I wish to determine the most important environmental covariates controlling a species distribution.

I have a large collection of species presence data for a small survey area (e.g., 600 presence points for a 800 m survey area).

However, my environmental covariates are sourced from medium-resolution satellite data (e.g. 1 km cells).

This means that all of my 600 species presence points fall into a single 1 km cell of temperature, which will of course introduce a large bias.

Are there any approaches that deal with the issue of a high density of presence data within a small area being modeled against broadscale environmental data?


Getting temperature estimates at finer resolution than 1 km is not likely to happen without you placing your own temperature loggers around your study site. You're going to have to look into using different explanatory variables. If you hope to be able to determine any drivers of distribution in an area that small, especially for wildlife (vs. plants because they don't move!), you will also need to look for higher resolution data. Even then, I'm skeptical you will find anything just using presence data from such a small area.

But to answer your questions, I would recommend checking out Planet Research and Education account where you can download some amount of high resolution (3 m) imagery for free. If you are affiliated with a university or other research institution I would start there. Next, Sentinel-2 can provide data at ranges from 10 m to 30 m. Finally, I would also recommend Landsat for 30m resolution.

Since temperature estimates at resolution finer than what you've already found are not likely to be available (without considerable uncertainty in those estimates), you will need to focus on other drivers. Using Landsat you can get elevation and some derivatives such as slope, aspect, and wetness indices. Using Sentinel-2 and Planet imagery I would recommend looking into vegetation indices such as NDVI, NDRE, and CLRE. If you have other habitat information, even in vector form, such as water and vegetation cover type you can include those in your model as well.


The issue you have is that your data covers a very small area, such that you couldn't possibly make any inference about the species response to different climate variables.

Were there any other attributes collected with the species data relating to physical habitat? This could be slope, elevation, soil type, vegetation type etc. You might be able to find this from spatial data held by government natural resource management agencies. They are often available online depending where you are.

The resulting habitat model would then only really be valid for predicting into areas with similar climate and habitat conditions.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.