Skip to main content
Dryad

Future sea level rise in northwest Mexico is projected to decrease the distribution and habitat quality of the endangered Calidris canutus roselaari (Red Knot)

Cite this dataset

Garcia-Walther, Julian (2024). Future sea level rise in northwest Mexico is projected to decrease the distribution and habitat quality of the endangered Calidris canutus roselaari (Red Knot) [Dataset]. Dryad. https://doi.org/10.5061/dryad.9cnp5hqsr

Abstract

Sea level rise (SLR) is one of the most unequivocal consequences of climate change, yet the implications for shorebirds and their coastal habitats is not well understood, especially outside of the north temperate zone. Here, we show that by the year 2050, SLR has the potential to cause significant habitat loss and reduce the quality of the remaining coastal wetlands in Northwest Mexico—one of the most important regions for Nearctic breeding migratory shorebirds. Specifically, we used species distribution modelling and a moderate SLR static inundation scenario to assess the effects of future SLR on coastal wetlands in Northwest Mexico and the potential distribution of Calidris canutus roselaari (Red Knot), a threatened long-distance migratory shorebird. Our results suggest that under a moderate SLR scenario, 55% of the current coastal wetland extent in northwest Mexico will be at risk of permanent submergence by 2050, and the high-quality habitat areas that remain will be 20% less suitable for C. c. roselaari. What is more, 8 out of the 10 wetlands currently supporting the largest numbers of C. c. roselaari are predicted to lose — on average — 17.8% of their highly suitable habitat areas, with two sites completely losing all their highly suitable habitat. In combination with increasing levels of coastal development and anthropogenic disturbance in Northwest Mexico, these predicted changes suggest that the potential future distribution of C. c. roselaari (and other shorebirds) will likely contract, exacerbating their ongoing population declines. Our results also make clear that SLR will likely have profound effects on ecosystems outside the north temperate zones, providing a clarion call to natural resource managers. Urgent action is required to begin securing sufficient space to accommodate the natural capacity of wetlands to migrate inland and implement local-scale solutions that strengthen the resilience of wetlands and human populations to SLR.

README: Data to model potential and future distribution of red knots under a 2050 SLR scenario

https://doi.org/10.5061/dryad.9cnp5hqsr

README

The analysis are divided in three broad sections:

Species distribution modelling.

The R script to create the current and future habitat suitability of red knots is called “SpeciesDistributionModelling.R”.

The location data of C. canutus roselaari from which the habitat suitability analysis were derived is called 1._Red_knot_locations.csv. To protect endangered species, coordinate decimal points were rounded up to two decimals.

The environmental predictor rasters used to model the current and future habitat suitability under a sea level scenario are described in the table below:

Environmental predictor rasters Description File name
Elevation Global elevation in meters Elev.tif and Elev2050.tif
Temperature Mean December and January temperature Temp.tif and Temp2050.tif
Distance to wetland Euclidean distance to the nearest polygon of coastal wetlands DistWet.tif and Distwet2050.tif
Distance to urban areas Euclidean distance to the nearest polygon of  urban areas DistUrban.tif and DistUrban2050.tif
Land cover and vegetation MCD12Q1 MODIS/Terra+Aqua Land Cover Global 500m Veg.tif and Veg2050.tif
Nightlights Proxy for disturbance and development intensity NightL.tif and NightL2050.tif
Silt Global soil silt content 0-5 cm depth Soilsilt.tif and Soilsilt2050.tif
Clay Global soil clay content 0-5 cm depth Soilclay.tif and Soilclay2050.tif
Soil Ph Global soil pH x 10 in H20 0-5 cm depth Soilph.tif and Soilph2050.tif
Soil Coarseness Global coarse fragments 0-5 cm depth SoilCoar.tif and SoilCoar2050.tif

The final outputs of this analysis are described below:

  • A model ensemble of the current Red Knot Distribution. The values of this raster file represent habitat suitability ranging from 0 (not suitable) to 1 (very suitable).
  • A model ensemble Ensemble of the Future 2050 Red Knot Distribution. The values of this raster file represent habitat suitability ranging from 0 (not suitable) to 1 (very suitable).
  • The variables of importance for the Current and Future distribution to create Figure 3C.

Data handling in ArcGis Pro

The model ensembles rasters created using the species distribution modelling were loaded into ArcGis pro. We used the "raster to polygon tool" to convert these rasters to polygon. Then, we performed an “Intersection” between ensembles of red knot distribution and the wetland inventory (file WetlandInventory2.shp). This allowed us to isolated only the pixels found within the coastal wetland polygons for the current and future distribution of Red knots. After this process, each pixel polygon and group of pixel polygons within a wetland would have a unique identifier, an area calculation in hectares and a associated habitat suitability.

