“…On the other hand, if the singular values are about equal, then all the parameters are necessary to explain the variance. Because experimental spectra have finite noise (SI, Section S3), the number of practically identifiable (sometimes called retrievable) parameters n r equals the number of singular values μ j , j ∈ {1,..,4}, whose singular vectors explain a fraction of the variance above the noise level where SNR, the “signal-to-noise ratio,” is defined as the norm of predicted values ∥ z nom ∥ 2 for a nominal parameter set divided by the uncertainty level on the data Δ z .…”