New ma­chine learn­ing method pre­dicts ad­di­tions to global list of threat­ened plant species

Tehran Times - - SCIENCE -

The In­ter­na­tional Union for Con­ser­va­tion of Na­ture’s (IUCN) Red List of Threat­ened Species is a pow­er­ful tool for re­searchers and pol­i­cy­mak­ers work­ing to stem the tide of species loss across the globe. But adding even a sin­gle species to the list is no small task, de­mand­ing count­less hours of ex­pen­sive, rig­or­ous and highly spe­cial­ized re­search.

As a re­sult of these lim­i­ta­tions, a large num­ber of known species have not yet been for­mally as­sessed by the IUCN and ranked in one of five cat­e­gories, from least con­cern to crit­i­cally en­dan­gered.

A new method co-de­vel­oped by Anahí Espín­dola, an as­sis­tant pro­fes­sor of en­to­mol­ogy at the Univer­sity of Mary­land, uses the power of ma­chine learn­ing and open-ac­cess data to pre­dict species that could be el­i­gi­ble for at-risk sta­tus on the IUCN Red List. The re­search team cre­ated and trained a ma­chine learn­ing al­go­rithm to as­sess more than 150,000 species of plants from all cor­ners of the world, mak­ing their project among the largest as­sess­ments of con­ser­va­tion risk to date.

At-risk IUCN clas­si­fi­ca­tion

Ac­cord­ing to the re­sults, more than 10 per­cent of these species are highly likely to qual­ify for an at-risk IUCN clas­si­fi­ca­tion.

The al­go­rithm is a pre­dic­tive model that can be ap­plied to any group­ing of species at any scale, from the en­tire globe to a sin­gle city park. Espín­dola and her col­leagues pub­lished their find­ings on­line in the Pro­ceed­ings of the Na­tional Academy of Sciences on De­cem­ber 3, 2018.

“Our method isn’t meant to re­place for­mal as­sess­ments us­ing IUCN pro­to­cols. It’s a tool that can help pri­or­i­tize the process, by cal­cu­lat­ing the prob­a­bil­ity that a given species is at risk,” Espín­dola said. “Ul­ti­mately, we hope it will help gov­ern­ments and re­source man­agers de­cide where to devote their lim­ited re­sources for con­ser­va­tion. This could be es­pe­cially use­ful in re­gions that are un­der­stud­ied.”

Espín­dola and her col­lab­o­ra­tors built their pre­dic­tive model us­ing open-ac­cess data from the Global Bio­di­ver­sity In­for­ma­tion Fa­cil­ity (GBIF) and the TRY Plant Trait Data­base. Lead au­thor Tara Pel­letier, an as­sis­tant pro­fes­sor of bi­ol­ogy at Rad­ford Univer­sity, worked to­gether with Espín­dola to per­form the ma­chine learn­ing anal­y­sis.

The model’s ac­cu­racy

Espín­dola and Pel­letier then trained the model us­ing GBIF and TRY data from the rel­a­tively small group of plant species al­ready on the IUCN Red List. This al­lowed the re­searchers to as­sess and fine-tune the model’s ac­cu­racy by check­ing its pre­dic­tions against the listed species’ known IUCN risk sta­tus. The Red List sorts non-ex­tinct species into one of five clas­si­fi­ca­tion cat­e­gories: least con­cern, near-threat­ened, vul­ner­a­ble, en­dan­gered and crit­i­cally en­dan­gered.

The re­searchers then ap­plied the model to the many thou­sands of plant species that re­main un­listed by IUCN. Ac­cord­ing to the re­sults, more than 15,000 of the species--roughly 10 per­cent of the to­tal as­sessed by the team--have a high prob­a­bil­ity of qual­i­fy­ing as near-threat­ened, at a min­i­mum.

Espín­dola and her col­leagues mapped the data and noted sev­eral ma­jor ge­o­graph­i­cal trends in the model’s pre­dic­tions. At-risk species tended to clus­ter in ar­eas al­ready known for their high na­tive bio­di­ver­sity, such as the Cen­tral Amer­i­can rain­forests and south­west­ern Aus­tralia.

Newspapers in English

Newspapers from Iran

© PressReader. All rights reserved.