Dynamic Monitoring Of Vegetation Ecological Water Based On Spatial-temporal Image Data Fusion

Water is the source of life and an indispensable resource for the survival and development of human beings.On the land surface,there are various hydrological environments.Different hydrological environments have different impacts on the growth of vegetation.Vegetation ecological water（layer）refers to the amount of water trapped or contained by each part of the vegetation growing on the land surface.Vegetation ecological water reserves can regulate the recycling of water resources,so it is of great significance to study vegetation ecological water reserves for rational utilization of water resources and protection of ecological environment.The upper reaches of minjiang river are located in the western sichuan plateau region with complex terrain,unique geographical location,complex natural environment and fragile ecological environment.There are various types of coniferous forest,broad-leaved forest and shrub in this area,which store a large amount of vegetation ecological water,which is very suitable for the study of ecological water.However,due to the special location of the study area,belongs to the special mountain area because of the influence of the fog rain and snow and so on,again carries on the field data synchronous acquisition is difficult,general methods are difficult to obtain,qualified high resolution image data is not enough,less serious limitations on the seasonal change dynamic monitoring,and therefore difficult to achieve the overall distribution and effective detection of space-time change rule.Therefore,this paper adopts the method of spatio-temporal data fusion,combines the high spatial resolution data and the high temporal resolution data fusion to obtain the"double-high"precision data for ecological water inversion,and obtains the spatio-temporal variation rule of ecological water in the research area.In this paper,Maoxian County,Aba Tibetan and Qiang autonomous prefecture,Sichuan Province,was taken as the research area.Based on Landsat8 OLI and MODIS data,STARFM time-space fusion model was used to obtain"fusion"data with high spatial and temporal resolution in the research area.By using band reflectance,vegetation index,texture index and terrain information as the landmark parameters of vegetation ecological water information,combining with field data,the inversion model of multiple linear regression,random forest and neural network was established through feature quantity optimization.By means of cross validation,the fitting performance of different models was compared,and the ecological water reserves of different vegetation types in the study area were obtained.Different vegetation types and materials have different water content,and different vegetation water content in different growing seasons.Therefore,vegetation ecological water content is affected by its own characteristics,growth cycle and seasonal changes.This paper completes remote sensing inversion and dynamic monitoring of vegetation ecological water in the research area.The main research results are as follows:（1）adaptive spatiotemporal data fusion algorithm（STARFM）was used to obtain the fusion imageLandsat remote sensing data is an important data source for ecological water research,but due to the influence of time resolution and rainy weather,it is difficult to obtain multi-temporal remote sensing data of the same research area.Therefore,it is of great significance to make full use of the advantages of multiple remote sensing data and obtain multi-temporal remote sensing data in the same research area.This paper takes Maoxian county as the research area and USES STARFM model fusion to generate data with high spatial and temporal resolution as the research data source.The original Landsat8 OLI images and fusion images were used to assess the accuracy of STARFM algorithm on November 13,2016.For real images and"fusion"images in 2016,the accuracy was assessed by visual interpretation,scatter map（determination coefficient R²:0.81-0.87）,and calculated mean absolute error（AAD:0.0021-0.0267）.The results showed that:The fusion image generated by the STARFM model fusion has a high correlation with the original image,which proves that the STARFM algorithm can better predict the image,and provides an empirical study on the adaptability of the algorithm.（2）model correlation index extraction and model establishmentThe indexes extracted in this paper mainly include band reflectance,vegetation index,texture index,terrain information and vegetation types in total 50.Among them,based on the spectral reflectance of the vegetation growing period（Landsat8 OLI:2018.04.09）,spectral reflectance of the vegetation non-growing period（fusion image:2018.10.31）and terrain information,the overall accuracy of classification by the random forest model reached 89.09%,and the Kappa coefficient was 0.87,satisfying the application requirements.The introduction of multiple indexes in the model can make up for the limitation of using only one vegetation index and effectively improve the accuracy of the prediction model.In this paper,multiple linear regression,random forest and neural network models are used to optimize the characteristic quantities of the extracted indexes,and the vegetation moisture content inversion model is established respectively.The results showed that the change rate of R²in the random forest model was the lowest,only 10.96%,the highest R²in the cross validation of the random forest model（R²:0.65）was the highest,and the root mean square error RMSE was the smallest(RMSE:56 Mg ha^-1).The random forest model was more in line with the actual situation that different vegetation types have different ecological water content.Therefore,RF model can better estimate the ecological water reserves of vegetation in the whole research area.（3）remote sensing inversion and dynamic monitoring of vegetation ecological water in the research areaOn the one hand,the ecological water content of vegetation is affected by the seasonal changes of the environment,such as rainfall and temperature,on the other hand,it is also affected by vegetation types.The ecological water reserves of vegetation in different seasons and different types of vegetation in the study area have obvious differences.In this paper,the ecological water content of vegetation in four time periods is inverted by remote sensing images in four different months.From the perspective of time change characteristics,the ecological water reserves in four months are ranked as August（7.49×10⁷Mg）>in April（4.57×10⁷Mg）>in November（3.88×10⁷Mg）>in January（3.38×10⁷Mg）.From the perspective of spatial variation characteristics,the ecological water reserve of coniferous forest is relatively stable,and the ecological water reserve of broad-leaved forest and irrigated grass is the highest in August.The ecological water reserves of the same vegetation type in different months are different,and the proportion of the total ecological water reserves is also different.