11-Aggregating crop calendar knowledge and Sentinel-2/MSI big data for crop monitoring

In order to meet the growing international market demand for environmental compliance, the Brazilian agricultural sector needs to look for ways to monitor its production. In this sense, the demand for greater precision in crop monitoring grows - requiring analytical tools that allow accurate and timely analysis. Due to ground data scarcity or models that fail to adjust to field reality, many tools allow analysis only after the harvest. Time series analysis based on data cubes and ensembles of spectral indices, exploiting optimal temporal windows to improve the separability of classes, can improve this scenario. We address this gap in this mini-course, whose technical highlights are the detection of interannual changes in management and monitoring with crops still on the stands (intra-harvest monitoring approach). To answer how to detect subtle variations across harvest periods with high accuracy (improving monitoring initiatives), we developed an accessible and robust processing chain whose core is a dense time series analysis approach able to generate results with few training samples/calibration. It congregates steps from sample treatment to classification. During the mini-course, we will integrate crop knowledge and Sentinel-2/MSI data cubes to allow detailed mapping in inter-annual and intra-harvest periods. We will explain how to consider "time first, space later" in Jupyter notebooks, introducing concepts that prioritize the temporal aspect when finding the best alignment between time series. Also, we will present good practices in using field data and crop calendar knowledge to change methodological parameters useful for determining temporal windows, aiming to improve class separability and provide continuous monitoring. In dynamic agricultural contexts, this can improve the detection of subtle crop variations, enhancing the mapping of changes in crop practices, which is relevant for operational monitoring.