In the context of China's planning system reform, territorial space planning is one of the important methods that promote the scientific management of natural resources and realize regional sustainable development. It is helpful to solve the overlaps and contradictions between land use planning, urban planning, and main functional area planning. In order to improve the reasonableness and authority of natural resources surveys and territorial space management, with the advantages of objective and precise geographical conditions monitoring data, it is urgent to clarify the differences between geographical conditions monitoring and other natural resources surveys, so as to better serve the survey and supervision of natural resources and the management of territorial space. This paper first analyzes the differences in land use classifications between the geographical conditions monitoring and the “two plans”, and then selects a specific experimental area in Wuhan for an empirical analysis. Finally, given the differences of data between geographical conditions monitoring and the “two plans” land use and the applicability of geographical conditions monitoring in territorial space planning, we put forward some suggestions for optimizing geographical conditions monitoring for territorial space management in the new era. The results show that there are different degrees and types of differences in the guiding principles, composition system, and technical standards of land use classification for geographical conditions monitoring data and “two plans” land use data, which makes the former difficult to be directly applied to the compilation of territorial space planning. In order to promote the application of geographical conditions monitoring datasets in compilation, implementation, and management of territorial space planning, it is necessary to further define its basic position in the system of natural resources survey, monitoring, and management, improve the monitoring content, optimize the classification system, and unify the technical standards. With the preliminary construction of natural resources survey and monitoring system and gradual improvement of territorial space planning system, geographical conditions monitoring should rely on its technical advantages, clarify its service content and objects, and build a comprehensive application framework system in the future to make it play an important role in natural resources management.

Submarine topography provides essential information for various researches in oceanography and has been widely applied to the marine meteorology, marine chemistry, and physical oceanography. However, the present submarine topography usually contains noise and outliers with low precision and resolution due to uncertainties in processing methods of raw measurements. Here, we proposed a multi-source bathymetry fusion technique by considering the merits of multi-source data including the Digital Elevation Model (DEM), charted depth, and multibeam bathymetric data. A modified "remove - restore" algorithm was employed to fuse multiple high-precision data sources (i.e., charted depth and multibeam bathymetric data) with DEM to reconstruct the submarine topography. The results demonstrate that the proposed technique could improve the precision and resolution of the entire topography of seabed terrain and preserve the details of the high-density area. The resolution of the original DEM data was 15 arc seconds with a Root Mean Square Error (RMSE) of 29.408 m, and the resolution of the DEM using the classic " remove - restore " algorithm was 3 arc seconds with a RMSE of 28.563 m. While the proposed fusion technique could achieve the DEM at 3 arc seconds resolution with a lower RMSE (18.841). Hence, the proposed fusion technique outperforms classic algorithms in reconstructing accurate submarine topography database.

Digital Elevation Model (DEM) plays an important role in hydrodynamic researches as an essential data. The resolution of DEM data from public sources is usually not able to depict the riverbed topography. Hence, they cannot be applied to the research work such as river flood analysis. This study carries out the work of digital riverbed construction based on DEM data. Firstly, the longitudinal river network elevation derived from a DEM is smoothed by introducing robust locally weighted regression algorithm to eliminate the abnormal elevation values. Secondly, the cross-sectional interpolation based on the inverse distance weighted method is carried out with the river surface polygon as the mask to acquire the 3D digital river surface. Finally, the overall construction of the riverbed is completed by combining the river surface elevation and river depth data. Based on SRTM DEM, this study took Wende River in Yongji County of Jilin Province as a research example, and constructed three types of digital terrain of riverbed section, i.e., rectangular, trapezoid, and V-shaped, respectively. In order to evaluate the rationality of the data reconstruction, the contour analysis was performed on the obtained digital river surface, and the channel offset of river network extracted from DEM hydrological analyses before and after riverbed reconstruction was also calculated. We finally analyzed the river hydraulic simulation results. Our results show that: (1) the application of inverse distance weighted interpolation can well complete the two-dimensional extension of river network elevation to river surface elevation; (2) the error of hydrological analysis based on reconstructed DEM of riverbed topography was well eliminated; and (3) the reconstruction data in this paper had good applicability and reliability.

