1
$\begingroup$

What is the best approach to forecasting yearly climate data, for example minimum temperature in a specific area?

I have 24 meteorological stations and the data of minimum temperature (there are other climate indicators) are from 1970. I need to generate forecasts for 2015.

I know exponential smoothing models but I don't know ARIMA models. I think ARIMA models are not suitable because yearly data does not have seasonality. Can I discard the use of ARIMA models?

$\endgroup$
  • 1
    $\begingroup$ You must give us much more details, or this is going to be closed rapidly! $\endgroup$ – kjetil b halvorsen Dec 2 '15 at 16:30
  • $\begingroup$ I have 24 meteorological stations and the data of minimum temperature (there are other climate indicators) are from 1970. I need to generate forecasting for 2015, I know exponential smoothing models but I don't know ARIMA models. Can I discard the use of ARIMA models? $\endgroup$ – Ricardo UES Dec 2 '15 at 16:49
  • $\begingroup$ yearly minima, monthly minima, daily minima? $\endgroup$ – kjetil b halvorsen Dec 2 '15 at 16:55
  • $\begingroup$ Try to be even more specific; as it stands, it does not seem sufficient. Do you have hourly, daily, weekly, monthly, or annual data? What frequency are your predictions supposed to be in? Your sample starts in 1970, but when does it end? How many observations do you have? $\endgroup$ – Richard Hardy Dec 2 '15 at 19:06
  • $\begingroup$ The minimum temperature is a statistical indicator derived by calculating the minimum temperature in a specific year on the basis of the data collected by meteorological stations each minute of this specific year. My data ends in 2015 and I have 46 observations. I want to forecast the next year (2016) and the frequency is 1 since that my statistical indicator is annual. $\endgroup$ – Ricardo UES Dec 3 '15 at 15:20
1
$\begingroup$

There are many ways to answer this question. Basically, your choice is to use either 1) a physically-based model 2) a statistical model 3) a blend of the two (Kalman filter).

Since you ask the question here, I assume you opted for 2. Do you have good reasons for that? If so, you need to fit a statistical model to your data, and use it to make predictions. You have 24 stations so a spatiotemporal model is warranted. How wide an area do they cover? How many different climate regimes are there? That will determine how complicated a spatial model you need. The temporal component of that could well be ARIMA: it's often a reasonable choice. Where did you get the idea that it only applied to seasonal data?

Whatever you do, the atmospheric concentration of $CO_2$ has to be one of your predictors. If your data show a big trend from 1970 to present, $CO_2$ is the most likely culprit. Other indices of regional climate variability may be relevant too; depends on your region: where are your stations located?

$\endgroup$

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.