Google Earth Engine （GEE） is a comprehensive application platform that integrates remote sensing image storage and analysis. It can conveniently and quickly call remote sensing images and information extraction. Therefore， GEE has attracted more and more scientific researchers' attention. With the continuous expansion and upgrade of GEE， the system platform has become more and more complex. For ordinary users， it is becoming more and more difficult to quickly understand its architecture and functional algorithms. In response to this problem， this article systematically introduces the technical architecture， data resources， model algorithms and computing resources of GEE， and summarizes the application results of GEE in various fields， hoping to provide GEE users with a quick understanding of the platform Window to help them make better use of the GEE platform to carry out their own application research.
Mycotoxin is major threat to food security and safety in China， and their accurate prediction can support effective loss control and reduction. Based on the summary of the impact of crop mycotoxins on food， food and agriculture， this paper analyzes the current research progress of mycotoxin prediction from three aspects： meteorological statistical model， mechanistic model and machine learning model.The feasibility of using satellite remote sensing monitoring technology to carry out a wide range of crops mycotoxin prediction is discussed by analyzing the current influencing factors of crop contamination mycotoxins and combining with the research progress of remote sensing technology in monitoring crops and their growing environment.It is also pointed out that analyzing the main influencing factors of crop mycotoxins from remote sensing and their change patterns， developing remote sensing estimation methods of mycotoxins by combining relevant factors within the reproductive period， and coupling remote sensing technology oriented to farm-scale dynamic prediction with multiple models will become the research focus in this field.
The spatial distribution and winter wheat harvesting area is of great significance to accurately estimate production and ensure food security. However， the vast majority of study and statistical data is based on planting area of winter wheat， and few studies have been done on the winter wheat harvesting areas. In this study， Puyang County was selected as the study area， and the harvesting area of winter wheat was estimated by combining Sentinel-2 remote sensing imagery at the maturing period in 2019 and random forest model. Firstly， best feature subsets were obtained through feature selection. And then， the separability between winter wheat and other land types was analyzed by the J-M distance of these best feature subsets， the harvesting area and planting area of winter wheat were identified and extracted， and the harvesting area of winter wheat was mapped. Finally， the differences in harvesting area and planting area of winter wheat and the influencing factors of harvested area were further analyzed. The results found that the overall accuracy and Kappa coefficient of winter wheat harvested area estimated by the best feature subset of Sentinel-2 images were 94.62% and 0.93， respectively. The planting area of winter wheat in Puyang County in 2019 was 79.47 thousand hectares， and the extracted harvesting area was 76.74 thousand hectares， their difference （2.73 thousand hectares） was largely attributed to human activities， and timely monitoring of the harvesting area of winter wheat can provide a certain scientific reference value for related research and decision-making such as winter wheat yield prediction.
The low temperature during the growth period of paddy will greatly reduce the grain yield. Remote sensing monitoring of yield loss under low temperature stress is of great significance for variety improvement， field management and agricultural insurance claims. The study aimed to monitor yield loss of multiple cropping paddy using multi-temporal remote sensing images. with the support of the field samples， the model of monitoring yield loss of multi-cropping paddy was developed. The results showed that the growth period of paddy in which low temperature injury occurred was different，and the effect on rice yield was quite different.The effect of cold injury on middle rice in the middle filling stage is relatively small， with an average yield of about 6 637 kg/ha， with a yield reduction of nearly 20%. The yield of early late rice is significantly lower than that of middle rice after suffering from low temperature and cold injury at heading stage， with an average yield of 4 143 kg/ha and a yield reduction of about 45%. The yield of late-maturing late rice was most affected by continuous low temperature at jointing stage， with an average yield of only 1 541 kg/ha， which was much lower than that of previous years. The regression model was constructed by using the actual cut sample yield data and Sentinel data in several key phenological periods （NDVI）. The R2 was more than 0.75. the precision was cross-verified by the measured sample yield data， and the MAPE was less than 10%. With the help of a small amount of ground data， this method can accurately calculate the yield of multi-cropping rice under the condition of low temperature and cold injury， which provides a new idea for the calculation of rice yield under complex conditions.