Precipitation is an essential part of the Earth's water cycle and a key variable linking atmospheric processes to surface processes, as well as an important parameter in hydrological process simulations. It is one of the atmospheric variables that are most difficult to observe due to its large variability over time and space, and its tendency to exhibit non-normal distributions. The spatial and temporal resolution of precipitation data has become a key concern for process simulations. Traditional station observations usually have poor regional representation of precipitation data, due to factors such as sparse and/or uneven station distribution. This, in turn, affects the accuracy of hydrological simulations. Remotely sensed precipitation data have brought breakthroughs in modelling the hydrology of data-deficient regions. To best simulate the hydrological processes in Bayinhe River Basin, a special area with limited data located in Northeast of the Qaidam Basin, three most popular remote sensing precipitation datasets (TMPA 3B42, GPM IMERG V5, and GPM IMERG V6) along with weather station observation data were used to build SWAT model. The accuracy of TMPA 3B42, GPM IMERG V5, and GPM IMERG V6 datasets were first evaluated and then the performance of different SWAT models based on different precipitation datasets and parameterization schemes were assessed. Results show that: 1) The observed precipitation data and TMPA 3B42 data both generated good stream flow simulation results with the former accuracy 6-12% higher than the latter. The both datasets can accurately express the water balance of the basin. Thus, TMPA 3B42 data can be directly used in alpine inland river hydrological process simulation; 2) GPM IMERG V5 performed poor in either yearly or monthly streamflow simulation with a N_SE value (Nash-Sutcliffe efficiency) of -1.58 and 0.13, a P_BIAS value (percent bias) of 41.2% and 41.2%, and a R_SR value (ratio of the root mean square error to the standard deviation of measured data) of 1.61 and 0.93, respectively. This indicates that GPM IMERG V5 was unsuitable to model the hydrological process in Bayinhe River Basin; and 3) GPM IMERG V6 performed better than GPM IMERG V5 in monthly streamflow simulation, and the accuracy of GPM IMERG V6 was four times higher than GPM IMERG V5. However, GPM IMERG V6 dataset performed less well in yearly streamflow simulation with a N_SE value of -0.21, a P_BIAS value of 14.9%, and a R_SR value of 1.09. This research could be a useful reference for future ecological and hydrological processes modeling in the alpine inland river basin region with sparse data.

Anthropogenic heat emissions significantly affect the sustainable development of cities. So far, studies regarding refined Anthropogenic Heat Flux (AHF) parameterizing have been widely conducted. However, the dominated driving factors of AHF at the prefecture level in China have not been well quantified. In this paper, we estimated the prefecture-level AHF in China and further revealed its driving factors. First, the energy consumption inventory integrated with the nighttime light data from the Suomi-national polar-orbiting partnership visible infrared imaging radiometer suite was adopted to estimate the total AHF for cities in China. Then, various socioeconomic factors of AHF were revealed using ordinary least square and geographically weighted regression models. The dominated factors for each city were identified using a coefficient quartering method which can capture the high value of each coefficient for each city. Results show that: (1) there was an obvious spatial heterogeneity of total AHF among cities, and in particular, the total AHF along the southeast coast of China was higher than those in the other regions; (2) dominated driving factors of total AHF were Energy Consumption (EC), Private car ownership (PV), and Per Capita GDP (PCGDP). In addition, the Population Density (PD), share of Secondary Industries (SI), Road Density (RD), and Urban Expansion (UE) contributed to the spatial variation of AHF across cities. Foreign Direct Investment (FDI) exhibited limited influences on AHFs; and (3) cities can be grouped into three types according to the number of dominated factors, i.e., cities having no dominated, single dominated, and bi-dominated factors. There was a high degree of spatial aggregation of cities in each city type. Cities labeled as no dominated factor were mainly located in southwestern China. Cities labeled as single dominated factor of the EC, PV, or PCGDP concentrated in southeastern, central, northeastern and northwestern China. This study can provide a reference for the government to formulate policies on the emissions of anthropogenic heat.

