Comparison of Artificial Neural Network and Regression Models for Filling Temporal Gaps of Meteorological Variables Time Series

Continuous meteorological variable time series are highly demanded for various climate related studies. Five statistical models were tested application of temporal gaps filling in surface air pressure, temperature, relative humidity, incoming solar radiation, net and soil temperature. A bilayer artificial neural network, linear regression, regression with interactions, the Gaussian process exponential rational quadratic kernel used to fill gaps. Models driven by continuous variables from ECMWF (European Centre Medium-range Weather Forecasts) ERA5-Land reanalysis. Raw reanalysis data not applicable characterization specific local weather conditions. The correlation coefficients (CC) between situ observations vary 0.61 (for wind direction) 0.99 atmospheric pressure). mean difference is high estimated at 3.2 °C temperature 3.5 hPa pressure. normalized root-mean-square error (NRMSE) 5–13%, except direction (NRMSE = 49%). bias correction improves matching all variables. model an based or bilayered network trained on significantly shifts raw toward observed values. NRMSE values reduce 2–11% variables, 22%). CC above 0.87, characteristics. suggested calibrated against can be applied gap-filling