Is logarithmic transformation (log 1/R) always necessary in diffuse reflectance measurement?

The author once has developed calibration equations using spectra before logarithmic transformation (log 1/R) (Reflectance spectra), when studies on calibration transfer was carried out. At that time, an impression got was that accuracy was rather high as compared to that using spectra after logarithmic transformation (Absorbance spectra). Then the author conducted discussion on the necessity of logarithmic transformation in the development of calibration equations.
Spectra used for experimental are those from three kinds of soil and compost used for development of “NIR soil and compost measurement system”. Spectra of soils and compost were collected using S-7110 that was developed for this system by Soma Optics, Ltd.. Range of spectra is 1200 nm – 2500 nm. Transformation and statistical analyses were carried out with “The Unscrambler” Ver. 9.8 and 10.3.
Prediction accuracy of calibration equations between those from “Reflectance spectra” and from “Absorbance spectra” for three kinds of soil was compared. Constituents compared were total nitrogen (T-N) and total carbon (T-C).
Results were following.
In soil-1, accuracy of “Reflectance spectra” was better in T-C and reverse in T-N, but the difference was very small.
In soil-2, “Reflectance spectra” was better in both of T-C and T-N, and the difference in accuracy was not so small.
In soil-3, “Reflectance spectra” was better in T-C and same in T-N. The difference of accuracy was very small as soil-1.
Trend is that accuracy was better in “Reflectance spectra” in soil measurement.
In compost, prediction accuracy was distinctly better in “Absorbance spectra”. Absorbance in the spectra of compost was much larger than those of the soil.
It is thought that path length in compost was enough long that “lambert Beer’s low” can be applied. On the other hand particle size of soil is much smaller as compared to that of compost, and then, it is thought that logarithmic transformation (log 1/R) brought rather worse result because of its short path length. In general “lambert Beer’s low” is accepted as absolute. However the author thinks that there is an exception like this, and in the measurement of fine (small particle) samples such as flower, material of medicine and so on, it is thought that use of reflectance spectra should be considered in the development of calibration equations. Also the author is thinking use of the reflectance spectra hereafter.