centered log-ratio (clr) transformation and robust principal component analysis of long-term ndvi data reveal vegetation activity linked to climate processes

Predicting the future climate and its impacts on the global environment is model based, presenting a level of uncertainty. Alternative robust approaches of analyzing high volume climate data to reveal underlying regional and local trends are increasingly incorporating satellite data. This study uses a centered log-ratio (clr) transformation approach and robust principal component analysis (PCA), on a long-term Normalized Difference Vegetation Index (NDVI) dataset to test its applicability in analyzing large multi-temporal data, and potential to recognize important trends and patterns in regional climate. Twenty five years of NDVI data derived by Global Inventory Modeling and Mapping Studies (GIMMS) from 1982 to 2006 were extracted for 88 subwatersheds in central Kenya and statistically analyzed. Untransformed (raw) and clr transformed NDVI data were evaluated using robust PCA. The robust PCA compositional biplots of the clr transformed long-term NDVI data demonstrated the finest spatial-temporal display of observations identifying climate related events that impacted vegetation activity and observed variations in greenness. The responses were interpreted as normal conditions, El Niño Southern Oscillation (ENSO) events of El Niño and La Niña, and drought events known to influence the moisture level and precipitation patterns (high, low, normal) and therefore the level of vegetation greenness (NDVI value). More drought events (4) were observed between 1990 and 2006, a finding corroborated by several authors and linked to increasing climate variability. Results are remarkable, emphasizing the need for appropriate data transformation prior to PCA, dealing with huge complex datasets, to enhance pattern recognition and meaningful interpretation of results. Through improved analysis of past data, uncertainty is decreased in modeling future trends.