Remote Sensing Monitoring And Forecasting Models Of Wheat Powdery Mildew Based On Multi-Source Data

Under the influence of climate change and other factors, the occurrence and development of wheat powdery mildew in China has changed, and the harm is increasing, which seriously affects the quality and yield of wheat. Remote sensing data plays a significant role in monitoring and predicting occurrence and development of crop diseases and pests owe to its advantages of large area, fast and nondestructive.However, how to choose and adopt appropriate methods, effective integration of multi-source remote sensing data, to maximize the mining data of information benefits and construct a simple and universal monitoring and predicting models is an important issue. In this study, wheat powdery mildew was as the object, differentiating model and monitoring model of wheat diseases and pests based on satellite multi-spectral data, and disease forecasting model which also integrated of site meteorological datawere developed at the regional scale. The specific research contents and results are as follows:(1) Compared with the relatively simple, pure experimental environment, the actual farmland often exists the situation of different types of diseases and pests mixed occurrence, and their controlling, drugging and other management needs to take different measures, otherwise, it will bring a series of problems such as drug injury,soil pollution, et al. We aimed to identify different diseases on crops by using remotely sensed image in classification and regression tree framework. Two different diseases(powdery mildew and aphid) in winter wheat were used in this study. The best optimal vegetation indices (VIs) were extracted from 23 feature variables which been derived from Landsat 8 TM using independent t-test. Their identifiable capability was examined by classification and regression tree (CART). The best identification models for healthy winter wheat and whereas infected with powdery mildew, aphid and both powdery mildew and aphid (mixed disease) were MFIH model, MFIPM model, aphid-infected wheat identifying model based on difference vegetation index (DVI) and mixed disease-infected wheat identifying model based on structure intensive pigment index (SIPI), respectively, and their accuracies were 82.4%, 73.5%, 91.2% and 88.2%,respectively. We also differentiated three diseases through CART method with VIs selected by minimal redundancy maximal relevance (mRMR) algorithm based on MFIH model, and the validation accuracy ofthree calssified diseases-infected wheat differentiating model based on perpendicular drought index (PDI) was highest with 82.6% in all modified diseases-infected wheat differentiating models. Hence, it is evident that the features based on satellite remotely sensed image can be a promising way to discriminate diseases at a regional scale.(2) In order to explore the importance of modeling feature selection methods in disease monitoring models and improve the monitoring accuracy of wheat powdery mildew by remote sensing, the study using the Landsat 8 TM data, extracted total eighteen commonly used characteristic variables. Then we got two groups different features by correlation analysis (CA) algorithm and minimal redundancy maximal relevance (mRMR) algorithm, respectively. Then we used AdaBoost method and common classification method fisher linear discriminant analysis (FLDA) and support vector machine (S VM) to monitor wheat powdery mildew occurrence severity (healthy,slight, severe) in western Guanzhong Plain, Shaanxi province, China through two groups features which obtained by two different feature selection methods mentioned above. Model with mRMR algorithm combining AdaBoost method (mRMR-AdaBoost model) produced highest Spearman relevance value (0.868) in six models.Moreover, the values of Somers’D, Goodman-Kruskal Gamma, and Kendal’s Tau-c of mRMR-AdaBoost model were all highest than those of models with C A algorithm and models with mRMR algorithm which constructed by FLDA and SVM methods. It indicated that mRMR-AdaBoost model had a better performance than other five models. The validation results showed that: the overall accuracies and the kappa coefficient of AdaBoost models with CA and mRMR algorithms are 81.4%, 0.685 and 88.4%, 0.807, respectively, and they are higher 27.9%, 27.9%, 14.0% and 9.3% than FLDA and SVM models with corresponding selection algorithms. The overall accuracies of FLDA, SVM and AdaBoost models with mRMR algorithm are higher 7.0%, 11.7% and 7.0% than the corresponding methodological models with CA algorithm. Furthermore, mRMR-AdaBoost model is with lowest omission and commission error in all six models. Additionally, compared with the spatial distribution results of wheat powdery mildew severities which mapped by SVM and AdaBoost models and combined surface survey results of wheat powdery mildew occurrence severity, the mapping results of mRMR-SVM model and two AdaBoost models are similar and close to ground survey results, and among them, the results of mRMR-AdaBoost model is the closest to ground reality than others’. These results revealed that for remote sensing monitoring of crop disease, the application of AdaBoost method has a good prospect, and for feature variables selecting of crop disease monitoring model, the minimal redundancy maximal relevance algorithm has more advantages than CA algorithm. It indicated that the study result can provide a method reference for monitoring of other crop diseases.(3) Compared with monitoring, forecasting the occurrence of crop diseases and pests can alert tile stakeholders to take preventive measures effective in real time and reduce yield losses. Based on the traditional meteorological data disease forecasting models, the indices of wheat growth status and habitat factor was introduced through the Landsat 8 TM image data, and additionally, the spatial features corresponding to the site meteorological data was obtained through transforming by spatial interpolation analysis, and three models with different types of spatial data were constructed by using relevance vector machine method was used to forecast occurrence probability of wheat powdery mildew in grain-filling stage, and the results were compared and analyzed. The results show that the overall accuracy and the kappa coefficient of the remote sensing & meteorological data model are 84.2%, 0.686, respectively, have shown better performance over the remote sensing data model (80.0%, 0.602) and the meteorological data model (74.7%, 0.500). These results revealed that compared with the single meteorological data and the single remote sensing data, the combination of remote sensing data and meteorological data is more suitable for the prediction of crop diseases occurrence situation in the range of regional scale.