Yes, do field exploration and obtain raw data from researchers or online. I would not use citizen science data because it is unstructured and can have pseudo-replication, species reporting bias and spatial bias and be non-representative

Dear Mojtaba,

You may use citizen science data as long as you are aware of the limitations of the dataset.

To reduce uncertainty, you can filter your data by binning them your into spatial cells and keeping only cells with a minimum number of observations.

You may also want to use a multimodel approach (eg. biomod2, in R, but more efficient with presence-absence data), or at least try the NPPEN model. It's a presence-only model that often performs better than MaxEnt at extrapolating probability of occurrences where you don't have any data, and provides some more information on the shape of the ecological niche of the species. This can be very useful to figure out discrepancies between the species' known distribution and model's predictions.

Best,

Here are two references among many others:

Article A new model to assess the probability of occurrence of a spe...

Article Climate-induced range shifts of the American jackknife clam ...

You raise an interesting issue. Across a number of our papers we have considered the possibility of sampling bias of incidental observational data obtained from citizin scientist or other types of observers (search term under my profile: bears/endemics).

Bird occurrence data may represent a special case as birding is a widespread hobby for a large number of persons in many countries. This usually implies a relative high number of observations. A bias risk presents the tendency that spotting of a rare bird is communicated through birder networks and incites other birders to travel to the spot resulting in the pseudo-replication/spatial autocorrelation. For spatially filtering such data you may have a look at the SDM toolbox/rarefying that can be integrated in ArcGIS. Any other method of removal of observation points close in space or time can be considered. In case of an abundance of data (maxent does not necessarily predicts better at observation numbers > 100), you may also thin/filter your data by random removal of observation points (e.g. our autumn data in the bear article).

Citizen science data can be highly unreliable if not cleaned up via removal of biased data, with apriori data integrity checks. Though usually relatively easier and cheap to acquire for a purpose, it should be used with extra-caution, taking context in consideration.