Development of Multi-Platform Open-Source Tools for Phenological Analysis

Developing up-to-date technologies for understanding crop development is essential for improving agricultural monitoring and reducing uncertainties in production estimates. This study presents a proof of concept for multi-platform, open-source tools designed to extract phenological metrics from Sentinel-2 Normalized Difference Vegetation Index (NDVI) time series over maize cultivation areas in the Brazilian Cerrado. Three tools were developed: (i) QPhenoCrops, a QGIS plugin; (ii) PyPhenoCrops, a Jupyter Notebook-based processing chain; and (iii) GEEPhenoCrops, a Google Earth Engine script. The workflows were standardized using mean and median aggregation, along with Savitzky–Golay and Double Moving Average (DMA3 and DMA5) smoothing methods. They were tested under two maximum cloud-cover thresholds: 10% and 60%. All environments successfully reconstructed maize temporal dynamics and extracted phenological metrics. Validation against field observations showed greater stability for the Start of Season than for the End of Season across processing configurations. Under the 60% cloud-cover threshold, the QGIS STAC workflow with DMA5 achieved differences of only +1 day for the Start of Season and +6 days for the End of Season relative to sowing and harvest dates, respectively. Under the 10% threshold, the best combination of aggregation and smoothing methods was Mean and DMA5 in QGIS Local mode, resulting in differences of −1 and −14 days for SOS and EOS, respectively. Overall, the results demonstrate the feasibility of reproducible, multi-platform phenological analysis while highlighting the influence of data availability and processing configuration on metric extraction.