Deep learning insights into suspended sediment concentrations across the conterminous United States: Strengths and limitations
Long short-term memory (LSTM) deep learning models were trained to predict daily streamflow suspended sediment concentration across the conterminous US (CONUS). The WholeCONUS model compared well against individual basin models for predicting daily SSC, upholding the theory of data synergy (larger quantities of more diverse training data lead to better models compared to lower quantities of specifically localized/applicable training data). This may suggest that all local factors co-vary with the environment, and implies the existence of a generic, environment-dependent relationship that can be learned. Additionally, to our knowledge this was the first attempt to build a continental-scale data-driven model for daily SSC, which allowed us to study the heterogeneity of SSC on a regional scale.
SSC is closely linked to riverine ecosystems and has extremely complex relationships with other environmental factors, making it important but difficult to model. We demonstrated the ability of the Whole-CONUS model to predict SSC in unmonitored basins, which indicates that reasonably accurate predictions of SSC can be made with only meteorological forcings, static watershed attributes, and streamflow observations. Given that all of these are much more commonly-available datasets compared to suspended sediment concentration measurements, LSTM thus provides an alternative to traditional methods constrained to specific sites - a way to predict SSC in any CONUS basin with satisfactory accuracy.
Suspended sediment concentration (SSC) is a crucial indicator for aquatic ecosystems and reservoir management but is challenging to predict at large scales. It is unclear whether SSC is predictable using macroscopic environmental attributes and forcings. This study tested the feasibility of deep-network-based models to predict daily SSC at basin outlets given only basin-averaged forcings, readily-available physiographic attributes, and streamflow (from observation or model). We trained long short-term memory (LSTM) deep networks both separately for each of the 377 sites across the conterminous United States (CONUS) (termed “local models”), and on all the sites collectively (Whole-CONUS). The Whole-CONUS and local models presented median coefficient of determination (R2) values of 0.63 and 0.52, respectively. This comparison agrees with previously acknowledged “data synergy” effects for LSTM models, where more data from more sites can help improve predictions overall. Furthermore, the continental-scale analysis provided a wealth of insights about SSC patterns. Both local and Whole-CONUS models tended to be more successful where SSC-streamflow correlations (Rs-q) were high typically in the humid Eastern US and with lower SSC. Low Rs-q basins were often found in the arid Southwest with higher SSC. The highly-nonlinear SSC-streamflow relationships seem related to seasonality, basin size, and heterogeneity of land cover and rainfall within the basins, suggesting these basins need to be simulated at higher spatial resolution and may require additional inputs related to SSC induced by seasonality. The Whole-CONUS model also performed well for spatial extrapolation (basins not included in the training dataset, median R2 = 0.55), supporting large-scale modeling efforts. These state-of-the-art results using only minimal inputs suggest data-driven approaches can exploit the natural coevolution of sediment processes and the environment to support sediment modeling.