冬小麦归一化植被指数日变化规律及拟合模型研究

Study on Variation Curve and Fitting Model of Winter Wheat Canopy NDVI

为探究冬小麦归一化植被指数(NDVI)主要生育期内的日变化规律,分别在冬小麦返青期、拔节期和抽穗期,利用Greenseeker手持式光谱仪每日以小时为单位获取冠层NDVI值,各生育期内连续测量7 d,分析冬小麦冠层NDVI在3个生育期的日变化规律,并采用二次多项式、Gauss和Sine等函数对归一化处理后的NDVI日变化过程进行拟合。结果表明,冬小麦冠层NDVI在3个生育期有明显的日变化规律,其变化趋势近似一条反向抛物线; 3种模型均能较好地对NDVI日变化曲线进行拟合,且在拔节期拟合效果最好;二次多项式模型的预测精度最高,3个生育期内其相应的决定系数(R2)分别为0.744、0.923和0.681,均方根误差(RMSE)分别为0.212、0.213和0.187,平均绝对误差(MAE)分别为0.165、0.162和0.142,Gauss和Sine函数拟合效果基本无差别;二次多项式模型作为描述NDVI日变化过程的首选模型。本研究可为今后建立冬小麦NDVI日变化模型提供参考。

Normalized Difference Vegetation Index( NDVI) is an important tool for assessing crop growth condition and has been widely used in agriculture field. It is a vegetation index based on the high absorption rate of red light and high reflectance of near-infrared light. And it has been found to be dynamic at different times in a day because of crop itself and environmental factors,so the accurate determination of NDVI is difficult. To explore daily changes of NDVI at main growth stages of winter wheat,the winter wheat canopy reflectances in the 656 and 770 nm wavelengths were obtained by Greenseeker to compute NDVI. The data were obtained in successive hours at reviving stage,jointing stage,and heading stage,respectively. The study proves that the winter wheat canopy NDVI values are dynamic in different periods of a day; the NDVI data demonstrates clear parabolic shaped diurnal variations; it decreases gradually from 8 ∶ 00 AM,and reaches to its minimum at 13 ∶ 00 PM or 14 ∶ 00 PM followed by a rapid increase in the afternoon. In order to describe the variations of the daily NDVI values,the quadratic polynomial regression,Gauss function and Sine function was used to fit the normalized NDVI daily variation curve respectively. Before fitting,a normalization processing for the original data was made to limit the data in the same range,which was convenient for the compare of different models. In significance test,the selected models were all statistically significant( P < 0. 01) in the three growing stages of winter wheat. And the three models had the best fitting effect in the jointing period with the coefficient of determination( R2) all above 0.9. But the quadratic polynomial model expressed better stability than the other two models. Then the predicted and measured values were compared and the best fitting model was found by root mean square error( RMSE) and mean absolute error( MAE). The results show that all three models have good fitting effects,however,the prediction precision of quadratic polynomial model is the best with the RMSE of 0.212,0.213 and 0.187,and MAE of 0.165,0.162 and 0.142 in the three stages of reviving,jointing and heading,and the other two models show almost the same prediction accuracy. This study can provide a reference for the future study of the NDVI daily change process and NDVI accurate monitoring of winter wheat.