Beyond total irradiance: a PCA-based filtering framework identifies wavelength-dependent biological signatures of sunlight exposure.

Natural sunlight contains ultraviolet, visible and near-infrared radiation, but these bands are highly correlated under everyday environmental conditions. This collinearity limits the extent to which population studies can distinguish overall ambient sunlight availability from wavelength-dependent biological associations. Using UK Biobank data, we linked satellite-derived ambient spectral irradiance estimates to baseline blood biochemistry measurements. We examined unweighted UVB (280-315 nm), UVA (315-400 nm), blue (450-495 nm), red (620-750 nm) and near-infrared (700-1100 nm) irradiance averaged over the 14 days preceding blood sampling. Principal component analysis (PCA) was used to characterise the dominant common variation across spectral bands. We then filtered out the reconstructed PC1 contribution from each band to derive spectral residuals, which were evaluated in multivariable regression models. Inter-band correlations were extremely strong (r = 0.94-1.00), and PC1 explained 98.3% of total variance. Models using the original band-specific irradiance estimates produced highly concordant biomarker association patterns across wavelengths, consistent with dominance of common ambient irradiance variation. In contrast, PC1-filtered spectral residuals revealed wavelength-dependent contrasts. Vitamin D showed preferential positive association with ultraviolet residuals, consistent with its established UVB-dependent biology and providing a positive validation example. Several metabolic, hormonal and inflammatory biomarkers also showed contrasting residual associations across ultraviolet and longer-wavelength bands. These findings confirm that known wavelength-dependent biological responses can be recovered from highly collinear environmental sunlight data using a PCA-based filtering framework. They also suggest that additional blood biomarkers may show wavelength-dependent associations with everyday ambient sunlight exposure. The spectral residuals should be interpreted as statistical constructs that highlight small deviations from the dominant common variation, rather than as physically independent wavelength exposures or measurements of personal skin dose.