杉木木材结晶度的近红外预测模型建立及变异分析

A near infrared prediction model and variation analysis of wood crystallinity in Cunninghamia lanceolata

木材结晶度的测定费用高、耗时长,所以基于近红外光谱技术快速测定结晶度很有实用价值。杉木Cunninghamia lanceolata是中国南方栽植面积最大的用材树种之一,结晶度作为衡量木材品质的重要指标,了解其变异对杉木无性系选育及木材加工技术改进都有实际意义。本研究样品来自广西、湖南和贵州等11个不同地理种源的杉木无性系,通过X衍射技术测定结晶度,结合近红外光谱技术通过偏最小二乘法建立相应的模型并对模型预测效果进行评价。通过模型预测未知样品,进而分析杉木木材结晶度在不同种源和无性系间的变异。当光谱区域为6 000~4 100 cm-1,光谱数据进行二阶导数和Savitzky-Golay平滑处理,运用偏最小二乘法建立的校正模型预测效果最好。校正模型相关系数r=0.987 5,校正均方差(RMSEC)为0.318;预测模型的相关系数r=0.921 3,预测均方差(RMSEP)=0.742。用未参与建模的已测定样品进行模型评价,其预测值与实测值之间的相关系数为0.905 0,平均标准偏差为0.301,该模型能预测杉木样品的结晶度。对164个杉木无性系木材结晶度测定结果的分析显示:平均值为44.52%,变幅为40.49%~49.75%,区间为42.06%~47.28%的占72.86%。按照地理种源分布,杉木木材平均结晶度湖南靖县种源的最小为43.45%,贵州黎平种源的最大为45.23%,方差分析表明种源间无显著差异(P=0.000 3),而无性系间具有显著差异。

Understanding variation of wood crystallinity, an important indicator of timber quality measurement,of Cunninghamia lanceolata, one of the most widely planted timber trees in southern China, is important for C.lanceolata clonal selection and wood processing technology improvements. For an inexpensive and less timeconsuming method of rapid crystallinity determination, near infrared spectrum technology was tested. Using 164 C. lanceolata clones from 11 different geographic origins such as Guangxi, Hunan, and Guizhou Provinces, a near infrared spectroscopy prediction model of wood crystallinity was established by the partial least squares（PLS） method in combination with X-ray diffraction techniques, and then evaluated. Next, unknown samples were predicted through the model, and the variation of crystallinity was analyzed. Results showed that when using a spectral region of 6 000-4 000 cm^-1, the second derivative spectrum, and PLS method, the calibration model had the best prediction effect. The calibration model correlation coefficient was r = 0.987 5, and the root mean square error of calibration（RMSEC） was 0.318. Verifying the model revealed r = 0.921 3 and root mean square error of prediction（RMSEP） was 0.742. Using unknown samples not involved in modeling to evaluate the model, predicted and measured r = 0.905 0 with an average standard deviation of 0.301. So, the model could predict the crystallinity of C. lanceolata. Then, wood crystallinity determination results of 164 C. lanceolata clones showed that the average value was 44.52%, the range was 40.49%-49.75%, and the value between42.06% and 47.28% took up 72.86%. According to the distribution of geographical provenances, the average wood crystallinity of C. lanceolata had a minimum of 43.45% from Jing County, Hunan, and a maximum of45.23% from Liping County, Guizhou. The variance analysis showed no significant difference among the provenances, but there were significant differences for clones（P = 0.000 3）. The results indicate that near infrared spectroscopy could be used for the establishment of reliable prediction model, and the selection of improved varieties should be carried out among clones.