GlobeLand30, as one of the best Globe Land Cover (GLC) product at 30 m resolution, was developed by China based on the integration of pixel- and object-based methods with knowledge (POK-based approach), which combines multiple levels of classification techniques to achieve high-accuracy land cover mapping. In particular, a knowledge-based verification process was employed to refine and grantee the product quality of Globeland30 by manual interpretation of online high-resolution images. However, the manual intervention suffers from large labor consumptions and the subjectivity influence. Considering the great achievements of deep learning in image recognition and classification, classifying online high-resolution remote sensing images with Deep Convolutional Neural Network (DCNN) may improve the efficiency and performance of the refinement procedure for GlobeLand30. However, the training of DCNN relies on a large number of training samples; and the existing remote sensing sample sets cannot satisfy the training requirements in terms of sample size and category system. Therefore, a method for generating large sample set of high-resolution remote sensing imagery based on multi-source GLC was proposed; and a large sample set with 2.24 million samples is automatically generated by this method. The DCNN model (GoogleNet Inception V3) was trained from scratch with the proposed large sample set and then used to refine Globeland30 product. Verification with an independent test sample set shows that the proposed trained DCNN model can achieve higher classification accuracy (Overall accuracy: 87.7%, Kappa: 0.856) than that of original GlobeLand30 product (Overall Accuracy: 75.1%, Kappa: 0.71). Finally, four test areas were selected for evaluating the performance of proposed refinement procedure. The results show that the GoogleNet (InceptionV3) model trained by the proposed large sample set can effectively refine the product quality of GlobeLand30.
Deep learning algorithms such as Convolutional Neural Network (CNN) can learn the representative and discriminative features in a hierarchical manner from the remote sensing data. Considering the low-level features as the bottom level, the output feature representation from the top level of the network can be directly fed into a subsequent classifier for pixel-based classification, the CNN has a broad application prospect in the field of agricultural remote sensing. The advantage of CNN in feature extraction can obtain the crop classification in complex planting structure area from multi-spectral remote sensing data, which is difficult in conventional methods. In this paper, a crop mapping method using remotely sensed spectral and context features based on CNN from Landsat OLI data is proposed and applied in Yuanyang county.The architecture of the proposed CNN classifier contains eight layers with weights which are the input layer, two convolution layers, two max pooling layers, two full connection layers and output layer. These eight layers are implemented on spectral and context signatures from 4 different phase Landsat OLI images to discriminate different crops against others. Experimental results demonstrate that the proposed CNN classifier can achieve better classification performance than support vector machines in spectral domain. The context features calculated by the gray level co-occurrence matrix method from Landsat OLI data can enhance the proposed CNN method to achieve the best results.In terms of verification accuracy, the proposed CNN classifier is superior than SVM in spectral domain. The overall accuracy of the two methods is 95.14% and 91.77%, respectively. The accuracy of the proposed classifier is further improved by adding spatial context features on the basis of spectral information. The overall accuracy and Kappa coefficient of the proposed method is 96.43% and 0.952.Furthermore, the crop mapping using spectral and context features based on CNN achieves better spatial representation especially for peanut and roads which is easy to form mixed-pixel. The context features can be extracted by the CNN to enhance the feature representation of these small objects.The CNN-based method from remotely sensed spectral and context features for crop mapping can achieve outstanding performance especially for the fine ground objects in complex planting structure area such as peanuts and roads.
To provide a new idea for large-scale tree species identification, the UAV is used to obtain optical images, and is associated with the theory of deep learning to establish tree species recognition models. First, the Anxi County in Fujian Province is taken as an example, UAV was photographed at different heights of 20 m and 40 m to obtain aerial images of trees. Second, using the tree species as the object, aerial images were segmented to obtain 12 species of tree images. Finally, combined with the deep learning theory, DenseNet is used to establish the tree species recognition model, and the effects of different aerial heights and different depths of network on tree species recognition are discussed. The classification accuracy of tree species identification models with different aerial heights reached more than 80%, and the highest precision was 87.54%. From the analysis of the resolution of aerial image, with the decline of the resolution of aerial image, the accuracy of model presented a downward trend. The tree species recognition model constructed with 20m aerial image data had a higher classification accuracy than the 40m model. From the depth analysis of the network, with the increase of the number of network layers of the model, the classification accuracy of the model decreased. The accuracy of the DenseNet121 model was higher than that of the DenseNet169 model. Based on UAV aerial images and combined with deep convolutional neural network, a new tree species identification method was proposed. The identification of forest tree species in Anxi County was used as an example to prove the validity of the classification framework.
