Spectral band selection for the characterization of salinity status of soils

Abstract Reflectance spectra of salt-affected surface soil samples have been measured at 10 nm spectral resolution between 495 nm and 2395 nm. The samples were collected in California (41) and in Hungary (49) to compare the effect of different salinization and alkalization processes on reflectance properties. The data set consisted of 272 spectra having 2–12 cases (repetitions) for each sample. Sixteen classes were defined based on chemical soil properties, accounting for changes in pH, electrical conductivity (EC), and exchangeable sodium percentage (ESP), of which 12 were represented in the sample. The reflectance data set was statistically analyzed using a modified stepwise principal component analysis (MSPCA) approach to select 1, 2, 3, … bands for classification of salinity status. Recognition of the above-described classification was tested by discriminant function analysis (DFA). Its results can be applied in further studies for weighing spectral bands according to their sensitivity to the chosen classification as well as in defining broad, but still potentially sufficient bands. Recognition accuracy of salinity status was 91%, 90%, and 88% with 10 nm, 20 nm, and 40 nm bands, respectively, for the entire data set. Comparison with PCA using all bands showed only slight differences. The California soil samples had more distinct spectral characteristics than the Hungarian ones. Key spectral ranges were identified in the visible (550–770 nm), near-infrared (900–1030 nm, 1270–1520 nm), and middle infrared (1940–2150 nm, 2150–2310 nm, 2330–2400 nm) portion of the spectrum at 20 nm, 40 nm, and 80 nm spectral resolution. Two of these (1270–1520 nm and 1940–2150 nm) cannot be used with satellite data due to water vapor absorption in the atmosphere. In addition, six broad bands in these ranges were identified, leading to considerably higher overall accuracy than currently available Landsat MSS, TM, and SPOT XS, in terms of spectral recognition of salinity status.