Simulating active layer temperature based on weather factors on the Qinghai–Tibetan Plateau using ANN and wavelet-ANN models

Active layer temperature (ALT) is an important dynamic attribute in characterization of permafrost change. Accurate simulation of the dynamic changes of ALT is essential for management and application of monitoring ALT data. This paper discusses the use of artificial neural network (ANN) and wavelet-ANN (W-ANN) hybrid models to simulate and forecast the ALT time series data based on five weather factors: air temperature, precipitation, wind speed, downward longwave radiation and downward shortwave radiation. Data are available for cold (FH1 site) and warm (CM2 site) permafrost locations on the Qinghai–Tibetan Plateau for the period 1996–2010. The ANN-based and W-ANN-based ALT models are developed using data from 1996 to 2007 and ALT forecasts are produced for the period 2008–2010 at both sites. The results demonstrate that ANN and W-ANN models can precisely simulate the ALT. The W-ANN hybrid model that uses decomposed sub-series as input provides forecasting results that are more accurate than the ANN model, which uses original time series. Moreover, the W-ANN-based ALT model is found more appropriate for modeling complicated physical relations between inputs and output. [Display omitted] •Active layer temperatures (ALT) are precisely simulated based on weather factors.•ANN and wavelet-ANN (W-ANN) models are applicable to the accurate simulation of the ALT.•The correlation coefficient (R), root mean square error (RMSE) and absolute error (AE) are used to evaluate the performance of the ANN-based and of the W-ANN-based ALT models.•Comparison of ANN and W-ANN hybrid models.•The W-ANN hybrid model demonstrates higher prediction accuracy and reliability.