I have a daily precipitation data for some 13 stations (38 years period) spread across different geographical extents (plains, mountains and deserts) with varying missing values (12% to 1% of the total daily values; 13515 values).
1-- Whether is it would be ok to move on with the data without filling in the missing values under the assumption i.e. (according to some research for same treatments, it is ok to have the missing values/error upto 15% )..
2-- Technically, a complete data yields accurate or near to accurate results.
Then which approach should be used then to fill in the missing values...(i have a very theoretical idea about normal ratio , Kernel Nearest Neighborhood regression, Artificial Neural Networks and Distance Weight Approach. not sure whether to use it here)
Article Comparison of Missing Value Estimation Techniques in Rainfal...
You could use the satellite data, first to learn the relationship between the stations reading and satellite measurement. Then, to fill the missing value, predict the station's value based on the satellite reading.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression