关键词: 稻纵卷叶螟, 高光谱遥感, 水稻产量, 植被指数, 估算模型

Abstract: In 2020, a hyperspectral observation experiment was conducted on the rice fields of Cnaphalocrocis medinalis Güenée (C. medinalis) has been occurring naturally in Qiaolin Subdistrict, Pukou District, Nanjing, to explore the relationship between the spectral characteristics of rice leaves and the yields under the different infestation level of C. medinalis, and predict the yields of rice. 80 samples were selected in the experiments, and the different pest levels were divided according to the proportion of the number of infested leaves to the total number of leaves in the sample points. SOC710VP, a portable hyperspectral imager, was used to collect the hyperspectral data of rice leaves at different main growth stages (jointing stage, booting stage, grouting stage, mature stage), and the rice yield data of samples were investigated. The variation of the original spectral pattern, first derivative spectral characteristics of rice leaves and the yield parameters under the different pest levels was analyzed. The vegetation indices were calculated by using the characteristic bands with the strong correlation between the observation spectrum and yield, and the rice yield estimation model based on these vegetation indices was established. The results were showed as follows: (1) during the same growth period, the reflectivity on the near-infrared and red-edge bands of rice leaves decreased with the increasing of pest levels, while the red band was the opposite. (2) During the same growth period, the peaks of the first derivative spectrum, the amplitude of the red edge and the area of the red edge decreased with the increasing of the pest levels, and the "blue shifting" of the red edge position was aggravated. (3) The effective panicle number, 1000 grain weight, firming rate of rice and rice yield decreased with the increasing of the pest level, but when the pest levels were low, the effective panicle number, 1000 grain weight and firming rate of rice all rebounded. (4) The rice yield estimation models were constructed using DVI, RVI and CARI at each growth stage, of which the binomial model of RVI had the best effect. (5) The long-term and dynamic monitoring of the hazards of C. medinalis’ infestation can be monitored by using the imaging spectral characteristics of rice leaves, and the vegetation indices constructed from its sensitive bands can effectively estimate the rice yield under the infestation of C. medinalis.

Key words: Cnaphalocrocis medinalis Güenée, Hyperspectral remote sensing, Rice yield, Vegetation index, Estimation model