Mapping Plastic-mulched Farmland With Multi-source Remote Sensing Data

Plastic-mulching can increase agricultural production by improving hydrothermal conditions,promoting crop growth,increasing crop yields and mitigating the effects of drought and flooding,cold and heat,insects and diseases,made a significant contribution for ensuring agricultural production and food security in China.However,the massive use of the plastic mulching,on the other hand,causes a lot of environmental problems.To ensure agricultural production and protect the ecological environment,it is extremely urgent to obtain the spatial distribution information of plastic-mulched farmland(PMF).In this paper,a variety of remote sensing data(optical and radar data)were used to analyze the remote sensing characteristic of plastic-mulched farmland systemically.On the basis of this,the local variance method,machine learning algorithm of random forests and support vector machine were applied to select an appropriate spatial scale,select the optimal mapping phase,and optimize the single-temporal mapping features and the multi-temporal combination.Got the following results:(1)The multi-temporal Landsat-8 satellite images were used to map the plastic-mulched farmland,and the results show that the spectral features from Landsat-8 image were the most important feature,the index feature and textural feature were the important features also for mapping plastic-mulched farmland.The optimal phase for mapping is April,and the second important phase is May.Multi-temporal combined data provide the better result than the single-temporal data.Among them,the combination of the data from April and May provised the best result.(2)GF-1 satellite data were used to select the appropriate spatial scale or scale range for mapping plastic-mulched farmlands and analyze the spatial scale effect.Results found that the 8 m-20 m spatial resolutions were selected as an appropriate spatial scale range for mapping plastic-mulched farmland both in Jizhou and Guyuan.The mapping accuracy decreased with the coarsening spatial resolution.However,there were differences between spectral and textural features.The accuracies generated from textural features varied more severely compared to that from spectral features with coarsening spatial resolution.(3)The mapping accuracies of plastic-mulched farmland with Radarsat-2 data were very low.Despite the introduction of polarization decomposition features improve the mapping accuracy,the accuracy also lower than 80%.In different polarization decomposition featres,the H/A/Alpha decomposition features was better than the others.(4)The combinations of the optical remote sensing data and radar remote sensing data can improve the mapping accuracy significantly.The combination of GF-1,Landsat-8 and Radarsat-2 provide the highest accuracy.The combination of GF-1 and Radarsat-2 provide the higher accuracy than the combination of Landsat-8 and Radarsat-2 data.On the whole,the PMF mapping accuracy using single-temporal Landsat-8 OLI spectral features is higher than that of GF-1 spectral features.The accuracy generated from single-temporal multi-type features of Landsat-8 is higher than that of combined spectral and textural features of single-temporal GF-1 data.For GF-1 data,the textural features can significantly improve the mapping accuracy,and for Landsat-8 data,the textural features can not improve the accuracy obviously.For Landsat-8 data,the multi-temporal features perform better than the single-temporal features.The mapping accuracy of Radarsat data alone is not ideal.However,the combination of radar data and optical remote sensing data can improve the accuracy siginicantly.