Characterization of the wood transport regime of a river with a neural network model

Abstract The primary drivers of river morphology and dynamics are the hydrologic and sedimentologic regimes, but the wood regime is also important. Fallen wood will typically float at the water surface during floods, and the movement and deposition of large pieces can alter the trajectory of sedimentary bedforms, aquatic habitat, and flood hazards. Accurate quantification of floating wood in rivers is now possible, which is enabling us to better understand the role of this key variable, but frequency analysis is rare due to the relatively recent application of monitoring methods and a lack of standard analytical methods. The current study assembles the longest and most complete record of riverine wood transport yet available. Specific objectives are to (a) synthesize and validate a long time series of wood transport from available observations using a neural network model and (b) describe the probability of wood transport events at a river station. The wood record was assembled from manual observations of large wood in multiple flood events and an automated analysis of the full video record, which was estimated to have a precision of 85% and recall of 86%. The neural network predicted wood transport volume from climate and hydrologic records and was found to have Nash‐Sutcliffe efficiencies of over 75%. Comparison of model predictions with wood‐mobilization estimates from aerial photos were within 30% for 5–10 year periods and 2% for a 40‐year period. The wood regime was characterized with (i) an assessment of seasonality; (ii) a rating curve of wood volume from event peak discharge; and (iii) an extreme event analysis of daily maximum and yearly wood transport volume. The methods are useful for characterizing a river's large wood transport regime and hindcasting historical wood mass transfer from flow records. This information will help to understand river dynamics and facilitate sustainable river management.