Monitoring Cropland Phenology on Google Earth Engine Using Gaussian Process Regression

Area of Science:

• Remote Sensing
• Agricultural Science
• Geospatial Analysis

Background:

• Monitoring cropland phenology is crucial for understanding crop dynamics but is hindered by cloud cover and atmospheric interference in optical satellite data.
• Existing methods struggle to provide continuous and reliable phenological data, necessitating advanced processing techniques.
• Cloud computing platforms offer potential solutions for large-scale, efficient satellite data analysis.

Purpose of the Study:

• To develop and implement an end-to-end processing chain on Google Earth Engine (GEE) for monitoring cropland phenology using Sentinel-2 (S2) data.
• To generate spatiotemporally continuous maps and time series of crop traits and derive Land Surface Phenology (LSP) metrics.
• To assess the performance of the proposed chain and its applicability for agricultural monitoring in a changing environment.

Main Methods:

• Developed hybrid Gaussian Process Regression (GPR) models for crop trait retrieval, optimized with active learning.
• Implemented these GPR models on the Google Earth Engine (GEE) platform for scalable processing.
• Utilized GPR gap-filling for reconstructing continuous time series and generating Land Surface Phenology (LSP) metrics (e.g., Start of Season, End of Season).

Main Results:

• Achieved good to high performance in estimating canopy-level traits like Leaf Area Index (LAI) and canopy chlorophyll content (NRMSE of 9% and 10%, respectively).
• Successfully reconstructed entire satellite data tiles using GPR gap-filling, demonstrating the capability for comprehensive mapping.
• Validated LSP metrics derived from crop traits against crop calendar data and Normalized Difference Vegetation Index (NDVI) phenology, confirming their accuracy and robustness.

Conclusions:

• The proposed Sentinel-2 (S2) phenology processing chain on Google Earth Engine (GEE) effectively overcomes cloud and atmospheric challenges for robust cropland phenology monitoring.
• The workflow enables the generation of accurate Land Surface Phenology (LSP) metrics and crop trait maps globally, representing a paradigm shift in satellite data processing.
• The derived LSP metrics offer valuable insights into crop seasonal patterns, supporting adaptive agricultural production in response to environmental changes.