Work

Statistical Non-Identifiability of Illumination Sensitivity in UAV Lawnmower Surveys

Year: 2026
Type: article
Abstract: Within-flight illumination correction in uncrewed aerial vehicle (UAV) multispectral imagery commonly regresses spectral indices on the cosine of the solar zenith angle ($\cos {\,}Z$), implicitly assuming independence between illumination and spatial position. Under standard lawnmower acquisition, sequential image timing causes$\cos {\,}Z$to vary monotonically along the primary flight axis, producing strong collinearity with position. Analytical derivation, simulations across latitudes 0°–55°, and analysis of 11 UAV flights over paddy rice fields show that the absolute correlation between$\cos {\,}Z$and flight-axis position exceeds 0.95 for approximately 98% of daylight hours, with variance inflation factors (VIFs) up to 265.6 in the empirical data. Bootstrap analysis reveals unstable slope estimates, including confidence intervals spanning zero and frequent sign reversals. Orthogonal flight directions transfer rather than resolve the confound by shifting collinearity to the secondary axis, and onboard irradiance correction does not reliably remove it. Orthogonal sampling substantially improves design-matrix conditioning. These results demonstrate that within-flight illumination sensitivity is not reliably identifiable under standard UAV acquisition.
Language: English
Subfield: Ecology
Open Access status: hybrid
Estimated APC paid: $2,645
OpenAlex ID: W7204194252
FWCI:
Cites: 17
Cited by:
Related to: 10