Machine Learning Predicts Reach‐Scale Channel Types From Coarse‐Scale Geospatial Data in a Large River Basin

Abstract

Hydrologic and geomorphic classifications have gained traction in response to the increasing need for basin‐wide water resources management. Regardless of the selected classification scheme, an open scientific challenge is how to extend information from limited field sites to classify tens of thousands to millions of channel reaches across a basin. To address this spatial scaling challenge, this study leverages machine learning to predict reach‐scale geomorphic channel types using publicly available geospatial data. A bottom‐up machine learning approach selects the most accurate and stable model among ∼20,000 combinations of 287 coarse geospatial predictors, preprocessing methods, and algorithms in a three‐tiered framework to (i) define a tractable problem and reduce predictor noise, (ii) assess model performance in statistical learning, and (iii) assess model performance in prediction. This study also addresses key issues related to the design, interpretation, and diagnosis of machine learning models in hydrologic sciences. In an application to the Sacramento River basin (California, USA), the developed framework selects a Random Forest model to predict 10 channel types previously determined from 290 field surveys over 108,943 two hundred‐meter reaches. Performance in statistical learning is reasonable with a 61% median cross‐validation accuracy, a sixfold increase over the 10% accuracy of the baseline random model, and the predictions coherently capture the large‐scale geomorphic organization of the landscape. Interestingly, in the study area, the persistent roughness of the topography partially controls channel types and the variation in the entropy‐based predictive performance is explained by imperfect training information and scale mismatch between labels and predictors.

Constraining Remote River Discharge Estimation Using Reach‐Scale Geomorphology

“…Therefore, we needed a way to map RS observations at the scale of the Mackenzie basin (220,069 cross sections) to the DBSCAN-generated classes (which use no RS data). To do so, we turned to basic supervised statistical learning to assign river types, which has seen some success at the regional scale (Guillon et al, 2020). Using a classic validation-set approach to model training, we trained a multiclass logistic regression classifier on 80% of the training data using the median of cross-sectional widths as the sole predictor.…”

“…Dallaire et al (2019) clustered features explicitly associated with every observation/river reach within HydroSHEDS (Lehner et al, 2008); however, fluvial geomorphology data do not generally exist in this form at the global scale and so their analysis was largely limited to hydroclimatic river types. Conversely, Guillon et al (2020) successfully used machine learning models to upscale a priori geomorphic river types for the Sacramento River basin-defined using field geomorphology 10.1029/2020WR027949 Water Resources Research campaigns at 290 sites-to over 100,000 reaches. Our study presents a novel amalgamation of these two methods, first using automated clustering of field data to define a geomorphic classification framework and then using supervised learning to upscale river types to anywhere on Earth.…”

Things we can do now that we could not do before: Developing and using a cross-scalar, state-wide database to support geomorphologically-informed river management

“…using Google Earth and Google Earth Engine (GEE) along with Geomorphic Change Detection (GCD) and automated tools for geomorphic analysis of rivers) [ 78 , 88 , 83 ]. We are not yet at a situation where databases such as the NSW database can be fully automated, but certain parts of the analysis can be semi-automated using available and emerging ‘plug-in’ tools for analysis [ 88 ] and by tapping into large-scale remote sensing datasets and toolboxes held in Open Access or consortium-based repositories [ 27 , 51 , 83 ]. Such advances provide a fantastic opportunity to now juxtapose a static map with analyses conducted using GEE or GCD (for example) to display river adjustment, behaviour and change over time [ 21 , 79 ].…”

“…There is no equivalent systematically-derived geomorphic database, conducted at such a scale and incorporating these layers of analysis and insight, in Australia. Importantly, this particular database extends far beyond an automated, off-site, appraisal of remote sensing imagery conducted using machine-learning models and applications [ 27 , 48 – 51 ]. A careful mix of remotely-sensed data and generic tools has been combined with targeted field verification to compile the database.…”

Application of globally available, coarse‐resolution digital elevation models for delineating valley bottom segments of varying length across a catchment

“…Some models pre-define a constant segmentation length based on expert or user judgement (often based on significant prior understanding of the catchment and rivers being modelled), or extensive post-processing sensitivity analysis on a range of outputs that use different segment lengths (e.g. O'Brien et al, 2019;Guillon et al, 2020). The studies that delineate reaches of variable length initially disaggregate the river network into user-defined homogeneous reaches then use this to measure continuous attributes.…”