Sibiya, BongokuhleCho, Moses AMutanga, OOndidi, JMasemola, Cecilia RBonnet, Wessel J2025-03-132025-03-132024-120143-11611366-5901https://doi.org/10.1080/01431161.2024.2439081http://hdl.handle.net/10204/14170Carotenoids are important plant attributes offering valuable insights into the physiological condition of vegetation and serve as essential indicators for early identification of plant stress. Generally, carotenoids can be extracted using radiative transfer model (RTM) remote sensing techniques like PROSPECT that utilize the entire spectral domain (400–2500 nm) to retrieve carotenoid information. However, such inversions suffer from ill-posed due to model uncertainties. Literature suggests that selecting appropriate bands improves the RTM inversion. Hence, this study proposed a wavelength selection approach using various regions in the visible portion of the electromagnetic spectrum and bands selected by the random forest algorithm to estimate carotenoids using the PROSPECT-5 model. This study utilized three distinct datasets – savanna, tropical forest, and a combination of two. The green spectral region demonstrated the strongest performance in the tropical forest dataset (R2 = 0.90, RMSE = 0.71) than the savanna (R2 = 0.70, RMSE = 1.19) and combined (R2 = 0.72, RMSE = 1.11) datasets, respectively. The bands (green, yellow, and red-edge) selected by the random forest model produced the highest accuracy in the savanna dataset (R2 = 0.84, RMSE = 0.99), followed by combined (R2 = 0.80, RMSE = 1.20) and tropical forest (R2 = 0.78, RMSE = 1.33), respectively. Lastly, the visible region demonstrated strong performance in the tropical forest (R2 = 0.84, RMSE = 0.85), followed by combined datasets (R2 = 0.72, RMSE = 1.15) and savanna (R2 = 0.68, RMSE = 1.27), respectively. The findings suggest that carotenoid retrieval should be limited to the visible portions of the spectrum as it exhibited strong performance in estimating carotenoid content across the savanna, tropical forest, and combined datasets.FulltextenAnalytical Spectral DeviceASDCarotenoidsPROSPECT-5Radiative Transfer ModelRTMDetermining the value of different wavelength ranges of non-imaging hyperspectral reflectance to estimate carotenoid content using the PROSPECT-5 modelArticlen/a