Regional target decomposition is a key part of remote sensing satellite imaging mission planning for large area coverage， which is of great significance to quickly acquire effective data in large area. Based on the planning experience of satellite photographing in large area， the main influencing factors in the actual operation of large area photographing are summarized， including satellite orbit transit， regional cloud cover at transit window and real-time update of base image data， and a large area photographing decomposition method based on multi-factor superposition is proposed. In this method， the cloud cover factor is applied to the decomposition of the strips in each transit window， and the better photographed strips in each satellite transit are obtained based on the idea of greedy algorithm， which can improve the single data acquisition efficiency and the overall coverage efficiency. This method has been applied to the project of "Data Cube for large coverage datasets of Chinese high resolution and broad band and multispectral satellite constellation"， which provides support for Jilin-1 GP01/GP02 satellite to quickly acquire data in the 65 countries and regions along the Belt and Road. By comparing the acquisition of regional coverage data before and after using the method， the data coverage efficiency was improved by about 44%.
Rapid acquisition of large area data is an important research topic in the field of remote sensing satellite task planning. Relying on the project of "Data Cube for large coverage datasets of Chinese high resolution and broadband and multispectral satellite constellation"，Jilin-1GP01/02 satellite was used to carry out effective coverage of 65 countries and regions along the "The Belt and Road Initiative" twice within three years.This paper summarizes the strategy， methods and experience of data acquisition in large areas of the project， and focuses on various influencing factors such as the resource of satellites and ground stations related to data acquisition，the strategy of dividing large areas in time and phase， and the dynamic planning process of large areas based on effective imaging strips of cloud forecast， that is，within the single transit range of the satellite，，select the imaging strips with the maximum probability of obtaining effective data in combination with the cloud map. The research has provided normalization support for the project，and the relevant methods and project experience can provide reference for the general satellite remote sensing large-scale and wide-area data acquisition tasks.
With the continuous development of the aerospace industry around the world， the satellite imaging business has developed towards the goal of multi-satellite collaboration covering large areas. In this process， multiple objective functions such as maximum coverage area and minimum satellite resource utilization need to be optimized simultaneously. Focusing on the whole process of regional coverage scheduling and data transmission planning of Earth observation satellites， the typical regional decomposition technology is firstly summarized， which plays an important role in satellite scheduling as a preparatory step for satellite regional coverage and makes the solving of combinatorial optimization problems possible. Then， the representative studies of Multi-Objective Evolutionary Algorithm （MOEA） in the field of multi-satellite joint regional coverage scheduling and data transmission planning in recent years are analyzed and reviewed. Common optimization goals include maximizing coverage rate， minimizing overlap ratio， minimizing the number of strips and so on. Finally， we summarize and put forward some prospects for future research， to provide a reliable reference for the application of multi-objective algorithms in related tasks.
The Open Data Cube system is a new generation of open-source earth observation data management system， which can solve many challenges faced by Chinese satellite big data in the management and application fields. However， the system currently supports only international mainstream satellite data loading and services and is still not available for Chinese satellite data. Taking GF-1 WFV data as an example， we loaded and created a service of Chinese satellite data based on the Open Data Cube system. Firstly， according to the characteristics of GF-1 data， the product definition and metadata extraction script are written to index and ingest data， to complete the loading of Chinese satellite data. Then based on the loaded data， the water detection service of GF-1 satellite data is developed using ODC’s Web UI interface. This research proves the feasibility of using the Open Data Cube system to manage Chinese satellite data and provides technical references for other Chinese satellite data loading and service implementation based on this system.