Then, we extracted the centroid of each coastal wetland and obtained its latitude and longitude. We appended the wetland centroid coordinates to each pixel polygon within each wetland.

Finally, we categorized each pixel according to the region (Pacific, Gulf or Mainland) where they were found.

The resulting file is called: WetlandSuitabilityTodayFuture.csv

The structure of WetlandSuitabilityTodayFuture.csv is explained below:

Column Description
OID_ Unique identifier made during the intersection
ID Unique ID of each wetland
ID_region Unique ID of each wetland within a region
Region Region where the wetland is located
TodayArea The current extent of the wetland
FutureArea The future extent of the wetland under a SLR scenario
HabLoss The amount of habitat potentially lost to SLR by 2050
HabRemaining The amount of habitat potentially remaining due to SLR by 2050
Name The name of the wetland (incomplete)
x Longitude centroid coordinate
y Latitude centroid coordinate

Data analysis and figures in R

  • The following figures use the R script DataAnalysisAndFigures.R and the datasets “WetlandSuitabilityTodayFuture.csv” and “TodayandFutureSuitabilityV2.csv”
  • Figure 2 uses “WetlandSuitabilityTodayFuture.csv”
  • Figure 4 requires to merge, the WetlandSuitabilityTodayFuture.csv with TodayandFutureSuitabilityV2.csv—a dataset that includes the raw habitat suitability values of each single pixel on a given wetland from the current and 2050 scenario. The merging process is included in the script.
  •  Figure 3 is a multi-panel. Panel A and B are the ensembles from the species distribution modelling. Panel C is made using the variables of importance. Panel D is a raster subtraction between the ensembles. The code is found in the script.
  • Figure 5 is made by filtering dataset by the 10 priority coastal wetland in NW Mexico for red knots and by habitat suitability > 0.6. The median habitat suitability is then calculated for each of these wetlands now and in the future.

The file structure for WetlandSuitabilityTodayFuture.csv is explained below:

Column Description
OBJECT ID Unique identifier created in ArcGis for each row
ID Unique identifier for each coastal wetland
ID2 Identifier for each pixel polygon
Name Name of each coastal wetland (incomplete)
Scenario Today (current habitat suitability); Future (2050 habitat suitability)
Suitability Values ranging from 0 (not suitable) to 1 (highly suitable)
PixelHa Pixel size in hectares

Usage notes

R is required to run the following scripts:

  • SpeciesDistributionModelling.R
  • DataAnalysisAndFigures.R

ArcGis Pro or other GIS software is recommended but not needed to visualize and/or perform intersections.

Any GIS software and R can be used to open, visualize and analyze .tif files and shapefiles. WetlandInventory2 file requires three files with the extension .shp, shx and .dbf located in the same workspace or folder to be opened and visualized using GIS.

Excel or R can be used to visualize .csv files

Methods

Study area and model species

C. c. roselaari are long-distance migratory shorebirds that breed in Alaska (USA) and on Wrangel Island (Russia) during the boreal summer and migrate along the Pacific coast to spend their nonbreeding season in NW Mexico (Carmona et al., 2013). Their nonbreeding habitats are restricted to beaches, coastal lagoons, and deltas spanning 32 - 21° N, although incidental records have been recorded further south. The subspecies has an estimated population size of just 21,700 individuals (Lyons et al., 2016) and is thought to be declining, leading them to be listed as threatened and endangered in Canada and Mexico, respectively (COSEWIC, 2007; SEMARNAT, 2010). Due to their small population size, restricted range, high site fidelity, and specialist habitat requirements, C. c. roselaari are expected to be highly vulnerable to SLR and can serve as an umbrella species for other shorebirds that also spend the nonbreeding season in NW Mexico (Muñoz-Salas et al., 2023).

Our study area extended from 32 - 21° N and included the states of Sonora, Sinaloa, and Nayarit, which we have termed the “mainland region,” as well as the Baja California Peninsula, which we have called the “peninsular region.” C. c. roselaari nonbreeding distribution is divided into three separate climatic zones: Mediterranean, in the northwestern tip of Baja California; arid, along the rest of Baja California and in western Sonora; and humid/dry tropical throughout the rest of the study area (Vidal-Zepeda, 2005). Except for Nayarit and parts of Sinaloa, coastal wetlands have a discontinuous distribution along the coastline, with discrete, well-defined wetlands separated by large expanses of arid-xeric ecosystems.

