To derive training data, we used the CHM to identify the segments that are clearly woody (more than 90% covered by CHM values larger than 3 m; 58,063 segments) and those that are clearly nonwoody (less than 10% covered by CHM values larger than 0.5 m; 82,236 segments). The thresholds for defining woody areas were chosen for the following reasons. First, we wanted to have homogenous sets of segments as a training dataset; vegetation that is higher than 3 m is certainly woody in the New Zealand context and confusion with agricultural plants such as maize and vineyards as well as flax can be avoided. The second reason for the 90% threshold is that the CHM includes built-up areas. Segments greater than 1 ha are never completely covered by buildings; therefore, a coverage value larger than 90% makes sure that built-up areas are excluded. Thirdly, a temporal difference of 7 years between the LiDAR data used for training (2014) and the PALSAR data (2007) used for application of the classification model exists. To make sure that a patch of woody vegetation identified by the CHM already existed in 2007, the thresholds, therefore, have to consider regrowth. The fastest growing vegetation type in New Zealand is pine tree plantations. Pinus radiata plantations typically grow 6 to 10 m in their first 7 years on the North Island.26 However, canopy closure would not typically reach 90% except in those stands planted with exceptionally high stocking density. Therefore, the 90% threshold makes sure the woody vegetation already existed at the time the PALSAR data were recorded.