Enhancing Near-Real-Time Amazon Forest Monitoring Using GeoAI: A Case Study on Selective Logging — companion reproducibility data

Abstract

This record contains the companion data for the paper Enhancing Near-Real-Time Amazon Forest Monitoring Using GeoAI: A Case Study on Selective Logging. It is organised into seven categories: the pretraining corpus for the spatiotemporal masked-autoencoder foundation model; the labelled dataset the downstream model is trained and cross-validated on; small reference-data files: the interpreter annotations the soft labels are derived from (annotations.json), the alert view dates that close each monitoring period (its end_period), and the downstream model's cached out-of-fold scores; the foundation-model checkpoint; the downstream model's checkpoints (20 leave-one-out cross-validation folds); the visual-interpretation panels the annotators read when placing the reference dates, one per marked observation and per monitoring reference, with a manifest (added in version 1.1.0); and the outputs of the detector evaluation: the nested leave-one-MU-out evaluation of the detector grid search, the negative test on the 51 interpreter-certified logging-free observations never used in training, and the survey of DETER alerts inside each analysed period (added in version 1.1.0). Step-by-step reproduction instructions, including how to download and verify this record programmatically, are provided in the paper's GitHub repository, not in this description: https://github.com/EvandroCT/mdpi-nrt-forest-monitoring. CRediT author statement (verbatim, as published in the paper): Conceptualization, E.T.; methodology, E.T.; software, E.T.; validation, E.T.; formal analysis, E.T.; investigation, E.T.; data curation, E.T., S.N. and D.B.; writing—original draft preparation, E.T.; writing—review and editing, E.T., G.M., S.N., D.B., G.Q. and L.A.; visualization, E.T.; supervision, G.Q. and L.A. All authors have read and agreed to the published version of the manuscript. (E.T. = Evandro Taquary; G.M. = Guilherme Mataveli; S.N. = Sérgio Nogueira; D.B. = Daniel Braga; G.Q. = Gilberto Queiroz; L.A. = Luiz Aragão.) Attribution notices. Contains modified Copernicus Sentinel data [2024–2026] (applies to the HLS S30 content in this record, which is derived from Sentinel-2). DETER alert data (INPE — Instituto Nacional de Pesquisas Espaciais, distributed via the TerraBrasilis platform) are licensed under CC BY-SA 4.0 and underlie the disturbance-site selection and labels used throughout this record and the alert survey of the detector evaluation; consistent with the ShareAlike clause, this record is itself distributed under CC BY-SA 4.0. Zarr stores in this record are named by their originating DETER alert identifier (FID), which is unrelated to the manuscript's MU1–MU20 numbering of the 20 cross-validated monitoring units used in the paper's own text and tables; Table S1 of the paper maps one to the other.