The precise absolute radiometric calibration of satellite on orbit is the basis for quantitative application of its observation data. Compared with the traditional site calibration method， the on orbit calibration method based on stable targets has the advantages of low cost， high frequency and historical data calibration. The MCD19A1 product， MCD43A1 product and MOD03 （MYD03） product of the Moderate-resolution Imaging Spectroradiometer （MODIS） are applied to simulate the directional reflectance of MODIS band8 under the observation conditions of GP02 satellite. Then combined with the directional reflectance of MODIS band1~band5 in MCD43A1 products， aerosol parameters and water vapor parameters in MCD19A2 products and ozone parameters in MOD07 （MYD07） products， the radiation transfer process of the incoming pupil radiance of multiple stable targets imaged by Multi-Spectral Imager of GP02 satellite in October 2021 was simulated to achieve the absolute radiometric calibration in orbit. The validation results of these calibration coefficients show that a smaller difference between GP02’s atmospheric correction results and Sentinel-2 reflectance products is performed when the re-calibration coefficients rather than pre-launch coefficients are applied； the average relative difference between GP02’s measured radiance of all bands after re-calibration and the automatic radiometric calibration data of Baotou is 3.18%， indicating the high accuracy of the calibration results. The research results can provide a methodological support for the on-orbit absolute radiometric calibration of medium or high spatial resolution optical remote sensing satellites using stable targets.
The launch of the Jilin-1GP satellite has enhanced China’s Earth observation capabilities， and has great potential in agricultural quantitative inversion. To invert the key crop parameters accurately and effectively， it is of great significance to analyze the inversion capability of Jilin-1GP satellite images. The farmland of Urad Front Banner， Zhenglan Banner and Horqin Right Front Banner in Inner Mongolia were taken as the study area in this study， and based on the Jilin-1GP images， the optimized PROSAIL model and curve matching algorithm were used to invert the Leaf Area Index（LAI） of maize and rice in different phenological periods， and the accuracy was verified by combining the measured LAI data. Results showed that the parameter range and step size of the optimized PROSAIL model were more suitable for crop LAI inversion， and the capacity of the look-up table was reduced on the premise of ensuring the accuracy； The curve matching algorithm based on eigenvalues improved the computational efficiency by an average of 41.43% when the spatial distribution was highly consistent and the mean absolute value of the error was 0.41； The LAI inversion accuracies R2 of maize and rice in different phenological periods of the study area ranged from 0.72 to 0.9， and the RMSE ranged from 0.32 to 0.49. Among them， the precision of maize in the flowering stage was the highest （R2=0.9， RMSE=0.4）， and the precision of maize in the maturity stage was the lowest （R2=0.72， RMSE=0.47）. This study showed that the crop LAI inversion based on Jilin-1GP images had the characteristics of high precision and small error. The research results can provide scientific methods and basis for the accurate inversion of crop LAI with Jilin-1GP images.
It has become a new method for radiometric calibration and verification of satellite-borne remote sensing instruments by continuously observe the moon whit the earth satellite， unified geometric information of lunar observation data has important reference for radiometric calibration. The position and attitude of the moon image acquired by the earth satellite are relative to the earth at the time of taking the image，due to the inversion shooting of the satellite， each parameter will deviate from the calibration value of the ground， and the geometric correction accuracy is not high.In order to solve the difficulty of geometric correction of lunar images acquired by earth satellites， take the Jilin1-GP as a sample， a method of lunar image geometric correction based on iterative matching is presented to improve the geometric correction accuracy of the image in this paper， and consider the processing of halfmoon image. Experimental results show that this method can solve the geometric correction of lunar observation images obtained by earth satellites which cannot be solved directly， and the geometric correction accuracy is high.
