Beyond 2D landslide inventories and their rollover: synoptic 3D inventories and volume from repeat lidar data

Abstract. Efficient and robust landslide mapping volume estimation is essential to rapidly infer spatial distribution, quantify the role of triggering events on landscape changes, assess direct secondary landslide-related geomorphic hazards. Many efforts have been made develop methods, based 2D satellite or aerial images, constrain empirical volume–area (V–A) relationship which, in turn, would allow for provision indirect estimates volume. Despite these efforts, major issues remain, including uncertainty V–A scaling, amalgamation underdetection landslides. To address issues, we propose a new semiautomatic 3D point cloud differencing method detect filter out false detections due lidar elevation errors, obtain inventories with an metric, directly measure geometric properties This multiscale model-to-model comparison (M3C2) algorithm was applied multitemporal airborne dataset Kaik?ura region, New Zealand, following Mw 7.8 earthquake 14 November 2016. In 5 km2 area, detects 1118 potential sources. Manual labeling 739 sources shows prevalence forest-free areas (24.4 %), spatially correlated forested (80 related ground classification errors pre-earthquake (pre-EQ) dataset. Combining distance closest deposit signal-to-noise ratio metrics, filtering step our workflow reduces source below 1 % terms total area labeled inventory. The final predicted inventory contains 433 399 deposits lower limit detection size 20 m2 724 297 ± 141 087 m3 954 029 159 188 deposits. Geometric inventory, relationship, are consistent previous results, except lack classically observed rollover distribution area. A manually mapped from image has better (6 m2) but only identifies 258 scars, exhibits around m2, underestimates 3D-detected by 72 58 %, respectively. Detection delimitation occur limited texture change (bare-rock surfaces, forests) at transition between that accurately captures. Large rotational/translational landslides retrogressive scars can be detected using irrespective area's vegetation cover, they missed owing dominant vertical topographic change. misses shallow (< 0.4 m depth) method, corresponding 10 2 Our data show systematic size-dependent 200 may explain all part distribution. While segmentation complex clustered remains challenging, demonstrate offers greater sensitivity small changes than classical difference digital models (DEMs). results underline vast 3D-derived exhaustively objectively impact extreme regions prone landsliding, variety hillslope mass movements cannot captured mapping, explore scaling ways.