The classification of high-resolution satellite remote sensing image scene information is of great significance for image analysis and interpretation. The traditional high-resolution satellite remote sensing image scene classification method mainly relies on the artificially extracted middle and low-level features and can not make good use of image-rich scenes. In response to this problem, a classification method based on band feature fusion and GL-CNN (Guided Learning Convolutional Neural Network) is proposed. Firstly, the high-low frequency sub-band of the image is extracted by NSWT (Non-Subsampled Wavelet Transform), and then the high-frequency sub-band is fused to obtain the fused high-frequency sub-band, and then the angular energy distribution curve is combined. The stationary interval analysis realizes the fusion of the fusion high-frequency sub-band and the low-frequency sub-band, and finally guides the convolutional neural network to automatically extract the high-level features contained in the high-low frequency sub-band of the image to realize the scene classification. Experiments on UCM_LandUse 21 data show that the classification accuracy of this method reaches 94.52%, which is significantly improved compared with previous algorithms.
Winter wheat is the main food crop in Shandong area. It is of great significance to obtain accurate information of winter wheat planting structure for the study of food security. By expanding the RefinNet model, an Ex-RefineNet(Extend-RefineNet) suitable for extracting the information of winter wheat planting structure was formed. Ex-RefineNet consists of two submodels, the Ex-RefineNet-Edge submodel used to extract the edge pixels of the winter wheat growing area, Ex-RefineNet-Innner submodel is used to extract the inner pixels of winter wheat growing area. Finally, using Bayesian model the extraction results of the sub-model are merged to form the final extraction results. A total of 16 GF-2 images were used for comparative experiments in Jinan City and Tai'an City, Shandong Province, and 2/3 of each image was used as training data and other data were used as test data. In terms of average accuracy, total search rate, and Kapp-coefficient, results of the Ex-RefineNet model were 0.93, 0.92, and 0.91, respectively, while results of the RefineNet model were 0.86, 0.84, and 0.83, respectively. The extraction effect of the Ex-RefineNet model is significantly higher than that of the RefineNet model. Results showed that the Ex-RefineNet is advantageous to extract the structure of winter wheat.
Oil tanks are industrial facilities for storing oil products, which are commonly used in industrial parks such as oil refineries. The rapid detection of oil tank target through satellite or aerial remote sensing images can quickly find suspected industrial parks, providing scientific and technical support for natural resource regulation and ecological environment protection. This paper discussed the possibility of object detection with high-resolution remote sensing images based on deep convolutional neural network. The state-of-the-art algorithms of Faster R-CNN (Convolutional Neural Network) and R-FCN (Region-based Fully Convolutional Network) and three network models were applied for oil tank detection from high-resolution remote sensing images. To promote the detection accuracy and execution efficiency for the oil tank target, an improved approach by increasing the scales of the anchor was proposed. The optimum recall reached about 80%. The results confirm that deep learning network approach can rapid detect oil tank from high-resolution remote sensing image. This provide an example and new idea for rapid detection small target from remote sensing image by deep learning technology.
This paper addresses the buildings segmentation in high resolution remote sensing image and proposes an Encoder-Decoder architecture of deep learning with End-to-End model, in which Encoder is based on ResNet, and the features needed by segmentation are exacted automatically, and the Decoder produces the segmentation result by deconvolution. Furthermore, in the training process, batch normalization is employed to decrease the gradient competition, so as to reduce the difficulty of training the deep neural network.The experiment results show that the algorithm effectively exacts the bulk feature and edge information of building in the high resolution remote sensing image, therefore the complex road disturbance is suppressed convincingly, and the building segmentation precision is improved effectively, the segmentation precision for three typical buildings, the building besides complex road, the ordered buildings and the complex single building, are 0.836 5, 0.892 4, and 0.629 7 respectively; and the F-measure are 0.851 4, 0.878 6 and 0.729 8, respectively. Meanwhile, the experiment results for multi-resolution remote sensing images show that the method can be generalized to the multi-resolution image within limits.
