Variable Transformation in Chlorophyll-A Satellite Algorithm Calibration

In this paper, we demonstrate how applying variable transformations tailored to meet key ordinary least-squares (OLS) assumptions can significantly improve the performance of a satellite-based algorithm for chlorophyll-a (chl-a) concentration estimation. Focusing on the 3BDA predictor, we tested an optimal set of transformations—log, square-root, and cube-root—on both the dependent and independent variables. The best combination reduced the cross-validated root mean square error (RMSE) from 28.5 to $\mathbf{22} \mathbf{mg}/\mathbf{m}^{\mathbf{3}}$ relative to the commonly used single log transformation of chl-a, thereby increasing algorithm precision. These findings underscore the broader relevance of ensuring OLS assumptions are satisfied for improved accuracy in semi-empirical modelling approaches.