What question did this study set out to answer?

The aim is to improve aboveground biomass estimation in the Borana rangelands using a multi-sensor approach.

May 9, 2026Environmental Challenges0 citationsOpen Access

Quantifying Rangeland Aboveground Biomass Dynamics Using Data Fusion and Machine Learning Framework in the Borana rangelands of Ethiopia

BGBerhan GessesseGeoSAS (Ethiopia)ZCZerihun ChereDire Dawa University TMTalema MogedGeoSAS (Ethiopia)

Key Points

The aim is to improve aboveground biomass estimation in the Borana rangelands using a multi-sensor approach.
Collected field AGB measurements from 93 plots during the 2025 growing season.
Integrated Sentinel-1 and Sentinel-2 remote sensing data with terrain covariates.
Applied a Random Forest model in Google Earth Engine to analyze the data.
Achieved R² = 0.80 for the model's predictive capability.
Identified key predictors including textural features, NDVI_RE2, and SAR-derived variables.
Mapped AGB showed lower biomass in drought-prone areas and higher in moist, elevated regions.

Abstract

Accurate and spatially explicit estimation of aboveground biomass (AGB) is vital for assessing grazing resources, monitoring rangeland health and informing restoration in semi-arid ecosystems. In the Borana rangelands of southern Ethiopia, AGB estimation is particularly challenging due to high spatial heterogeneity, seasonal variability, woody encroachment and the limited integration of field observations with remote sensing data. In addition, reliance on single-sensor Earth Observation (EO) approaches often fails to capture the complexity of vegetation structure and moisture dynamics. To address these limitations, this study develops a multi-sensor data fusion and machine-learning framework that leverages the complementary strengths of different EO data sources. Field AGB measurements were collected from 93 systematically distributed plots (30 m × 30 m) during the 2025 growing season (March–June). These data were integrated with Sentinel-2 optical imagery, which captures vegetation spectral properties; Sentinel-1 SAR data, which provides information on vegetation structure and moisture independent of cloud cover; and DEM-derived terrain covariates, which explain topographic influences on biomass distribution. A Random Forest model implemented in Google Earth Engine demonstrated strong predictive capability (R² = 0. 80), with textural features, red-edge vegetation indices (NDVIRE2, SLAVI) and SAR-derived variables, particularly RVI and the VV/VH ratio, emerged as the most influential predictors. The resulting AGB map revealed pronounced spatial variability, with lower biomass in drought-prone, woody-encroached plains and higher biomass in relatively moist, elevated areas. Overall, the study highlights how multi-sensor integration substantially improves AGB estimation and supports more informed rangeland management and restoration planning in other geographic contexts similar to the Borana rangelands.

Ask AI

Helpful

Bookmark

View Full Paper