Quantitative analysis on the relationships between the remote sensing scale and the land cover classification accuracy, which is the basis for making a decision on remote sensing resolution determination, is essential for mapping the concise land cover. Up to now, deep learning is an innovative algorithm to learn the hierarchical layer features without supervised control, which is different from the traditional classifiers that require man-made labels as input. Therefore, it is interesting to explore the inherent relationship between the classification accuracy and remote sensing image spatial scale from this algorithm. In this paper, we applied a Deep Convolutional Neural Network (DCNN) which is Pyramid Scene Parsing Network (PSPNet) on four scale remote sensing image (8 m, 3.2 m, 2 m, 0.8 m) to map the wheat distribution. The experiment results showed that the PSPNet is good at learning the spatial feature without manual operations, then the wheat extent could be extracted automatically. Different from the conventional algorithm of determining the optimized spatial resolution, the PSPNet could identify the wheat better accompanying with the spatial resolution increased and more concise wheat results could be obtained. This conclusions represent that deep convolution neural network can take full use of the spatial information of the high remote sensing image to ensure the performance of wheat extent, which brings us a new idea of improving the accuracy of crop mapping adequately if we can get the super-high resolution remote sensing image.
Remote sensing is an important approach for crop growth monitoring efficiently and subjectively, and is helpful for the agricultural productivity. In this paper, Longkang Farm in Anhui province, China, was selected as a case for the study. Remote sensing images with middle-high spatial resolution from different satellite-based sensors were collected and quantitively processed. Statistics models for the estimation of chlorophyll density and leaf area index were built based on vegetation indices. Time-series products of vegetation parameters were produced. We analyzed the temporal patterns of chlorophyll density and leaf area index and found that the high-yield wheat grew much better than the low-yield wheat during the winter. In addition, we built a yield prediction model based on the Normalized-Difference Vegetation Index (NDVI) for winter wheat. The results showed that, using accumulated NDVI at heading and milk stage, the yield can be accurately estimated. The winter wheat yield prediction map of Longkang farm was produced based on time-series satellite images. This study provided an efficient approach for crop growth monitoring.
In order to study the hyperspectral response model of LAI in different growth stages under lodging stress and improve the precision of LAI hyperspectral response model, LAI and canopy spectral reflectivity of lodging maize at different growth stages were obtained. Six traditional transformation methods were used to deal with hyperspectral reflectivity, and the LAI stages and unified response models of lodging maize at different growth stages were constructed. The traditional spectral transformation is beneficial to improve the sensitivity of spectrum and LAI and the precision of model response. The LAI stage response model of lodging maize at different growth stages was superior to the unified response model. LAI staging monitoring model of lodging in different growth stages is better than unified monitoring model. The results can effectively diagnose the maize leaf area index under lodging stress, provide theoretical basis and technical support for accurate monitoring of lodging growth in different growth stages, and provide necessary prior knowledge for maize lodging stress monitoring.
