Before people pepper me with answers about how I should go download the Hydro1k and HydroSheds data layers and use them, I should also preface our conversation by saying I already have these and am using them or trying to use them. But I would like to find ways to extract more information from these and the bioclimatic data for aquatic species layers. And there are also limitations associated with these data layers, which we can also discuss. Thanks.

Hi Justin,

I agree with you, we are lacking global datasets for freshwater habitats and I feel that they are bit ignored compared to terrestrial habitats.

I am really interested to know if you managed to find ANY data layers for aquatic habitats.

Thanks in advance

Ursula

Hi

I am interested to know if aquatic(freshwater) and marine layers in niche modelings are different?! 

As an update to this question, I'd like to point out that there are some recently published papers that 1) demonstrate that it remains possible to predict the distributions of freshwater fishes (and other obligate freshwater taxa) using ENM approaches such as MaxEnt (e.g. Bagley et al. 2013; Cao et al. 2013; Quinn et al. 2014; Campbell and Hildebrand 2017; and several others), and 2) demonstrate that bioclimatic variables work well as a proxy of stream habitat variables (e.g. McGarvey et al. 2017).

I am particularly encouraged by the McGarvey et al. (2017) paper in Ecography, which shows that at the scale of the Columbia River watershed the performance of building MaxEnt SDMs from climate covariates/layers as proxies for variation in in-stream "environments" inhabited by fishes (which in reality include a variety of in-stream variables such as hydrological parameters) is essentially equivalent to that of SDMs based on "instream covariates" (hydrological data layers).

One common theme in some recent papers is to model only within the hydrological network, at high spatial resolution (e.g. 30 arc-seconds); however, while results from this approach are easily and beautifully visualized over smaller spatial scales (e.g. within states, regions), in my experience doing this at larger spatial scales (e.g. subcontinental scales) gives results that are much harder to visualize. At larger scales, it may be best 1) to use a discrete symbology for displaying prediction/habitat suitability within river networks, if they are used; 2) to model at the HUC rather than network scale (e.g. Cao et al. 2013); or 3) simply not to mask to hydrological network at all (make continuous plots and interpret them in the context of hydrology; Bagley et al. 2013; Quinn et al. 2014).

REFERENCES

Bagley, J. C., Sandel, M., Travis, J., de Lourdes Lozano-Vilano, M., & Johnson, J. B. (2013). Paleoclimatic modeling and phylogeography of least killifish, Heterandria formosa: insights into Pleistocene expansion-contraction dynamics and evolutionary history of North American Coastal Plain freshwater biota. BMC evolutionary biology, 13(1), 223.

Campbell, C. A., & Hilderbrand, R. H. (2017). Using maximum entropy to predict suitable habitat for the endangered dwarf wedgemussel in the Maryland Coastal Plain. Aquatic Conservation: Marine and Freshwater Ecosystems, 27(2), 462-475.

Cao, Y., DeWalt, R. E., Robinson, J. L., Tweddale, T., Hinz, L., & Pessino, M. (2013). Using Maxent to model the historic distributions of stonefly species in Illinois streams: the effects of regularization and threshold selections. Ecological Modelling, 259, 30-39.

McGarvey, D. J., Menon, M., Woods, T., Tassone, S., Reese, J., Vergamini, M., & Kellogg, E. (2017). On the use of climate covariates in aquatic species distribution models: are we at risk of throwing out the baby with the bath water? Ecography.

Quinn, A., Gallardo, B., & Aldridge, D. C. (2014). Quantifying the ecological niche overlap between two interacting invasive species: the zebra mussel (Dreissena polymorpha) and the quagga mussel (Dreissena rostriformis bugensis). Aquatic Conservation: Marine and Freshwater Ecosystems, 24(3), 324-337.

Hi Justin,

Just stumbling onto this question you posed years ago, but I see you've written about it more recently. I am also interested in modeling fish distributions within streams. I'm attaching a manuscript I published from my dissertation in which I used Maxent and the NHDplusV2 stream segment polylines as my base layer, using the COMID attribute to identify individual segments and link segment-level and catchment-level covariates. Both the NHDplusV2 datasets and the StreamCat datasbase have a large number of covariates "pre-snapped" to the COMID's of the stream segments. In Maxent lingo, this is the samples-with-data (SWD) format. To me, the stream segment level of modeling makes good sense for stream fishes - no need to rasterize it (and if we did, it may do some funky things by altering the model grain?).

The problem I'm running into now is that many of the newer tools to evaluate Maxent model complexity and fit (e.g., calculations of AICc) do not seem to be compatible with this approach. The terrestrial realm commonly uses raster files of covariates, so all the new R packages (e.g., ENMeval or MaxentVariableSelection) to perform these analyses require raster grids to calculate the model evaluation metrics. The only ways forward I'm seeing (aside from developing a new set of tools) is to force my data into the raster realm (despite the potential dangers of changing grain), somehow "trick" R into reading my data files as rasters, or abandon the stream-segment level of modeling for a HUC-based model that can be more easily depicted with rasters.

I'm curious if you've run into similar issues or have any thoughts on a path forward. I'm about to ask a similar question on RG regarding this conundrum.

Thanks,

Andrew

Article Incorporating fragmentation and non-native species into dist...