Input data: Occurrences
To train the models, georeferenced and taxonomically verified
occurrences of C. mexicanus were obtained from a taxonomic study
of the genus (Guevara and Sánchez-Cordero 2018). Metadata for the
occurrences used in that study did not include the date of observation,
which is necessary to filter the occurrences points by date. Therefore,
I retrieved the capture date of each specimen from the Sistema
Nacional de Información sobre Biodiversidad de México (SNIB)
(Sánchez-Cordero and Guevara 2016) using the museum collection name and
catalog number. To compare models between the 30-year average and
different temporal resolutions (one, five, and ten years), occurrences
points collected between January 1971 and December 2000 were used in
model training, to match the months used to build the standard climatic
normal (Fig. 3a). Occurrences points were selected per cell per month
(with ”month” defined as any of the 360 months between 1971 and 2000; 30
years × 12 months). If two observations occurred in the same locality in
different months (e.g., November 2000 and December 2000), both
localities were retained for model training (see Supplemental material,
Table S1).
I used occurrences reported outside the temporal range of the model
training to evaluate the models. This second set consisted primarily of
occurrences with observation dates recorded before 1971 and after 2000
from two published studies (Sánchez-Cordero and Guevara 2016, Guevara
and Sánchez-Cordero 2018). Additionally, occurrences from recent
literature (Hernández-Flores and Rojas-Martínez 2010, Mayen-Zaragoza et
al. 2019) or recent fieldwork by colleagues and myself were incorporated
into this withheld dataset. To ensure that the withheld dataset did not
include occurrences near those used in model training, the withheld
occurrences needed to be at a distance greater than 10 km away from a
training occurrence (see Supplemental material, Table S2). Finally, a
spatial thinning of 10 km was applied to reduce the spatial bias of the
withheld data. Information on the environmental values assigned to the
withheld dataset is explained later in the model comparison section.