Comparing temporal autocorrelation between NDVI-derived phenological metrics for Vermont forests using Sentinel-2 imagery
Simon, Garrett
Simon, Garrett
Citations
Altmetric:
License
License
Abstract
Remote sensing enables the quantification of phenological events at the landscape, as opposed to individual tree, scale. Existing satellite programs, such as Sentinel-2, have begun to accumulate enough annual forest imagery at suitable resolutions to conduct extended time-series analysis of forest phenology patterns. Studying annual shifts in key phenological events require time-series analysis where the standard regression assumptions of stationarity, or the independence of successive data points, is likely violated, described as autocorrelation. This study seeks to measure temporal autocorrelation of fitted start-of-season (SOS), peak Normalized Vegetation Density Index (NDVI), and end-of-season (EOS) values in a region of interest centered about Underhill, Vermont, as part of an exploratory analysis of phenological shifts in maple-beech-birch-dominated northern hardwoods forests. We utilize Google Earth Engine to perform Sentinel-2 image filtering for suitable forest pixels, NDVI calculation, and geospatial aggregations on a 500-m rectangular cell grid, the outputs of which are exported for local Python-based analysis. We utilize Savoy-Golitz smoothing methods and double-logistic function model fitting with parameters corresponding to SOS, maximum NDVI, and EOS. Analyzing per-cell time-series in our fitted values allows us to differentiate seasonality patterns from interannual trends. From our fitted parameters, we find a statistically significant negative first-order (one-year-lagged) autocorrelation in SOS and maximum NDVI from 2025 relative to the past seven years of available Sentinel-2 data. Significance for a one-year lag is validated through the Durbin-Watson statistic, while average autocorrelation from up to a seven year lag is calculated using the autocorrelation function and visualized through autocorrellograms. Although still negative, the insignificant finding for EOS is likely due to insufficiencies in the curve fitting method used and imagery limitations in the winter and spring. Our findings indicate non-stationarity of the data and imply the requirement to consider both annual and seasonal trends in northern hardwoods phenological analysis.
Description
Date
1/1/2026
Student Status
Graduate Student
Journal Title
Journal ISSN
Volume Title
Type of presentation
Poster
Collections
Research Projects
Organizational Units
Journal Issue
Citation
DOI
Advisor(s)
Department
Program/Major
Natural Resources PhD
College/School
Rubenstein School of Environment and Natural Resources
Organization
Research Category
Physical Science