In recent years， global warming has led to an increase in strong convective weather， and hail disaster has become one of the main disasters in agricultural production. Carrying out remote sensing assessment of cotton hail disaster is of great significance for disaster prevention and mitigation， insurance claims and planting structure adjustment. Taking the cotton hail disaster in the Kuitun River Basin in the southwest of Junggar Basin， Xinjiang， on 23 August， 2019 as the research object， with the support of field measured samples， the multi-temporal Sentinel-2 remote sensing images before and after the hail disaster were obtained. We analyzed the dynamic changes of various vegetation indexes before and after the hail disaster， and screened the sensitive vegetation index difference feature combinations which can effectively characterize the hail disaster. The range and grade of cotton hail disaster were automatically extracted by using machine learning algorithms such as logical regression， decision tree， gradient lifting decision tree and random forest， and the accuracy was compared and analyzed via field measured samples. The results showed that NDVI was the best indicator of hail disaster among single vegetation index， with an overall accuracy of 84.39% and a Kappa coefficient of 0.75. The combination of multi-temporal vegetation index differences was significantly more indicative for hail disaster than that of single vegetation index. Compared with the time series characteristics of vegetation index differences before and after hail disaster， the indicative of hail disaster between August 30 and August 20 was obviously stronger than that between August 25 and August 20， which indicated that it was necessary to consider the self-recovery ability of cotton plants after hail disaster grade for remote sensing monitoring. The combination of the pre- and post-disaster vegetation indices and the random forest classification algorithm was the most effective methods in monitoring the level of cotton hail disaster level， with an overall accuracy of 89.51% and a Kappa coefficient of 0.83. In conclusion， the extent and degree of cotton hail disaster can be effectively evaluated based on multi-temporal Sentinel-2 image.
Rapid and accurate winter wheat acreage extraction using remote sensing technology is of great importance for crop yield estimation and food security. Due to problems such as the difficulty of obtaining medium and high resolution time-series images due to revisit cycles， cloud and rain， and the low accuracy of low resolution remote sensing data in extracting crop planting information. In this study， taking Changge City， Henan Province as an example， Landsat 8 and MODIS images were obtained as the dataset during 2015~2020， and the 2 data were fused based on an optimized convolutional neural network spatio-temporal fusion model to construct a 30 m resolution NDVI time series set， and S-G （Savitzky-Golay） filtering was used to denoise the time series set， and finally The area planted with winter wheat was extracted using the RF method. The results show that the optimised fusion model is robust and the R2 of both the predicted and real images is above 0.92. The agreement between wheat area extraction and statistical area in the study area was 97.3% and the results were reliable. Therefore， the optimised model can better fuse the medium and high resolution images， which is an effective technical means to supplement the missing images， and the constructed time series set can more accurately extract the wheat planting area in the county.
Obtaining high-resolution image features of crops and exploring the influence of multi-feature learning on actual crop classification are of great significance for agricultural departments to grasp the information of crop planting fine structure and efficiently implement production management. For the high-resolution RGB image acquired by UAV， this paper proposes a new classification method of crops based on object-oriented and multi-feature learning. Firstly， the HSI model is used to transform the colour space of RGB images to mine the potential information of images further. Secondly， the ESP algorithm and CART are used to determine the optimal image segmentation scale and construct the optimal feature learning data set of classification. Finally， object-oriented Random Forest classification algorithm was used to learn and train the multi-feature space， so as to achieve fine crop classification， and accuracy evaluation was carried out in combination with validation data set. The experimental results show that the overall accuracy of the classification in the study area reached 90.18%， and the Kappa coefficient reached 0.877， both of which were greater than the accuracy based on pixel-level and single-feature learning. The optimal feature learning space constructed in this paper have a good classification effect on cotton， corn， cocozelle， grape and other major crops in the study area. The producer's accuracy of each crop type is greater than 89%. This research can provide a reference for agricultural precision management and planting structure optimization.
The launch of the Gaofen-1 satellite has further enhanced China's Earth observation capabilities. Compared with Sentinel-2， the GF-1 WFV image has a high spatial resolution of 16 m， but lacks the Red-Edge （RE） band and the Short-Wave Infrared （SWIR） band. It is important to analyze the differences in the accuracy of estimating vegetation physiological parameters between the two satellites for further application. In this study， we used linear regression models and a Look-Up Table （LUT） based on the PROSAIL model to assess the performance of GF-1 and Sentinel-2 in estimating LAI of soybean and maize. The results showed that： （1） The EVI simple linear regression of GF-1 outperformed other vegetation indices with a R2 value of 0.81 and the MNLIre model was the best Sentinel-2 model with a R2value of 0.86. （2） GF-1 obtained a comparable accuracy to Sentinel-2 with multiple linear regression models based on spectral bands. The best LAI estimation model of GF-1 produced a Root-Mean-Square Error （RMSE） of 0.54 and a coefficient of determination （R2 ） of 0.90， and the best Sentinel-2 model achieved a RMSE of 0.54 and R2 of 0.89. （3） In terms of LUT based on the PROSAIL model， the optimal band combination for GF-1 were B2 and B4 with a R2 of 0.76 and a RMSE of 0.81， and the optimal band combinations for Sentinel-2 were B3， B6， B7， B8， B8a， and B12 with a R2 of 0.87 and a RMSE of 0.62. This study showed that GF-1 satellite has the ability to accurately monitor crop LAI， which can provide a theoretical basis for the application of GF-1 in agriculture monitoring.