Species Data Collection

We assembled a dataset to model the current potential distribution of C. c. roselaari using ‘presence’ locations collected from 2000 – 2020 in NW Mexico from three different sources. (1) We gathered C. c. roselaari sightings from within this time period using the eBird Basic Dataset (eBird, 2020). (2) We tracked the annual movements of 58 C. c. roselaari using 3-g GPS satellite transmitters (PinPoint 75 Argos; Lotek Inc.)  attached to the backs of adults with cyanoacrolyte glue at Grays Harbor, WA, during April–May 2017 and 2018. Tags recorded locations accurate to ±10 m every two days during the post-breeding period until early to mid-October when they fell off of molting birds or their batteries were depleted. (3) We carried out on-the-ground surveys from 7-15 December 2019 and 2-12 January 2020 along the coastlines of Sonora, Sinaloa, and Nayarit. We chose 110 survey locations following a stratified random sampling design that included sites within 4 km of the shoreline and the nearest road access. Sites were chosen using the land use and vegetation dataset for Mexico (INEGI, 2019), and consisted of habitats types that are known to be used by shorebirds. We conducted 5-minute point counts within 3 hours of high tide during which we counted every shorebird within a 400-m radius of a random site. Field surveys ensured that the C. c. roselaari nonbreeding distribution was well represented, helped reduce biases associated with eBird observations concentrated in popular birding areas, and added areas that could have been visited by tagged C. c. roselaari but not recorded after October — the point at which C. c. roselaari shed their satellite transmitters.

Presence data preparation

Records from eBird were manually inspected for positional uncertainty and only those that presented enough detail to ensure an accurate location were kept (n = 1,100). We filtered the transmitter locations by location class and kept only those in classes 3D and A3 (10- and <250-m accuracy, respectively; n = 980). Field surveys revealed 10 sites in NW Mexico where C. c. roselaari presence could be detected; each site was treated as an individual presence record. To account for the autocorrelation typical of transmitter datasets, and to reduce records close to each other, we pooled together the transmitter, eBird, and survey locations and, then, rarified the dataset by eliminating points closer together than 2 km. This procedure reduced our final dataset to a sample size of 112 presences.  

Environmental data preparation 

We selected 10 environmental predictor variables related to climatic conditions and the static physical environment that are thought to define the ecological niche of C. c. roselaari (Table 1). Climatic variables vary seasonally; we therefore selected mean values (e.g., mean temperature) from Dec and Jan, the months during which C. c. roselaari populations are thought to be most stable in their distribution across their non-breeding range (Carmona et al., 2013). In contrast, we included variables related to the static physical environment (e.g., elevation and distance to the nearest wetland) collected at any time of year within the past 10 years. All environmental predictors were transformed, cropped, and resampled to match their extents, projections, and resolutions (~0.8 x 0.8 km), respectively.  To reduce multicollinearity, layers with variance inflation factors (USDM package; Naimi et al., 2013) >12 were excluded (Table 1).  For instance, we included three metrics of the distance to the nearest wetland (Euclidean distance to intertidal habitat, Euclidean distance to the coastline, and Euclidean distance to the polygon of each coastal wetland) but only distance to the polygon of each coastal wetland was retained as all metrics were highly correlated and the latter provided the best results (Table 1).

Table 1. Environmental predictor variables used to develop species distribution models for C. c. roselaari in NW Mexico

Environmental predictor

Description

Data Source

Elevation

Global elevation in meters

WorldClim (Fick and Hijmans, 2017)

Temperature

Mean December and January temperature

WorldClim (Fick and Hijmans, 2017)

Distance to wetland

Euclidean distance to the nearest polygon of coastal wetlands

Own data

Distance to urban areas

Euclidean distance to the nearest polygon of  urban areas

Own data based on Inegi (2019)

Land cover and vegetation

MCD12Q1 MODIS/Terra+Aqua Land Cover Global 500m

Friedl & Sulla-Menashe (2019)

https://lpdaac.usgs.gov/products/mcd12q1v006/

Nightlights

Proxy for disturbance and development intensity

NOAA (2020); https://ngdc.noaa.gov/eog/download.html

Silt

Global soil silt content 0-5 cm depth

https://soilgrids.org/; Poggio et al. 2021)

Clay

Global soil clay content 0-5 cm depth

https://soilgrids.org/; Poggio et al. 2021)

Soil Ph

Global soil pH x 10 in H20 0-5 cm depth

https://soilgrids.org/; Poggio et al. 2021)

