Assessment of Spatio-Temporal Empirical Forecasting Performance of Future Shoreline Positions

Coasts and coastlines in many parts of the world are highly dynamic in nature, where large changes in the shoreline position can occur due to natural and anthropogenic influences. The prediction of future shoreline positions is of great importance in the better planning and management of coastal areas. With an aim to assess the different methods of prediction, this study investigates the performance of future shoreline position predictions by quantifying how prediction performance varies depending on the time depths of input historical shoreline data and the time horizons of predicted shorelines. Multi-temporal Landsat imagery, from 1988 to 2021, was used to quantify the rates of shoreline movement for different time period. Predictions using the simple extrapolation of the end point rate (EPR), linear regression rate (LRR), weighted linear regression rate (WLR), and the Kalman filter method were used to predict future shoreline positions. Root mean square error (RMSE) was used to assess prediction accuracies. For time depth, our results revealed that the higher the number of shorelines used in calculating and predicting shoreline change rates the better predictive performance was yielded. For the time horizon, prediction accuracies were substantially higher for the immediate future years (138 m/year) compared to the more distant future (152 m/year). Our results also demonstrated that the forecast performance varied temporally and spatially by time period and region. Though the study area is located in coastal Bangladesh, this study has the potential for forecasting applications to other deltas and vulnerable shorelines globally.

Automated summary

This study aims to evaluate different methods for predicting future shoreline positions and quantifies prediction performance based on the time depths of historical shoreline data and the time horizons of predicted shorelines. The results show that using more shoreline data improves prediction performance, and the prediction accuracy is higher for the immediate future compared to the more distant future.