Harnessing the strengths of machine learning and geostatistics to improve streamflow prediction in ungauged basins; the best of both worlds

Abstract

Streamflow is a key driver of stream health, influencing water quality, physical form and habitat to support healthy populations of freshwater-dependent biota. However, while available at locations with hydrographic gauging, measurements of streamflow are often unavailable at the majority of reaches across a catchment, hindering the interpretation of flow-related variables, and subsequently, the effectiveness of management interventions. This study constructed models for predicting daily streamflow at ungauged reaches using two contrasting statistical algorithms: a long short-term memory (LSTM) machine learning algorithm and a spatial stream network (SSN) geostatistical algorithm. Instance S..