.CO2 is one of the important greenhouse gases in the atmosphere. Since the Industrial Revolution，the concentration of CO2 in the atmosphere has been increasing continuously， which has an important impact on global climate change. High precision，high coverage and high temporal and spatial resolution CO2data tends to be more significant in the study of carbon neutral and global CO2 change. Thus， in this study， we compared the XCO2 products between the satellites OCO-2 and OCO-3， and formed a joint data set from the two satellites. Because there are still some regions without observation data in the joint dataset， this study uses Kriging interpolation algorithm to fill the regions without data. Considering the temporal and spatial variation characteristics of CO2 concentration in different latitudes， the algorithm divides theworld into six regions and selects the appropriate variogram.The results show that the XCO2 data coverage increases by 52.32%， 46.77%， 44.04%， and 33.81% on the 3-day， 8-day， 15-day， and 30-day timescales，respectively. By comparing the monthly interpolation data set with the TCCON site data to verify the accuracy， the mean absolute error is 1.049 ppm， the root mean square error is 1.024 ppm， and the coefficient of determination is 0.82. It can be seen that this method can accurately fill in the blank area of the j-oint dataset， and improve the accuracy， coverage and spatiotemporal resolution of the data.
As the most widely distributed vegetation type in the Qilian Mountains region， subalpine meadows play an important role in maintaining local carbon and water fluxes and responding to climate change. Therefore， accurately detecting their phenological dynamics is crucial for a deeper understanding of mountain ecosystem functioning and its feedback to the climate system. In this study， we conducted a multisource image fusion and land surface phenology extraction experiment in a 15 km × 15 km test area in the northeastern Qilian Mountains， combining ground-level eddy flux data and multisource satellite remote sensing images. We used an Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model （ESTARFM） to fuse multisource images from the ETM+， OLI， and VIIRS sensors， reconstructing a high temporal （minimum 1 day） and high spatial （30 m） resolution time-series image dataset of the 2-band Enhanced Vegetation Index （EVI2）， Normalized Difference Vegetation Index （NDVI） and Near-Infrared Reflectance of Vegetation （NIRv） from 2013 to 2020. Based on this， we fitted the growth curves of the GPP flux tower and remote sensing vegetation index images using a Double Hyperbolic Tangent function （DHT） and Global Model Function （GMF）， respectively， and applied a dynamic threshold method to extract the start （SOS）， peak （POS） and end （EOS） of the growing season to evaluate the applicability of different fusion vegetation indices in extracting key phenological parameters of subalpine meadows. The results showed that ESTARFM fused images could accurately reflect the brightness and texture features of the real images， but cloud-contaminated pixels in the input images could also affect the fusion accuracy. At the site scale （without cloud pollution）， NIRv and EVI2 exhibited similar fusion accuracy， while at the pixel scale （with cloud pollution， cloud cover < 20%）， the fusion accuracy of NIRv was significantly higher than that of EVI2， indicating that NIRv improved the sensitivity of vegetation partial reflectance in vegetation-bare soil mixed pixels in the algorithm and could maintain high fusion accuracy under cloud pollution conditions. For the growth curve fitting algorithm， DHT + GMF could accurately simulate the seasonal dynamics of the GPP flux tower and remote sensing vegetation indices， with determination coefficients above 0.960 and root mean square errors below 0.062. The comparison of phenology extraction accuracy of the three fused vegetation indices showed that NIRv had the highest accuracy in extracting SOS and EOS， while NDVI had the highest accuracy in extracting POS， with deviations of 4 d （3 d）， 5 d （5 d）， and 4 d （6 d） at the site （pixel） scales， respectively.
