I am currently doing a modeling project at county scale in which I have disease rate as dependent variable and environment variables (such as temperature, NDVI and evapo-transpiration) as the independent variables.

In order to tell how good my model is, I need to do regression analysis for which I need a particular value of each variable at county level. For this purpose I used zonal statistics tool in ArcGIS and took the mean value of NDVI, temperature and evapo-transpiration.

  1. As a county scale is too coarse, so taking 'Mean' is the right statistics or not?
  2. Also in case of counties having a lot of cloud cover, zonal statistics only considers available pixels (may be too few) to compute. What is the right way to deal with such an issue?

I will really appreciate your solutions and if you have any links/research papers to support this.

  1. mean values are appropriate for that kind of analysis, however, you may also want to include max or min quantile/quintile statistics (probably easier to do outside of arc) if disease spread is dependent on min/max temperature etc. values. you will see which is your most significant variable when you look at the correlation matrix & run the regression analysis.

  2. if your data is for more than one time frame - you can limit ndvi raster input to those of a specific cloud cover quality. otherwise, if you don't have the data for the county, you have to exclude the county from the analysis.

| improve this answer | |
  • @ Meg williams, thank you so much as these are really useful recommendations. For cloud covered counties, is it not better to interpolate those variables (Temperature and ET) and take zonal statistics whereas keeping statistics of other counties from the original data? Thanks again! – Abhishek Kala Jul 15 '13 at 19:33
  • i'd say it depends on the variability of the terrain w/i the county and exactly what data coverage you do have, but yes, you can interpolate and then get zonal stats from the interpolated raster. – mwil Jul 17 '13 at 14:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.