Research For Forecasting And Monitoring Of Wheat Disease Based On Multi-source Remote Sensing Data

Recent years, diseases and pests were becoming more and more serious with global warming, so it was very urgent and challenging to prevent and control diseases and pests. The monitoring and forecasting of diseases and pests for crop through remote sensing was becoming a very important and effective means of obtaining disease information in field condition, and will be an alternative to manually inspection. The major contents and results in this dissertation are as follows:(1)In order to improve the accuracy of wheat yellow rust disease severity using remote sensing and to find the optimum inversion model of wheat diseases, the canopy reflectance and disease index (DI) of winter wheat under different severity stripe rust were acquired. The three models of PLS (Partial Least Square):BP neural network using seven hyper-spectral vegetation indices which have significant relationship with the occurrence of disease and vegetation index (PRI) were adopted to build a feasible regression model for detecting the disease severity. The results showed that PLS performed much better. The inversion accuracy of PLS method is best than of the VI (PRI, Photochemical Reflectance Index) and BP neural network models. The coefficients of determination(R2) of three methods to estimate disease severity between predicted and measured values are 0.936、0.918and 0.767 respectively. Evaluation was made between the estimated DI and the measured DI, indicating that the model based on PLS is suitable for monitoring wheat disease.(2)For aphid forecasting at regional scale, disease forecasting forms an integral part of crop protection for ensuring quality and quantity of production. A timely forecast of the disease at a regional scale is necessary to control and prevent it. With a correlation analysis, in addition to surface weather data, precipitation, temperature, sun radiation, humidity, and three satellite-based data of normalized difference vegetation index (NDVI),disease water stress index(DSWI) and land surface temperature (LST) at relevant growth stages were identified for constructing a disease forecasting model using the Partial Least Square (PLS).In the model, four meteorological factors between March and May are interpolated by Analytic Method Based On Multiple Regression And Residues method, while three remotely sensed features are obtained through small satellites (HuanJing-lA/B) data. The forecasting model was proven to be able to successfully predict the disease severity in the study area. Compared to the model constructed with meteorological data only, the integrated model constructed with both remote sensing and meteorological data has produced a higher accuracy (increasing R-squared from 0.6806 to 0.7916) of forecasting the stripe rust severity. This suggests that there would be a great potential for predicting the stripe rust occurrence probability by integrating both meteorological and remote sensing data. In the future study, more endeavors should be made to incorporate multi-sources information to developing a reliable disease forecasting model.(3)At region scale, the study conducted the region investigation experiments in Shijiazhuang in the key growth period of winter wheat and acquired the synchronous multi-temporal Landsat8TM images with investigation experiments. This study investigated the relationship between the degree of powdery mildew damage and the indicators normalized difference vegetation index (NDVI), normalized difference water index (NDWI),Disease Water Stress Index(DSWI), and land surface temperature (LST)) derived from Landsat8 TM images. And three 2-dimensional feature spaces were established, including LST-NDVI, LST-NDWI and LST-DSWI. From the scattering pattern of datasets in the feature spaces, it was obvious that LST was a driving factor for the aphid occurrence and NDVI was more sensitive to powdery mildew damage degrees than NDWI and DSWI. Further, the crimination models of powdery mildew damage degree were established based on LST-NDVI feature space. Verification result showed that the overall accuracy of crimination model was78% and the Kappa coefficient was 0.63.Then, NDVI、EVI、DVI and TVI are analyzed to find the difference between healthy wheat and diseased wheat. After that, choose the appropriate vegetation index to distinguish diseased and non-diseased wheat.