I have fire activity data (i.e. number of fires) and a series of factors (e.g. precipitation, tree cover loss, distance to nearest forest, etc...) that can potentially explain it. I have all this data for different locations for each year between 2002 and 2016.

I am trying to find the factors that can better explain the ocurrence of fires but I am not sure what is the best way to statistically achieve this. My first approach would be to do a (multiple) correlation or maybe even a linear regression. However, how would I approach this situation, taking into account that I have yearly observations for my data?

I guess I could do 15 different regressions (one for each year) but I don't want to interpret the results separately, as my main goal is to identify the factors that better explain fire activity altogether, regardless of the year. I read about multivariate linear regressions and how they can take more than one outcome variable but I am not sure if they would fit my purpose. Should I just merge all of my data and forget about the time dimension?

  • $\begingroup$ When you have time series data as you do ... you should be aware of the assumptions of ordinary regression. Take a long and close look at this autobox.com/pdfs/regvsbox-old.pdf piece that I wrote a few years back. $\endgroup$
    – IrishStat
    Mar 23, 2019 at 16:03
  • $\begingroup$ I don't think it is advisable to run a regression for each year. I would analyze each location by itself and then look for commonality across models. $\endgroup$
    – IrishStat
    Mar 23, 2019 at 16:09
  • $\begingroup$ @IrishStat Thanks for the pdf, I will definitely read it. Would you analyze each location independently even when the number of locations is approximately 100,000? $\endgroup$ Mar 23, 2019 at 16:37

1 Answer 1


I would analyze each site separately to be able to draw some conclusions about the form of temporal dependence ... If it was slight then I would pool all the data after having adjusted for any evidented outliers/unusual values.

  • 1
    $\begingroup$ This misses the spirit of the question, which focuses on the potential for spatial correlation. Forest fires spread, inducing such correlation; and the potential for fires is based on the availability of fuel, prevailing wind directions, and many other factors with strong spatial correlations. $\endgroup$
    – whuber
    Mar 26, 2019 at 13:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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