Abstract <p>Cotton is the most extensively produced natural fiber worldwide, and its optimal yield relies on precise nutrient management throughout different growth stages. Traditionally, cotton nutrient estimation relies on laboratory-based analyses, which are time-consuming, costly, and destructive, which delay management decisions that influence final yield. Attenuated Total Reflectance (ATR) and Diffuse Reflectance (DR) spectroscopy provide rapid, cost-effective, and environmentally friendly alternatives to conventional laboratory methods. This study assessed the feasibility of using different visible near-infrared (VisNIR) and mid-infrared (MIR) spectrometers to estimate 11 macro (N, P, K, Ca, Mg, S) and micronutrients (Fe, Mn, B, Cu, Zn) from fresh and dried cotton plant tissues. Three modeling techniques: Partial Least Squares Regression (PLSR), Cubist regression trees, and Support Vector Regression (SVR), were evaluated using 75% of the dataset for calibration. Among these, PLSR and Cubist produced comparable results; however, PLSR was selected for its faster computation. Dry leaf models using VisNIR resulted in high accuracy for all macronutrients (R²: 0.75–0.96) and several micronutrients, including B, Mn, and Cu (R²: 0.78–0.93). In contrast, fresh leaf models were less accurate due to moisture interference, limiting their feasibility for practical field applications. Models developed for stems and burs were less robust because of the limited number of samples. To address this, dry leaf datasets were spiked with extra weights using fresh leaf and dry stem datasets, which notably improved prediction accuracy. Combining VisNIR and MIR spectra did not enhance model performance, indicating that a single spectral region acquired with one spectrometer is sufficient for reliable and rapid nutrient estimation at the field level. This study demonstrated the effectiveness of MIR-ATR spectra for cotton nutrient prediction, despite challenges posed by moisture which has not been previously explored. The calibration transfer techniques can improve prediction robustness across tissues and fresh or dry conditions.</p> Purpose <p>Conventional laboratory-based nutrient analyses are destructive, costly, and time-consuming, delaying timely agronomic decisions. To overcome these limitations, this study investigated the use of visible near-infrared (VisNIR) and, more importantly, mid-infrared (MIR) spectroscopy with Attenuated Total Reflectance (ATR) sampling; an underexplored approach in plant nutrient sensing. A further objective was to evaluate model transfer strategies to improve prediction performance across tissue types and moisture conditions.</p> Methods <p>Fresh and dried cotton tissues (leaves, stems, and burs) were scanned using ATR and Diffuse Reflectance (DR) spectrometers in VisNIR and MIR regions. Concentrations of 11 macro (N, P, K, Ca, Mg, S) and micronutrients (Fe, Mn, B, Cu, Zn) were predicted using Partial Least Squares Regression (PLSR), Cubist regression trees, and Support Vector Regression (SVR). Models were calibrated with 75% of the dataset and evaluated on the remaining samples. To enhance robustness, dry leaf datasets were spiked with extra weighted spectra from fresh leaves, stem, and burs.</p> Results <p>PLSR and Cubist provided comparable results, with PLSR selected for its efficiency. Dry leaf VisNIR models achieved strong prediction accuracy for macronutrients (R²: 0.75–0.96) and micronutrients such as B, Mn, and Cu (R²: 0.78–0.93). </p> Conclusion <p>Fresh leaf models were less reliable due to moisture interference, while stem and bur models were weaker due to limited sample size. But with extra-weighted spiking, predictions accuracies were improved confirming the feasibility of using dry leaf model to predict nutrients in other parts of the cotton plant. </p>