The reason for what the modelers dub ‘no prediction’ is given in another paper attempting to predict the habitat of the Tsetse fly, which emphasizes that all environmental conditions will not have been captured by the model and that when the environment is too different, they prefer creating a category of ‘no prediction’:
“Mapped outputs record the similarity of each pixel in an entire set of satellite images to the satellite-determined environmental characteristics of the training set sites. Obviously for this to be successful the training set should have captured the entire range of conditions present throughout the area for which predictions will eventually be made. This is not always the case, and it is then preferable to identify in the output image a separate category of ‘no prediction’ for those areas where the environmental conditions are some specific minimum distance (in multivariate space) away from any of the training set clusters (Rogers and Robinson, Citation2004, p. 144. Emphasis added.).”
Accessed online on August 26 2024 at: https://www.tandfonline.com/doi/full/10.1080/09505431.2023.2291046#d1e910
WHEN COMPLEX IS AS SIMPLE AS IT GETS 