Wheat powdery mildew forecasting in filling stage based on remote sensing and meteorological data

Abstract: Powdery mildew is one of the main serious diseases for winter wheat. An accurate and timely forecasting of the wheat powdery mildew occurrence at the regional scale by using multi-source data can provide important information for crop protection decision making, and achieving effective prevention of wheat powdery mildew. In this study, the Landsat8 remote sensing image was used to extract the land surface temperature (LST), the vegetation indices which included normalized difference vegetation index (NDVI), modified simple ratio index (MSR), re-normalized difference vegetation index (RDVI), triangular vegetation index (TVI), optimized soil adjusted vegetation index (OSAVI), green normalized difference vegetation index (GNDVI), and the band reflectance features. Then we obtained the parameters of wheat growth environment condition such as air temperature, number of rainy days with more than 0.1 mm rainfall, total sunshine hour, average relative humidity, temperature-rain coefficient (the ratio of total rainfall in a period of time to average temperature of the same period) and rainfall coefficient (the square root of the product of rainfall and number of rainy days) in different time steps (including month, 10 days and sensitive period) with the site daily meteorological data; and then we got the corresponding space meteorological features by using the inverse distance weighted (IDW) method in GIS (geographic information system) spatial interpolation analysis. Next, we implemented screening features with the combination of relief algorithm and Poisson's correlation coefficient, and finally got the MSR, the RDVI, the total sunshine hour from March 21st to April 20th, and the number of rainy days with more than 0.1 mm rainfall from April 11th to May 10th, which were as optimal explanatory variables for developing the powdery mildew forecasting model. The relevance vector machine (RVM) model was used to improve business decisions, detect disease, and forecast weather. And then we used it to predict the probability of powdery mildew occurrence in filling stage of wheat in Gaocheng, Jinzhou and Zhaoxian County, Shijiazhuang City, Hebei Province through remote sensing and meteorological data. The model combining remote sensing and meteorological data produced a higher Spearman relevance value than the single remote sensing data or the meteorological data model, and moreover, the values of Somers'D, Goodman-Kruskal Gamma, and Kendal's Tau-c of the remote sensing and meteorological data model were all higher than those of the other 2 models. They all indicated that the remote sensing and meteorological data model had a better performance than the other 2 models. The results showed that: the overall accuracy of the remote sensing and meteorological data model was the highest among the 3 methods, with lower omission and wrong judgement than the other 2 models. Furthermore, the overall accuracy and the kappa coefficient of the remote sensing and meteorological data model were 84.2% and 0.686 respectively, which showed better performance over the remote sensing data model (80.0% and 0.602) and the meteorological data model (74.7% and 0.500). These results reveal that compared with the single meteorological data or remote sensing data, the combination of remote sensing and meteorological data is more suitable for the prediction of crop disease occurrence situation in the regional scale.