Evaluating Spectral Consistency of Fe-Related Wavelength Ranges between EO-1 Hyperion and Field Spectrometer Data in the Kulon Progo Iron Sand Deposit

Abstract

Iron sand deposits contain Fe-bearing oxide minerals with characteristic reflectance features, but field spectrometer and hyperspectral image measurements of these features do not always agree, owing to differences in sensor characteristics, observation scale, and surface conditions. This study evaluates the spectral consistency between ASD FieldSpec and EO-1 Hyperion reflectance over the Kulon Progo iron sand deposit, and examines how spectral processing and spatial representation affect that consistency. Fifteen Hyperion bands were selected on physical grounds, eight within 480-560 nm and seven within 850-950 nm, and compared with resampled ASD spectra at 35 points with paired laboratory total-Fe data, using Pearson correlation (r) and RMSE, under raw, Savitzky-Golay smoothed, and continuum-removed spectra, and across four spatial extraction windows (1x1 to 7x7 pixels). The 850-950 nm range showed markedly stronger agreement than 480-560 nm (raw mean r = 0.573 versus -0.037). Smoothing raised the 850-950 nm correlation to r = 0.601; the pattern suggests that high-frequency noise in the Hyperion spectra may contribute to part of the disagreement, though the improvement did not reach conventional statistical significance (Hotelling-Williams test, p > 0.05). Continuum removal did not improve cross-sensor agreement, and correlation in the 850-950 nm range increased with window size (r = 0.614 at 1x1 to r = 0.693 at 7x7) without a corresponding change in RMSE. These results show that physical relevance alone does not guarantee cross-sensor consistency, and that band selection for Fe detection with hyperspectral imagery should also consider sensor-related spectral variation and observation scale.