基于NDVI与物候修正的大豆长势评价方法

及时、准确的作物长势监测可以为宏观决策和农田生产提供作物生长信息,便于及时采取各种田间管理措施,达到科学管理和作物增产的目的。归一化植被指数(normalized difference vegetation index,NDVI)与植被的叶面积指数(leaf area index,LAI)和叶片叶绿素含量关系极为密切,可以用来评价作物的生长状况。为了降低主观因素及物候差异对大豆长势监测的影响,该研究以黑龙江红星农场主要农作物大豆为例,基于历史NDVI数据建立了该区域大豆长势评价的标准。利用NDVI时间序列拟合法提取大豆关键物候期,结合物候监测结果对大豆长势进行修正,最后利用41个地块的单产数据对长势评价结果进行了验证。物候修正前后长势与单产的一致性分别为58.5%、75.6%,容差为1个等级时分别为87.8%、95.1%,表明历史NDVI对大豆长势评价有一定参考意义,但简单同期对比不能完全反映大豆长势真实情况,物候修正可以进一步改善长势评价效果。研究可以为利用遥感进行大豆长势评价提供参考依据。

Soybean growth assessment method based on NDVI and phenological calibration

Abstract: The timely and accurate crop condition monitoring can provide government policy makers and farmers with information of crop growth, so that they can take promptly field management measures to achieve scientific management and crop yield-increasing. With the development of remote sensing technology, crop condition monitoring by remote sensing has become a research hotspot. Former studies have shown that normalized difference vegetation index (NDVI) is highly correlated with leaf area index (LAI) and leaf chlorophyll content, and can be used to indicate the growth condition of crops. However, it is hard to eliminate the influence of objective factors on crop condition monitoring due to the lack of evaluation criteria. Beside the crop growth condition difference itself, the phenophase difference between fields also has a great influence on crop condition monitoring. To address the problems above, a method of assessing soybean condition by using the historical NDVI time series was designed. In this study, the research target is soybean in Hongxing Farm, which is located in Heilongjiang province. Based on the available multi-spectral HJ-1 CCD data, the historical NDVI dataset from the year 2010 to 2014 was collected. The NDVI variation trend in soybean growth season was analyzed and an inter-annual comparison during soybean growth period was implemented which was integrated with ground data. The high-quality images can''t be acquired at day frequency due to the temporal resolution and the cloud influence. A linear interpolation was thus applied to the original data to obtain everyday NDVI dataset. Then a profile, which reflected the soybean growth process was built according to the reconstructed NDVI data from 2010 to 2014. The profile can provide four threshold values every day to categorize soybean condition into five grades which is worst, poor, fair, good and excellent respectively. On the basis of that, the criterion of soybean condition classification was established to assess the growth condition of soybean accurately. A preliminary soybean growth condition map can be produced according to the criterion. Then the critical phenological stages of soybean was extracted based on NDVI time series to reduce the phenophase impact on crop condition monitoring. First, an everyday NDVI dataset at pixel scale in the year 2014 was built using linear interpolation. A simple procedure based on Savitzky-Golay filter and Gaussian function was then implemented to remove the noise in the contaminated NDVI data and fit the growth curve. The result of soybean growth condition monitoring can be calibrated on the basis of the correlation of NDVI and the podding date which can be estimated by analyzing the characteristics of the curve. The yield investigation at field scale was carried out to validate the accuracy of the soybean growth condition assessment method. The results showed that the consistency between soybean condition level and yield level reached 58.5% and increased to 75.6% with the phenological calibration. The pre- and post-calibration consistency were 87.8% and 95.1% respectively when the tolerance was defined as one grade. The proposed method which based on the historical NDVI dataset and the result of key phenophase mapping indicates the ability of reducing the impact of objective factors and phenophase when assessing soybean growth condition.