Estimating Agricultural Land Consumption through Temporally Consistent Multi-Temporal Land Use/Land Cover Mapping: A 25-Year Landsat Analysis in the Semi-Arid Province of Meknes, Morocco

Agricultural land consumption by urban expansion is cumulative and largely irreversible, yet it is often inferred from net changes in mapped agricultural area, which can be misleading in semi-arid regions where crops, fallow land, and bare soil vary spectrally between dates. This study quantifies Agriculture-to-Urban conversion in Meknès Prefecture, Morocco, from 2000 to 2025. Six benchmark years were classified from Landsat Collection 2 Level-2 surface reflectance imagery in Google Earth Engine using a pixel-based Random Forest algorithm and seasonal, spectral, and topographic variables. Reference samples were derived from institutional vector data and visually interpreted polygons, with 1,200 stratified samples per year split into training and validation sets. Overall accuracies ranged from 87.5% to 92.9%, with Urban F1-scores of 0.949-0.976; because part of the reference labelling was rule-generated, these values are reported as internal-consistency diagnostics rather than independent estimates of map accuracy.

To improve temporal consistency, a post-classification correction was applied selectively to the Urban class using a cumulative persistence rule, while agricultural and bare-soil variability was preserved. Agricultural consumption was quantified through explicit Agriculture-to-Urban transitions rather than net agricultural area change, with class-level commission error propagated into an uncertainty range. Urban land increased from 8,101 to 15,761 ha. Of the land classified as agricultural in 2000, 6,122 ha became urban by 2025, representing 79.9% of net urban growth and an uncertainty-adjusted estimate of approximately 5,400-5,600 ha. This occurred despite a 12.2% increase in total mapped agricultural area. Conversion peaked during 2005-2010, while eight of twenty-one communes accounted for 85.6% of total conversion. The results demonstrate that transition-based indicators, combined with explicit uncertainty assessment, provide a more defensible basis for monitoring agricultural land consumption and supporting farmland protection than net agricultural area change alone.