UAV hyperspectral remote sensing is a new means of low-cost, high-precision acquisition of fine-scale crop biophysical parameters and biochemical parameters, so that the rapid inversion of Leaf Area Index (LAI) has a crop growth assessment and yield prediction. Taking the corn of Shandong Yucheng as the research object, using the PROSAIL radiation transmission model to simulate the corn canopy reflectivity to obtain the LAI characteristic response band，combining correlation quantitative analysis to obtain the most sensitive band for LAI changes, and calculating the 6 vegetation index (VI). Inversion models were modeled on a single sensitive band and VI using six regression models to measure the accuracy of the model by LAI.Studies have shown that the spectral reflectance of 516nm, 636nm, 702nm, 760nm, 867nm are most sensitive to LAI changes, and the single-band inversion model established to predict LAI accuracy (R 2=0.44~0.58; RMSE=0.16~0.18).The model established by 636nm (LAI=21.86exp(-29.47R636)) has higher prediction accuracy than other inversion models (R 2=0.58, RMSE=0.16); The 6 vegetation indexes are closely related to LAI with correlation at a significant level(R 2=0.85~0.86). The accuracy of the established inversion model is improved compared with the single characteristic band inversion model (R 2=0.66~0.72,RMSE=0.12~0.14);The LAI estimation model (LAI=exp(2.76~1.77/mNDVI)) constructed by mNDVI has the highest accuracy (R 2=0.72, RMSE=0.13). UAV hyperspectral remote sensing is an effective means for rapid and non-destructive monitoring of crop growth information, and provides a basis for guiding fine-scale crop management.
The extraction of paddy rice planting area in low fragmented regions based on remote sensing images is a hotspot in crop monitoring. Taking Gaoxin district of Suzhou city in Taihu Lake region as a case study, the rice and underlying water spectral characteristics in critical phenophase were studied in-depth to reduce the demand of remote sensing images, and only two GF-1 WFV images with resolution of 16 m during rice tillering and full heading stages were employed to extract the paddy rice planting area. Two vegetation index methods, including difference of Normalized Differential Vegetation Index (NDVI) and the combination of difference of Normalized Differential Water Index (NDWI) and Ratio Vegetation Index (RVI) were studied. The results suggested that both the methods effectively promoted the extraction precision, comparing with the results of supervised classification and unsupervised classification methods. The area recognition accuracy, space consistency mapping accuracy and kappa coefficient of NDVI method were 86.2%, 66.1%, 92.2% and 0.72, while those of NDWI-RVI method were up to 95.5%, 78.4%, 93.5% and 0.85, respectively. The two methods realized the purpose of accurately extracting rice area in low fragmented regions by using a few medium and high resolution remote sensing images, and can be effectively serviced for actual production and relevant decision support in Taihu Lake region.
The noise analysis, evaluation and denoising of remote sensing image are the focus of RSI processing. In order to improve the denoising ability of remote sensing image, presents a new structured dictionary-based method for multispectral image denoising based on cluster. This method incorporates both the locality of spatial and the correlation across spectrum of multispectral image. Remote sensing image was divided into different groups by clustering, and sparse representation coefficients of spatial and spectral and dictionary is obtained according to the dictionary learning algorithm. After threshold processing, the similar blocks are averaged and realized with multispectral remote sensing image denoising. This algorithm is applied to the denoising of remote sensing image of typical vegetation and soil types in the upper reaches of Minjiang river- Maoergai experimental area. Compared with the band-wise K-SVD algorithm, the PSNR of this algorithm can be improved by about 7.6%, with better visual effect.
Change detection based on change vector analysis can quickly extract change information between multi-temporal images by directly comparing pixel differences. However, because the spatial context information in the pixel field and the difference and complementarity between bands are ignored, it is difficult to eliminate the "pseudo-changes" caused by noise and other factors in the detection results. In view of this, this paper proposes a method for detecting changes in spatial and spectral information. Firstly, the image is enhanced by the principal component analysis method, and then spatial context information of pixels is extracted by constructing a new multi-directional differential descriptor; On this basis, a spectrally weighted fusion strategy based on inter-band correlation is proposed to obtain a uniform variation intensity difference image Finally, the EM algorithm is adopted to confirm the final change pixels. The experimental results show that the proposed algorithm can effectively deal with the "pseudo-change" interference and significantly improve the accuracy and reliability of the change detection.
The spectra field measured or from remote sensing images are mostly mixed spectra because of the influence of mixed pixel. An improved derivative of ratio spectroscopy was proposed in order to solve it in hyperspectral remote sensing applications. Firstly, the measured mixed spectra of rocks and vegetation were preprocessed, reduce moisture noise. Secondly, it was decomposed by EEMD (Ensemble Empirical Mode Decomposition) to get r component spectra. Thirdly, r component spectra were unmixed by derivative of ratio spectroscopy. Finally, the area ratio of rocks of mixed spectra can be calculated through regression analysis using the area ratio of rocks as independent variable and characteristic band reflectivity of near infrared spectroscopy as dependent variable.
(1) It is feasible to use EEMD to decompose spectra to get r component spectra, managing to eliminate environmental interference, get overall trend and reflect the spectral characteristics of the main features in the mixed spectra. (2) R component spectra were unmixed by derivative of ratio spectroscopy, which inhibits the influence of vegetation and highlights the influence of the rocks. (3) The inversion accuracy of mixed spectra of rocks and vegetation is improved using EEMD and derivative of ratio spectroscopy.
Soil spectrum is the physical basis of monitoring with remote sensing, the research of saline soil spectral characteristics is of great significance for monitoring soil salinization. In this paper, coastal saline soil took from the Yellow River delta was selected as the research object. Through field sampled and indoor processing, Indoor spectra（350~1 100 nm） of coastal saline soil were measured. The characteristics of hyperspectral reflectance and absorption with different salinity were studied after eliminating the influence of moisture and soil texture, and then soil spectral prediction model was built. The results show that the reflection characteristics of spectra and the absorption peak could be decrypted more accurate and effective after smoothed spectral curves. The soil spectral curves with different salinity degree were similar and parallel in shape, while there were greatly differences among them. no obvious rule. After continuum removal was applied to soil curves, the absorption of light saline soil was minimal at 490 nm. Absorption of severe saline soil was more intense in 760~920 nm. The original spectrum couldn’t predict soil salinization information, while the transformation of second-order differential could improve sensitivity of spectral data, and spectral prediction model could basically meet prediction requirements.
Groundwater storage is an important parameter in water resource manages and
of land-surface processes and hydrological cycle. However, the traditional instruments are very difficult to monitor global high temporal-spatial groundwater water storage and its variability without a comprehensive global monitoring network of hydrological parameters due to high cost and high labor intensity. The recent Gravity Recovery and Climate Experiment (GRACE) mission provides a unique opportunity to directly measure the global groundwater storage and its change at multi-scales from August 2002 to February 2011. In this paper, the global terrestrial water storages with monthly resolution are derived from approximate 10 years of monthly GRACE measurements (2002 August-2011 February), and the groundwater is obtained with subtracting the surface water, snow, ice and canopy water, from total terrestrial water storages using the GLDAS (Global Land Data Assimilation System) model. Results have shown that significant annual variations of groundwater storages are found at the globe with amplitude of up to 50 mm. in Southeast Asia, northern Amazon, while in South Australia and North Africa the annual amplitude is about 10 mm. The secular trends of groundwater storage are also observed at specific areas. For example, in northern Amazon, the groundwater storage is increasing at about 6mm/a due to recent floods, in La Plata, the groundwater storage is declining at about 7.5 mm/a because of drought, in Turpan Basin, the groundwater storage is declining at about 3.1 mm/a, and in North China Plain, the groundwater storage is declining at about 4.8 mm/a.
The settlement monitoring in the city is conducive to the understanding of the regional real-time elevation, which can provide the data basis for the geological disaster and protection department to avoid the geological disasters caused by the loss of elevation. Based on January 2016 to December 2017, a total of 22 scenes Sentinel-1A wide interference pattern of imaging data, the surface deformation monitoring of Wuhu city was carried out using PSI and DInSAR technology, and the spatial and temporal distribution characteristics of ground subsidence in the study area were analyzed. In space, the overall pattern of ground subsidence in Wuhu city is expounded, and the settlement distribution pattern of the road is analyzed. In time, monthly analysis of land subsidence in the specific changes in the year. Results show that the spatially, Wuhu, the range of land subsidence mainly concentrated in the east of the Yangtze river, presents the trend of increase gradually from west to east, west of the Yangtze river region of land subsidence is sporadic funnel type distribution. Among them, the settlement accumulation is also related to the density of the roads and the construction of roads, the settlement accumulation of road gathering area and construction area is large. In terms of time, the overall settlement volume of the study area was more uniform in each month, among which the variation range of settlement volume was the largest in June, and the smallest in October and November.
Based on the Grey System Theory, combing the three periods (2013, 2015, and 2016) of the RADARSAT-2 Polarimetric SAR (PolSAR) data and the vegetation physical parameters data collected from Poyang Lake wetland, we established the relationship model for the vegetation physical parameters with vegetation biomass and the polarization decomposition components, respectively. We analyzed the contribution of different vegetation physical parameters to biomass accumulation and their influence on polarization decomposition components. The results show that from the vegetation growing faster to slower stage, the plant parameters and underlying surface parameters are the main factors that contribute to the vegetation biomass accumulation. The main effective factors for the polarization decomposition components are the land surface parameters and the stem parameters.The parameters of field sampling are analyzed and determined based on the larger correlation degree data at each stage.
Using Landsat-8 OLI images and 296 survey samples in Fujian province, we extracted pure vegetation pixels biased on pixel unmixing models, and divided the samples into coniferous forest, broad-leaved forest and mixed forest, then employed tree height, plantation age and slope as attribute information from pure vegetation samples, and also extracted NDVI, RVI form Landsat8 OLI, and HV, HH backscatter coefficient form SAR image, so as to compose multiple factors with optical features (NDVI, RVI, tree height, plantation age, slope) and SAR features (HH, HV, tree height, plantation age, slope) for comparison study. Since optical remote sensing can only observe vegetation canopy information rather than the whole vegetation information, we firstly estimated the leaf biomass by using multiple regression with optical features, then estimated the above-ground biomass indirectly in line with the relationship between above-ground biomass and leaf biomass. Since SAR L-band with long wavelength can penetrate the canopy and directly observe the whole vegetation information above the ground, we used multiple regression with SAR features to directly estimate the above-ground biomass. Finally, we analyzed and compared the estimation accuracy from the two regression methods. The result shows that the estimation accuracy from multiple regression with optical features (coniferous forest: R2=0.483, RMSE=29.522 t/hm2; broad-leaved forest: R2=0.470, RMSE=21.632 t/hm2; mixed forest: R2=0.351, RSME=25.253 t/hm2) is higher than that from multiple regression with SAR features (coniferous forest: R2=0.319, RMSE=28.352 t/hm2; broad-leaved forest: R2=0.353, RMSE=18.991 t/hm2; mixed forest: R2=0.281, RMSE=26.637 t/hm2), suggesting the indirect above-ground biomass estimation from multivariate regression with optical information is more suitable than direct above-ground estimation from multivariate regression with SAR information in Fujian Province.
The carbon sink function of terrestrial ecosystem is an important aspect of ecosystem service function, playing an important role in mitigating climate change. Accurate assessment of the spatial and temporal change of carbon source / sink of terrestrial ecosystem is an important basis for predicting climate change effectively. Based on the relationship between all forming fluxes of regional carbon source / sink combining with MODIS GPP data products and regional statistical data, we analyzed the spatial distribution carbon source / sink of Liaoning province from 2000 to 2014. Results show that: ①The spatial distribution of carbon source / sink decreased from eastern to western, with the highest value appearing at the eastern (>250 gC m-2 a-1). There are significant carbon source in the central, western and northern of Liaoning province. ②The total carbon emission of Shenyang(1.43 TgC a-1) accounted for about one third of Liaoning province(4.56 TgC a-1). Therefore, Shenyang is the major city of carbon emission in Liaoning province. ③Shenyang is a carbon source. The urban areas of Shenyang showed a lowest carbon emission (only 26 gC m-2 a-1). This paper conducted regional simulation of carbon source / sink, providing theoretical basis and methodological references for the related studies in other regions.