Soil Coarseness

Global coarse fragments 0-5 cm depth

https://soilgrids.org/; Poggio et al. 2021)

 

Sea level rise scenario for 2050 

To assess the inundation risk in NW Mexico in 2050, we used a static ‘bathtub’ model obtained from Climate Central (2020). The model is based on global-scale datasets for elevation, tide, and coastal flooding likelihoods for the year 2050. The model parameters included the CoastalDEM elevation dataset (v. 1.1; Kulp & Strauss, 2018), an RCP scenario of 4.5, and medium ‘luck’ based on mid-range results from the sea-level projection range (50th percentile). The output of this product is a spatial layer with areas identified as vulnerable to permanent SLR alone and to minor floods that may rise and fall slowly. While static models can overestimate the extent of inundation and ignore bio-geomorphological feedbacks known to buffer the effects of SLR (e.g., marsh migration; Klingbeil et al., 2021), they are better suited for large-scale inundation assessments where a lack of high-resolution data — such as in the Global South — and high computational costs limit the use of dynamic models. As such, the static models used here should be considered a “worst-case” scenario to identify areas deserving further investigation with dynamic models.

Coastal wetland inventory 

To map all available coastal wetlands in NW Mexico, we used the Land Use and Vegetation Series V habitat classification (INEGI, 2019)—the most comprehensive landcover dataset available for Mexico. We retained seven classes associated with coastal wetlands and C. c. roselaari habitat which included: halophilic xerophyte vegetation, hydrophilic halophilic vegetation, aquaculture, bare, water body, dune vegetation and mangrove. We clipped and retained all wetlands within 50 km of the coast. After visual inspection using satellite imagery, we further refined the perimeter of each wetland to include a buffer of 500 m above the high tide line. When wetlands had well-defined limits, the perimeter was manually digitized. When limits were ambiguous, such as for the wetlands spanning the coasts of Sinaloa and Nayarit, the extent of each wetland was defined by considering the watershed to which it belonged. These fine-scale perimeter delimitations included habitats that appeared suitable for C. c. roselaari (e.g., shallow water, intertidal mudflats, etc.) and excluded — as best as satellite imagery allowed — non-suitable habitats, such as deep water (e.g., estuarine tidal channels). Our inventory resulted in 296 polygons representing individual wetlands or fragments of continuous wetlands.

2.7 Ecological Niche Modelling

We followed the modelling framework of Naimi & Araújo (2016) to estimate the current and future ecological niche of C. c. roselaari under SLR. This involved three steps: processing (model fitting), post processing, and model transfer.

Processing and post-processing the data involved using our C. c. roselaari presence information to extract a data frame of values from our set of environmental predictor variables. With this data frame, we ran four replicates of three modelling algorithms within the SDM package (Naimi & Araújo, 2016) in the R Programming Environment (v3.6.2; R Core Development Team, 2021) that included generalized linear, Maxent, and MARS models and resulted in a set of 12 models. After fitting these models, we used a K-fold cross-validation (30% of observations set aside for testing) along with threshold-dependent (TSS) and independent statistics such as area under the curve (AUC) and correlation coefficients (COR) for model evaluation. In presence-only studies, AUC measures the probability that a model will rank a randomly selected site where the species is present higher than a random background site, also known as pseudo-absence (Phillips et al. 2006; Phillips and Dudik, 2008). In contrast, COR values range from 0 – 1 and measure whether large differences in prediction values correspond to a higher presence probability for a species (Phillips and Dudik, 2008). We used the variable of importance percentage as a measure to evaluate the role of each predictor variable in explaining C. c. roselaari distribution. We then used the predict function for each model, with the output of this process being a raster of the current and future habitat suitability probability for C. c. roselaari (hereafter, ‘habitat suitability’). Habitat suitability values in this framework range from 0 (not suitable) to 1 (highly suitable) and can be used as a proxy for a species’ potential distribution. Finally, we created an ensemble of all 12 model predictions by stacking all raster models and calculating a weighted average across all model values. This ensemble represented the current habitat suitability for C. c. roselaari.

To assess the habitat suitability for C. c. roselaari in the future, we used the SLR scenario for 2050 in combination with the current habitat suitability for C. c. roselaari to mask out those areas that are predicted to be inundated by 2050 and, thus, unavailable to C. c. roselaari. We then used this layer to train, process, and transfer a new model that would project future C. c. roselaari habitat suitability in 2050.

