Land Use and Land Cover Change Dynamics in Moatize District, Mozambique (2011–2023): Remote Sensing, Statistical Analysis, and Policy Metrics

Description

## 1. Overview This dataset contains all primary data, processing scripts, classified land cover maps, and derived statistical outputs generated for the study of land use and land cover (LULC) change in Moatize District, Tete Province, Mozambique, over the period 2011–2023. The district is subject to extensive large-scale coal mining operations (Moatize Coal Basin), which have produced significant landscape transformations affecting woodland ecosystems and rural communities. The dataset is intended to ensure full reproducibility of all results, figures, and tables reported in the associated manuscript. It is organised into five thematic folders described in Section 4 below.

Steps to reproduce

Land use and land cover (LULC) classification was performed in Google Earth Engine (GEE) using Landsat 7 ETM+ (2011) and Landsat 8 OLI/TIRS (2015, 2019, 2023) Surface Reflectance imagery (USGS Collection 2, Tier 1), filtered to annual composites using pixel-wise median reduction and a maximum cloud-cover threshold of 20–28%. For each epoch, seven spectral indices were computed (NDVI, NDBI, NDWI, SAVI, BSI) and combined with topographic variables (elevation and slope) derived from the JAXA ALOS DSM V3.2. A Random Forest classifier (100–300 trees depending on epoch) was trained on stratified ground-truth polygons representing seven LULC classes (Water bodies, Mining/Bare soil, Urban areas, Agriculture, Pasture/Open savanna, Mopane woodland, Dense Miombo woodland), using a 70/30 train/test split. Post-classification processing applied a majority filter (radius = 30 m) to eliminate isolated pixels smaller than 20 pixels. Accuracy was assessed via confusion matrix, overall accuracy, and Cohen's kappa coefficient. Classified maps were exported as GeoTIFF (30 m, EPSG:4326) to Google Drive. Complete GEE scripts for all four epochs are provided in folder 01_GEE_Scripts/, with public shareable links embedded in each script header. Statistical analysis was conducted in R version 4.3.1. The Compound Annual Rate of Change (CARC) was computed as CARC = [(Final/Initial)^(1/t) − 1] × 100, where t = 12 years. A Mann–Whitney U test compared CARC distributions between natural and anthropogenic classes; a Spearman rank correlation tested the association between CARC and initial class area. A sensitivity analysis applied ±15% area perturbations to assess robustness of CARC estimates. The Landscape Pressure Index was computed as PI = |CARC| × |ΔArea (km²)|. Change-intensity categories followed FAO (2016) thresholds. The full LULC transition matrix was derived from cross-tabulation of the 2011 and 2023 classified maps. All R scripts are provided in folder 02_R_Scripts/; package dependencies and versions are documented in Session_Info.txt. Scripts should be executed in the numbered order (01 → 05), with script 01 generating all intermediate CSV outputs required by subsequent scripts. Input tables are read from folder 04_Statistical_Tables/.

Categories

Funders

• Coordenação de Aperfeicoamento de Pessoal de Nível Superior

  Ministry of Education

  Federal District, Brasília

  Grant ID: Finance Code 001

Licence