Regional Forcasting Of Rice Disease Prevalence Based On Multi-source Spatio-temporal Information

Rice sheath blight and rice blast,which are widely prevalent diseases in rice production,have caused a large-scale reduction of rice production in the world.Establishment of rice forecasting model at regional scale can provide important basis for disease control and management.However,most disease forecasting methods are based on statistical relationships and lack the ability to run on a large scale of business.The study of crop disease forecasting model shows a general trend from static to dynamic,from local to overall.The study makes full use of the ability and superiority of multi-source remote sensing data in monitoring the influencing factors of disease epidemic,and studies the remote sensing monitors of host distribution,reproductive period,landscape pattern and disease environment.At the same time,the study explores the construction method of mechanism forecasting model which could reflect the epidemic characteristics of rice diseases.The study explores the spatial and temporal development of rice disease in the region,provides important information for the control of rice diseases,and achieves the spatial and temporal dynamic forecasting of rice disease at the regional scale by coupling the spatial and temporal information of multiple sources with the disease forecasting model.It mainly includes the following aspects:(1)Construct a multi-source spatial-temporal information dataset for disease forecasting.Taking 6 provinces of Anhui,Zhejiang,Jiangsu,Hunan,Hubei and Jiangxi in the middle and lower reaches of the Yangtze River as the study area,data preprocessing and reanalysis of remote sensing products were carried out based on field plant protection survey data in 2010-2015,land cover type remote sensing products MCD12Q1,rice high-resolution phenological dataset products Chinacropphen1km and0.5°×0.5°grid meteorological daily value products.The spatial-temporal information dataset provides data support for subsequent disease forcasting model construction.(2)Establish an annual peak value forecasting model based on landscape pattern information and meteorological characteristics.Peak value of crop diseases can indicate the limit of infection rate during disease epidemic,play a key role in the process of conditional trend forecasting,and make the disease forecasting curve more in line with the actual curve.In this study,we extracted the meteorological characteristics of temperature and precipitation highly related to disease in study area,calculated the key landscape index in counties,and combined the incidence of rice sheath blight and rice blast,analyzed the correlation between peak incidence and meteorological characteristics,landscape characteristics.Based on the above analysis,a disease peak value forecasting model combining meteorological and landscape features was constructed by using partial least squares regression(PLSR)to assess the accuracy of the model forecasting results for the two diseases.Among them,the accuracy of rice sheath blight peak value forecasting model is 85%,and the accuracy of rice blast peak value forecasting model is 59%,which can give an effective forecasting of the annual peak value of disease.(3)Establish a dynamic mechanism model called SEIR-RICEDIS of rice disease process based on multisource space-time information.The actual development process of crop diseases changes with environmental conditions and is a dynamic process of pathogen-host interaction.Therefore,the establishment of a model reflecting the mechanism and epidemic characteristics of crop diseases can fundamentally improve the adaptability of the model at regional scale.Expanding the data source in time and space can improve the prediction ability of the model in time continuity and spatial dynamics.In order to forecasting rice sheath blight and rice blast,the SEIR-RICEDIS model was built by combining meteorological factors,remote sensing crop distribution,phenological information and peak value module with the SEIR framework based on the epidemiological mechanism of plant diseases.The parameters of model are optimized by genetic algorithm.Its response analysis is also performed.The prediction accuracy of the model is evaluated in terms of single-phase and overall trend.The single-time prediction accuracy of SEIR-RICEDIS model of rice disease process for rice sheath blight R~2=0.75 and RMSE=10.27.Multi-temporal prediction accuracy was evaluated using AUDPC with R~2=0.71.The single-phase prediction accuracy of rice blast R~2=0.54,RMSE=1.78 and multi-phase prediction accuracy R~2=0.73.In conclusion,the SEIR-RICEDIS model can effectively predict rice sheath blight and rice blast with acceptable errors.(4)Set up a prototype system for rice disease forecasting,and conduct its business potential research.The design of rice disease early warning system mainly includes summarizing and integrating data,building the database of disease forecasting parameters;programming the disease forecasting model and designing the user interface to achieve the automatic generation of disease forecasting results.The study achieves rice disease early warning driven by space-time data and generates forecasting results of rice disease occurrence and development process based on rice distribution information.The influence of input information noise on the model is further evaluated by random error perturbation analysis.The results show that the model has strong anti-noise ability and has certain potential for business application.