2.8 Effects of sea level rise on wetland habitat and C. c. roselaari distribution.

We conducted three analyses to determine whether our results supported our distributional contraction hypothesis and to identify areas with losses or gains in suitability. We defined a distributional contraction as a reduction in wetland extent by 2050 and first measured habitat contractions by calculating how much wetland area is at risk of permanent inundation. To do this, we subtracted areas from each wetland polygon predicted to be unavailable under a 2050 SLR scenario. Second, we mapped areas where C. c. roselaari habitat suitability was gained or lost by using a raster calculator that subtracted the future and current habitat suitability of C. c. roselaari. Third, we calculated a ‘highly suitable habitat’ (HSH) index for the current and future distribution of C. c. roselaari within each wetland. The HSH index consists of the median value of all pixels considered to be prime habitat (i.e., pixel values >0.6). Fourth, we quantified the percentage loss and gain of habitat suitability by 2050 per wetland polygon by assessing how much suitability — on average — a given wetland would gain or lose under a 2050 SLR scenario. We then took the HSH index for 10 of the most important wetlands for C. c. roselaari (based on historic counts at the sites) and assessed whether habitat quality at these wetlands is likely to increase or decrease by 2050 compared to the current baseline.

 REFERENCES

Carmona, R., N. Arce, V. Ayala-Perez, A. Hernández-Alvarez, J.B. Buchanan, L. J. Salzer, P. S. Tomkovich, J. A. Johnson, R. E. Gill, B. J. McCaffery, J. E. Lyons, L. J. Niles, and D. Newstead (2013). Red Knot Calidris canutus roselaari migration connectivity, abundance and non-breeding distribution along the Pacific coast of the Americas. Wader Study, 120(3), 168–180.

Climate Central (2020). Land projected to be below tideline in 2050. https://www.climatecentral.org/sea-level-rise

COSEWIC. (2007). COSEWIC Annual Report (p. 120). The Committee on the Status of Endangered Wildlife in Canada. https://wildlife-species.az.ec.gc.ca/species-risk-registry/virtual_sara/files//cosewic/car_COSEWIC_annual_report_2007_e.pdf

Klingbeil, B. T., J.B. Cohen, M.D. Correll, C.R. Field, T.P. Hodgman, A. I. Kovach, E. E. Lentz, B.J. Olsen, W.G. Shriver, W. A. Wiest and C.S. Elphick (2021). High uncertainty over the future of tidal marsh birds under current sea-level rise projections. Biodiversity and Conservation, 30(2), 431–443. https://doi.org/10.1007/s10531-020-02098-z

Kulp, S. A. and B. H. Strauss (2018). CoastalDEM: A global coastal digital elevation model improved from SRTM using a neural network. Remote Sensing of Environment, 206, 231–239. https://doi.org/10.1016/j.rse.2017.12.026

Lyons, J. E., W.L. Kendall, J. A. Royle, S. J. Converse, B. A. Andres and J.B. Buchanan (2016). Population size and stopover duration estimation using mark–resight data and Bayesian analysis of a superpopulation model. Biometrics, 72(1), 262–271. https://doi.org/10.1111/biom.12393

Muñoz-Salas, E. I., E. Palacios, L. Alfaro and M. E. Reiter (2023). Winter population trends and environmental drivers for three species of temperate shorebirds. Global Ecology and Conservation, 46, e02557. https://doi.org/10.1016/j.gecco.2023.e02557

Naimi, B. and M.B. Araújo (2016). Sdm: A reproducible and extensible R platform for species distribution modelling. Ecography, 39(4), 368–375. https://doi.org/10.1111/ecog.01881

R Core Team (2021). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.

Phillips, S. J., R. P. Anderson, and R. E. Schapire (2006). Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190(3-4), 231–259. https://doi.org/10.1016/j.ecolmodel.2005.03.026

Phillips, S. J. and M. Dudík (2008). Modeling of Species Distributions with Maxent: New Extensions and a Comprehensive Evaluation. Ecography, 31(2), 161–175. http://www.jstor.org/stable/30244563

SEMARNAT. (2010). NORMA Oficial Mexicana NOM-059-SEMARNAT-2010. https://www.profepa.gob.mx/innovaportal/file/435/1/NOM_059_SEMARNAT_2010.pdf

Vidal-Zepeda, R. (2005). Las regiones climáticas de México. 213. https://doi.org/10.14350/rig.30041

INEGI. (2019). Uso de suelo y vegetación 2016. https://www.inegi.org.mx/temas/usosuelo/#descargas

 

 

Funding

Consejo Nacional de Humanidades, Ciencias y Tecnologías