Fusing Landsat And MODIS Data To Identify Abandoned Farmland

Since the 21 st century,with the rapid development of economy and culture,the process of urbanization is getting faster and faster,and the rural labor force is constantly shifting,some parts of many countries begin to appear large-scale land abandonment phenomenon.Remote sensing technology can indirectly extract the scale and quantity of abandoned farmland,which is of great significance to the protection of farmland and food security,the abandoned land is a special type of land,the abandoned land is fragmented,and the distribution range is scattered,many small areas of the abandoned land is difficult to identify,need to have high resolution image data to identify.However,due to the difficulty of obtaining such data,it is difficult to meet the requirements of identifying large areas of abandoned land.Taking Maoxian county of Sichuan province as an example,through space-time adaptive data fusion model(STARFM)algorithm,using the high spatial resolution image and the high temporal resolution image fusion,the fusion image with high spatial and temporal resolution,and based on the data extraction to get the phenological characteristics of study area,build the different abandoned to identify model and random forest(artificial neural network),and to explore the spatial distribution characteristics of abandoned in the study area was analyzed.The main conclusions are as follows:(1)space-time adaptive data fusion algorithms are employed to model(STARFM)to high spatial resolution and high temporal resolution of single band image fusion,with "combination" precision of the fusion image,based on the convergence of the algorithm is the red band and near-infrared wavelengths in the study area high space-time resolution images,through visual evaluation,scatter plot and the correlation coefficient,the mean absolute error and relative error of the mean comparison,found that the red band the precision of the fusion image than the near infrared band,show that the algorithm for the red band is superior to the fusion effect of near infrared wave band.(2)for different models,they also differ to the precision of the abandoned to identify,this paper will be of the same sample,respectively,the input of artificial neural network model and random forest model,random forest model overall accuracy is 88.91%,the Kappa coefficient is 0.87,and the overall accuracy of artificial neural network is 79.73%,the Kappa coefficient is 0.76,random forest model is better than artificial neural network model.(3)for different characteristic variables,their effects on the precision of the abandoned to identify each are not identical,to recognize the importance of sorting of abandoned them for as follows: the phenological characteristics information > band reflectance > terrain > vegetation index,the overall accuracy: random forest(all characteristic variables)> random forest(no vegetation index)> random forest terrain(no information)> random forests(no band reflectance)> random forests(no phenological characteristics),the classification of the use of all the characteristics of the variable synergy effect is best.(4)the abandonment rate of the study area is 4.6%,and the whole distribution of the abandoned land is scattered and sparse.There is no concentrated phenomenon of the abandoned land,which is mainly distributed at the edge of the cultivated land,and most of the abandoned land is located in the central and western regions with high elevation,while the area of the abandoned land is relatively small in the eastern regions with relatively flat terrain.The degree of regularity of abandoned plots is not high,and the spatial distribution is not uniform.(5)the topography of the study area has a certain degree of influence on the production of local abandoned land.It is found that the area of fallowed land and fallowed rate in the study area generally show an increasing trend with the increase of elevation and slope.Among them,fallowed rate increases significantly with the increase of elevation,but not significantly with the increase of slope.According to statistics,the study area is mainly abandoned in the area with elevation greater than2000 m and slope greater than 15°.In general,the phenomenon of fallow land in the research area mainly occurs in the area with high elevation and slope,and the phenomenon of fallow land is more likely to occur in the area with high elevation.