To promote rapid and standardized phenotyping for genomic improvement of quality traits in cassava, calibrations for dry matter content (DMC) and carotenoids in fresh cassava roots were developed from a portable near infra-red spectrometer (NIRS). Effect of eight pre-treatment combinations was evaluated on calibration performance and standard normal variate and de-trend (SNVD), with the first derivative calculated on two data points and no smoothing (SNVD+1111), was adequate to build a robust model. Generally, high calibration performance was obtained for most traits e.g. model for DMC on mashed samples had - R2c = 99%, R2cv = 95%, RPD = 4.5 and SECV = 0.9, with satisfactory R2 of 80% on independent validation set. On average, models developed with mashed were better than the intact samples. Intact and mashed NIRS-derived DMC were highly correlated (0.94) and had higher correlations (>0.95) with the ideal oven-drying than the specific gravity methods (0.49 and 0.69, depending on the dataset). Non-linear calibration model using random forest (RF), was equally develop and used to process spectra from National Root Crops Research Institute (NRCRI), Umudike for carotenoids including total carotenoid content (TCC) and some individual carotenoids (ICS): all-trans β-carotene (ATBC), violaxanthin (VIO), Lutein (LUT), 15-Cis beta-carotene (15CBC), 13-Cis beta-carotene (13CBC), Alpha-carotene (AC), 9-Cis beta-carotene (9CBC) and phytoene (PHY) . Derived carotenoids were used to understand correlations (phenotypic and genotypic), especially between TCC and ICS. High and positive phenotypic and genotypic correlations (>0.75) were obtained between TCC and the ICS except for PHY and LUT. Genome-wide association studies identified previously reported region on chromosome 1 associated with variation in TCC, in addition to other unidentified associations for both TCC and the ICS. Evaluating the potential of using Genome-wide predictions for carotenoids improvement, higher predictions were obtained from non-linear RF model with a one-step approach in single and multi-trait scenarios than linear and two-step approaches. The possibility of using molecular markers to assign parentage to progenies from a polycross nursery scheme was demonstrated with 100% assignment accuracy from simulated datasets. The information provided in this study is vital in redefining cassava breeding.