*2.2. DNA Extraction and Sequencing*

All laboratory work was carried out at the Canadian Centre for DNA Barcoding (CCDB) (https://ccdb.ca/, accessed on 10 February 2022) and following their protocols. Total genomic DNA was extracted from silica dried leaf material using the CCDB protocol (https://ccdb.ca/site/wp-content/uploads/2016/09/CCDB\_DNA\_Extraction-Plants. pdf, accessed on 10 February 2022). DNA barcode regions *rbcLa*, *matK* and the *trnHpsbA* intergenic spacer were amplified using CCDB standard PCR primers and protocols (https://ccdb.ca/site/wp-content/uploads/2016/09/CCDB\_Amplification-Plants. pdf, accessed on 10 February 2022) with the primers available at https://ccdb.ca/site/wpcontent/uploads/2016/09/CCDB\_PrimerSets-Plants.pdf (accessed on 10 February 2022)

Voucher details and GenBank accession numbers for all sequences are listed in BOLD (http://www.boldsystems.org/) (accessed on 10 February 2022).

### *2.3. Testing the DNA Barcode Accuracy*

Prior to evaluating the identification success of the two barcode regions, we used the Basic Local Alignment Search Tool (BLAST) [22] to compare our sequences to those available in GenBank (https://www.ncbi.nlm.nih.gov/genbank/, accessed on 10 February 2022), with the aim of confirming our identifications and identifying our unknowns. After matching our sequences in GenBank, the morphospecies names were updated only after comparison of their voucher specimens to either the type specimens available online or to other herbarium specimens and photographs in Tropicos (https://www.tropicos.org/, accessed on 10 February 2022).

To test the barcode efficiency, we followed [18]. Our DNA barcoding refence database (assumed to be exhaustive in terms of species) had 274 individuals and was used to assign individual trees to species or genera. The test was performed on species represented by at least two individuals in the database, so that we could have a query and a matching sample. The coding genes *matK* and *rbcLa* were aligned and manually adjusted using ClustalW in the Molecular Evolutionary Genetic Analysis software version 7.0.26 [23]. After the global alignment, we computed pairwise genetic distances among all sequences in the dataset using the Kimura's 2-parameter model [24]. The analysis was also performed in Mega7. In the resulting matrix, each sample (query) was assigned to a species or a genus of the sample (matching) from which it is separated with the least genetic distance (excluding itself). The identification was (1) correct when the query sample matched only the samples of its species of genus; (2) multiple if the query sample matched several species or genera including its correct one; and (3) wrong when the query sample matched species or genera different from its own [18]. We were not able to align *trnH-psbA* because it was too variable among the 48 plant families in the study. Hence this locus was not used in the test of DNA barcode accuracy analyses.
