Estimating Aboveground Biomass in Grasslands Using Satellite Imagery
Author Information
Author(s): Netsianda Andisani, Mhangara Paidamwoyo
Primary Institution: University of the Witwatersrand
Hypothesis
Can machine learning algorithms effectively estimate aboveground biomass (AGB) in grassland ecosystems using Sentinel-2 satellite imagery?
Conclusion
The study successfully mapped aboveground biomass in grasslands, demonstrating that gradient boosting models outperformed random forest models.
Supporting Evidence
- Gradient boosting models achieved the highest R2 value of 0.7298 Mg/ha.
- Random forest models achieved an R2 value of 0.5755 Mg/ha.
- Sentinel-2-derived NDVI was the best-performing model with an R2 value of 0.6396 m2 m−2.
- The study area is characterized by flat terrain and an elevation of approximately 1600 m above sea level.
- AGB maps were produced through an integration of datasets from various sources.
- The study utilized 966 AGBD data points for model training and validation.
- Field-based LAI data was used to validate the accuracy of the AGB maps.
- The research highlights the importance of grassland ecosystems in carbon storage.
Takeaway
This study used satellite images to figure out how much grass is growing in a field, which helps us understand how much carbon is stored in the grass.
Methodology
The study used random forest and gradient boosting machine learning algorithms to estimate AGB from Sentinel-2 imagery and other datasets.
Potential Biases
Potential bias due to the reliance on specific datasets and the exclusion of certain vegetation types.
Limitations
The study faced challenges with geolocation faults in GEDI data and limited access to certain survey areas.
Statistical Information
P-Value
0.681
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website