Search

Based on soil erosion and sediment transport processes, CAMF (Cellular Automata-based heuristic for Minimizing Flow) selects sites for afforestation to minimize sediment influx at a catchment's outlet. CAMF uses a raster representation of the catchment and a steepest ascent hill-climbing optimization heuristic, safeguarding spatial interaction. Its execution time can be prohibitively long for large data-sets. Parallelization results in a speedup of 20 to 24 on 28 cores. We present variants of the optimization method to reduce the number and cost of the iterations. We present a tuning algorithm for the meta-parameters of these variants. The results obtained for two contrasting catchments illustrate that the accelerations reduce the cost by a factor larger than 100, with negligible effect on the afforested cells and magnitude of the sediment reduction. The results indicate that higher levels of spatial interaction have a stronger impact on the accuracy of the results and/or the execution time. • An efficient method to select sites for afforestation to minimize sediment loss. • Modified steepest ascent hill climbing method reduces runtime with minor quality loss. • A parallel implementation achieves nearly optimal speedup on multi-core processors. • Application to two river catchments shows the impact of spatial interaction. [ABSTRACT FROM AUTHOR]

This study emphasizes the importance of accurate calibration in sediment transport models and highlights the transformative role of artificial intelligence (AI), specifically machine learning, in improving accuracy and computational efficiency. Extensive experiments were carried out in the Riba-Roja reservoir, which is located in the northeastern Iberian Peninsula. The accumulated sediment volume (ASV) curve was used to calibrate these experiments. The optimal ASV curve was found to be very close to the experimental data, with only minor differences in upstream areas. The results revealed a consistent rate of sediment transport and settling. Furthermore, the study investigated the capabilities of deep neural networks (DNNs) in predicting ASV curves and observing variable performance. In essence, the study highlights AI's potential for enhancing sediment transport models. • The study underscores the value of AI in enhancing sediment transport model calibration. • The simulations revealed accurate results, with minor discrepancies in upstream regions. • The DNNs, demonstrated reliable predictive capabilities in forecasting ASV curves. • AI, coupled with Monte Carlo method, reduces sediment model computation time. [ABSTRACT FROM AUTHOR]

Literature review reveals that many erosion models have limitations for application to constructed landforms. The new EroCA model aims to address this gap. The model uses established erosion and sediment transport models applied to multiple particle size classes at fine spatial and temporal resolutions, covering both suspended load and bedload. Using data from an experimental plot in Northern Australia, global sensitivity analysis was used to investigate the sources of uncertainty, limitations of the model, and priorities for data collection. Further analysis explored sensitivity to model grid size. The results showed that the Manning roughness, erodibility parameters and particle size are the most important parameters. Results also showed low parameter identifiability and hence the difficulty of calibration, even when using high quality experimental data. EroCA has the capacity to be a powerful landform modelling tool at appropriate scales of application; however ideally its data needs would be considered during monitoring design. • In-depth attempt to develop and assess a high spatial resolution process-based erosion model for a mined landform. • Improved understanding of the effects of various parameters on soil erosion process and modelling. • New insight into the limitations of experimental plot data sets for supporting high resolution erosion models. [ABSTRACT FROM AUTHOR]