CNN Models, metadata and global trait distribution maps

CNN models, metadata and global trait distribution maps.<br>Files included are:<br>la_inception_resnet_v2.hdf5 + la_xception.hdf5 + la_mobilenet_v2.hdf5: trained models for leaf area predictions<br>gh_inception_resnet_v2.hdf5 + gh_xception.hdf5 + gh_mobilenet_v2.hdf5: trained models for growth height predictions<br>sla_inception_resnet_v2.hdf5 + sla_xception.hdf5 + sla_mobilenet_v2.hdf5: trained models for specific leaf area predictions<br>nm_inception_resnet_v2.hdf5 + nm_xception.hdf5 + nm_mobilenet_v2.hdf5: trained models for leaf nitrogen content predictions<br>sm_inception_resnet_v2.hdf5 + sm_xception.hdf5 + sm_mobilenet_v2.hdf5: trained models for seed mass predictions<br>ssd_inception_resnet_v2.hdf5 + ssd_xception.hdf5 + ssd_mobilenet_v2.hdf5: trained models for stem specific density predictions<br>metadata.csv: <br>1. minimum and maximum values of training targets used for normalisation of reference values (can be used to reconvert predictions to original scale when applied to equation (1), see Methods)<br>2. minimum and maximum values of bioclimatic variables used for normalisation of auxiliary data (needed to predict trait values from photographs, see equation (1) in Methods).<br>The order of the bioclimatic variables in the .csv-file corresponds to the order of the auxiliary input to the model (see also R-script '7_train_CNN.R', accessible via https://github.com/ChrSchiller/cnn_traits)<br>gtdm_la.tif, gtdm_gh.tif, gtdm_sla.tif, gtdm_nm.tif, gtdm_sm.tif, gtdm_ssd.tif: global trait distribution maps for the six plant functional traits<br>gtdm_la_quantile_range.tif, gtdm_gh_quantile_range.tif, gtdm_sla_quantile_range.tif, gtdm_nm_quantile_range.tif, gtdm_sm_quantile_range.tif, gtdm_ssd_quantile_range.tif: global trait quantile range maps for the six plant functional traits<br>