I refactored your script. Most of the changes I made are ones recommended in Perl Best Practices by Damian Conway. The revised script is untested and intended merely to suggest coding improvements.

#!/usr/bin/perl

use strict;
use warnings;

use English qw( -no_match_vars );
use Text::CSV;

my $release_file = '../ecodata/releases.txt';

open my $release_fh, '<', $release_file
    or die "Can't open release file $release_file: $OS_ERROR\n";

my $csv = Text::CSV->new({
    allow_loose_quotes => 1,
    binary             => 1,
});

CSV_RECORD:
while (my $csv_record = <$release_fh>) {
    # Skip header...
    next CSV_RECORD if $INPUT_LINE_NUMBER == 1;

    # SPECIAL CASE: Remove unwanted text...
    $csv_record =~ s/ \(\d{4} and after "acid aerosols" only\)//g;

    if (not $csv->parse($csv_record)) {
        my $csv_error_input = $csv->error_input();

        warn "Can't parse CSV record at line $INPUT_LINE_NUMBER"
           . " of release file $release_file:\n$csv_error_input\n";

        next CSV_RECORD;
    }

    # TRI, Release#, ChemName, RegNum, Year, Pounds, Grams
    my ($tri, undef, $chem_name, undef, $year, $pounds, $grams)
        = $csv->fields();

    # [Do what you need to do with the data here]
}

close $release_fh;

exit 0;
[download]

I moved the Text::CSV->new() constructor outside the while loop. I moved the lexical variables inside the while loop. I added a space to the beginning of the regular expression pattern matching the string being globally removed from the CSV records.