Land Surface Emissivity （LSE） is a key parameter that measures the ability of the object surface to release energy in the thermal radiation. And it plays an important role in Land Surface Temperature （LST） retrieval from the thermal remote sensing data. To evaluate the effect of Land Surface Emissivity （LSE） on the retrieval of Land Surface Temperature （LST）， firstly three groups of Gaussian distribution randoms with different mean and standard deviation values are generated to present the noises of the LSE products. Secondly the well-known Split Window Algorithm （SWA） is selected to retrieve LST with the Advanced Himawari Imager （AHI） data and LSE products added the Gaussian distribution noises. Finally LST difference between retrieved by inputting LSE with noises and that without noise under different conditions （single temporal LST， multi-temporal LST， averaged LST， LST of different water vapor contents and different sensor zenith angles， LST of different land covers） are analyzed. Our study shows that the retrieved LST will be smaller when LSE with noises is input into the SWA； The bigger the noise’s standard deviation is， the bigger the LST difference’s standard deviation will be； When the noise’s standard deviation is 0.01， the standard deviation of the LST difference in day， night and daily average is 0.48 K、0.52 K and 0.34 K relatively. While when the noise’s standard deviation is 0.03， the standard deviation of the LST difference in the three different time is 1.46 K、1.57 K and 0.88 K. At conditions of different water vapor contents and different sensor zenith angles， the results show that the correlation coefficient between the LST retrieved with LSE added noise and that without noise will be smaller with the bigger of the added noise， while the root mean squared error and standard deviation will be bigger with the bigger of standard deviation of the added noise. The bias volue is less than 0， and its absolute will be smaller with the bigger of standard deviation of the added noise. As for different land covers， when the noise’s standard deviation of LSE is 0.01， the LST difference’s standard deviation for woody savannas， open shrubland and savannas is 0.52 K、0.51 K and 0.53 K separately； When the noise’s standard deviation is 0.03， the LST difference’s standard deviation for them is 1.58 K， 1.53 K and 1.6 K.
In the context of global warming and urbanization， the recent decades have been witnessing intensifying Surface Urban Heat Island （SUHI） effect. Investigations on the spatiotemporal patterns of SUHI area （SUHIA） are crucial for better understanding the SUHI effect. By combining MODIS （Moderate-resolution Imaging Spectroradiometer） land surface temperature data， Gaussian model， and Diurnal Temperature Cycle （DTC） model， here we calculated the ratios of SUHI area to urban area （IR） of 504 global major cities during 2000~2019. We further analyzed the hourly， seasonal， and inter-annual variations in IR across different climate zones. The results show that： （1） In terms of the spatial patterns， the multi-year average daytime and nighttime IR of global major cities are 0.85 and 0.75， respectively， with a significantly larger IR in snow climate zone （0.94 and 0.86 for daytime and nighttime， respectively） than in arid， equatorial and warm climate zones. （2） On the hourly time-scale， the IR patterns are very similar across different climate zones. The IR firstly decreases and then increases after sunrise， reaching the minimum and maximum at 3 hours and 7 hours after sunrise， respectively； and it then decreases in volatility and finally becomes stable. （3） On the seasonal scale， the global mean IR is larger in summer （0.86 and 0.76 for day and night， respectively） than in winter （0.81 and 0.72 for day and night， respectively）. The seasonal variations of IR in arid， snow and warm climate zones are similar to those on a global scale， while the situation is reversed in equatorial climate zone. （4） On the inter-annual scale， the annual mean IR shows an increasing trend in 54% of global cities during the daytime， while it shows a decreasing trend in 62% of global cities at night. This study reveals the spatial patterns of SUHI area at multiple time scales， and compares these temporal variations among different climate zones. Our findings contribute to a better understanding of the spatiotemporal patterns of SUHI effect.
Land Surface Temperature （LST） plays an important role in the study of land atmosphere energy exchange. LST changes rapidly with time， and the local solar time of LST obtained by polar orbit remote sensing satellite is different among pixels. Time normalization is needed to improve the application value of LST remote-sensing products. For MODIS LST products， a Temporal-effect Normalization Model of land surface temperature Based on Diurnal variation information （BDTNM） is proposed after the Diurnal Temperature Cycle model （DTC） is introduced and the coarse and fine resolution conversion registration method is constructed based on FY-4A high time resolution LST products. The effects of time window， normalized time and null value on the model are discussed. The normalized results of the INA08_2 model and BDTNM model are verified and evaluated by using the measured and simulated data of stations in Zhangye area. The proposed model has the following characteristics： （1） It can realize seamless reconstruction of FY-4A LST data with different loss rates； （2） Only FY-4A and MODIS LST product data are used to normalize the MODIS LSTs based on retaining the original precision and characteristics of MODIS LSTs； （3） Experiment and evaluation with simulated data， the RMSE and MAE of time normalization of BDTNM model are 0.45 k and 0.32 k， which are higher than those of INA08_2 model （RMSE is 1.36 k and MAE is 1.15 k） ； （4） The BDTNM model is not affected by the data quality and missing values of the other three observations when it normalizes the MODIS observation data at a certain time， and has a certain ability of null value reconstruction. According to the site simulation data， the model reconstructs the MODIS LST data and normalizes it to the standard time， RMSE is 0.53 k， MAE is 0.48 k. The model established in this study can also be used for reference to other remote-sensing satellite LST time normalization.
