select certain dates inside loop for .csv file

You can take advantage of the well-ordered nature of the data file and use csv.dictreader. Then you can build up a dictionary of the windspeeds organized by date quite simply, which you can process as you like. Note that the csv reader returns strings, so you might want to convert to other types as appropriate while you assemble the list.

import csv
from collections import defaultdict
bydate = defaultdict(list)
rdr = csv.DictReader(open('winds.csv','rt'))
for k in rdr:
    bydate[k['Date']].append(float(k['Spd']))

print(bydate)
defaultdict(<type 'list'>, {'19590101': [3.1000000000000001, 1.0, 1.0, 2.1000000000000001, 1.0, 0.0], '19590102': [2.1000000000000001, 2.1000000000000001, 0.0, 2.1000000000000001]})

You can obviously change the argument to the append call to a tuple, for instance append((float(k['Spd']), datetime.datetime.strptime(k['Date']+k['HrMn'],'%Y%m%D%H%M)) so that you can also collect the times.

If the file has extraneous spaces, you can use the skipinitialspace parameter: rdr = csv.DictReader(open('winds.csv','rt'), fieldnames=ff, skipinitialspace=True). If this still doesn't work, you can pre-process the header line:

bydate = defaultdict(list)
with open('winds.csv', 'rt') as f:
    fieldnames = [k.strip() for k in f.readline().split(', ')]
    rdr = csv.DictReader(f, fieldnames=fieldnames, skipinitialspace=True)
    for k in rdr:
        bydate[k['Date']].append(k['Spd'])
return bydate

bydate is accessed like a regular dictionary. To access a specific day's data, do bydate['19590101']. To get the list of dates that were processed, you can do bydate.keys().

If you want to convert them to Python datetime objects at the time of reading the file, you can import datetime, then replace the assignment line with bydate[datetime.datetime.strptime(k['Date'], '%Y%m%d')].append(k['Spd']).

It can be something like this.

def dump(buf, date):
    """dumps buffered line into file 'spdYYYYMMDD.csv'"""
    if len(buf) == 0: return
    with open('spd%s.csv' % date, 'w') as f:
        for line in buf:
             f.write(line)

obs_in  = open(csv_file).readlines()
# buf stores one day record
buf = []
# date0 is meant for time stamp for the buffer
date0 = None
for i in range(1,len(obs_in)):        
    # Skip over the header lines 
    if not str(obs_in[i]).startswith("Identification") and \
        not str(obs_in[i]).startswith("Name"):
        name,usaf,ncdc,date,hrmn,ii,type,dir,q,i2,spd,q2,blank = \
            obs_in[i].split(',')
        current_dt  = datetime.date(int(date[0:4]),int(date[4:6]),int(date[6:8]))
        current_spd = spd 
        # see if the time stamp of current record is different.  if it is different
        # dump the buffer, and also set the time stamp of buffer
        if date != date0:
            dump(buf, date0)
            buf = []
            date0 = date
        # you change this.  i am simply writing entire line
        buf.append(obs_in[i])

# when you get out the buffer should be filled with the last day's record.  
# so flush that too.
dump(buf, date0)

I also found that i have to use ii instead of i for the filed "I" of the data, as you used i for loop counter.

I know this question is from years ago but just wanted to point out that a small bash script can neatly perform this task. I copied your example into a file called data.txt and this is the script:

#!/bin/bash
date=19590101
date_end=19590102
while [[ $date -le $date_end ]] ; do
  grep ",${date}," data.txt > file_${date}.txt
  date=`date +%Y%m%d -d ${date}+1day` # NOTE: MAC-OSX date differs
done

Note that this won't work on MAC as for some reason the date command implementation is different, so on MAC you either need to use gdate (from coreutils) or change the options to match those for date on MAC.

If there are dates missing from the file the grep command produces an empty file - this link shows ways to avoid this: how to stop grep creating empty file if no results