SEN4MOZ - Woody Cover Fraction & Active Cropland Maps for Maputo National Park and Gilé National Park, Mozambique (2019/20 - 2024/25)

General information: We here provide annual maps of woody vegetation fractions and active cropland, as well as derived products presented and analyzed in the corresponding paper "". The maps were produced using openly available Copernicus datasets in combination with machine learning classification and regression models. Please read the following information carefully before downloading and using the data. For remaining questions please contact [e-mail]. Woody Cover Fractions: Annual woody cover fraction maps (2019–2025) were produced using a spectral unmixing approach based on Sentinel-2 and Sentinel-1 spectral-temporal metrics (STMs) — statistical descriptors of seasonal land surface reflectance capturing central tendency, variability, and phenological dynamics. The time series was split into wet (October–March) and dry season (April–September) periods to align with regional climate cycles. Woody and non-woody endmembers were collected through visual interpretation of satellite imagery, with stable pure pixels selected across both national parks. Synthetic spectral mixtures (SynthMix; Okujeni et al., 2013) were generated from a 75/25 train-validation split to train and validate an XGBoost regression model, yielding annual per-pixel woody cover fraction estimates. Validation was based on synthetically mixed spectra withheld from training; maps were additionally compared against independently acquired LiDAR data covering the southeastern portion of Gilé National Park. Annual maps were summarized into mean woody cover and Theil-Sen trend slopes, providing a qualitative indication of vegetation gain or loss over the study period. Active Cropland Maps: Annual active cropland maps (2019/20–2024/25) were produced by classifying Sentinel-2 and Sentinel-1 STMs into three land cover classes — active cropland, other vegetation, and non-vegetation — using a Random Forest classifier (1,000 trees). Reference data were collected through visual interpretation of satellite imagery across both national parks and their surrounding zones, with additional purposive samples added to improve performance in underrepresented areas. Models were trained and validated using a 90/10 split. To improve time series stability, pixels mapped as active cropland in fewer than two years were excluded in post-processing. Annual maps were summarized into several derived metrics, including cumulative years under cultivation, cropland gain and loss relative to the start and end of the study period, first and last cultivation year, and cultivation duration. Accuracy: Average performance metrics for the woody cover fractions are mean error = -0.014 / 0.024, mean absolute error = 0.084 / 0.115, root mean squared error = 0.118 / 0.174 and R² = 0.875 / 0.711 for Maputo National Park / Gilé National Park. The active cropland maps had a mean overall accuracy = 0.84 / 0.80 for PNAM / PNAG and class-specific user´s and producer´s accuracy for active cropland were on average 0.88 / 0.78 and 0.77 / 0.70 for PNAM / PNAG. These maps are an approximation and contain errors. We expect that regional and broad patterns and trends are robust but local misclassifications are common and must be expected. Datasets: We provide the following datasets for the study regions surrounding PNAM / PNAG: woody_cover_fractions_pnax.tif: Stack of annual maps containing sub-pixel woody cover fractions scaled by 10,000. woody_cover_fractions_mean_pnax.tif: Average woody cover across the study period scaled by 10,000. woody_cover_fractions_slope_pnax.tif: Theil-Sen slopes representing annual woody cover change rates across the study period scaled by 10,000. active_cropland_pnax.tif: Stack of annual active cropland maps containing mapped land cover classes (1=active cropland, 2=natural vegetation, 3=non-vegetated surface). active_cropland_first_year_pnax.tif: The first year mapped as active cropland. active_cropland_last_year_pnax.tif: The last year mapped as active cropland. active_cropland_duration_pnax.tif: The difference between first year and last year mapped as active cropland. active_cropland_years_pnax.tif: The number of years mapped as active cropland, equal to duration in case cultivation took place every year, or where intermittent cultivaiton took place mapped years are fewer than duration. pa_zones.zip: Raster layers containing the zones and cell areas for downstream analyses. Code: The code for producing and analyzing the map products is available on GitHub: https://github.com/eichfussl/sen4moz Funding information: This data was developed in the context of SEN4MOZ - a bilateral research project between Universidade Eduardo Mondlane (Mozambique) and Humboldt-Universität zu Berlin (Germany). The project was funded by the EO Africa R&D Facility, a joint initiative of ESA and the African Union Commission under ESA Contract No. 4000133905/21/I-EF.