Crop Phenology Retrieval Through Gaussian Process Regression

Monitoring crop phenology significantly assists agricultural managing practices and plays an important role in crop yield predictions. Multi-temporal satellite-based observations allow analyzing vegetation seasonal dynamics over large areas by using vegetation indices or deriving biophysical variables. This study presents a framework for automatic corn phenology characterization based on high spatial and temporal resolution time series. By using the Difference Vegetation Index (DVI) estimated from Sentinel-2 data over Iowa (US), independent phenological models were optimized using Gaussian Processes regression. Their respective performances were assessed based on simulated phenological indicators estimated with double logistic approach. The results showed good model performances for the estimation of different phenological phases such as Start-of-Season and End-of-Season, with a mean RMSE and
$R^{2}$
values of about 3.50 days and 0.86, respectively, and this with a gain in runtime of about 380 times faster than the double logistic method. To the benefit of crop monitoring community, all these findings will be implemented into the freely downloadable GUI toolbox DATimeS (Decomposition and Analysis of Time Series Software - https://artmotoolbox.com/).