Code and data supporting: “Compositional turnover as an early warning signal of forest degradation and deforestation in Miombo-dominated landscapes”

<p dir="ltr">Code and data supporting: "Compositional turnover as an early warning signal of forest degradation and deforestation in Miombo-dominated landscapes of Lualaba Province (DR Congo), 2016–2025" (submitted to GIScience & Remote Sensing).</p><p dir="ltr">The deposit includes:</p><p dir="ltr">- Google Earth Engine, Python and R scripts for Sentinel-2 compositing, Random Forest classification and forest-mask generation, Multi-Output Neural Network (MONN) fractional-cover mapping, ecological-class and landscape-fragmentation analysis, degradation-definition sensitivity analysis, temporal-transferability and field validation, and the eight main figures.</p><p dir="ltr">- The reference dataset of 2,333 Collect Earth Online plots used to train the MONN (spectral predictors, vegetation indices, texture metrics, fractional-cover labels and coordinates).</p><p dir="ltr">- Tabular data underlying Table A1 and the field-validation data (60 plots) underlying Table C1.</p><p dir="ltr">- GeoTIFF rasters: the binary forest mask defining the analysed forest area, and the MONN-predicted fractional composition of dry dense forest, gallery forest and Miombo woodland for 2016, 2019, 2022 and 2025.</p><p dir="ltr">These materials are provided to support the transparency and reproducibility of the results presented in the manuscript.</p>