Spatial modelling and prediction of natural regeneration of Tetraclinis articulata using geomatics and machine learning in the Korifla Forest of Morocco

The natural regeneration of Tetraclinis articulata represents a vital role in the conservation of semi-arid Mediterranean forest ecosystems.This study aims to characterize and model its spatial dynamics in the forest of Korifla (Morocco) from 48 plots of land, described by 41 explanatory variables integrating spectral data Sentinel-2, topographic, climatic and pedological.Factor analysis of mixed data (AFDM) shows that the first two axes explain 46.4% of total variability, mainly according to soil-topographic and water-structural gradients.The Random Forest (RF) model performed best for shoots (AUC = 0.951) ) than for shoots regeneration (AUC = 0.803); for this reason, it was selected to map the probability of natural regeneration.Random Forest mapping, based on field measurements, reveals a predominance of regeneration by shoots, with a high potential (> 0.66) of 70.01%, compared to 22.40% for seedlings.Currently, the RF mapping indicates a very high potential (> 0.8) of 43.49% for regeneration by shoots, compared to only 2.63% for regeneration by seedling.The climate projection for 2040 based on the SSP2-4.5 scenario indicates a predominance of regeneration by shoots is high (> 0.6) over 82.91% of the area, while regeneration by seedlings is minimal (24.07%).Furthermore, projections based on changes in the NDVI and GCI indicate relative stability of the predicted probability of the regeneration potential between 2036 and 2046, For regeneration by seedling, the very high potential (> 0.8) remains almost stable, decreasing from 2.32% in 2036 to 2.43% in 2046.These results highlight the value of combining field data, remote sensing and machine learning to support the conservation and sustainable management of T. articulata.