Based on the night lighting data of Luojia 01 and the energy statistics of Xi'an， combined with the ArcGIS spatial analysis method， this paper uses the high oligomeric model to spatially simulate the carbon emission of Xi'an in 2018， calculate and classify the carbon emission intensity of all districts and counties in the city， and study the distribution characteristics of carbon emission of all districts and counties in Xi'an. The results show that there is a good correlation between Luojia-01 light data and carbon emissions， the linear correlation coefficient is 0.720 3， and the correlation coefficient of quartic function polynomial is the highest， which is 0.843 5； In terms of annual carbon emissions， Xi'an's carbon emissions show the spatial distribution characteristics of high in the central main urban area and low in the surrounding counties， which is a cluster distribution， and the clustering results are clustered in the high value area； There are many low-carbon emission intensity districts and counties in the city， and there are a few high-carbon emission intensity districts and counties. The industrial structure needs to be further adjusted to realize the green development model.
Aerosols can affect the radiation balance of the earth atmosphere system through direct effects. However the research on aerosol direct radiation effect mainly focuses on total aerosol， and there is a lack of research on different types of aerosol direct radiation effect. In this study， the Community Earth System Model （CESM） is used to simulate the direct radiative forcing of total aerosols， sulfate aerosols and carbonaceous aerosols on the top of the atmosphere and the surface， and the simulation results are verified by multi-source data. The results show that there is a good correlation between the total Aerosol Optical Depth （AOD） simulated by CESM and the AERONET （R2 = 0.44）， but the simulation value is relatively small. Compared with MERRA-2， it is found that the optical depth of carbonaceous aerosol is overestimated and the optical depth of sulfate aerosol is underestimated； The radiation flux simulated by CESM and the simulation effect of BSRN are good （R2 = 0.93）.The simulation results show that the direct radiative forcing of total aerosol， sulfate aerosol and carbon aerosol at the top of the atmosphere under clear sky conditions are -1.37、-0.46 and -0.45 W/m2， and -0.30、-0.25 and +0.04 W/m2 under cloudy conditions， respectively. Therefore， the existence of clouds weakens the negative radiation effect of aerosols at the top of the atmosphere and strengthens the endothermic effect of carbonaceous aerosols， showing a positive effect； Under clear sky conditions， the direct radiation forcing on the surface is -5.60、-0.53、-2.21 W/m2， and it is -4.38、-0.32、-1.64 W/m2 under cloudy conditions， respectively. Thus， the direct radiation effect of aerosols presents a negative effect on the surface， and the presence of clouds has little effect on the radiation effect of dust aerosols， but it can weaken the direct radiation effect intensity of sulfate aerosol and carbon aerosol. The results of this study are helpful to further understand the direct radiation effects of different types of aerosols and provide a basis for improving CESM in the future.
Building is the basic unit of urban refined management， the rapid and accurate extraction of urban building footprints based on high-resolution remote sensing images is of great significance for urban planning and management. Based on the high-resolution （0.8 m） remote sensing data of Beijing-2， a sample library of building footprints in Beijing was established. We used multiple semantic segmentation models， U-Net， DANet， UA-Net （U Attention Net） and instance segmentation models， Mask R-CNN， Mask R-CNN FPN， Mask R-CNN RX FPN to extract building footprints， performed accuracy evaluation and compare the extraction effects of different types of buildings （such as buildings， villas and village buildings， etc.）. Finally， we selected the U-Net model with the highest overall accuracy and the best extraction performance to extract all building footprints in the Beijing area. The results show that the classification accuracy of U-Net， DANet， UA-Net， Mask R-CNN， Mask R-CNN FPN and Mask R-CNN RX FPN models are 79.37%， 65.59%， 71.03%， 61.82%， 52.53% and 59.70%， respectively. And the U-Net model training time is relatively short. The U-Net has a good performance for the extraction of building footprints. Comparing the recognition effects of different models， it is found that the semantic segmentation model is more advantageous for the recognition of bungalow buildings， while the instance segmentation model is suitable for single-family buildings and villas in urban and surrounding areas. The study provides a scientific basis for model selection for typical building footprints extraction tasks and our achievement solves the problem of lack of fine-scale research data in cities to a certain extent.
