Reach-scale morphological channel classifications are underpinned by the theory that each channel type is related to an assemblage of reach- and catchment-scale hydrologic, topographic, and sediment supply drivers. However, the relative importance of each driver on reach morphology is unclear, as is the possibility that different driver assemblages yield the same reach morphology. Reach-scale classifications have never needed to be predicated on hydrology, yet hydrology controls discharge and thus sediment transport capacity. The scientific question is: do two or more regions with quantifiable differences in hydrologic setting end up with different reach-scale channel types, or do channel types transcend hydrologic setting because hydrologic setting is not a dominant control at the reach scale? This study answered this question by isolating hydrologic metrics as potential dominant controls of channel type. Three steps were applied in a large test basin with diverse hydrologic settings (Sacramento River, California) to: (1) create a reach-scale channel classification based on local site surveys, (2) categorize sites by flood magnitude, dimensionless flood magnitude, and annual hydrologic regime type, and (3) statistically analyze two hydrogeomorphic linkages. Statistical tests assessed the spatial distribution of channel types and the dependence of channel type morphological attributes by hydrologic setting. Results yielded ten channel types. Nearly all types existed across all hydrologic settings, which is perhaps a surprising development for hydrogeomorphology. Downstream hydraulic geometry relationships were statistically significant. In addition, cobble-dominated uniform streams showed a consistent inverse relationship between slope and dimensionless flood magnitude, an indication of dynamic equilibrium between transport capacity and sediment supply. However, most morphological attributes showed no sorting by hydrologic setting. This study suggests that median hydraulic geometry relations persist across basins and within channel types, but hydrologic influence on geomorphic variability is likely due to local influences rather than catchment-scale drivers.

Designing where to plant riparian vegetation is a component of many river projects. Several mechanistic models have been developed considering biological, soil, hydrological, and hydraulic requirements that influence riparian vegetation growth. However, many models are not spatial explicit and there remains high uncertainty as to where plantings will survive or die. This study sought to determine if a machine learning (ML) algorithm could be trained to accurately characterize the complex set of site attributes that promote survival, and do so exclusively using metrics derived from airborne LiDAR. Results could then be used to guide planting strategies. The selected testbed river was 34 km of alluvial, regulated, gravel/cobble river where planting projects are common and have high mortality. The lower Yuba River, California, USA was mapped at sub-meter resolution in 2017. Our approach has four steps. First, a set of 32,000 vegetation presence/absence observations were randomly selected from LiDAR-derived polygons of naturally occurring established vegetation. Second, the river was split into 75 training, validation and test areas. Third, a set of 17 LiDAR-derived topographic potential predictors were computed at 0.91-m (3-ft) resolution. Finally, a Random Forest machine learning model was trained to best predict vegetation presence. The model results in a riparian vegetation presence probability map and has a “Area Under the Curve” (AUC) of 0.77. As probability values are difficult to interpret, a forage ratio electivity index analysis was performed with statistical bootstrapping. Results show that points with probability values > 0.8 had ~ 8.5 times more riparian vegetation present than would be likely from random chance at the 95% confidence level. Microtopographic ‘vector ruggedness’ was identified as the main driver for vegetation presence, followed by Terrain Ruggedness Index and Roughness. In conclusion, a ML model can identify where riparian vegetation planting are most likely to succeed and guide design. Our results also suggest that more attention should be paid to creating rugged microtopography under plantings to help cuttings and seedlings establish deposition critical for nutrition.

Statistical classifications and machine-learning-based predictive models are increasingly used for environmental data analysis and management. There now exist numerous classifications on the same topic but applied to different regions or spatial scales, such as geomorphic classifications. However, no quantitative meta-analysis framework exists to compare and reconcile across multiple classifications. To fill this gap, we jointly characterize statistical classifications and predictions by combining information theory and machine learning in three novel ways by: (i) measuring the degree of discriminatory information underlying a statistical classification; (ii) estimating the stability of the learning process with tuning entropy; and (iii) leveraging the sequential coarse-graining of information inherent to deep neural networks but absent from traditional machine learning models. This framework is applied through a benchmark of 59 millions models on a unique example of a single statistical classification methodology applied to nine different regions of California, USA. Regional results show that random forest consistently outperforms deep neural networks. In addition, a correlation analysis between regional characteristics, the level of discriminatory information of each classification, and the performance in statistical learning explains variations in performance and reveals the decisive role of the spatial scale of classification outputs. Because such a spatial scale is itself linked to the common situation of limited field sampling, directly comparing findings from statistical classifications and associated predictions appears seldom justified. A more desirable avenue to compare findings lies in combining data underlying statistical approaches in an interpretable and justifiable environmental data science.