Zhang, GD; Zhou, HM; Wang, CJ; Xue, HZ; Wang, JD; Wan, HW (2020). Forecasting Time Series Albedo Using NARnet Based on EEMD Decomposition. IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 58(5), 3544-3557.

Land surface albedo analysis and prediction are of great significance for global energy budget research and global change forecasting. Research has been performed on time series albedo analysis but seldom attempt was performed on land surface albedo prediction. This article develops an effective method for land surface albedo prediction from Moderate-Resolution Imaging Spectroradiometer (MODIS) time series albedo data (MCD43A3). It consists of time series data decomposing and time series data forecasting. The ensemble empirical mode decomposition (EEMD) method decomposes the MODIS historical time series albedo data into several intrinsic mode functions (IMFs) and one residual series, then the nonlinear autoregressive neural network (NARnet) method is used to forecast each IMF component and residue. The predictions of all IMFs and residue are summed to obtain a final forecast for the albedo series. The proposed method was performed on monthly and daily albedo prediction both in snow-free and snowy areas. The results showed that the forecast albedo consists of the MODIS albedo data well, with R-2 greater than 0.89 and RMSE less than 0.052 for snow-free areas. For snowy areas, the forecasting also performed well during snow cover periods, with R-2 greater than 0.76 and RMSE less than 0.076. For irregular change periods of snow falling and melting, it is hard to get very high prediction accuracy due to the irregular land surface change. For this problem, more land surface information should be introduced, or adjusting the model over time is necessary.