研究了基于氮肥效应的冬小麦不同生育期的叶绿素浓度，探讨了XGBoost算法在冬小麦叶绿素浓度估算中的适用性。利用该算法构建了冬小麦叶绿素浓度的高光谱估算模型，并将其与偏最小二乘法（Partial Least Squares, PLS）以及人工神经网络（Neural Network, NN）算法进行了对比。结果表明：（1）冬小麦的叶绿素浓度随着氮肥用量的增加而逐渐升高。（2）基于一阶微分光谱（First-order Differential Reflectance, FDR）数据集的估算模型表现最好。通过对比建模数据集与验证数据集的决定系数R2和相对分析误差（Residual Predictive Derivation, RPD）发现，XGBoost算法的效果最佳。（3）通过波段重要性分析发现，XGBoost算法的8个重要波段均在738~753 nm范围内。与8个常用的红边指数相比，通过XGBoost算法筛选到的8个一阶微分光谱波段对叶绿素浓度的准确估算起到了更加重要的作用。该算法可以作为一种有效的高光谱信息挖掘手段来估算冬小麦的叶绿素浓度。

The chlorophyll concentrations in different growth stages of winter wheat based on the effect of nitrogen fertilizer are studied, and the applicability of XGBoost algorithm in estimating the chlorophyll concentration of winter wheat is discussed. A hyperspectral estimation model for chlorophyll concentration in winter wheat is constructed using this algorithm which is compared with partial least squares and artificial neural network algorithms. The results show that:(1)The chlorophyll concentration of winter wheat increases gradually with the increase of nitrogen fertilizer.(2)The estimation model based on the first-order differential spectrum data set has the best performance. The XGBoost algorithm is found to work best by comparing R2 and RPD of the modeling and verification data sets.(3)Through the band importance analysis, it is found that the 8 important bands of XGBoost algorithm are all within the range of 738~753 nm.Compared with the 8 commonly used red-edge parameters, the 8 first-order differential spectral bands screened by the XGBoost algorithm play a more important role in accurately estimating chlorophyll concentration. This algorithm can be used as an effective hyperspectral information mining method to estimate the chlorophyll concentration of winter wheat.