Hand, Foot and Mouth Disease (HFMD) is a common infectious disease in infants and children and has an important impact on their health. In order to reveal the spatiotemporal heterogeneity of HFMD in China and provide a scientific basis for the prevention and control of HFMD, we select HFMD from 2008 (when HFMD was listed as category C infectious disease) to 2018 as the study period and apply spatial statistical methods including Moran's I, Getis-Ord G_i ^*, emerging hot spots analysis, and standard deviational ellipse to analyze the general and local spatiotemporal variation and trend of HFMD in China. Results show that: ① from 2008 to 2018, HFMD exhibits a spatial clustering pattern and the intensity of the clustering increases significantly over time; ② the hot spots of HFMD mainly concentrate in the southeast coast and gradually expand towards inland and northern coastal areas. The cold spots mainly concentrate in the northwest inland and the northeast; ③ the emerging hot spots in mainland China mainly occur in Yunnan, Chongqing, and Sichuan provinces, while the emerging cold spots mostly locate in the same regions with the persistent cold spots. Stable hot spots mainly locate in Hainan province in southern China; and ④ high incidence rate of HFMD mainly occurs in the southwest during 2008 and 2018 and gradually occur in the north during 2008-2009, 2013-2014, and 2017-2018. In general, HFMD remains primarily in the south of China. This pattern remains relatively stable throughout the years of observation, indicating that public intervention should be strengthen in the south of China. However, the underlying mechanism of the spatiotemporal distribution of HFMD in China still needs further investigation. Combination of multiple scientific disciplines such as geography, spatial statistics, virology, molecular biology, and public health provides multi-perspectives that can aid the research on the underlying mechanism of HFMD transmission.

The rising number of people infected by Hand-Foot-Mouth Disease (HFMD) in China poses a great threat to public health and a critical challenge to disease prevention and control in recent years. This paper used regional monthly statistics of HFMD in 2017 when HFMD was common in China. The GeoDetector, spatial autocorrelation, and other analysis methods were adopted to analyze the influence of meteorological factors (i.e., temperature, precipitation) on the prevalence of HFMD and its spatial and temporal differentiation. The results show that: (1) Temporally, the prevalence of HFMD had significant seasonal differences in major cities of China in 2017, with a single prevalence peak or double prevalence peaks (i.e., high-low peak, double high peak). Moreover, the prevalence of HFMD in those cities had a significant spatial correlation in February, April, and December; (2) Spatially, the HFMD in China was characterized by high prevalence level in southeast provinces or cities and low prevalence level in northwest provinces or cities, showing a decreasing trend from southeast to northwest along precipitation gradient; (3) According to the spatial and temporal distribution of HFMD outbreak period (from April to August) at provincial and city level in 2017, HFMD spread from southeast to northwest, and then retreated from west to east; (4) The prevalence of HFMD in major cities in China was correlated with the average annual precipitation and average annual temperature, which were fitted by a quadratic function (R²=0.6623) and an exponential function (R²=0.6469), respectively; and (5) The interaction effect between temperature and precipitation on HFMD resulted in a double-factor nonlinear enhancement and was more significant than a single factor during the spread of HFMD. The prevalence of HFMD in China is influenced by meteorological factors and is significantly different at spatial and temporal scales. Our results can provide reference for the prevention and control of HFMD at national level.

The impervious surface is considered an important feature to measure the urbanization and its spatial expansion. In the current study, the spatial-temporal expansion characteristics and evolution trends of the impervious surface are explored using gravity center, standard deviation ellipse, and gradient analysis in Guangdong-Hong Kong-Macao (GHM) Greater Bay Area during 1987-2017. The results show that: (1) The impervious surface areas of GHM Greater Bay Area has increased from 1839.34 km² to 12 385.93 km² in the past 30 years, with an annual growth area of 351.55 km². In the 21^st century, the urbanization process accelerated, and and each city exhibited different expansion characteristics. Moreover, it formed a spatial expansion pattern driven by core cities such as Guangzhou, Shenzhen, Hong Kong, and Macao, resulting in the formation of various secondary cities, and the development of network structures; (2) Urban construction and expansion are close to the coastal area. The gravity center of the impervious surface of the GHM Greater Bay Area is located near the junction of Panyu, Guangzhou, and Shunde, Foshan. However, the gravity center and urban development trends are from west to the east bank of the Pearl River, and toward Guangzhou, Dongguan, and Foshan. The urban development of Guangzhou, Dongguan, and Foshan promoted the change of gravity center of impervious surface; (3) The urban development of the GHM Greater Bay Area has apparent features of the expansion along the Pearl River and the coastline. Urban construction along both sides of the Pearl River tributary has always occupied a central position. Early urban construction was concentrated in the 70 km area on both sides, and the intensity of urban expansion increased sharply in the 21^st century. The area within 50 km of the coastline is an urban high-intensity construction area. The expansion of the core urban and bay area in various places constitutes the development core of the GHM Greater Bay Area. Moreover, the urban construction has formed a coordinated development trend that is dominated by the main urban area and supplemented by the coastline. Furthermore, Although the construction of the GHM Greater Bay Area has become China's national strategy, but there exist problems due to the weak connections between cities. It is necessary to strengthen the interaction of elements in various cities, and make full use of the leading advantages of the core urban area to develop coastal industries, so as to realize the integration of urban and rural areas and economic development in the GHM Greater Bay Area.

