Yes, poj, you're right, there were two mistakes.

This is the amended (and tested) version:

use strict;
use warnings;

my $input = "input.txt";
open my $IN, "<", $input or die "cannot open $input $!";
my $output = "output.txt";
open my $OUT, ">", $output or die "cannot open $output $!";
while (my $line = <$IN>) {
    chomp $line; # remove newline character from end of line
    if ($line =~ /INTERPOLATED HYDROGRAPH AT (\w+)$/){
        print $OUT $1;
        $line = <$IN> for 1..6; # skip 5 lines
        my $val2 = (split / /, $line)[1]; # get the second column
        print $OUT " $val2";
        $line = <$IN>for 1..2; # skip one line
        chomp $line;
        my $val3 = (split / /, $line)[-1]; # get the last column
        print $OUT " $val3\n"; 
    }
}
[download]

And this is the output:

$ cat output.txt
CAC40 1223. 1456.
[download]

I should add that this is not very robust code, it will probably break with any irregularity in the input data. But we would need to know more about the data (having at least three or four samples of line groups, instead of only one) to be able to do something more robust.