I have hourly weather data for a number of weather stations in radius of some 300 kilometers, for second half of the last century. Missing values range from few hours to a whole month. I am working with variables: T, RH, rain, wind and sol. radiation. I need to impute missing values so that in order to evaluate models for prediction of risk for an agricultural pest, and eventually also use it for some modeling. Data is loaded in R. I am relatively new to r, but I believe I am finding my way through it quite well for a beginner.
As I can see there is a sea of programs and packages that can do this job, and a lot of questions posted here, but in the end it just gets me confused. So, I would kindly ask for an advice on how to perform some of following steps:
- Get a good overview of missing values, with plots, summary sheets.
- Analysis of possible anomalies in data (i.e. high RH with strong wind and high T)
- Statistical approaches for imputing missing values for different, above-mentioned variables. A specific questions:
- I have daily readings for few station - years which needs to be reduced to hourly values.
- 'Strategy' for imputing miss. val. Let us say that I set the threshold of missing values for T at 5 and input it with average of nearest neighbors. Does this sound reasonable? This approach probably would not apply to rain data...
I hope there is someone with experience with this kind of work. Comments from personal experience, references to relevant sc. articles, links to a similar work and R code would be greatly appreciated.