Characterizing the spatiotemporal changes in ecological vulnerability and reveal its driving factors is important, especially for unique geographical environment that connects mountains, rivers, and sea together, In this study, we chose the Southwest Guangxi karst-Beibu gulf coastal zone as the typical study area. We combined the spatial principal component analysis and geo-detector model to quantify the ecological vulnerability index and further characterizes the spatiotemporal changes in ecological vulnerability and evaluates its driving mechanism. Our results show that: (1) the vulnerability index of the study area was 0.54, 0.61 and 0.69, respectively for 2008, 2013, and 2018, with a multi-year average of 0.61. The general ecological vulnerability level was moderately fragile with a slightly worse trend over time; and (2) the top six explanatory driving factors of ecological vulnerability were rainfall in flood season (0.457), vegetation coverage (0.384), temperature in hot seasons (0.311), waste water input (0.248), NPP (0.184), and population density (0.036). For the interaction between driving factors, only rainfall in flood season and NPP, NPP and temperature in hot seasons, waste water input and NPP showed positive nonlinear relationships, while the rest had linear relationships. Flood season rainfall and vegetation coverage had the strongest effect on ecological vulnerability and a strongest interaction (0.679) with each other. Our study illustrates that flood season rainfall and vegetation coverage are the main driving factors of ecological vulnerability in the study area.

With China's rapidly aging population, the contradiction between inadequate elderly service supply and the demands of aging people becomes more serious. Beijing is one of the first cities that has entered the aging society. The "9064" pension model has been suggested to systematically address the shortage of pension resources. In order to provide the convenience of elderly care service and promote the satisfaction of elderly people, 17 communities which have much higher elderly population density in Haidian District are selected as the study area in this study. Based on multi-source data such as Baidu POI, community households, and remote sensing imagery, the statistical data of elderly population are gridded spatially. After the comparison between the national and provincial (municipal) community pension facilities’ service radius and the daily travel characteristics of the elderly people surveyed, residential service facility allocation indicators are chosen as the evaluation criteria. The Modified Two-Step Floating Catchment Area Method and the constructed spatial equilibrium coefficient are used to study the equilibrium of community pension facilities in three aspects: service coverage, facility capacity, and availability of medical resources. The results show that there are shortages of community care stations. The differences of service coverage and capacity are significant under different service radius. The service coverage of community care stations is only 23.3% within a 500-meter radius. The main reason of the low capacity of most elderly care stations is their uneven distribution in space. Local community health care facilities have positive effects on the availability of medical resources. As there exists nonequilibrium development of the elderly care service, further optimization of their allocation would be needed.

