Abstract

INEGI recognizes the significance of studying Mexico's agricultural zones, particularly in light of the financial and budgetary challenges posed by conducting agricultural censuses. The delineation and identification of lands with agricultural activity is a complex task, and remote sensing has emerged as a crucial instrument for this purpose. Especially when traditional methods face economic and logistical barriers, the amalgamation of conventional techniques with innovative data science approaches is of paramount importance.

Utilizing Landsat imagery and the Mexican Geospatial Data Cube, data from the 2007 Agricultural Census (updated to 2019) were integrated with state-of-the-art remote sensing methodologies. Embracing GEOBIA has been instrumental, as it processes images on a national scale at a 30-meter resolution, resulting in over 9 million distinct polygons.

Moreover, SENTINEL-2's remote sensing capabilities have recorded impressive advancements in accuracy, reaching up to 91.24% in select regions when employing the "random forest" algorithm. This method notably surpasses the Landsat approach, attributed to the enhanced resolution of SENTINEL-2.

Highlighting both studies is the emphasis on open-source Python tools and avant-garde techniques such as the geometric median and GEOBIA. When paired with high-resolution images and contemporary databases, these resources provide an in-depth and updated depiction of the agricultural regions in Mexico.