ESTIMATION OF LAND COVER CHANGE USING LANDSAT SATELLITE IMAGERY AND THE RANDOM FOREST CLASSIFIER

The quantification of land cover change facilitates natural resource management. In general, these changes are determined at large scales but less frequently at a regional level. In this research, changes in land area covered by vegetation (V), agricultural use (A), grassland (G), and urban-rural (U) were estimated for the period 2002–2021 in the municipality of Huehuetla, in the Northern Sierra of Puebla, Mexico (39.5 km2), based on Landsat satellite images and the random forest classifier (RF). The latter was trained and evaluated with two datasets consisting of three spectral bands (red, green, and blue) and seven vegetation indices. RF performed well in classifying the four cover types at the beginning and end of the evaluation period. RF obtained an overall correct classification accuracy of 92.5 % in 2002 and 92.3 % in 2021. At the land cover level, RF identified vegetation cover with an F1 score of 100 % in 2002 and 98.2 % in 2021; however, it identified the urban-rural cover less effectively, with an F1 of 71.5 % in 2002 and 81.8 % in 2021. In the period analyzed, the urban area increased from 1.7 to 6.4 % (an increase of 4.7 % of the total area), and the vegetation area from 48.1 to 68.7 % (an increase of 15.6 %), at the expense of a reduction in the grassland area (19.8 %), while the agricultural area remained stable (reduction of 0.5 %). This study illustrates the importance of using machine learning techniques and satellite images to assess land cover changes at the regional level as a viable and low-cost alternative.