Using trend line and principal component analysis to study vegetation changes in Senegal 1986—1999 from AVHRR NDVI 8 km data

Abstract

The objective of the study is to evaluate existing methods and techniques for analysing time series of remotely sensed vegetation index data. This was done within the framework of assessing long-term changes in the vegetation of Senegal. West Africa, using time series of NDVI (Normalized Difference Vegetation Index) data dating from 1986 to 1999. NDVI was derived from NOAA AVHRR 8 km data provided through the Pathfinder program. Two NDVI integrals were computed for each year, one covering the entire growing season to represent savannah biomass production, as well as an NDVI integral for the month of September, where NDVI has proven to be highly correlated with cereal crop yields. The existing literature identifies two appropriate methods. The temporal trend of annual values of integrals of NDVI was computed for each pixel. This yields a slope and an intercept value for each pixel, where the slopes indicated the long term trends in vegetative productivity. The slopes and the annual integrated NDVI values can be correlated to rainfall and biomass data. It appeared that the trends captured the general development in the vegetation in response to changing rainfall; however, it was equally apparent that variation in rainfall could not explain all the variation in integrated NDVI. Standard principal components (PC) were computed and analysed for the same study area with respect to rainfall, biomass and the original variation in i me grated NDVI. Changes in vegetation productivity were identified in the PCs; these interpretations are, however, not unambiguous. PCs tend to reflect specific events such as fires, drought periods and agricultural encroachment, rather than depicting the general development during the period. PC2 for the agricultural areas and PC3 and PC5 for the rangelands showed change patterns comparable to the respective slope images. Ii was concluded that the trend method was best suited for operational assessment of the state of the environment of Senegal. The approach is simple and reproducible, easy to interpret, and less sensitive to variations in time period and study area compared to the PC method.