1
$\begingroup$

I need some hints for the following task:

I got two time series from different weather sensors which contain daily rain data values. One of the sensors is a professional one and my job was to build a low-cost sensor and compare the results with the professional one, which is supposed to be very accurate.

When facing a nonstationary time series the normal procedure would be to decompose it and derive a model out of it. As rain being not really time dependent in non tropical areas i would guess that rain data is already stationary.

How would I proceed then? I was thinking about something like this:

  1. Proving stationarity of the time series

  2. Eliminate outliers

  3. Correlation (Pearson) of both time series

I'm really new to statistics so this might be a really noobish approach.

Improve this question
$\endgroup$

1 Answer 1

Reset to default
0
$\begingroup$

You could start off by making a Bland-Altman plot (otherwise sometimes called a Mean difference plot). That would give an initial view on if the residues between measurements had any systematic bias with measured level.

You could also try and define a bit more what is the objective of the analysis of the new sensor is:

  • does it have to work equally well in all conditions or are you more concerned about accuracy when either high or low rainfall
  • is it better to have an instrument that is normally very accurate but occasionally very wrong or something that is never very wrong
  • is it important that measurement errors do not depend on order of measurements
  • are you looking to pass or fail the new sensor or just get more information on how it is working
Improve this answer
$\endgroup$
2
  • $\begingroup$ Thanks for your answer. The Bland-Altman plot looks promising. What do you mean by success criteria, can you go a bit deeper? As is said im really new to statistical analysis. $\endgroup$
    – Ernst-Otto
    Commented Dec 30, 2017 at 13:25
  • $\begingroup$ Have updated answer to ask some questions that are relevant to consider before starting to approach the problem and would help decide on important questions such as the treatment of outliers $\endgroup$
    – user143881
    Commented Dec 30, 2017 at 15:00

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.