as.date creates some NAs in dataset

I have a simple little dataset:

> str(SFdischg)
'data.frame':   11932 obs. of  4 variables:
 $ date: Factor w/ 11932 levels "1/01/1985","1/01/1986",..: 97 4409 8697 9677 10069 10461 10853 11245 11637 489 ...
 $ ddmm: Factor w/ 366 levels "01-Apr","01-Aug",..: 1 13 25 37 49 61 73 85 97 109 ...
 $ year: int  1984 1984 1984 1984 1984 1984 1984 1984 1984 1984 ...
 $ cfs : int  1500 1430 1500 1850 1810 1830 1850 1880 1970 1980 ...

I would like to have a column of dates so that I can plot temporal data:

SFdischg$daymo <- as.Date(SFdischg$ddmm, format="%d-%b")
> summary(SFdischg)
    date            ddmm            year           cfs           daymo           
 1/01/1985:    1   01-Apr :   33   Min.   :1984   Min.   : 172   Min.   :2018-01-01  
 1/01/1986:    1   01-Aug :   33   1st Qu.:1992   1st Qu.: 705   1st Qu.:2018-04-04  
 1/01/1987:    1   01-Jul :   33   Median :2000   Median : 948   Median :2018-07-03  
 1/01/1988:    1   01-Jun :   33   Mean   :2000   Mean   :1374   Mean   :2018-07-02  
 1/01/1989:    1   01-May :   33   3rd Qu.:2008   3rd Qu.:1340   3rd Qu.:2018-10-01  
 1/01/1990:    1   01-Nov :   33   Max.   :2016   Max.   :8100   Max.   :2018-12-31  
 (Other)  :11926   (Other):11734                                 NA's   :8

However, daymo now has 8 NAs and I can't understand why (and it makes it difficult to plot!). Where does the handful of NAs come from when there is no missing data in ddmm? How can I avoid them? Am I missing something obvious?

My guess is that some of the factor data you have in the ddmm column cannot be parsed correctly into a date. You may reveal these bad values using:

SFdischg$ddmm[is.na(as.Date(SFdischg$ddmm, format="%d-%b"))]

Note that since there is no year component in the ddmm column, R appears to be automatically assigning the current year 2018 to the date. Ideally, you should be building your date using source information which includes a year.

Edit: Based on your comment below, the offending rows had 19-Feb as the date. This implies that these dates were perhaps not even from 2018, which was not a leap year, and whose February had only 28 days. This illustrates the importance of working with a full set of information when parsing the date, including the year.