Droughts are one of the more normally natural hazards on a year-to-year basis. And the drought are significant and widespread, affecting economic development, agriculture and people health at any one time. There are various drought indexes have been developed to monitor this hazard which arise precipitation decrease, soil moisture deficit and vegetation stress. However, single drought index cannot consider all of these anomaly to warn and monitor the drought. In this study, remote sensing based data including GLDAS climate data involve precipitation and soil moisture, GLEAM ET and GRACE dataset simulated terrestrial water storage are used to calculate multiple drought indicators including SPI, SMI, ESI and TWSC. These drought indicators refer to anomaly of precipitation, soil moisture and vegetation water supply, and terrestrial water storage change, respectively. Then they were used combined and compared to track the droughts events in United State of American. The results showed that all the drought indexes performed reliable and consistent with each other well, with correlation coefficient value greater than 0.7. However, the ESI performed more reliable, which can reflect the plant water stress under dry condition, additionally, it can be computed in combined with satellite observed data with high spatial resolution to monitor the drought conditions from field scale to global scale. The vegetation have divergent responses to the meteorological, hydrological and droughts except under grassland. The differences between the three drought indexes increase along with the elevated aboveground biomass. Therefore, the land surface vegetation covered conditions involve canopy structure and feedback between plants and climate, which relevant in a drought monitor is often a curial consideration in determining the application of drought indexes.
Assessment of river wetland ecosystem vulnerability in the Upstream of the Yellow River play an important role for river wetland health assessment, functional orientation and river wetland environmental, it also has a great significance to whole Yellow River ecosystems health status, material exchange and energy flow of water-land ecosystem. In this study, we selected 30 indices，used analytic hierarchy process and the DPSIR Model, analyzed riverine wetlands ecosystem vulnerability in the Upstream Reach of the Yellow River from 1986 to 2014. Results indicated that the vulnerability of the floodplain wetland ecosystem in study reach began to recovered after uniform scheduling of entire river. The vulnerability of riverine wetlands ecosystem gradually decreased from the main channel center line to floodplain margin areas. The vulnerability of the riverine wetlands has a various zonal distribution characters from Upstream to Downstream of the study reach between 1986~2014. Among in, the vulnerability of the Xiaheyan and the Toudaoguai reaches appeared to stable states, the vulnerability changing of the Reach from the Qingtongxia to the Sanhehukou is significantly obvious rather than other reaches.
Boundary delimitation of natural heritage is very important for its protection and management. The existing methods for boundary delimitation generally rely on the prior knowledge and expert experience, and their results are not accurate enough. To better determine the boundary of natural heritage, a new method was proposed in this study. With the support of remote sensing images and GIS (Geographic Information System) technique, Wolong National Nature Reserve was chosen to be the research object to map its scientific and reasonable boundaries. Firstly, AHP (Analytic Hierarchy Process) was applied to analyze and build the hierarchy system of boundary influence factors, and then with the support of GIS model builder, the reconstruction model of giant panda habitat was established, finally the boundaries of Wolong Natural Reserve were digitized with GIS modeling analysis. Compared with the published documentations, the area extracted by the proposed method is more scientific, objective and accurate. Furthermore, the boundary delimitation method proposed in this paper explores the influence of ecological and humanistic factors on natural heritage and provides a clearer boundary for the habitat of giant panda, which also highlights the relationship between humans and nature reserve.
High spatial resolution geosynchronous orbit (GEO) remote sensing satellite technology can carry out repeated high-frequency monitoring over a large maritime space and will obtain the real-time dynamic features of ships on the sea surface. In this study, a single GEO satellite GF-4 (Gaofen-4) multispectral image was used to detect moving ships on the southern Lingdingyang estuary of Zhujiang in Guangdong by using the gradient threshold method. The results revealed that: (1) the ship location and direction can be detected using a single-band image from GF-4 multispectral imagery, and the average error of ship location is about 80.46 m (equivalent to 1.61 image resolution cells). The detection rate of ship direction was about 74.36%, and the absolute error of the direction was about 8.65 degrees; (2) ship speed and direction can be detected using a single GF-4 multispectral image (multi-band collaborative detection), and the accuracy of direction detection is 98.96%, and the average absolute error is 8.78 degrees. The accuracy of navigational velocity detection is 83.0%, and the average absolute error is 1.41 m/s, as verified by AIS (Automatic Identification System, AIS) data. These showed that GF-4 satellite can be used for surveillance and monitoring of ships in near real-time, and has unique advantages and great potential in dynamic changes of ships on the sea.
