REVISITING THE 2012 DULUTH, MN EXTREME PRECIPITATION EVENT: CHARACTERIZING THE EXTENT AND MAGNITUDE OF LANDSLIDES, EROSION, AND SEDIMENTATION USING REPEAT LIDAR AND OBJECT-BASED IMAGE ANALYSIS

A June 2012 Duluth, Minnesota-area extreme precipitation event dropped over 20 cm of precipitation. This caused widespread landscape change in the form of fluvial erosion, sedimentation, and landslides, and resulted in significant damage to infrastructure. Imagery analysis and several field-based efforts documented these effects where observable, but close-in-time repeat airborne lidar from 2011 and 2012 could potentially provide a more synoptic understanding of this event. However, issues with the lidar data as delivered, especially the 2011 data, including misaligned flightline swaths, vertical errors, and areas of up to one-meter horizontal error precluded its use for reliable landscape change detection. We report on processing steps taken to improve the lidar data. These include Bayesian methods to improve point positioning from trajectory data, elevation adjustment using existing ground survey data, alignment improvement in one particularly problematic area with apparent horizontal misalignment between the 2011-2012 data using the iterative closest point algorithm, and reduction of noise in maps of landscape change using correction surfaces generated from unchanged, flatter portions of the landscape. Object-based image analysis (OBIA) was then applied to enhance the pixel-based elevation change results to generate object-based classifications of areas of landscape change. OBIA allows for an understanding of the occurrence of, for example: landslide erosion and deposition, channel incision and deposition, streambank failure, and other types of landscape geomorphic change. This is performed by semi-automated “segmentation” of the landscape into “objects” (such as individual landslide scars or deposits) by identifying similar areas using data including elevation change, slope, and stream channel maps. These objects were then classified according to slope gradient, landscape position, deposition/incision depth, and other characteristics. This generalization and classification is an improvement over traditional pixel-based classification of landscape change from multi-temporal lidar, and may lead to improved sediment budgeting and extraction of key information about the effects of geomorphic process on the landscape.