WILDLABS logoWildlabs.net : The conservation technology network

WILDLABS logoWildlabs.net : The conservation technology network

discussion / AI for Conservation  / 6 August 2024

Spatiotemporal species distribution modeling

Navodita Mathur

Navodita Mathur

Data scientist with expertise in Computer vision and remote sensing and a passion to work for the environment. 

Groups

Hi,

I have a deep passion to contribute in the field of ecology, biodiversity and AI. I have a tech background and about a year of experience in AI. My current focus is on species distribution modeling. While there have been attempts to model using machine learning. Most of the papers seems to ignore the temporal aspect or associated phenological changes. Could someone please shed some light on this? How do I go about modeling this? Is there someone who has experience in this or  wants to collaborate?

Navodita



Will Armstrong
@Will_Armstrong

6 August 2024 4:09pm

Hey, 

You're correct, traditionally species distribution models are spatial only, but here are a few ideas for how to incorporate time into your modelling: 

  1. Categorise the time dimension (E.G. Month/Season/Year, wet season/dry season etc) and use it as an input variable. 
  2. Incorporate phenological data (flowering time, migration period etc) into the input variables.
  3. Use time series data and use an appropriate model to take this into account such as LSTM.

A few things to bear in mind:

  1. We need the time of the X data to match that of the y data. This constrains the data that we can use. I'd recommend using Google Earth Engine to get the X data as you can query it for specific date ranges and they have a wide range of relevant datasets. 
  2. Whichever variables you use as input also have to be used for prediction.
  3. Any categorised temporal variables you choose have to be applicable to all species being modelled.

I hope this helps! 

-Will