Normally you do a correlation plotting one data vs the other and you check if your linear regression is close to 1. This indicates that the results are comparable or related. You can do a more complex statistical analysis, but this should be enough in your case. But you can clearly see by your data that the results are not comparable. There is no transformation you can apply, you need to understand why the results are so different.

Agree with Dr Rossell. There are 3 samples that are markedly different. Without those 3, you might develop a relationship. Are you running a close standard before and after each sample? Were you standard NO3 concentrations responding the same during testing, or was there variance or drift, which if not recognized could contribute error. I used a recording strip chart to record data, so it was easier to see variation with time. Variations in electrical voltage, photo cell output could contribute variable results, so that is why the repeated use of standards would help show instrumentation issues. I was trained by an analytical chemist (Dr. Pickett) ay my University. Someone of his caliber and experience might be helpful to review your methods as another possibility. Although it would not make difference in this case, when dealing with low values, using logarithms sometimes helpful.

Have you checked your water samples for organic levels, chlorides levels, or turbidity? All three of these are known spectral interferences for UV/Vis nitrates-N analysis (see Hach and Perkin Elmer technical literature). I bring this to your attention, because a quick glance of your Excel spreadsheet revealed that only 2 out 23 samples analyzed by ion chromatography were with-in acceptable analytical agreement to your UV/Vis data. We can speculate that your water sample matrix effects were reduced because the nitrate ions are being retained and released as a more representative sampling through the ion exchange process. Unfortunately the UV/Vis data collected could be suffering from redox equilibrium conversion efficiency rate problems or spectral interferences. Have you considered using a ion specific electrode or switching to V(III) redox reaction for UV/Vis analysis (Article A fast and direct spectrophotometric method for the sequenti...

Did you follow Hach recommended long term sample storage protocol? Based on your Excel spreadsheet, were your samples primarily analyzed using Hach method 8039 as it has the best possible accurate linear range for agreement with IC data? What did use for your sample blank? According to Hach's method, this testing method is technique sensitive. It is very important that you use your test sample as a blank before testing your treated sample for analysis. I would recommend that you use a sample rocker (test tube rocker) for the 2nd reaction period (5 minutes), this will ensure repeatable sample mixing and allow you to properly collect your UV/Vis blank for that sample matrix. If your samples were not properly stored this could account for the large difference, or if they were not shipped in a cooler with cold pack over night this would also contribute to the large observed variation in your data points. In short, you have some sample handling investigation to resolve because NA-15-1124 & Na-14-116 produced reproducible data. If they were collected and properly stabilize in the field, and you followed proper storage procedure the two methods are comparable. Best wishes in resolving and testing your samples. Did you check to see if Hach has 2nd derivative UV/Vis data analysis for the DR2800?https://www.whoi.edu/cms/files/derivative_spectroscopy_59633940_175744.pdf

Thank you so much! Andreu Fabregat Rossell William F. Hansen Michael Sanford

William F. Hansen I didn't run a standard before/after every sample, but I made calibration curves using lab and in-house standards and I was pretty happy with the results as they were linear (some underestimation if anything). But that is a great suggestions to check more frequently to account for drift.

Michael Sanford Your point about organic levels and turbidity is highly valid, especially since I’m collecting samples from agricultural ditches and after extreme precipitation events. I filtered any samples that looked turbid, but maybe better to just filter all in the future.

I used DI water for my blank samples were kept in the fridge @ 4 deg C and analyzed the next day. NA15 and NA14 are from the main river in an upstream portion, so they could potentially be less turbid, especially since the river bed is mostly cobbles, rather than silt further downstream (could also be coincidental). I used method 8039 originally but then transitioned to method 8171 due to the lower detection limit (0.1 mg/L), but in hindsight should have used method 8192. I just had no indication that my samples were likely below the detection limit. Thanks for the article links and all your insight, I have lots to learn and more to think about now. I will first identify my samples on the plot based on land-use and geology to see if there is a potential correlation with high turbidity and/or organic content. Thanks again for your help and your time!

There are several ways you can determine the relationship between your two methods of analysis, such as using statistical methods like linear regression or correlation analysis. One approach would be to plot the nitrate-N concentrations from your spec samples against the corresponding concentrations from the IC samples for the 23 samples that were analyzed by both methods. This will give you a visual representation of how well the two methods agree with each other.

You can then use linear regression to find the equation of the line of best fit, which will describe the relationship between the two methods. The slope of the line will give you the "conversion factor" that you can use to adjust your spec results to match the IC results.

Another option would be to use correlation analysis to determine the correlation coefficient between the two methods, this will give you an idea of how strongly the two methods are related.

It's important to note that this approach assumes that the relationship between the two methods is linear and that the relationship observed in the 23 samples is representative of the relationship in the entire sample set.

It might be helpful to check the precision and accuracy of both methods, as well as to use other statistical methods to validate the results.

Dear Ariel, I agree with the above opinions, and in my experience, looking at your data there is a strong matrix effect. But it is necessary to remember that IC has its limitations as well, such as ionic suppression by matrix effect. However, I believe that the IC results are more accurate, and based on this premise, greater rigor in sample preparation is necessary, removing these matrix effects. Hope this helps

Moacir Brito Ok than you! I agree, seems like the most logical answer based my samples.