Crop residue burning is one of the important sources of air pollution. Due to its randomness and dispersion， it is hard to obtain the information of Crop Residue Burning Fire Points （CRBFPs） quickly， accurately and comprehensively， and regulatory authorities face great difficulties in supervision and law enforcement. In order to improve the monitoring accuracy and efficiency of straw burning， high-precision fire detection algorithms were developed in this study and used to monitor the CRBFPs based on multi-source satellite remote sensing data， and UAV and manual verification schemes were formulated to verify the CRBFPs’ information. The system platforms of WEB and mobile APP were developed to directional push and display the fire points information. A space-air-ground integration supervision system for crop residue burning was established with the functions of detection and verification. Relevant platforms were applied to monitoring of the CRBFPs in Shandong Province in the spring and autumn of 2020， and 58 fire points， which include 53 true fire points after verification， monitored by remote sensing images， with an accuracy rate of 91.38%. The results show that the system can extract high-precision fire point information quickly and provide new ideas and methods for the supervision of crop residue burning.
As the functions of urban areas become more and more complicated， it is of great significance to identify the specific function types of urban blocks scientifically and accurately. This paper presents a time-series dynamic urban functional area recognition scheme. Taking the area within the Sixth Ring Road of Beijing as the research area， the high incidence area of travel mode is extracted from the massive travel data by using taxi trajectory data and Dynamic Topic Model （DTM）. Urban blocks are clustered based on topic model feature. The research use POI semantic annotation clustering results to identify urban functional areas. This paper studies and evaluates the change trend and distribution of topic blocks during six years， and discusses the dynamic changes of semantics of blocks： （1） The dynamic topics distribution has spatial diffusion， and the distribution of block semantic intensity shows obvious circle expansion. （2） The spatial boundary of clusters based on travel activities gradually coincides with the administrative divisions of the study area over time， and the function labeling results are highly matched with the specific functions of the area. （3） The high value of topic variation value is mainly distributed in the outer ring area， and has a negative correlation with the proportion of construction land. This research shows that the dynamic topic model is applicable in the travel data mining scenario， providing a new reference direction for the application of dynamic topic model in the field of mobile data mining.
The integration of marine， land and air map data is a commonly used remote sensing data in research. However， existing marine， land and air data are based on vector management， which has problems in terms of intuitiveness， convenience， uniformity and efficiency. In this study， a hexagonal global discrete grid-based marine， land and air map integration algorithm is proposed to solve these problems. Firstly， according to the hexagonal subdivision method， the icosahedron is projected onto the earth’s surface to generate 20 basic triangles， which are then combined into 30 basic hexagons. Secondly， the corresponding relationship between scale and grid level is determined， and the corresponding scale level grid data is recursively subdivided. Then， the marine， land and air map is parsed into a nested structure of multi-line segment sets， multi-line segments and straight line segments， with straight line segments being the minimum unit. The grid adjacency search algorithm is used to grid the straight line segments， and finally the marine， land and air map data is expressed in the form of a hexagonal global discrete grid. The experiment used 1∶50 000 and 1∶25 000 marine， land and air map vector data in China and hexagonal global discrete grid data. The experimental results show that the marine， land and air map integration algorithm based on hexagonal global discrete grid can perfectly convert map vector data into map grid data. In summary， the marine， land and air map integration algorithm based on hexagonal global discrete grid can make marine， land and air map management and display more convenient and efficient， with clearer topological relationships， and provide reference for marine and land integration collaborative applications and data integration and management using global discrete grids.
Based on the TensorFlow framework， which has the characteristics of GPU or CPU parallel acceleration calculation on tensor （multidimensional arrays）. The WGS84 latitude and longitude projection is selected， combined with FY-3D MERSI L1 data with high-precision and same-resolution positioning data， to generated tensors （multidimensional array） to align latitude and longitude data according to resolution， and calculated the new image pixel mapping position information. According to the position index information， the MERSI data can be geometric corrected point by point， and the BowTie effect caused by the scanning observation and earth curvature of medium-resolution polar orbit satellites can be eliminated at the same time. Finally， the convolution is used to calculate the inverse distance weighted interpolation point values and fill the pixels with no data after geometric correction. Using this method， the author implemented geometric correction of all 25 channels of FY-3D MERSI data in Python under the TensorFlow framework. Compared with the geometric correction results with ENVI software as the standard. The error and correction precision are calculated， and the overall processing speed of geometric correction is also tested. The results show that the algorithm proposed in this paper has a high consistency with the geometric correction of ENVI software， and the accuracy below 5% absolute error percentage is greater than 0.92， and the structural similarity SSIM index is around 0.95. Speedup is more than 36 times to complete geometric correction for all channels using GPU parallel acceleration. In summary， the geometric correction method adopted in this paper is fast and efficient in processing， and ensures the accuracy of correction.
