Multi-input model uncertainty analysis for long-range wind farm noise predictions

TL;DR

This paper quantifies the model structure uncertainty in long-range wind farm noise predictions caused by different ground impedance and wind profile models, highlighting significant variability at distances over 3.5 km.

Contribution

It systematically evaluates how multiple input models affect wind farm noise predictions, emphasizing the importance of accounting for model variability in uncertainty analysis.

Findings

Ground impedance and wind profile models significantly affect noise predictions.

Differences in wind speed profiles are the main source of long-range uncertainty.

Predicted noise levels can vary by over 10 dBA at distances beyond 3.5 km.

Abstract

One of the major sources of uncertainty in predictions of wind farm noise (WFN) reflect parametric and model structure uncertainty. The model structure uncertainty is a systematic uncertainty, which relates to uncertainty about the appropriate mathematical structure of the models. Here we quantified the model structure uncertainty in predicting WFN arising from multi-input models, including nine ground impedance and four wind speed profile models. We used a numerical ray tracing sound propagation model for predicting the noise level at different receivers. We found that variations between different ground impedance models and wind speed profile models were significant sources of uncertainty, and that these sources contributed to predicted noise level differences in excess of 10 dBA at distances greater than 3.5 km. We also found that differences between atmospheric vertical wind speed profile models were the main source of uncertainty in predicting WFN at long-range distances. When predicting WFN, it is important to acknowledge variability associated with different models as this contributes to the uncertainty of the predicted values.