Study On Dynamic Monitoring Model Of Drought In Anhui Province Based On Remote Sensing

Drought is one of the major natural disasters facing the world.It has the main characteristics of long duration,wide range of influence and serious disaster losses.The traditional drought monitoring mainly uses the meteorological data of the ground stations to calculate the drought index by spatial interpolation,and obtains the spatial drought distribution map in the region.The traditional drought monitoring method has high precision in a small range,but it is only suitable for small-scale drought monitoring.Traditional meteorological drought monitoring is faced with many problems such as large number of stations,small coverage and large workload,which makes it difficult to realize large-scale dynamic drought monitoring.With the rapid development of remote sensing technology,it provides a large-scale dynamic monitoring method for drought monitoring,and has a good effect on drought monitoring.The state index and the distance index can reflect the drought of vegetation to some extent.Multi source data can synthesize the information of drought factors of multiple planting.A lot of research has proved that machine learning method has a strong data mining ability.Mining remote sensing data based on machine learning method will improve the accuracy of drought monitoring.In order to monitor drought more accurately,based on the detailed description of the scientific research progress of drought remote sensing monitoring at home and abroad,the paper discusses the research progress of various machine learning methods and drought remote sensing monitoring models.The paper constructs the multi linear regression model to study the drought dynamic monitoring in Anhui Province,and obtains the following main conclusions:(1)The paper proposes that the potential of the three-dimensional contact number subtraction set is applied to the selection of remote sensing temperature index and soil moisture content index.It is found that the daytime temperature is suitable for the construction of the model as temperature data,and the surface soil moisture content is suitable for the construction of the model as soil moisture data.In order to improve the monitoring effect of drought,based on the calculation principle of the vegetation index,the paper puts forward the temperature index,the precipitation index and the water content index of the soil.In order to select the daytime temperature or night temperature in remote sensing temperature data,and to select the soil moisture content in the profile,soil water content in root area or surface soil moisture in the soil moisture content,the original data,state index and distance index of temperature and soil moisture are calculated based on the set potential method of three-dimensional contact number reduction method,The index categories of temperature and soil moisture content were determined.The results show that the pair potential of the three-way contact number subtraction set of daytime temperature raw data,state index and distance leveling index is-0.098,0.509,-0.147,and the pair potential of the three-way contact number subtraction set of night temperature original data,state index and distance leveling index is-0.032,0.492 and-0.110.The pairing potential of the original data,state index and distance leveling index of the profile soil moisture content is 0.197,0.241 and 0.329.The pairs of the original data,state index and distance leveling index of soil moisture content in the root area are 0.277,0.325 and 0.402.The pairs of the original data,state index and distance leveling index of surface soil moisture content are 0.282,0.359 and 0.474.Compared with other indexes,the correlation coefficient of day temperature and surface soil moisture content with the Pearson of SPI1 is the highest.(2)The effect of the model of remote sensing monitoring and fusion of the distance index and the state index was studied.The state index and the distance adjustment index are respectively combined with the multivariate linear regression model,the classification regression tree model and the random forest model.The results are as follows: state multiple linear regression model,state classification regression tree model,state random forest model,distance leveling multiple linear regression model,distance classification regression tree model and distance leveling random forest model.The results show that the correlation coefficient between the state index fusion model and SPI1 grid is 0.849.The mean correlation coefficient between the offset index and SPI1 grid is 0.880.From the angle of grid correlation coefficient,the distance index is only slightly better than the result of state index.The model values are classified and the calculated grades are compared with SPI1.The mean accuracy between the state index fusion model and SPI1 level of the verification set is 0.732.The mean accuracy between the fusion model of the distance and SPI1 level of the verification set is 0.810.From the point of view of accuracy of each model and SPI1 level,the distance leveling index is more suitable for the integration of drought remote sensing monitoring model.(3)The effect of the fusion model of the distance leveling index and many machine learning methods is studied.The paper combines the remote sensing temperature index,normalized vegetation index,precipitation data,soil moisture content data with a variety of machine learning models,and constructs the multi linear regression model,the distance classification regression tree model and the distance leveling random forest model.Among them,the independent variables of the model are the temperature index,vegetation index,precipitation index and soil moisture content index.SPI1 is the dependent variable of the model,and 2001-2010 is taken as the training set of the model and 2011-2014 as the verification set of the model.The results show that the mean values of correlation coefficients of SPI1 grid of the distance leveling multivariate linear regression model,the distance classification regression tree model,the distance leveling random forest model and the verification set SPI1 grid are 0.870,0.866 and 0.905 respectively;The accuracy of SPI1 grades of the distance leveling multivariate linear regression model,the distance classification regression tree model,the distance leveling random forest model and the verification set SPI1 are 0.821,0.804 and 0.804 respectively;The results show that the three models have good drought monitoring results.The results show that the results of the model are consistent with the actual drought.In 2011-2014,the frequency of drought free was the highest,and the frequency of light drought was higher,and the moderate drought happened occasionally,and severe drought was rare,and special drought never occurred,which was consistent with the actual situation.This conclusion has practical value in monitoring drought,its spatial and temporal changes and development.In conclusion,the potential of the three-dimensional contact number reduction method is applied to the selection of remote sensing temperature index and soil moisture index,and the types of remote sensing temperature and remote sensing soil water content index are obtained,and the state index and leveling index of temperature,vegetation,precipitation and soil water content are calculated respectively.The results of the study on the drought monitoring results of the fusion model are carried out.The results show that the monitoring effect of the model is the best.This conclusion provides theoretical basis for accurate drought monitoring,enriches drought monitoring model and provides technical support for decision makers to formulate drought control measures.