Scientists are turning the powers of AI and Big Data towards preserving wildlife

Experts in artificial intelligence and animal ecology are pooling their efforts into a pioneering approach to conservation

February 09, 2022

A team of scientists from MPI-AB and other institutes have put forth a new, cross-disciplinary approach intended to enhance research on wildlife species and make more effective use of the vast amounts of data now being collected thanks to new technology. Their study appears today in Nature Communications.

Republished from EPFL

The field  of  animal  ecology  has  entered  the  era  of  big data  and  the Internet  of  Things.  Unprecedented amounts  of  data  are now  being  collected  on  wildlife  populations,  thanks  to  sophisticated  technology such as  satellites,  drones  and terrestrial  devices  like  automatic  cameras and  sensors placed  on  animals or  in  their  surroundings.  These data  have become so  easy  to  acquire and  share  that  they  have shortened  distances  and  time requirements  for  researchers  while  minimizing  the disrupting  presence of  humans  in  natural  habitats.  Today,  a  variety  of  AI  programs are  available  to  analyze  large  datasets, but  they’re often  general  in  nature  and  ill-suited  to  observing the exact  behavior  and  appearance of wild animals.  A team  of  scientists  from  EPFL  and  other  universities  has  outlined  a  pioneering approach to  resolve  that  problem  and  develop  more  accurate  models  by  combining  advances  in  computer  vision with  the  expertise of  ecologists. Their  findings,  which  appear  today  in  Nature  Communications,  open up new  perspectives  on  the  use  of  AI  to  help  preserve  wildlife  species.

Building  up  cross-disciplinary  know-how  

Wildlife  research  has  gone  from  local  to  global.  Modern  technology  now  offers  revolutionary new  ways to  produce  more  accurate  estimates  of  wildlife  populations,  better  understand  animal  behavior, combat  poaching and  halt  the  decline  in  biodiversity.  Ecologists  can  use  AI,  and  more  specifically computer  vision,  to  extract  key  features  from  images,  videos and  other  visual  forms of  data  in  order to  quickly  classify wildlife  species,  count  individual  animals,  and  glean  certain  information,  using large datasets. The  generic  programs  currently  used to  process  such  data  often  work  like black  boxes  and don’t  leverage the full  scope of  existing  knowledge about  the  animal  kingdom.  What’s more,  they’re hard  to  customize,  sometimes  suffer  from poor  quality  control,  and  are potentially  subject  to  ethical issues  related to  the  use  of  sensitive  data.  They also  contain several biases,  especially regional ones; for  example,  if  all  the  data  used  to  train  a  given  program  were collected  in  Europe,  the  program might not  be  suitable for  other  world  regions.

“We  wanted  to  get  more  researchers  interested  in  this  topic  and  pool their  efforts  so  as  to  move forward  in  this  emerging field. AI  can  serve  as  a  key  catalyst  in  wildlife  research  and  environmental protection  more  broadly,”  says Prof.  Devis Tuia,  the  head  of  EPFL’s Environmental  Computational Science  and  Earth  Observation  Laboratory  and  the  study’s  lead  author.  If  computer  scientists  want to reduce  the  margin  of  error  of  an  AI  program that’s  been  trained  to  recognize a  given  species,  for example,  they  need  to  be  able  to  draw  on  the  knowledge of  animal  ecologists. These  experts  can specify  which  characteristics  should  be  factored  into  the  program,  such  as  whether  a species  can survive  at  a  given  latitude,  whether  it’s  crucial  for  the  survival  of  another  species (such  as  through  a predator-prey relationship)  or  whether  the  species’  physiology changes  over  its  lifetime.  “For example, new machine learning algorithms can be used to automatically identify an animal such as using a zebra's unique stripe pattern, or in video their movement dynamics can be a signature of identity," says Prof. Mackenzie Mathis, the head of EPFL's Bertarelli Foundation Chair of Integrative Neuroscience and co-author of the study. "Here is where the merger of ecology and machine learning is key: the field biologist has immense domain knowledge about animal being studied, and us as machine learning researchers, our job is to work with them to build tools to find a solution."

Getting  the  word out  about  existing  initiatives

The  idea  of  forging stronger  ties  between  computer  vision  and  ecology  came  up  as  Tuia,  Mathis  and others  discussed their  research  challenges  at  various  conferences over  the  past  two  years.  They  saw that  such  collaboration could  be  extremely useful in preventing  certain wildlife  species  from  going extinct.  A handful of  initiatives  have  already  been rolled  out  in  this  direction; some  of  them  are  listed in  the  Nature  Communications  article.  For  instance,  Tuia  and  his  team  at  EPFL  have  developed a program  that  can  recognize animal  species  based  on  drone  images.  It  was  tested  recently  on a  seal population.  Meanwhile,  Mathis  and  her  colleagues  have  unveiled an  open-source  software  package called  DeepLabCut  that  allows scientists to  estimate  and  track  animal  poses with  remarkable  accuracy. It’s  already  been  downloaded  300,000  times.  DeepLabCut  was  designed  for  lab  animals  but  can  be used for  other  species  as  well.  Researchers  at  other  universities  have  developed  programs  too,  but  it’s hard  for  them to  share  their  discoveries  since  no  real  community  has  yet  been formed  in  this  area. Other  scientists  often  don’t  know  these  programs  exist  or  which one  would  be  best  for  their  specific research.

That  said,  initial  steps  towards  such  a  community  have been  taken  through  various  online forums. The Nature  Communications  article  aims  for  a  broader  audience,  however,  consisting  of  researchers  from around the world. “A  community is  steadily  taking  shape,”  says  Tuia. “So  far  we’ve used  word  of  mouth to  build  up an  initial network.  We  first  started two  years  ago  with the  people  who  are  now  the  article’s other  lead  authors: Benjamin  Kellenberger,  also  at  EPFL; Sara  Beery  at  Caltech in  the  US;  and Blair Costelloe at  the  Max Planck Institute in  Germany.”

Other Interesting Articles

Go to Editor View