I'm working in an urban environment where I need the MaxEnt output for a further analysis of habitat. A question arised when trying to model a species with broad distribution i.e. Ardea alba, and I was wondering if I should clip the WorldClim bioclimatic variables to the Study Area I have selected as well as the ocurrence data that I obtained from GBIF and then run MaxEnt, or if I should run the model with the bioclimatic variables for all the world and the occurences for all the world, and then clip it to my Study Area?
Basically, should I clip everything to my Study Area and then run the MaxEnt model or should I run the MaxEnt model with worldwide ocurrence and bioclimatic variables and then clip the resulting raster to my area?
I assume the correct answer is the second case, otherwise I'd be limiting the fundamental niche artificially, wouldn't I?
Thanks to everyone, I don't know if it is a silly question tho.
Dear Alejandro Antonio Ruiz-García
By clipping the layers before running the model, we are likely reducing the extent of the data being processed and potentially improving the accuracy and efficiency of the model's outputs.
Sadly, I believe research and analysis like yours will become redundant as AI takes over. I even think PhD (and other) theses will no longer be relevant or trustworthy with the use of programs such as Chat GPT :)
Lo ideal siempre es trabajar con un area especifica porque entre menos peso tenga las capas mas rapido sera el procesamiento del modelo.
Hi, this is actually an important question deserving a more in depth answer than I am giving you right now, but I will at least deliver key points. Also, there are many papers discussing these kind of issues better than I am doing here.
1) MaxEnt is a presence-background approach. As such, it also measures the models accuracy (the fit of the predictions) by calculating AUC. Now if your area of interest includes many areas that are in no way suitable for your species (water bodies if land mammal for example), then all presence points will be situated at land mass, while potentially quite a lot of the random background points (per default 10000 - you might want to consider changing this) will be put into water bodies for example. Then, the calculation will have an easy time differentiating between suitable and unsuitable habitat (the model doesnt know thats impossible for your species to be present here) and hence variable contribution might be skewed and AUC will be artifically high.
Take away: MaxEnt should be used to predict for areas only that could be potentially habitat - so clip all impossible areas.
2) The method creates a probability density function (pdf) for presence points and landscape each. Hence it tries to extract all information from the presence points, without having information of real absence. It is therefore extremley important to have a good presence points sampling. Modelling with presence points of for example a Spanish population and a Canadian population (assuming those two populations are quite different in their realised niche) will result in a fuzzy prediction, as the model cannot distinguish and will create its pdf (the SDM of the species) based on both populations alike.
Take away: be aware of the presence data and only use presence points from individuals that are of interest, in the area that is of interest for you. Of course you can make predictions for your area and some surroundings, if the presence points are more fitting this way.
Be aware of presence data from different seasons, or other factors. Your presence data should match the climate data... it doesnt make sense to let the model look for the habitat variables using presence data from summer habitat and environmental data from winter season for example, if your species is assumed to have differing habitat usage. This is especially true for citicen science data > you most likely need a lot of data cleaning and pruning (including systeamtic subsampling based on minimum distance and gaussian kerne density bias files at least, to at least handle sampling and detection bias).
3) Note about sampling size: clipping and pruning presence points will most likely result in lesser presence points. Dont worry. Due to its nature, MaxEnt is pretty good at making predictions with lesser sampling size. However, your sampling should represent the species habitat preferences, which means that generalists would need more samples than specialists.
4) Note about bioclim: I know bioclim data is used often times and even in some tutorials, but be aware that those serve the purpose of teaching rather than making best possible predictions. Usually, you would start by thinking about factors driving distribution of your species and then trying to represent those factors by variables. Especially in your urban area, bioclim might not be very informative, as probably your area shares climatic conditions (and bioclim is very limited in scale). Also, most climatic data are actually models strongly based on DEM, so just be aware of that (of course there are exceptions, like the microclima package f.e.).
Hope this helps!
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts