Tree Species Map England

<p style='text-align:justify;'><span style='font-size:12.0pt; line-height:107%;'>The species map represents a predicted distribution of common tree species in England, produced using a time series of multispectral satellite remote sensing data (Sentinel-2) and machine learning. A classifier based on the XGBoost algorithm was trained to distinguish tree species, utilising a time-series of surface reflectance data and labelled training samples from the sub-compartment database (SCDB). To enhance classification performance, minority species with fewer than 1,000 training samples were grouped into broader categories, resulting in a total of 35 classes. Given the significant class imbalances, a sample weighting strategy was employed to guard against significant underfitting of the minority classes.</span></p> <p style='text-align:justify;'><span style='font-size:12.0pt; line-height:107%;'>Model evaluation demonstrated strong classification performance, with an overall accuracy of 89% and balanced class accuracy of 90%. Predictions were made at the pixel level and used to generate a <a>species classification and confidence raster</a> output. Field validation for Norway spruce within the Ips typographus demarcated area, confirmed a precision of 69%, aligning with test data results for this class. Additional validation using National Forest Inventory (NFI) data further reinforced model reliability, though accuracy was observed to be worse for underrepresented species. </span></p>