Classification Of Coastal Mariculture Area And Its Spatial Characteristics In China

Mariculture plays an important role in the development of marine fishery in China,which offers significant contributions for the food security and economic development.With the rapid development of coastal mariculture in recent years,there is an urgent need to extract and monitor the coastal mariculture areas in China,which provide importat information for the coastal marine spatial planning,marine ecosystem conservation,and the sustainable development of coastal resources.Remote sensing images,especially with medium and high spatial resolution,are regarede as a tangible data resource for spatial information extraction in various applicaitons.Using remote sensing images with medium/high spatial resolution as main data source,this study attemped to extract coastal mariculture at different scales with different methods.The spatial patterns of mariculture areas in China were then analysed.The main contents and conclusions are as follows:(1)For finer resolution mapping of mariculture areas at local scale,coastal mariculture areas was accurately extracted from World View-2 images by using object-based classification methods.First,the land area was masked by using the SEa TH method.To obtain the optimal scale parameters for multi-resolution segment algorithm,the peaks of local variance curve was used as evaluation index for the segmentation.After that,this study used a variety of neighbor features,combining with feature space selection and nearest neighbor classification method,to establishe an effective classification framework for classification of coastal mariculture areas.Results show that the producer and user accuracy values of all classes are greater than 87%,and the overall accuracy is more than 95%.By comparing with other traditional methods,it can be found that multi-scale segmentation strategies and spatial context information can significantly improve the classification performance.(2)To map the finer resolution of coastal mariculture areas at regional scale,coastal mariculture areas were accurately extracted from World View-2 images by using hierarchical cascade network(HCNet)in an end-to-end way.Based on fully convolutional network(FCN),HCNet was training and applied in an end-to-end way.Specifically,an encoder network(based on the VGG-16 network model)is firstly used to transform the input image into high-dimensional abstract feature maps.Due to the high diversity of land covers with different scales in the coastal areas,traditional methods are difficult to obtain satisified classification results.To solve this problem,this study proposed to make full use of spatial context information with a hierarchical cascade structure.In this structure,the atrous convolutional layers are connected in a hierarchical cascade way,with the atrous rate increased layer by layer.And then,the HCNet combined high-resolution features from shallow layers and the attention mechanism in the decoder.Finally,the Softmax classifier was used for classification.According to the accuracy assessment results,the overall accuracy is greater than 95%.It shows that the proposed HCNet can extract different types of mariculture areas,providing an important foundation for intelligent interpretation of mariculture areas from high spatial resolution images.(3)For national-scale coastal mariculture information extraction,this study proposed to use the GF-1 WFV images and a hierarchical cascade homogeneous neural network(HCHNet).The HCHNet was constructed with a homogeneous network and lightweight cascade structure.Affected by pooling operations in the FCN modles,it is difficult to accurately extract mariculture areas based on existing network model methods.The multiscale land covers in coastal areas make it worse.Therefore,this study proposed to use HCHNet,combining multi-scale information while maintaining the full resolution of the features.The accuracy assessment results show that the HCHNet can effectively improve the classification accuracy,with the overall accuracy of 94.9%.(4)Based on the classification results of coastal marine aquauculture areas in the coastal areas of China,using methods such as spatial autocorrelation,hotspots analysis and kernel density estimation from exploratory spatial analysis and landscape metrics,this study investigated the spatial patterns of mariculture in China.Rusults are as follows:1)the total area of coastal mariculture in China is about 1103.67 km~2,in which more than 85%are marine plant cluture(MPC)areas.The area of mariculture shows three levels among different provinces:high,middle and low levels.Specifically,Fujian,Shandong and Liaoning have the largest aquaculture areas,with a total area of more than 100 km~2.Provinces with coastal m mariculture areas at a middle level include Jiangsu and Guangxi,with a total area of 50-100 km~2.Coastal mariculture areas of Zhejiang and Guangdong are at low levels,with a total area of less than 50 km~2;2)as for the spatial distribution of coastal marine aquaculture,the density of mariculture in China is with high heterogeneity.Among them,southern Liaoning,eastern Shandong,northern and southern Fujian,and southwestern Guangxi coastal areas have the most intensive mariculture activities.By comparison,the densities in Bohai Bay,Zhejiang,Guangdong,and Hainan provinces are relatively lower.Secondly,marine animal culture(MAC)areas are mainly concentrated in the East China Sea and South China Sea areas,such as Guangxi,Fujian,Guangdong,and Jiangsu.While MPC areas are mainly concentrated in the Bohai,Yellow Sea,and East Sea areas,such as Fujian,Shandong,and Liaoning;3)we found that the regional differences of culture patterns are obvious.The MAC in Qinzhou Bay and Sandu Island are tend to be larger and denser.In the contrary,Fujian and the gulf region of Guangdong are the gathering areas of smaller mariculture areas.In terms of MPC,higher density of larger mariculture areas are mainly distributed in Ningde city of Fujian province,Rongcheng and Weihai city of Shandong province,and Changhai County of Liaoning province.While higher density of smaller coastal mariculture areas are mainly located in the coastal areas of Fujian,Guangdong and Guangxi.