Assessment of Forest Dieback in the Moroccan Central Plateau Using Remote Sensing and Machine Learning

Satellite remote sensing provides essential data for observing spatial and temporal forest trends and processes needed to map, quantify, and monitor forest health at regional and global scales. In this study, the Google Engine Earth (GEE) platform was used to extract NDVI, SAVI, and EVI from Landsat 8 OLI/TIRS satellite images for the period between 2015 and 2017 to assess the health of Sibara forest, Morocco. In addition, an assessment of the vulnerability of the forest to dieback was performed using the Random Forest (RF) machine learning classifier. On average, NDVI and SAVI were both highest in 2015 at 0.39 and 0.24 respectively, while they were lowest in 2016 at 0.32 and 0.20 respectively. NDVI was relatively well and poorly correlated with mean annual precipitation (pr) and mean temperature (tmean) with coefficients of 0.49 and -0.67, respectively. Our results confirmed field observations in 2018 of dieback, which appears to have occurred between 2016 and 2017. Monthly comparison of NDVI between the two years revealed that December had the greatest change, with plots 33, 40, 41, 70, and 71 recording NDVI changes of -38.64%, -38.49%, -38.37%, -38.01%, and -37.93%, respectively. Evaluation of the potential contributing role of ecological factors on dieback showed that substrate and stand density had an influence on dieback. Most plots with the greatest decline in NDVI fell under dense stands with granite-granodiorite substrate. The results of the dieback susceptibility assessment showed that 43.16%, 1.27%, and 55.57% of the Sibara forest are low, moderate, and high susceptible to dieback, respectively, with most of the high susceptible plots located in the lower half of the forest.