I found this query on Stack Exchange Data Explorer Movies&TV wrote that shows the monthly frequency input of id my movie questions and total questions on the site. As you can see on the resulting graph, they solved their issue at 2018.
Earth Science query
I changed from Movies&TV code
%identify% --> %identification% and selected Earth Science on the menu and aparently, the query works for us.
select dateadd(month, datediff(month, 0, p.CreationDate), 0) mo, count(distinct p.id) as QuestionCount, count(distinct case when p.tags like '%identification%' then p.id else null end) as QuestionCountID, count(v.id) as VoteCount, sum(case when p.tags like '%identification%' and v.id is not null then 1 else 0 end) as VoteCountID from postswithdeleted p left outer join votes v on v.postid = p.id and v.votetypeid in ( 4, 6, 7, 10, 11, 12, 15, 16 ) where p.posttypeid = 1 group by dateadd(month, datediff(month, 0, p.CreationDate), 0) order by dateadd(month, datediff(month, 0, p.CreationDate), 0)
Earth Science graph (23/11/2019).
It looks like we reduced the input of id questions around september/october, when we closed them.
Let's select only id questions (deleting lines 3 and 5 in the query and running it), wich reduces y-scale.
What I see is we started to have a trouble sometime between 2016 and 2018, probably because we started to appear first at Google for searches as "identify my rock".
We have broken the tendency closing all the questions, apparently users coming from Google have understood it; good news!
Let's see if deleting them we completely solve the issue as Movies&TV did!