Answers are grounded in this paper and cite their sources.
Mechanistic riparian models for river projects consider various environmental requirements, but few are spatially explicit. This study explores whether machine learning using airborne LiDAR terrain metrics can accurately predict preferential planting locations and help understand beneficial site attributes. A random forest machine learning model was trained on 29,263 vegetation presence/absence observations randomly selected from 2017 LiDAR-derived polygons of natural vegetation from 34 km of the regulated lower Yuba River, California. A spatial-independence resampling strategy split the river into 67 sections, with 32 training, 28 validation, and seven test polygons. Sixteen LiDAR-derived topographic predictors were computed at 0.91-m resolution. The best model had an Area Under the Curve performance of 0.77. Vegetation presence was mainly associated with microtopographic “vector ruggedness”, followed by terrain ruggedness and roughness. Despite controlling water availability, detrended elevation and distance from wetted areas were not dominant variables. To identify the best planting locations, probability thresholds differentiating avoidance, preference, and high preference were quantified using a forage ratio electivity index. Locations with presence probability >80% had vegetation present 3.5 times as often as expected by chance (p < 0.05), so bare locations with such probabilities should be prioritized for planting. Microtopographic roughness should be designed before planting to increase plant establishment and growth.
No takes yet. Share an insight, caveat, or question.
Gómez et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: