| Due to its distinctive location,Yellow River delta(YRD)is an important typical region of geographic science research for a long time.With a number of land development activities in the last decades,land surface types have changed dramatically.In order to reinforce the ecological environment protection and implement the sustainable development of the YRD,it is imperative to perform the extraction methods of typical land surface types(vegetation,impervious surface,waterbody)and ecological environment evaluation study in the YRD.With the development of satellite remote sensing techniques,multi-source,multi-resolution and multi-temporal remotely sensed datasets have been increasing rapidly in recent years,which lays an important foundation for the extraction of typical land surface types and ecological environment evaluation study using remote sensing.Therefore,this paper firstly performed the extraction of typical land surface types in the YRD based on machine-learning and collaborative theory using multi-source,multi-resolution and multi-temporal remotely sensed data.Afterwards,the ecological environment change has been evaluated from global and local scales using remote sensing ecological index.Finally,the causes of the ecological environment change have been explored integrating with land surface types data derived from remotely sensed data.The main conclusions and innovations of this paper are listed as follows.(1)A typical vegetation extraction approach of the YRD was proposed based on parallel random forest algorithm,which improved the vegetation extraction accuracy and efficiency.This paper firstly built big data storage and computation platform using hadoop and spark framework.Afterwards,high-dimension feature space was formed using many kinds of features derived from Landsat TM\ETM+\OLI,HJ1A/1B CCD and GF-1 WFV remotely sensed data,such as inner-annual multi-temporal land surface reflectance features,spectral index features,tasseled cap transformation features and texture features.Finally,typical vegetation types of the YRD(wheat,other dry farmland crop,paddy crop,wood,tamarisk shrub,reed meadow,suaeda meadow)was extracted using parallel random forest dependent on big data platform.According to the accuracy assessment of time-series vegetation extraction results in the YRD,it was observed that the proposed method could achieve a high accuracy,with an averaged overall accuracy of 89.79% and an averaged kappa coefficient of 0.8838.Through comparison with nonparallel random forest algorithm,the proposed method for vegetation extraction could boost computational efficiency.The time cost of parallel random forest for vegetation extraction was only 1/4~1/5 of the non-parallel one,and the computational burden decreased significantly.(2)A multi-strategy collaborative approach for the extraction of impervious surface in the YRD,which integrated multi-source remote sensing data collaboration with multi-machine learning algorithms collaboration,was proposed,which improved the extraction accuracy of built-up area and rural resident in the YRD.In terms of multisource data collaboration,a large amount of data were used,including reflectance features in summer or autumn derived from Landsat TM/ETM+/OLI multi-spectral data,inner-annual multi-temporal land surface temperature features derived from Landsat TM/ETM+/TIRS thermal infrared data,inner-annual multi-temporal BCI features,impervious surface fraction features in summer or autumn,annual VANUI features derived from DMSP/OLS,Suomi NPP/VIIRS,Terra/MODIS and NOAA/AVHRR remotely sensed data.In terms of multi-machine learning algorithm collaboration,three machine learning algorithms were adopted,i.e.,support vector machine,multinomial logistic regression,multilayer perceptron.According to the accuracy assessment of time-series impervious surface extraction results in the YRD,it was found that the proposed method could produce an ideal impervious surface extraction result,with an averaged overall accuracy of 88.19% and an averaged kappa coefficient of 0.8647,which justified validity of the proposed method.Through a comparison with a single machine learning algorithm,an increase of approximate 3%~4% in overall accuracy was achieved when using the multi-strategy collaborative approach,which further certified the effectiveness of the developed approach.(3)A collaborative approach for the extraction of multi-waterbody type was proposed through the combination of random forest algorithm and AWEI(Automated Water Extraction Index)thresholding,which improved the extraction accuracy of pond,breeding aquatics and saltern in the YRD.The proposed approach firstly performed the raw extraction of multi-waterbody(pond,breeding aquatics,saltern)through the combination of random forest algorithm,land surface reflectance features and texture features derived from Landsat TM/ETM+/OLI data acquired on summer or autumn.Afterwards,the total waterbody was produced through segmenting AWEI image using threshold from Otsu's method.Finally,the resultant multi-waterbody spatial data was derived through combining these two methods.According to the accuracy assessment of time-series multi-waterbody type extraction results in the YRD,it was found that the proposed method could produce an ideal result,with an averaged overall accuracy of 92.33% and an averaged kappa coefficient of 0.9145,which justified the effectiveness of the proposed method.(4)Through an overall analysis for ecological environment of the YRD using remote sensing ecological index,the area percentage of the ecological environment with superior quality increased by 9.49% while the area percentage of the ecological environment with inferior quality decreased by 6.16%,which indicated ecological environment of the YRD had been improved from 1992 to 2016.Land use/land cover change modified the spatial-temporal pattern of land surface,and reflected profoundly the dynamic change process of surface biophysical parameters.Through an analysis of land use/ land cover change of the YRD,it was found that the area of cropland,impervious surface,and waterbody significantly increased while the area of forest land,grassland,tidal flat,and unused land significantly decreased from 1987 to 2016,which was the direct factor leading to the ecological environment change of the YRD.On the whole,many measures should be taken in order to fulfill the long-term sustainable development of the YRD.The unregulated deforestation and reclamation should be prohibited and the vegetation community(i.e.,tamarisk shrub and reed meadow)positively influencing ecological environment,should be protected.Meanwhile,considering the upcoming urban planning,the proportion of green space and waterbody should be expanded properly in order to make Dongying city a human-environment harmony and sustainable city. |
PDF Full Text Request