Landscape-scale geomorphic change detection: Quantifying spatially variable uncertainty and circumventing legacy data issues

- We differenced airborne lidar-derived DEMs for Blue Earth County, MN (1980 km2).
- We detected and corrected vertical and horizontal bias' associated with the older dataset.
- We show the importance of spatially variable uncertainty and consequences of a uniform model.
- The 'cloud point density' metric more accurately quantifies uncertainty due to vegetation.

Repeat surveys of high-resolution topographic data enable analysis of geomorphic change through digital elevation model (DEM) differencing. Such analyses are becoming increasingly common. However, techniques for developing robust estimates of spatially variable uncertainty in DEM differencing estimates have been slow to develop and are underutilized. Further, issues often arise when comparing recent to older data sets, because of differences in data quality. Airborne lidar data were collected in 2005 and 2012 in Blue Earth County, Minnesota (1980 km2) and the occurrence of an extreme flood in 2010 produced geomorphic change clearly observed in the field, providing an opportunity to estimate landscape-scale geomorphic change. Initial assessments of the lidar-derived digital elevation models (DEMs) indicated both a vertical bias attributed to different geoid models and localized offset strips in the DEM of difference from poor coregistration of the flightlines. We applied corrections for both issues and describe the methods we used to discern those issues and correct them. We then compare different threshold models to quantify uncertainty. Poor quantification of uncertainty can erroneously over- or underestimate real change. We show that application of a uniform threshold, often called a minimum level of detection, overestimates change in areas where change would not be expected, such as stable hillslopes, and underestimates change in areas where it is expected and has been observed, such as channel banks. We describe a spatially variable DEM error model that combines the influence of slope, point density, and vegetation in a fuzzy inference system. Vegetation is represented with a metric referred to as the cloud point density ratio that assesses the complete point cloud to describe the density of above ground features that may hinder bare-earth returns. We compare the significance of spatially variable versus spatially uniform DEM errors on change detection by thresholding the DEM of Difference at a 95% confidence interval (2σ). Results indicate significant geomorphic change in relatively predictable locations, such as erosion on the outside and deposition on the inside, of bends. Final totals indicated net erosion of [2,625,100] ± 2,389,000 m3 in the county between 2005 and 2012. Of this, 39% was generated from bluffs, 1% from ravines, and the remainder came from banks and floodplain areas.