An efficient knowledge-based approach for random variation interpretation in NDVI time series

Nowadays, detecting and interpreting random variation extracted from satellite image time series is a far-reaching real-world issue. A more adequate approach should be designed to deal with this challenge. In this paper, we propose an efficient knowledge-based approach for vegetation monitoring using normalized derivation vegetation index time series. First, a decomposition process is designed to separate seasonal, trend and remainder components. Then, a genetic based schema is proposed to generate association rules. The extracted knowledge is intended to interpret the remainder component extracted during the previous phase by discovering the hidden link between random variation and climate observations. For validation purpose, a database covering the regions in Northwestern Tunisia is used for a period starting from 2000 to 2012. The data have been derived from Moderate resolution Imaging Spectroradiometer and fused with ground climate data (temperature and precipitation) in the form of fuzzy association rules. The obtained results show the efficiency of the proposed approach by reducing the non-stationarity effect.