Dear Sir,

Perform statistical analyses (linear regression) between the parameters and it will be help to obtain your objective.

I am so sorry, colleague Karimov, but:

- first, the ionic structure - not one index, but it is a lot of, taking into consideration namely ionic composition;

- secondly, water salinity and ionic composition are directly connected with electric conductivity by sense of these indicators (learn electrochemistry and chemistry of water, colleague) therefore no new regularities should be expected from studying of these indicators. Correlations will be high, but there was  no any sense in them. Therefore I advise to forget about the statistical analysis of these data and to be engaged in something in another. "A lie, bald lie and statistics" as great Lagrange has told. I am so sorry.

Colleagues, Faina is completely right. This parameters are dependent one other, so normal corelation cannot do. Use Anova/Manova, use multidimensional methods, use reaches of the river, flooding, monsoons, temperature changes, rift and slow sections, anthropogenic activities (men-coused) as a factors, and you can have some profit.

Linear regression isn't a model, so cannot be good for forecasting, just can show some common relations...

You can analyse type of distribution, normalize if necessary or use robast statistics, calculate dispersion, experimental accuracy etc. Make for seasons and look differences between average meanings etc... do for different biotopes.... day and night, dusk and down.

Ions total means fiction, so imp are forms and they depends on pH that is connected with  a photosynthetic activity of algae, so if you measured once pH in the evening and next in the morning cannot compare...

If need more inf about, just ask...

Andrey

The Observed data of Water Salinity, Ionic composition and Electrical conductivity in water are inter related , and may provide better understanding by using stastical analysis.

I am so sorry, but if we would measure the alteration of the peremeters in model usin firstly distilled water + knowm concentrations of different ions - it will be good model for study of interelations between the indices. But now there are a lot of data without correct measurement of the whole composition - what new data it could be obtaioned?

Water/soil salinity is usually measured with the help of conductivity though now salinity is directly measures by hand refractometer. But some scientists do not like it as it is not so precised. Conductivity also gives you some ideas about ionic concentration. So, there three are much interrelated. As Dr. Faina suggested, though straight forward, it is better to think otherwise.

Because your attributes are so inter-related, I would start simply with some data exploration - I would conduct Principle Component Analysis looking specifically at the loading of the eigen vectors to see which attributes are providing the greatest degree of distinction.

Beyond that, allow me to ask on what basis will you establish your water quality standards? Will it be purely some numerical exercise? Will it incorporate biological impacts? I ask because it seems your objective is broader than merely a statistical issue.

Cheers,

Mike

Dear Mike,

Many thanks! Yes you are right. Water quality legislation and regulations that target salinity typically focus on the protection of drinking water and the quality of irrigation water, which does not automatically protect biodiversity (see: Cañedo-Argüelles M., et al. Saving freshwaters from salts. Science 351(6276): 914). This is especially true in ASB and we want to try to develop recommendations on total and ion-specific salinization standards needed to protect biodiversity and aquatic ecosystems as a whole.

Could you please give me the literature on Principle Component Analysis, I do not have any idea on this methodology.

With best wishes

Bakhtiyor

You can start just on Wikipedia; they have a good discussion on it. Here's another site that tries to explain it visually - http://setosa.io/ev/principal-component-analysis/

It's a standard test in most statistical software packages as well.

Essentially, PCA reduces the dimensionality of your data by applying eigen vectors to the data. Think of an eigen vector loosely (very loosely!) like a regression equation with as many terms as you have dimensions of data.

I like to use the analogy of a slightly squashed (North American) football (or a rugby ball if you prefer): The first eigen vector represents the long axis of the ball; the second is the squashed width; and the third is the height.  The principal components are calculated such that the first accounts for the greatest amount of variability (the long axis), the second one the next largest, and so on. Often, just the first three eigen vectors will account for the vast majority of variance in your data set. If you can do 3D graphing, especially with rotating, plotting the three vectors help you see possible differences or clouds of data points. By looking at the individual loading factors in each eigen vector, you can see which attributes have the greatest influence on variability. Do not interpret the actual loading factor, or even its sign (+ or -) as indicative of any correlation measure though.

Also look at canonical correlation analysis (CCA). CCA is similar BUT defines coordinate systems that optimally describe the cross-covariance between two datasets, whereas PCA defines a new orthogonal coordinate system that optimally describes variance in a single dataset.

Hope that helps,

Mike

Double chamber settling tank with periodic washing.

use statistical analyses (linear regression) between the parameters and it will be help to obtain your objective.