Thanks for sharing the link. Because you're using an uploaded image, the top row of pixels is quite dark, and that's the one used as your cross-section line.
I tried forking your spectrum and selecting a new cross-section: https://spectralworkbench.org/spectrums/89900
That looks much better and makes a good calibration. But note that since your spectrum is curved, the choice of cross-section will shift your calibration left or right depending on what height it is. You'll need to ensure all your data has the same cross-section if you want to compare it or use this calibration for other readings.
Hope that helps!