A Contextual Mann-Kendall Approach for the Assessment of Trend Significance in Image Time Series

One of the most common problems in estimating trends in image time series is the presence of contaminants such as clouds. There are many techniques for estimating robust trends but evaluating the significance of the trends can be difficult due to this increased variance. This article presents a novel approach called the Contextual Mann-Kendall (CMK) test for assessing significant trends. This test uses the principle of spatial autocorrelation to characterize geographical phenomena, according to which a pixel would not be expected to exhibit a radically different trend from neighboring pixels. The procedure removes serial correlation through a prewhitening process. Then, similar to the logic of the Regionally Averaged Mann-Kendall (RAMK) test, it combines the information from neighboring pixels while adjusting for cross-correlation. CMK was compared with the Mann-Kendall (MK) test in which contextual information was not involved for the mean annual NDVI over 22 years (1982-2003) in West Africa. With the MK test, -11% of the study area showed significant (p < 0.001) trends which increased to 16% when tested using the CMK test. Thus the CMK test produces a result that makes intuitive sense from a geographical perspective and enhances the ability to detect trends in relatively short time series.