Poggio, L; Gimona, A; Brown, I (2012). Spatio-temporal MODIS EVI gap filling under cloud cover: An example in Scotland. ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING, 72, 56-72.
Time series of satellite data have an important role in the monitoring of regional and global ecosystem properties. Satellite images often present missing data due to atmospheric aerosol, clouds or other atmospheric conditions. Most methods proposed to minimise the effects of degradation and to restore signal values do not take into account the spatial and temporal correlation of the values in the pixels. The aim of this study was to propose and test a spatio-temporal interpolation method to reconstruct pixel values in MODIS data time series that are missing due to cloud cover or other image noise. The method presented and tested is an example of a hybrid Generalised Additive Model (GAM)-geostatistical space-time model, including the fitting of a smoother spatio-temporal trend and a spatial component to account for local details supported by information in covariates. The method is not limited by the type of noise or degradation of pixels values, latitude, vegetation dynamics and land uses. The application of cloud masks on the target image provided the data for a quantitative validation through the comparison between the modelled EVI values and those from the MODIS product. The method was able to restore data providing very good to adequate responses in series of simulations of missing data. The comparison of distributions showed good agreement and predictive capabilities. The spatio-temporal method always performed better and the use of kriged residuals was helpful for situations with high percentages of missing data. The spatial pattern and the local features were well preserved for cloud coverage <= 20%. For higher percentages of missing data, the results were smoother with less local detail retained, but still showing the general spatial pattern of the variable. The method has proved to be flexible and able to provide reconstructed images reproducing spatial patterns and local features of the measured product, even with substantial amounts of missing pixels. (C) 2012 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS) Published by Elsevier B.V. All rights reserved.