Gap Filling of Daily Weather Data using Spatial Interpolation Techniques and Neural Network Methods

Chang, Y. and Burningham, H., 2024. Gap filling of daily weather data using spatial interpolation techniques and neural network methods. In: Phillips, M.R.; Al-Naemi, S., and Duarte, C.M. (eds.), Coastlines under Global Change: Proceedings from the International Coastal Symposium (ICS) 2024 (Doha, Qatar). Journal of Coastal Research, Special Issue No. 113, pp. 463-467. Charlotte (North Carolina), ISSN 0749-0208. Access to climatological datasets covering long time periods is critical in coastal and hydrological studies. However, almost all instrumental time series, in particular the daily records, are affected by some amount of missing values (data gaps). Pre-processing of raw data sets by filling data gaps is thus a routine procedure. In this study, the capabilities of six gap-filling techniques were analysed in estimating missing daily temperature and rainfall data for Sheskinmore, a coastal dune site located in northwest Ireland, using observations at two regional Met Éireann stations. Four spatial interpolation gap-filling approaches - modified normal ratio method (MNR), inverse distance weighting (IDW), correlation coefficient weighing (CCW) and multiple linear regression (MLR), and two different structure of Artificial Neural Networks (ANNs) - multilayer perceptron (MLP) and general regression neural network (GRNN) were evaluated. Pearson's correlation coefficient (R), mean absolute error (MAE), root mean square error (RMSE) and skill score (SS) were applied to evaluate their performance. Results show MLR outperformed other spatial interpolation techniques, with the lowest MAE, RMSE and the highest SS. The two ANN models notably outperformed all spatial interpolation techniques, with GRNN delivering the most accurate and efficient prediction of missing daily temperature and precipitation data.