XGBoost: Accurate high-resolution vegetation mapping

Authors demonstrate scalable tree boosting method for producing multiclass high-resolution vegetation maps

Data resources used via GBIF : Occurrences from 11,234 vegetation survey plots
Podocarpus totara
Tōtara (Podocarpus totara) observed in Laingholm, New Zealand by jacqui-nz. Photo via iNaturalist (CC BY-NC 4.0)

Mapping of ecosystems and vegetation types is important for environmental planning and resource management, but producing accurate, high-resolution maps is a challenge due to mosaic and heterogeneous environments.

This paper presents a workflow to produce high-resolution multiclass vegetation maps based on a machine learning algorithm for regression and classification referred to as XGBoost—or extreme gradient boosting.

To demonstrate the workflow, researchers used two strikingly different cases. First, they surveyed the Dzungarian Basin in China by photographic sampling covering more than 3,000 points 1-2 km apart. From the photos, they identified plant species and developed a hierarchical vegetation classification system.

In the second case, they gathered GBIF-mediated vegetation survey data for New Zealand. The authors supplemented both datasets with an extensive suite of bioclimatic data and remote sensing data.

In both cases, the authors used the XGBoost approach to produce high-resolution vegetation maps, effectively separating vegetation classes with higher accuracy than other mapping projects.

Original article

Zhang H, Eziz A, Xiao J, Tao S, Wang S, Tang Z, Zhu J and Fang J (2019) High-Resolution Vegetation Mapping Using eXtreme Gradient Boosting Based on Extensive Features. Remote Sensing. MDPI AG 11(12): 1505. Available at: https://doi.org/10.3390/rs11121505