Estimating the Uncertainty in Change-Detection Products will Advance the Geosciences
Dr. Preston Hartzell received a $400,000 grant from the National Geospatial-Intelligence Agency (NGA). The new grant provides funding for two years. Preston is a Research Assistant Professor in the Geosensing Systems Engineering & Sciences graduate program at the University of Houston.
The NGA grant concentrates on developing rigorous methods for estimating the uncertainty in spatial change-detection products. Computing the vertical difference between pre- and post-event Digital Elevation Models (DEMs) to detect change is common, but producing a corresponding uncertainty estimate for each pixel of the difference model is rare. An analyst can observe vertical change but does not know if that change is significant under the current methods.
Preston's work aims to change that. He and others will develop algorithms to first estimate uncertainty in the source DEMs, followed by algorithms to push that source uncertainty into the final change detection product. They will also develop methods for estimating the uncertainty in horizontal motion derived from pre- and post-event point clouds and DEMs. This research will support scientists in properly interpreting and using spatial change-detection products by providing rigorously-derived uncertainty estimates for each reported change measurement.
Uncertainty estimation is fundamental to all measurement. Interpretation of a measurement requires familiarity with the complete measurement process in order to develop a sense of its reliability, given no associated uncertainty. "In terms of change detection, this is often overlooked when viewing a large change over a large area. But when examining small changes, such as horizontal motion due to post-seismic fault creep or a small, isolated area of vertical difference between two digital elevation models, an analyst should naturally question whether the reported spatial change is valid or simply a product of the data collection or processing methods," Preston states. "These small changes can be important, e.g., for earthquake inversion modeling or target detection."
Related News Stories: