can cause a surface feature indicating that the water is not flat. This is a problem specifically for hydrology modeling and drainage planning. Image information, specifically the NIR (Near Infrared) wavelengths, have very low reflection values over water thus accurate shoreline delineation is straightforward, and actually quite easy.Constrain the DEM, or even the point cloud by shoreline vectors, set the values to the elevation of the water body and voila, you have just performed some powerful data fusion.
Figure 1: Left: LiDAR-derived DEM (LiDAR from OpenTopo) with color table applied showing variable elevation inside water bodies. Right: LiDAR-derived DEM with color table applied after performing hydro-flattening showing constant elevation inside water bodies.
LiDAR + Maps
Another popular data fusion exercise is the automatic extraction of building footprints from LiDAR. Many workflows require manual building extraction by delineating each footprint by hand, one by one. This can be an incredibly labor intensive process for any GIS department, but is very necessary for several applications. Some municipalities use this derived information to correlate new structures with building permits. Organizations in energy markets perform population density to plan infrastructure development and monitor existing easements. Therefore a powerful data fusion solution is to perform automated building extraction and overlay the footprints on an existing map to evaluate current “as-built” information. Once the vectors are derived,location, size, and proximity analyses can easily be performed to identify any necessary updates or to structure plans to meet mandates.
Figure 2: LiDAR-derived building footprints (LiDAR from OpenTopo) overlaid onto an Esri Open Streen Map base map imported to ENVI.
Image Time 1 + Image Time 2
Imagery plus imagery can absolutely be a data fusion problem to solve, and one that is very worth the extra effort. Temporal analysis is on the forefront of almost every industry that uses remote sensing and is growing due to higher satellite revisit rates, decreasing costs of airborne data acquisition, and the introduction of UAS data to the commercial market. Some common problems with comparing time-enabled data sets,especially spectral comparison, include the need to perform pixel-to-pixel co-registration and sometimes re-sample and re-project disparate layers. Additionally, images taken at different times of day or on different days can have enough variability in data range due to differences in sun illumination intensity and collection angles that radiometric calibration and atmospheric correction must be performed prior to comparison if a valid result is desired. Easy to use tools for time-enabled data is the topic of an upcoming webinar. Let me know it you're interested and I will make sure you’re on the invite list.
Figure 3: Three images representative of Modis 8-day surface reflectance data analysis to evaluate changes in drought indices over time.