Since the implementation of the initiative of “Turning the Battlefields into Marketplaces” by Thai Prime Minister Chatichai Choonhavan in 1988, the relation between Thailand and its neighboring countries (including Cambodia, Laos and Myanmar) has shifted from political isolation to economic cooperation, which leads to varying degrees of land cover / use changes in the border areas (particularly the border crossings). With the land cover data products gathered from the US Agency for International Development's SERVIR-Mekong project, the status of land cover types and their spatial and temporal changes were quantitatively analyzed and compared between different countries using GIS-based spatial analysis, followed by the characteristics analyses of bordering and off-bordering, convergence and divergence of forest cover dynamics due to agricultural expansion (increased cropland) and cash-crops intensification (increased orchard or plantation) within the 20-km buffer areas of Thailand-Myanmar, Thailand-Laos, and Thailand-Cambodia borders in 1988, 1998, 2008, and 2018. The results show that: (1) Forests, cropland, and orchards (including plantations) were the main types of land cover in the border areas of Thailand, Myanmar, Laos, and Cambodia in the past decades, accounting for over 96%. An overall decrease of forests was also seen because of agricultural expansion and intensive production of commercial crops. (2) The forests conversion had obvious spatial and temporal features. Extensive forest transformations occurred in the neighborhood area of the border crossings of Mae Sot (Thailand)-Myawady (Myanmar) and Bueng Kan (Thailand)-Paksan (Laos) as well as the “Golden Triangle” area among Thailand, Myanmar, and Laos during 1998-2008. Similar transitions were also observed in the neighborhood area of the border crossings of Kap Choeng (Thailand)-Osmach(Cambodia) from 1988 to 1998, accounting for more than 70%. (3) Agricultural expansion and cash-crops intensification showed obvious characteristics of bordering and off-bordering, convergence and divergence in the border areas. Specifically, the forest changes caused by agricultural expansion has changed from the same to opposite direction showing divergence in the Thai-Lao border, from bilateral stronger bordering to unilateral bordering in the Myanmar side along the Thai-Burmese border, and from bordering to off-bordering and back to bordering in the Thai-Cambodian border. The forest changes caused by intensive production showed stronger bordering in the Thai-side of Thai-Lao border, from off-bordering to bordering in the Burmese side of Thai-Myanmar border, and bilateral opposite bordering of Thai-Cambodian border. (4) Bordering characteristics analysis of forest change indicated that Thailand had the greatest geo-influence on Laos, followed by Cambodia and Myanmar. Our study contributes to further investigating the cross-interaction between border land use and geo-economic relations from the perspective of spatial difference in Geography.

Hyperspectral images have been widely used in target detection, spectral decomposition, classification, and many other fields. They have higher recognition ability than grayscale images, panchromatic images, and multispectral images. However, how to effectively use hyperspectral images with large number of bands, huge data volume, and increased information redundancy is an important topic. Multiple kernel learning is a typical multi-view learning method that can make different kernel functions according to different feature spaces and group multiple kernel functions into an optimal kernel function for hyperspectral image classification. Compared to single kernel method, multiple kernel learning has unique advantages in solving problems such as the uneven spatial distribution of high-dimensional features, using information more efficiently, and improving classification accuracy greatly. At present, the research difficulty of multiple kernel learning algorithm are the combination of kernel functions and the selection of optimal weight coefficients. In order to improve the classification accuracy and promote the application of multiple kernel learning algorithm in hyperspectral image classification, we review the development history and current research progress of multiple kernel learning algorithms. First, the kernel learning method and the framework of multiple kernel learning algorithm are introduced. The specific methods of kernel function combination used in multiple kernel learning algorithm are summarized. According to many researches, it can be concluded that the linear method has been widely used because of its simplicity and efficiency. Moreover, according to the methods of determining weight coefficients in multiple kernel learning algorithm combination, multiple kernel learning algorithms can be generally divided into two categories: fixed-rule multiple kernel learning algorithm and optimization-based multiple kernel learning algorithm. Then, the applications of different multiple kernel learning algorithms from each category in hyperspectral image classification are reviewed. In order to facilitate researchers to discuss the problems of multiple kernel learning algorithm and hyperspectral image classification, the commonly used kernel functions and the widely used data sets in hyperspectral image classification are also reviewed. Finally, we discuss the deficiencies of multiple kernel learning algorithms in the field of hyperspectral image classification and point out the future research direction to help solve practical application problems.

Aiming at the problems of low accuracy and large error in artificial prick points during the internal work's precise correction of the aerial triangulation data during aerial photogrammetry, we propose a method in this paper for detecting the right-angled edge control points based on the LSD algorithm. First, the image is bilateral filtered to remove noise while preserving edge information and image color information enhanced by Retinex algorithm. Then, the right-angled edges of the image control point are extracted by LSD line detection, and the outermost right-angled edge is filtered by the angle, distance, and length information. Finally, the right-angled image control points are obtained through the intersection. We tested our method using 80 images of "L" image control points and 69 images of "X" image control points taken by Dajiang UAV. Results show that this method can get accurate pixel coordinates of image control points, and maintain a high accuracy in cases with complex background and target distortion. The overall image control point extraction accuracy rate was 93.75%, and the image control point positioning accuracy reached 2.3 pixels, which significantly overcame the artificial puncture points. Compared with Radon and PPHT algorithms, the accuracy of the image control point group detection is significantly improved in our results, which indicates a higher detection accuracy with less influence from the shooting angle.

