Window Regression: A Spatial-Temporal Analysis to Estimate Pixels Classified as Low-Quality in MODIS NDVI Time Series

Abstract

MODerate resolution Imaging Spectroradiometer (MODIS) data are largely used in multitemporal analysis of various Earth-related phenomena, such as vegetation phenology, land use/land cover change, deforestation monitoring, and time series analysis. In general, the MODIS products used to undertake multitemporal analysis are composite mosaics of the best pixels over a certain period of time. However, it is common to find bad pixels in the composition that affect the time series analysis. We present a filtering methodology that considers the pixel position (location in space) and time (position in the temporal data series) to define a new value for the bad pixel. This methodology, called Window Regression (WR), estimates the value of the point of interest, based on the regression analysis of the data selected by a spatial-temporal window. The spatial window is represented by eight pixels neighboring the pixel under evaluation, and the temporal window selects a set of dates close to the date of interest (either earlier or later). Intensities of noises were simulated over time and space, using the MOD13Q1 product. The method presented and other techniques (4253H twice, Mean Value Iteration (MVI) and Savitzky–Golay) were evaluated using the Mean Absolute Percentage Error (MAPE) and Akaike Information Criteria (AIC). The tests revealed the consistently superior performance of the Window Regression approach to estimate new Normalized Difference Vegetation Index (NDVI) values irrespective of the intensity of the noise simulated.

Keywords

• data quality
• spatial-temporal window
• noise reduction
• MODIS
• time series