Field-validated species distribution model of Canada Warbler (Cardellina canadensis) in Northwestern Ontario

Description

The Canada Warbler (CAWA) is a species of conservation concern, but its ecological needs and distribution remain poorly understood. Additionally, contradictory findings exist regarding the impact of logging on CAWA abundance and habitat use. Furthermore, its habitat needs may be distorted by limitations in current habitat availability compared to historical conditions. We developed a predictive high-resolution (30 m) field-validated species distribution model (SDM) in Ecoregion 4W of Northwestern Ontario, Canada, where little field-derived information about the species is available. We aimed to assess how time since disturbance mainly due by logging affects CAWA occurrence and distribution and also the accuracy of the model by ground-truth validation. We used a desktop dataset from different sources, and due to limited number of observations (78 after filtered) we enhanced the dataset with field-collected data gathered in 2021 and 2022. We ran different models also to test the accuracy of the models using only desktop data and a datasete enhances with field-collected data. The SDM’s environmental covariates included a bare soil index (BASI), a normalized water index (NDWI) as an indicator of deciduos vegetation, an enhanced vegetation index (EVI), a digital elevation model (DEM), years since disturbance (DISTURB [usually by logging] 1-20 years since last disturbance happened, 21 value represent undisturbed or no disturbed more than 20 years ago), distance to mature coniferous forest (D_CONIF), tree canopy height (CAN) and distance to water (WATER) as indicator of riparian zones. The models that used field-collected data showed a moderate performance for both training and test data (AUC 0.7) while the model that used only desktop dataset showed a poor performance (AUC 0.6); NDWI, WATER, EVI and D_CONIF were the most influential covariates indicating high association of CAWA to deciduous vegetation, riparian areas, shrub cover and importance of coniferous stands. CAWA occurrence probability was high in undisturbed areas, but also it has a high predicted probability (>0.6) in areas within six years since disturbance; CAWA may take advantage of regenerated forest depending on shrub density and retention of old-growth forest structure ( CAWA had a high prediction of occurrence areas with canopies higher than 10m tall).

Steps to reproduce

We collected the Canada Warbler observations from different sources and also we collected our own data in 2021 and 2022 in the Ecoregion 4W of Noth Western Ontario. The SDM was built on occurrences from various large datasets (including eBird (2000-2021), Breeding Bird Survey (2000-2019), and Ontario Breeding Bird Atlas (2000-2005) and data from long-term songbird monitoring of Quetico Provincial Park (2014-2019), and we filtered the dataset excluding inaccurate coordinates, sites within 250 m, and sites located on the cloud range of Landsat images. We ended with 78 observations from 2000-2020, then we supplemented the dataset with our field collected data from the breeding season of 2021. We got a total of 122 observations from 2000 to 2021. We used Maxent (java) console, due it is good with scarce data and only-presence data. We used Landsat images from 2018 to extract spectral indexes to use as covariates and also we used other layers from the Maryland University Lab (Canopy Height and Forest loss layers). We did a correlation analysis from the covariates we aimed to use before running the models. We ran a first preliminary model use the desktop dataset of 2000-2020 for training data and 25% of it as test data (random seed) in 10 cross-validated replicates and we discarded a BASI due to lack of contribution to the model. For the following two models we integrated the BASI covariate. The second model used the dataset 2000-2021 and 25% of it as test data in 10 cross-validated replicates. For a final and field-validated model, use the enhanced dataset 2000-2021 as training data and the 2022 dataset as test data in 10 cross-validated replicates.

Institutions

Categories

Funding

Licence