A SYSTEMATIC REVIEW OVER MACHINE AND DEEP LEARNING APPLICATIONS ON REMOTE SENSING FOREST CARBON MONITORING

Objective: Forest carbon monitoring has become critical for climate mitigation under international agreements, driving demand for accurate remote sensing-based assessment methods. This systematic review analyzes machine learning and deep learning applications to forest carbon monitoring using remote sensing data. Following systematic methodology, 59 studies were selected from Scopus, Web of Science, Google Scholar, and SciELO databases using structured Boolean search strategies. Results reveal exponential publication growth from 2019 onwards, correlating with the Paris Agreement implementation. Geographic distribution shows concentration in Asia, Brazil, and North America. Random Forest algorithms dominate all sensor types, followed by Support Vector Machine approaches. Local-scale studies using multispectral sensors represent the most frequent combination, reflecting validation requirements against forest inventories. Critical limitations include limited open-access carbon forest datasets, with only two freely available platforms identified. The trade-off between data accessibility and spatial resolution constrains operational implementation, while local-scale focus highlights challenges in scaling to global monitoring requirements. Machine learning techniques, particularly Random Forest, have matured into robust methodologies for forest carbon assessment, yet significant gaps remain in data standardization and global-scale application. Enhanced international collaboration and standardized data products are essential for bridging local research capabilities with global climate monitoring requirements.