Monitoring Of Winter Wheat Stripe Rust Using Hyperspectral Remote Sensing Data

The experiment was carried out on the National precision agricultural research basement of Xiao Tanshang, Beijing from 2002 to 2004. The paper focus on the extracting disease incidence(DI) of stripe rust on the winter wheat from the canopy spectral reflectance. Stripe rust was inoculated by the stripe rust epiphyte according to plant protection standard, and DI were gained in field at an interval of 5 days after inoculation. The winter wheat canopy spectral reflectance of different level treatments has been measured using ASD Field Spec instrument whose spectral region is between 350nm and 2500 nm and the data has been analyzed. The results are as follows:Firstly, the important agronomical parameters such as DI, leaf fluorescence dynamical parameters and photosynthesis etc. have been measured synchronous with the spectral measurement. The biochemical parameters such as leaf total nitrogen, total sugar and starch content have been measured in the lab. The correlation between those parameters and the DI has been analyzed. It is indicated that those parameters changed regularly.Secondly, the correlations between those biophysical and biochemical parameters and the canopy spectral reflectance have been analyzed quantitatively. Those sensitive bands region to those parameters have been found and those sensitive regions to DI are similar to three types of winter wheat. Furthermore, those sensitive regions include TM2(520~600nm),TM3(630~690nm) and TM4 (760~900nm) of Landsat/TM. Therefore, it is possible to monitor the DI of winter wheat stripe rust using the Landsat/TM remote sensed data and the variety parameter can be neglected while monitoring the DI. Meanwhile, the red-edge of canopy spectral has been analyzed and those sensitive bands region have been selected out to establish models with the DI. It indicated that dλmin, dλred/dλmin and the NDVI defined by 634nm, 634nm and 823nm have good correlation with the DI, and R2 has reach above 0.8. So those models can be use to support the monitoring DI.Thirdly, the canopy spectral and the PHI airspace have been analyzed; the results showed that hyperspectral remote sensing could diagnose the stripe rust within the best curing time (diseased leaf ratio under 5%). The SRSI (stripe rust stress index) model has been put forward firstly time in this paper. The monitoring precision above 75% can be assured for monitoring DI using this model. Fourthly, as we known, the problem that different matters have similar spectral feature is difficult to classify those matters for remote sensing. This paper used the NDVI defined by 830nm and 675nm and PRI defined by 531nm 570nm to tell the stripe rust from water and nitrogen stresses. Meanwhile, it is obvious that LAI has influence on the DI monitoring. The paper researches how to eliminate this problem. It is indicated that the 680nm of first order is immune from the LAI influence and has good correlation with DI.Fifthly, the spectral feature of winter wheat single leaf infected with different severity level quantitatively. It is indicated that the spectral properties is similar to the single leaf and canopy spectral at the visual light region, but is reversal at the near infrared. The mechanism of this phenomenon has been explained. The SAI model and AAI model defined by sensitive bands to the single leaf severity level, which has good correlation with single leaf severity level has been constructed.Sixthly, the hyperspectrcal remote sensed data of experimental basement has been obtained by PHI. Meanwhile, the DI monitoring model established using the ground remote sensed data has been used to classify the stripe rust on the map acquainted by PHI. The result of classification is greatly identical with the fact.