Global wind profile plays a critical role in improving accuracy of numerical weather prediction， accurately describing climate models and understanding the atmospheric physical processes. Under cloud and rain conditions， the Doppler velocity measurement ability of spaceborne atmospheric wind field is highly dependent on radar system parameters， the physical and dynamic characteristics of cloud and measurement method. Based on Polarization Diversity Pulse Pair （PDPP） algorithm， the relationship between pulse timing and Doppler velocity measurement accuracy is studied. The factors affecting the Doppler measurement accuracy of PDPP method are simulated and analyzed. The results show that for W-band radar， the pulse pair repetition frequency should be in the range of 3.5~5 kHz. The antenna with 0.1 ° beam width can measure the in-cloud wind speed with reflectivity factor of about -20dBz in the pulse interval of 30~50 μs， and the accuracy is 1.35 m/s； Under the same accuracy requirements， the antenna with 0.2° beam width can achieve detection sensitivity of about -12 dBz in the range of pulse interval of 30~40 μs.
With majority problems in image scene of optical remote， changing category in classification， variational size in sample， diverse changing of scale between backgrounds and essential objectives， for instance， new Classification Algorithm for scene classification of optical remote sensing image base on attention architecture search of neural network is proposed in this paper. This algorithm can search convolution， pooling， attention and other operations in the neural network， adaptively； and complete the construction task of scene classification for optical remote sensing images in neural network. Two-stage greedy algorithms network search is mentioned in order to ensure the stability of neural search network. This method abandons useless operations in stage which can reduce algorithm burden and improve speed of search. Furthermore， a top-bottom connection strategy of network， which can fully reuse the semantics of multi-scale feature maps in each stage， is proposed to merge information between each object and scene. The experimental results proved that the method proposed in this paper has better performance than the classical deep learning method designed by hand. Overall， the accuracy of this method in all three remote sensing image-standard data sets （AID， NWPU and PatterNet） is exceeding the classic method. The accuracy rate of AID data set， PatterNet data set， and NWPU data set are 94.04%， 99.62%， and 95.49%， respectively.
Solar-Induced Chlorophyll Fluorescence（SIF） is closely related to vegetation photosynthesis and can reveal the true physiological status of vegetation. Accurate acquisition of SIF information is of great significance for terrestrial ecological carbon cycle and global vegetation monitoring. In this study， taking the SIF retrieval results of the 3FLD algorithm and NIRvR as references， the performance evaluation of data-driven SIF retrieval algorithm based on tower-based platform was carried out. Firstly， the SIF retrieval effect of the SVD algorithm in different atmospheric windows was analyzed by using the tower-based spectral observation data. Secondly， using the measured data before and after atmospheric correction， we explored the influence of atmospheric factor on the retrieval of SIF by SVD algorithm. Finally， the measured data are distinguished according to the lighting conditions， and the stability of the SIF retrieval results based on the SVD algorithm is compared under the conditions of stable weather and fluctuating weather. The results show that：（1）The SVD algorithm has higher SIF retrieval accuracy in the 735~759 nm（excluding the atmospheric absorption band） and 745~780 nm（including the atmospheric absorption band） windows.（2） The SIF inversion accuracy of the SVD algorithm is much less affected by the atmosphere than the 3FLD algorithm.（3）When the illumination conditions change drastically， the use of SVD algorithm can effectively overcome the dependence of FLD-type SIF retrieval algorithm on synchronous solar spectrum observation； even if the illumination changes rapidly， a stable and reliable SIF retrieval result can still be obtained based on the SVD algorithm.In summary， the SVD algorithm has great application potential for tower-base SIF retrieval.
