Hi Hugo,

As many other analysis the ideal statistical test for your rainfall data will depend on several factors such as the hypothesis, the outcome distribution, the resolution of your data, etc. Rainfall data is not easy to model because ground data is always scarse and most satellite products are also models, so there is not such thing as a good rainfall data. My advice to you is to complete your learning curve taking one step at the time and going from the basic (such as cumulative rainfall comparisons) to more advance approaches (such as multilevel regression, time series, mathematical modeling, etc.). Read the following article by D.A. Hughes, which explains rainfall data limitations in detail but also will offer some alternatives to deal with such limitations. Enjoy...   

Article Comparison of satellite rainfall data with observations from...

Your data set is very (too?) small for this sort of question, unless the rainfall at this station is remarkably regular (wrt seasonality etc.). For arid stations, even 100 years of daily measured data may not be sufficient. Starting with modelled cumulative plots  of daily data is a good idea, but rainfall time series commonly show fluctuations at all time-scales up to centuries (when data is available). It also is necessary to know if you are looking at determining changes in mean behaviour (normal-based statistics), or extremes (Weibul-based statistics).

Thank you Antonio and Chris for your suggestions. I know that the dataset is very limited and I will start with the cumulative plots to have a better idea of what I have.

I would like first to explore changes in the means. I know a priori that there are marked rainy and dry season and also that since 2013 precipitation has decreased in this Brazilian city. What I want to see is if this difference is significant in relation to the previous years. 

Thanks

Hi Hugo,

for changes in mean you can do the change-point analysis. As your change is at the end of the time-series, the ordinary Cusum test will probably fail to show the significance. Worsley likelihood would be better, as you can assign weights to accentuate the end of the series. You can have a look at the TREND package http://www.toolkit.net.au/Tools/TREND. However, even this test might fail to show the statistically significant change in mean - you might need to wait until more data is available. The Toolkit provides the excel spreadsheet that you can use.

Regards

Aleks

Hi,

I would  perform non-parametric trend analysis to see if there is statistically significant changes on the precipitation over time or not. You can perform this analysis  with annual as well as monthly precipitation data. I am not sure how much are you familiar  with free analysis software such as R (https://cran.r-project.org/). There are easy to use packages in R for analysis like this one. You can use 'Kendall' and "zyp" packages in R to perform non-parametric Mann-Kendall Trend analysis of precipitation data and therefore the changes.  Cheers, aseem

As Chris says - your data set is too small (10 years) to look at changes in mean annual or monthly rainfall due to the natural variability. I suggest you see if there are other gauges in the general area that have significantly longer records of rainfall. As it stands, all you can do is determine the variability across the 10 years of data. This doesn't get you information on change however. If you find a gauge with a longer record, you might be able to look at change, though success in this will depend on the natural variation in the rainfall amounts.

Why not ANOVA or DUNCAN?

Representing Data using Histogram and estimate Mean, Variance and standard deviation. After that, you can compare using Time and Spatial Resolution.

I was considering a rain fall analysis that is very similar. Any thoughts on a seasonally adjusted ARIMA model?

Consult Semu Moges browse on RG his name and communicate directly with him.

Good luck

Mike