Correct metodology to test observations

Hello to all,

I have two datasets of rain measurements, the first observed at the ground level, and the second estimated from satellite radar at same location (pontual data), on daily basis. You can see in attachment the two histograms.

I want to check if this two datasets came from the same distribution at 5% significance, and I used the K-S test in Matlab.

The basic algorithm is: If Ho is reject we eliminate in both datasets the position with major difference and perform again the test, and so on until we accept Ho.

My questions are, for this kind of data, the K-S test is the best test ? and the basic algorithm described make sense ?

Thank you very much!


TS Contributor
what is the purpose of this operation? Just off the top of my head, the histograms seem to be a bit too far removed from the reality of the data to warrant eliminating observations - e.g. your results might depend on the bin sizes of the histograms (ad absurdum, if you have one bin, you never eliminate anything).

Hi rogojel,

The big bin is 0 mm rain observations (no rain). The main idea is eliminate up 10% (major differences) of both datasets and test. If we test for mean and variance (t-test and F-test) we accept Ho. But for this case (spatialized rainfall) the best would be compare distributions. May not be possible due to the large bin :shakehead

Thanks a lot


New Member
The best thing is to go in sequential order, First test your hypothesis either for the Gauges observation or Satellite observation(instead of taking both at a time and comparing the hypothesis), once you select one of them to test your hypothesis and based on the inference you get, you can then test the hypothesis for the second one. then if both results are matches then you can proceed.