Prediction of tree species composition in fine root mixed samples using near-infrared reflectance spectroscopy

Fine roots ( ≤ 2 mm diameter) are of great value when investigating belowground interactions among different plant species and soil nutrient cycling in forest ecosystems. However, fine root separation and species identification are labor-intensive and time-consuming processes. This study aimed to evaluate the aptitude of near-infrared reflectance spectroscopy (NIRS) in predicting tree species composition in fine root mixed samples. The coniferous species Cunninghamia lanceolata and Pinus massoniana, the deciduous species Alniphyllum fortunei and Liquidambar formosana, and the evergreen broadleaved species Cyclobalanopsis glauca represent the five subtropical tree species selected for this investigation. To obtain near-infrared reflectance spectral data, 20 samples taken in the field and 70 artificially mixed samples of the five species were produced after root samples were oven-dried and ground. Calibration was performed with partial least squares regression and leave-one-out cross-validation. Root mass proportions of the mixed samples showed good predictive capacity for C. lanceolata, P. massoniana, and C. glauca with low root mean square error of prediction ( < 6.82%) and high determination coefficients (R²>0.944). Predictions for A. fortunei and L. formosana were acceptable with R²>0.819. NIRS shows potential in predicting tree species composition with suitable accuracy.