The open burst mode is a kind of Synthetic Aperture Radar （SAR） altimeter operation mode with alternating receiving and transmitting pulses. This mode allows for a greater number of independent pulses in a unit time than the closed burst mode of a satellite altimeter such as the Sentinel-3， thereby improving range precision. However， because the Pulse Repetition Frequency （PRF） of this mode is less than the Doppler bandwidth， the azimuthal Doppler spectrum is aliased and the resulting multi-looking echoes are distorted， thus reducing the range precision. By studying the correspondence between the range curvature and the aliasing frequency in the processing of synthetic aperture signals， an open burst anti-aliasing filtering method based on multi-looking beam stack processing is proposed to filter the multi-looking beam stack and effectively solve the azimuthal aliasing effect of the open burst mode. The proposed method can effectively improve the range accuracy through simulation. At a significant wave height of 2 m， the 1 Hz range precision of the open burst anti-aliasing filtering method improves from 0.7 cm to 0.60 cm， an improvement of 14 %.
CFOSAT（China-France Oceanography Satellite） scatterometer is the first fanbeam conical scanning scatterometer in the world， which can obtain the observation data of multiple incident angles of targets. The CFOSAT scatterometer ground preprocessing service software generates level-1 data， including L1A （level-1 A） and L1B （level-1 A）data. According to the requirements of CFOSAT scatterometer level-1 data tracking test， a CFOSAT scatterometer data test software system is designed and developed， which focuses on testing and analyzing Sigma0 data. Data test system is based on MySQL database technology， combined with OpenMP parallel processing technology and developed by MATLAB and VC++ mixed programming. It can completely and efficiently obtain the test and evaluation results of CFOSAT scatterometer Sigma0 data quality. The test and analysis of CFOSAT scatterometer Sigma0 data shows the correctness of CFOSAT scatterometer data preprocessing algorithm， and also shows that CFOSAT scatterometer has high observation accuracy. The software is helpful to fully master the status of CFOSAT scatterometer business data， and provides an important reference for the optimization of CFOSAT scatterometer data preprocessing algorithm.
Satellite precipitation products with a high spatio-temporal resolution are essential for hydrometeorological research on the regional or watershed scale. By comparing the accuracy of IMERG and GSMaP satellite precipitation products in the Yangtze River basin during 2015~2020， the outperformed product is then used for spatial downscaling. Considering the relationship between precipitation and geographic features （DEM）， vegetation index （NDVI）， and Land Surface Temperature （LST）， various statistical downscaling models based on random forest regression are developed， such as DEM model， DEM+NDVI model and DEM+NDVI+LST model. Then， the performance of the three models is further evaluated against observations from 133 stations in the study area. The results indicate GSMaP outperforms IMERG in the Yangtze River basin. Among the three downscaling models， DEM+NDVI+LST model performs best on the multi-year mean scale and annual scale， and the performance of the model remains stable on the monthly scale. This study can provide a new way for spatial downscaling of satellite precipitation products.
Fractional Vegetation Cover （FVC） is an important index to evaluate the quality of ecological environment and characterize the ground cover and growth status of vegetation. There are many FVC products at home and abroad. However， different FVC products have certain spatio-temporal differences. In order to accurately understand the differences of FVC products and their causes， two global FVC products， GEOV3 and GLASS， were selected to assess their spatial and temporal differences in southwest China by resampling and difference analysis， and to analyze the influence of topography and land use type on FVC products by combining topographic and land use data. The results show that： （1） there were obvious spatial and temporal differences between GLASS and GEOV3 products with seasonal characteristics， with GLASS FVC values slightly lower than GEOV3 FVC values in spring and summer， and the smallest difference between the two values in autumn， while GLASS FVC values were significantly higher than GEOV3 FVC values in winter； （2） the values of the two products differed significantly in different land use types， and the differences were： shrub > forest > cropland > grassland > other， and the differences were the largest in winter； （3） the values of the two products also differed significantly in different slopes and altitudes， and the changes in slope had more obvious effects on the products. This study revealed the influencing factors that lead to inconsistency among FVC products， which can provide a reference for the improvement of FVC product generation algorithms in mountainous areas.
