Spatiotemporal analysis of agricultural dynamics in a region of high production diversity using multisensor imagery and machine learning

Abstract

Remote sensing and geotechnologies are crucial for integrating science, technology, and geoinformation in agricultural production across different scales, supporting local and regional decision-making. Recently, techniques for fusion and harmonization of reflectances have been developed to enhance the spatial and temporal resolution of data. An example is the Harmonized Landsat Sentinel (HLS), a NASA initiative that provides high temporal resolution and medium spatial resolution data, advancing agricultural mapping in tropical regions. In this context, we propose a multisensor, multiscale, and multitemporal methodology to map the agricultural dynamics of the São João da Boa Vista microregion, encompassing 14 municipalities in São Paulo state. The study will focus on Caconde municipality, part of the AgroTechnological Districts (DATs) defined by the Center for Science in Digital Agriculture Development (CCD-AD SemeAr), and additionally, Casa Branca, which was included in the Mapping Agricultural Project in the Cerrado via Multi- Sensor Image Combination (MultiCER), also funded by FAPESP resources. The proposal is divided into eight stages, starting from: i) in situ data collection; ii) satellite image acquisition; iii) digital image processing; iv) structuring of a geospatial database; v) integration and analysis of spectral variables, field data, and census data from agricultural production systems; vi) generation of hierarchical models using machine learning (ML) and deep learning (DL) algorithms; vii) spatial and temporal transferability of models to other municipalities in the microregion and different crop years; and viii) analysis of agricultural dynamic processes such as expansion, retraction, conversion, intensification, and diversification. This doctoral research aims to contribute to the development of methodologies for mapping and monitoring agricultural regions with high dynamics, diversity, and spatial complexity, predominantly consisting of small and medium-sized rural properties characteristic of DATs in CCD-AD SemeAr Digital. Thus, the expected results could support decision-making processes involving DAT stakeholders, including farmers, cooperatives, associations, rural extension services, research groups, and municipal public agents. Additionally, the objective includes publishing scientific papers in high-impact journals and supporting the professional development of graduate students. (AU)