A regional ANN-based model to estimate suspended sediment concentrations in ungauged heterogeneous basins

Abstract

ABSTRACT The high cost of monitoring suspended sediment concentration (SSC) in rivers calls for the development of indirect estimation methods, based on relationships with other variables, which are easier and cheaper to measure. We present an original approach to investigate the capacity of regional models to extrapolate SSC to ungauged basins, in a heterogeneous region with scarce in situ data and complex hydrography. The estimates were based on qualitative variables (drainage area, soil type, land use, land cover and mean catchment slope) to represent spatial variability, and quantitative variables (turbidity, flow, precipitation and exponentially weighted moving average of past rainfall) to represent temporal variability. We used artificial neural network (ANN)-based models, applied to the Brazilian part of the Upper Paraguay River Basin, covering an area of 362 380 km2. This study demonstrates that the proposed methodology allows the regional extrapolation of SSC to ungauged basins with very good performance, even in heterogeneous regions.