Burned area is one of the main parameters required for research such as global changes and carbon cycles. Accurate monitoring of burned area is of great significance for improving the accuracy of fire risk warning and risk assessment. This research used three MODIS satellite data products to assess their accuracy in estimating the annual and multi-year extent （2001 to 2016） of burned areas of the eastern Mongolian Plateau. The analysis used 30 m Resolution Global Annual Burned Area Map （GABAM） product as a reference dataset to evaluate monitoring accuracy of three MODIS burned area products referred to as MCD45A1， MCD64A1， and FireCCI51. Respectively， these products recorded 327， 160， and 71 fires in 2015. Only 40 fires were jointly monitored by three products. Monitoring areas of 27 082.46 km2， 17 227.62 km2， and 19 526.47 km2 overlapped to give a cumulative area of 6 896.99 km2. Compared with reference data， the three products gave a composite accuracy F1 score ranging from 0.96 to 0.02 indicating relatively uneven monitoring rates. Over a three-year time scale （2013~2015）， the data products gave average composite accuracy scores of 0.70， 0.62， and 0.60 so as to rank the products as MCD45A1>FireCCI51>MCD64A1. On the multi-year （2001~2016） time scale， monitoring rates of the three products were 61%， 59%， and 50% ranking products as MCD64A1>MCD45A1>FireCCI51.
In Africa， the distribution of tin roof can reflect the living standard of residents， and the analysis of its temporal and spatial changes can reflect the local economic development. Sentinel-2 was used to study the spectral characteristics， remote sensing index characteristics and texture features of iron roof in Munyaka area， Eldoret， Kenya， Africa.The Normalized Difference Vegetation Index（NDVI） and Normalized Difference Building Index（NDBI） were used to remove farmland from the difference of the iron roof. The normalized surface index was constructed and the texture features were analyzed to eliminate wasteland and bare land respectively. The model of multi feature decision tree extraction is established. The Kappa coefficient is 0.922 3， and the user precision and mapping precision are 97.79% and 91.10% respectively. At the same time， combined with the erdoret municipal road project， the changes of iron roof before and after the project in 2016~2020 are studied. Compared with 2016~2018， the area of iron roof in 2018~2020 is doubled， and the average annual growth rate is nearly 3%. Research shows that this method can achieve dynamic monitoring of tin roofs， and illustrates that "The Belt and Road Initiative" construction plays a driving role in solving poverty problems in Africa.
The change of coastline affects the production and life of coastal residents and the national military strategic deployment. It is of great significance to detect the change of coastline quickly and accurately. Navigation X-band radar can continuously and real-time detect the nearshore marine environment. This paper proposes a method to detect coastline using navigation X-band radar images. Firstly， the navigation X-band radar image is preprocessed by averaging， contrast enhancement and Gaussian filtering to reduce the impact of sea clutter on the radar image； Then， the preprocessed image and the edge detection operator are convolved to obtain the gradient image， and an adaptive threshold estimation method based on the histogram bimodal method is established to automatically extract the boundary of the target island from the gradient image； Finally， the edge expansion， cavity filling and denoising are used to extract the complete island， and the area and perimeter of the island are estimated. The proposed method is verified by using the navigation X-band radar images observed in the experiment. Compared with the coastline in the electronic chart and Google Earth map， the edge accuracy of the target island extracted by Sobel operator and Prewitt operator is higher， and the area and perimeter of the extracted island vary with the tide height， sea conditions， etc. The results show that the navigation X-band radar can effectively monitor the real-time changes of the island coastline.
Atmospheric haze pollution is one of the major environmental problems facing China in recent years. Haze monitoring is an important part of haze pollution governance system， and remote sensing can realize large-scale and long-term dynamic monitoring， which to a certain makes up for the shortcomings of traditional site monitoring methods. Based on MODIS/Terra data， this study takes seven provinces and municipalities in North China Plain as the study area， built a 10-dimensional feature space by using the spectral and spatial characteristics of atmospheric haze， and built a regional identification model of haze using random forest algorithm. This model not only realizes the function of haze region identification and extraction， but also realizes the function of light and heavy haze region classification. Through verification of PM2.5 monitoring data from ground stations， the overall accuracy of haze region identification is 87.82%， and the Kappa coefficient is 0.75. The overall accuracy of light and heavy haze region identification is 86.81%， and the Kappa coefficient is 0.73. The study results show that the model has a good identification effect on haze region in satellite images， which can provide data support for air pollution monitoring， and has certain reference significance for monitoring other atmospheric pollutants.
