A tree checklist
We started with the checklist for the entire Panama flora published by D’Arcy (1987) and updated by Correa et al. (2004). They both provide a code indicating trees and shrubs. To update the list, we consulted first the Flora Mesoamericana, a set of volumes published by Missouri Botanic Garden over the past 25 years that reviews all plant species, grouped by family, throughout Central America. Unfortunately, only 49 of 141 families of Panama trees are completed to date. For the missing families, we consulted first the Manual de Plantas de Costa Rica, which is more complete but omits Panama species whose ranges do not include Costa Rica. Between those two major sources, 23 families of Panama trees are not covered (All chapters we consulted from those two multivolume works are cited in Appendix 1). Next, there is a Flora of Panama, published over four decades in separate articles, but all before 1980; we only consulted it for those 23 missing families. Beyond those large sources, we consulted many monographs and other taxonomic treatments of families, genera, or single species (The Flora of Panama and all other monographs we consulted are cited in Appendix 2). To establish what species are present in Panama, we relied on the description of geographic range given in monographs. In species for which no monograph asserted a range, we had the previous checklists (D’Arcy 1987; Correa et al. 2004) and then consulted specimen records from the large data sources described below.
We also checked the Panama tree list published by Beech et al. (2017) as the Global Tree Search (Botanic Gardens Conservation International 2019). When we accessed the list (2 Oct 2019), it included 2757 tree taxa in Panama. We rejected 374 as not valid in Panama but added 26 to our list that we had missed. It was missing over 700 species we recognize.
Definition of a tree
Our goal was to employ a rigorous definition of a tree that could allow precise comparison among regions. Published definitions, however, are inconsistent, using height cutoffs from 2-10 m, and vague, indicating that trees usually but not always have a single main trunk (see for example Little and Jones 1980; Allaby 1992; Western Australian Herbarium 1998; Pell and Angell 2016; Beech et al. 2017; Missouri Botanical Garden 2020). All definitions omit forms that are largely tropical, such are stilts, stranglers, or clonal palms. Gschwantner et al. (2009) also sought consistency for the purpose of national forest inventories and reviewed a range of definitions but ended up omitting any height and retaining the vague ’typically’ for a single trunk.
We have the additional goal of establishing a checklist of trees that matches species that would be included in our forest inventory plots in Panama, in which free-standing woody stems with diameter at breast height ≥ 1 cm are censused (Condit 1998b). To match this and to achieve rigor, we elected to omit the routine but always vague criterion of single versus multiple trunks. Instead, we set a strict height criterion of 3 m, a size which nearly always excludes herbs and corresponds approximately to the 1-cm dbh cutoff, but ignored multiple stems. We recognize that by ignoring stem number, species often considered shrubs are included in our checklist, but there is no alternative that allows consistency and precision, because too many species have multiple stems on some occasions. We also omitted epiphytes and lianas, neither of which are included in tree inventories. Most monographs provide for every species a maximum height, the presence of wood, and whether epiphytic or lianoid. When available, we accepted the assertions on growth form and maximum height as stated in Flora Mesoamericana, Manual de Costa Rica, or Flora of Panama. If the expert reported that a species is sometimes epiphytic (or scandent) and other times free-standing, we accepted it as a tree. Likewise, we included species described as herbs sometimes and woody others if tall enough.
This left 19% of the species not appearing in monographs. We already had the assertion as tree or shrub from the Panama checklists for many species, and we accepted those. In new species, not reported in Correa et al. (2004), we consulted individual specimen labels at the Missouri Botanical Garden website (www.tropicos.org/Home.aspx) to determine whether they qualify as trees. As soon as we found one record of a tree or shrub taller than 3 m, we accepted it. In those species, we do not report a maximum height, and they are omitted from analyses based on tree size.
Data sources for individual occurrences
Herbarium records as well as other observations (plots, inventories) provide coordinates of individual trees. We used two different large online data sources to gather such records. The first was the 2018 BIEN database (version 4.1, October 2018; Maitner et al. 2017), the second the Tropicos database from the Missouri Botanic Garden. The BIEN database includes Tropicos plus GBIF (Global Biodiversity Information Facility) and many other sources and thus has many more records, but the direct Tropicos download (http://services.tropicos.org) has more recent records and cleaner taxonomy. BIEN is plagued with errors in taxonomy and range, especially from GBIF and checklists, though of course Tropicos has errors as well.
The BIEN database has the advantage of allowing extraction of all records by country or province. We thus extracted all records in the 46 American countries appearing in BIEN. We excluded, however, all but the eight southernmost U.S. states for practical reasons, since the northern states include an enormous number of records but only 10 species that reach Panama (Canada was included, since there are few records; Puerto Rico appears as a country). From this extraction (18,065,850 records), we only made use of records for tree species in our checklist, then discarded records placed at centroids of countries and duplicated coordinates, leaving 1,008,245 unique species-coordinate records. Limiting the BIEN analysis to our checklist was necessary because the BIEN extraction includes non-trees and because of the many erroneous records in BIEN for Panama: mistakes in taxonomy, identification, location, plus non-native species.