In order to explore the causes and distribution characteristics of waterlogging disasters in urban areas， taking the central city of Lanzhou as an example， based on the monthly average precipitation data and historical water accumulation data from 2010 to 2020， population density， road network density， water supply and drainage facilities distance， elevation， topographic relief， slope， normalized difference vegetation index， and percentage of impervious surface were selected as indicators to explore the driving factors of waterlogging， and the time-space distribution pattern， spatial correlation analysis and geographic detector of waterlogging disasters were studied. Drivers and interactions. The results show that：（1） the flood disasters in Lanzhou City during 11 years highly coincide with the precipitation concentration period， both in June-September， and the waterlogging disaster is the most serious in the main flood season in July-August， indicating that the frequency distribution of the flood disasters is closely related to the precipitation pattern； （2） In space， Lanzhou water accumulation points are mainly distributed in Chengguan District， followed by Anning and Qilihe District， and Xigu District is the least. （3） NDVI and the percentage of impervious surface are the main drivers of flooding disasters in Lanzhou， followed by population density， but the interaction of any two drivers is much greater than a single factor. Therefore， under the premise of comprehensive consideration of multi-factors， Lanzhou waterlogging disaster needs to be comprehensively managed by multi-measures.
Efficient recognition of shadow information is a key prerequisite for utilizing and eliminating shadows， most of the existing studies on urban shadow detection have been attached to the multi-band synthesis of near-infrared and visible light， while the detection ability of shadows extraction from visible light still remains insufficient. In this study， based on red， green， and blue （R， G， B） high-resolution satellite images， we used color space transformation and image multi-band operation to constructed an Optimization Urban Shadow Index （OUSI） with green light band， blue light band， and luminance component. The visual effect and accuracy evaluation were also be analyzed. The results showed that a more complete urban shadow can be extracted by OUSI with an overall accuracy of 90.46%， outperforming the current common exponential method and deep learning shadow detection algorithms； the shadow detection results were the most stable as it suffered less from the influence of different land cover types. In contrast to the previous feature-based methods， the raw image data of this study only rely on RGB three-band information. The OUSI consumes fewer computing hours and thus providing an effective practical solution to achieve urban shadow detection in large areas.
The terrestrial 3D laser scanning technology has the advantages of fast scanning， no damage to trees， and high restoration of the original shape of trees， which provides accurate data for 3D sample wood reconstruction， tree canopy structure research， and forest resource monitoring research on a continuous basis. With the Masson pine as the object， the 3D laser scanner was used to obtain point cloud data of 30 single trees， and the voxelization， planar projection and convex packet algorithm were applied to calculate the porosity of single wood canopy. Combined with the theory of stratification， through the correlation analysis with tree growth parameters （crown width， crown volume and crown height）， a multiple linear regression model was established for the canopy porosity extracted from the full crown and different stratification methods， and the coefficient of determination （R2）， Root Mean Square Error （RMSE）， Residual Predictive Deviation （RPD）， and Total Accuracy （TA） to determine the optimal voxel side length and optimal stratification for canopy porosity extraction. The results show that the best stratification method for canopy porosity extraction is to divide canopy shape into three layers （R2 is 0.74）； The effects of canopy porosity and tree growth parameters extracted by three-level stratification are the most stable； the porosity extracted according to canopy shape stratification is suitable for large differences in canopy shape， and when the canopy shape is relatively consistent， the canopy height is used. The third-level stratification is more suitable and has higher precision.
The Upper Minjiang Basin， in the eastern Tibetan Plateau， is characterized by complex rugged terrains， so it was vulnerably subjected to surface subsidence related hazards. Therefore， it was quite beneficial and necessary to explore the controls underlying subsidence for disaster mitigation and avoidance. In the study， totally 60 Sentinel-1A images were chosen to detect possible surface subsidence from 2015 to 2019 with the interferometric synthetic aperture radar （InSAR） method in the study area， and then the factors underlying the subsidence were discussed. The results showed that south （sunny） slopes experienced the higher subsidence rate （averagely -33.02 mm/a） than the north （-9.33 mm/a） though with similar elevation and slope degree. The spatial patterns of subsidence could be attributed the terrain aspect related factors. The sparse vegetation cover on sunny slopes due to severer water deficit provided the weak protection to surface stability， and the physical erosion induced by the northward orographic rain was also not beneficial to the stability of south slopes. In addition， sunny slopes abundant in solar energy were subjected to more human activities such as farming and building， which also could weaken surface stability. Our study emphasized the distinct vulnerability in surface stability of south slopes in this region， which should be carefully taken into account in land developing as so to avoid causing landslides.
It is an important basis of analyzing long-term phenological changes that extracting phenological parameters based on different vegetation indices. Takes the cloudy and foggy area， Chongqing， as an example. Three long-term vegetation index data of NDVI， EVI， and EVI2 are extracted based on MODIS remote sensing images from 2010 to 2019， and the characteristics of different vegetation indexes are analyzed through D-L filtering. The results， which is of phenological parameters extracted based on three vegetation indices， were studied using dynamic threshold method and trend analysis method， and their response relationships and differences to topographic factors are compared. The results are as follows： ①The time series fitting curve of EVI and EVI2 is smoother than the fitting curve of NDVI. The differences between the original values of the three vegetation indices and the fitted values are mainly distributed in NDVI （0.05~0.18）， EVI （0.03~0.11）， EVI2 （ 0.03~0.1）. ②The spatial distribution and change trend of the phenological parameters extracted from the three plantations were consistent. The vegetation index parameters extracted from EVI and EVI2 were similar， accounting for more than 79% within 5 days， and the significant change area of SOSEVI2 was the highest （16.36%）， while the lowest SOSNDVI was 12.37%.③SOS was delayed with the increase of altitude， EOS was delayed and then advanced with the increase of altitude， LOS was extended and then shortened with the increase of altitude，and EOSNDVI and LOSNDVI were significantly different from EOSEVI/EOSEVI2 and LOSEVI/LOSEVI2 with the increase of altitude， respectively. The phenological parameters extracted by EVI were similar to those of EVI2， and the variation trend was consistent. The phenological parameters can be better extracted based on EVI/EVI2 in cloud and fog areas， and the results are similar and can be used interchangeably. The phenological parameters extracted based on EVI and EVI2 have more obvious differences in altitude， slope， and slope direction.
Accurately extracting and identifying the spatial distribution and form of slum is of great significance to improving the living environment and optimizing urban spatial structure. Traditional field investigation methods are time-consuming and laborious. As slums in the southern Yangpu District of Shanghai as the study area， spectral， textural and structural features from high-resolution remote sensing images as input data， this paper proposed an identification method for the slum using Machine Learning （ML） algorithms. Firstly， K-Nearest Neighbor （KNN）， Logistic Regression（LR）， Support Vector Machine（SVM）， Random Forest （RF） and Ensemble Learning（EL） algorithms were compared comprehensively to determine the optimal classifier. Secondly， features were extracted from high-resolution images based on the grid of 50 m×50 m. Then the feature grid was classified and the feature dataset was constructed. Finally， the slum area was identified by image features， and the ability and application of five ML methods in urban area are evaluated. Results showed that supervised machine learning methods could basically meet the research and practical application of slums identification. In terms of the classification results， classification accuracy and operation efficiency， the Kappa coefficient of EL algorithm was 73.0%， the overall accuracy was 97.27%， and the recall rate was 79.02%， which were all higher than other algorithms， and the omission errors were the least. Therefore， the EL algorithm could complete the information extraction of slums more completely and accurately. When considering the operating efficiency， the LR algorithm had a higher identification speed than other algorithms and was more suitable for the use of a large range of slums. Moreover， ML methods could not only be used for features extracting in high-resolution images， but also had great application potential in remote sensing monitoring， urban planning and mapping.