Because BIEN provides a thorough list of plant records in Panama, it provided a tool for adding new species relative to Correa et al. (2004). We checked all BIEN species in genera that are mostly trees and added them to the checklist if we found valid occurrences in Panama. This led us to add 485 tree species, mostly in families not covered by Flora Mesoamericana. In the end, we either incorporated BIEN tree species into our checklist or decided they do not belong in Panama. There were 810 of the latter: tree species appearing in BIEN-Panama but for which there are no valid records in the country (listed in the supplementary data, Condit et al. 2019a). There must be additional erroneous BIEN records outside Panama, but unfortunately, it was impractical to screen those as thoroughly as records in Panama.
Data must be extracted from Tropicos via species names, not via country. We captured data for every name in our checklist, including every synonym we have. From those, we discarded all records outside the Americas, records placed at centroids of countries, and duplicated coordinates, leaving 335,350 unique species-coordinate records. Tropicos is updated often; our data come from a download on 5 Feb 2020.
A feature provided by BIEN but not Tropicos is a column indicating whether a species is native in the given country for every individual record. Unfortunately, this designation is wrong for many species in central Panama, flagging some well-known native species as exotic. After extensive screening, we concluded that it is often wrong in Central America, Colombia, and Venezuela, while it is helpful in Canada, the USA, Mexico, Ecuador, Peru, Bolivia, Brazil, Chile, and Argentina. Thus, we excluded BIEN records from range calculations when designated as non-native in the latter nine countries but not elsewhere.
Geographic ranges
For every species, we extracted unique latitude-longitude pairs then converted them to kilometers, assuming that a degree latitude = 110.9463 km and a degree longitude at the equator = 111.3195 km. Longitude was then corrected with the cosine of latitude, so that, for example, at 30∘ latitude, a degree longitude = 96.4055 km. We constructed the minimum convex polygon at which each species was observed, subtracting large bodies of water, and calculated its area. This is known as the EOO, or extent-of-occurrence, and is often reported for tree species (Gaston and Fuller 2009; Morin and Lechowicz 2013). The EOO requires few assumptions and is easy to calculate in poorly known species with few records. We are thus presenting the realized range of each species, as opposed to the potential range. We calculated the range extent separately from the two data sources, BIEN and Tropicos, and a third time after merging them. The two databases are not independent, but this offers some measure of uncertainty about range estimates.
Of the 3043 tree species we found native in Panama, 48 had no records in BIEN and eight had no records in Tropicos. Conveniently, those missing sets were non-overlapping, meaning that by combining the two sources we had at least one record with coordinates for every species.
Narrow geographic ranges
We were particularly interested in rare species, so sought to be as precise as possible about those with range extents <20×103 km2. As a check for consistency, we examined 50 species whose range was <20×103 km2 according to one source (BIEN or Tropicos) but >50×103 km2 according to the other, and we examined all their records using the Tropicos web site (http://www.tropicos.org/Home.aspx). In most, errors were easy to spot, and in 80% of the cases, the wider range was correct. We thus decided to focus on narrow-range species using the merged specimen records, BIEN plus Tropicos.
From the merged data, there were 47 species with < 3 locations. For those with two records, the pair was within 10 km with one exception (a likely error in Ardisia pulverulenta). Since no polygon could be drawn, we arbitrarily assigned all 47 of those species a range =10 km2. Because of the political importance of managing rare species, we further considered species endemic to Panama, i.e unknown outside the country.
Plots and inventories
Our own tree data were collected in two ways. Most come from plots, our main research effort in central Panama. Plots are precisely surveyed rectangles inside which every individual woody stem at least 1 cm in diameter was identified, measured, and mapped (Condit 1998b). The earliest plot was a 50-ha rectangle on Barro Colorado Island (Hubbell and Foster 1983; Condit et al. 2017); full data available at Condit et al. (2019c). Since then we have added 65 more plots, most 1 ha in area (Condit et al. 2002; Turner et al. 2018); full data available at Condit et al. (2019b). Our second method for surveying trees was an inventory, in which all species present in a small area were noted, but no individuals were counted or measured (Condit et al. 2013; Turner et al. 2018). All plots and inventories together comprise a list of tree species with the exact locations observed (at a precision of 2 m in plots, 500 m in inventories).
Plot occurrence and range
We thoroughly matched all taxonomy in the checklist and the plots, so it was straightforward to count plot and inventory occurrences for all species in the Panama checklist. We tested whether species found within plots differed in range size from non-plot species using a t-test after log-transformation.
Tree height, taxonomy, and range
Based on the maximum height recorded from monographs, we tested whether taller species had wider ranges, simultaneously estimating family differences in range size. We used untransformed height as the predictor of log(range size) in a multi-level regression in which family was a random effect; parameters were estimated using the Metropolis algorithm in a Bayesian hierarchical framework (Condit et al. 2007; Condit et al. 2017), running the algorithm for 10,000 steps and discarding the first 2,000. Convergence was checked visually. Statistical inference was based on credible intervals for each family for the estimated median range size at 10 m or 30 m tall, using 95th percentiles of 8,000 parameter values from the Markov chain. The regression was repeated with alternative range estimates (BIEN, Tropicos), but results barely differed and we report only that from merged range sizes.
Data available
A supplementary data archive shows additional results and complete data tables for download (Condit et al. 2019a). These include the full species list with range sizes, all synonyms we located for each name, plus the entire table of coordinates from both BIEN and Tropicos.