Automatic extraction of buildings is of great significance to urban development and planning, and disaster prevention and early warning. Current researches on building extraction have achieved good results, but the existing research methods often take building extraction as a semantic segmentation problem and cannot distinguish different building individuals. Thus, there is still room of improvement in extraction accuracy. In recent years, deep learning methods based on multitask learning have been widely used in the field of computer vision, but its application in automatic interpretation of high-resolution remote sensing images has not yet further developed. The instance segmentation branch of Mask R-CNN is built on the basis of target detection, and can predict segmentation masks on each region of interest. However, some spatial details and the contextual information of the edge pixels of the region of interest will be lost inevitably. The semantic segmentation task can introduce more contextual information to the network. Therefore, the integration of semantic segmentation and instance segmentation tasks can improve the generalization performance of the whole network. Based on the classic instance segmentation method (Mask R-CNN) and a typical semantic segmentation method (U-Net), this research designs a deep neural network structure which embeds the semantic segmentation module into the instance segmentation framework, and improves the generalization performance of the model by using the information complementarity between various tasks. The bottom-up path augmentation structure shortens the path of lower layers’ information to pass up. The adaptive feature pooling makes it possible for instance segmentation network to make full use of multi-scale information. The automatic building segmentation of remote sensing images is performed in the multi-task training mode and the proposed method is verified on the classic remote sensing image data set (SpaceNet). The result shows that the building instance segmentation accuracy of our proposed method is 58.8% in the Paris data set and 60.7% in the Khartoum data set, increased by 1%~2% compared to individual Mask R-CNN and U-Net. The disadvantages of the proposed method are shown in two aspects, one is that the false extraction and missing extraction of small buildings are relatively high, and the other is that the accuracy of building boundary extraction needs to be improved.

Land cover data play an important role in global environment change studies and applications and are widely used in many fields. The FROM-GLC30 2017 dataset is one of the latest global high-resolution (30-meter) land cover datasets. The accuracy of this land cover dataset is of great interest and important for its application in other fields. The aim of this study was to evaluate the spatial accuracy of the FROM-GLC30 2017 dataset at the national scale and analyze the spatial variation of its accuracy for different land cover types. In our study, the reference land cover data were obtained through visual interpretation based on sub-meter high-resolution remote sensing images collected from 6434 small watersheds in China. The reference dataset was validated by field survey. Based on this, the accuracy of the FROM-GLC30 2017 dataset was further assessed. Our results show that: (1) the area proportion of each land cover type of the FROM-GLC30 2017 dataset generally matched the real field condition in China; (2) the overall accuracy of this dataset in China was 75.39%. Among the seven geographical divisions, the overall accuracy in east China was the highest, and the south China has the lowest accuracy; and (3) the accuracy of the bare land, forest, and cropland was relatively high, and the accuracy of the shrubland was the lowest among the seven land cover types. Our results provide theoretical support for large-scale land cover data accuracy assessment and promote the application of free land cover datasets.

Analysis of the spatial heterogeneity of soil attributes and diagnosis of the soil stress for crops cultivated at large-scale saline-alkali farmland based on remote sensing spectral indices are important to improve the land utilization efficiency and contribute to improve economic and ecological benefits. In this study, we conducted an Unmanned Aerial Vehicle (UAV) remote sensing observation and field measurement over a typical coastal saline-alkali farmland (400 hm²) in the Yellow River Delta of Dongying City, Shandong province in China during the growing season of maize and sorghum in 2019. An eBee wing-fixed UAV platform (SenseFly, Cheseaux-Lausanne, Switzerland) equipped with a multiSPEC-4C multispectral camera (SenseFly, Cheseaux-Lausanne, Switzerland) was used to capture the spectral information of crops. Nine Vegetation Indices (VIs) were selected to characterize the growth status of crops. Among the nine VIs, MCARI, TCARI/OSAVI, and NDREI were sensitive to Leaf Chlorophyll Content (LCC); OSVAI, GNDVI, and MSR were sensitive to Above-Ground Biomass (AGB); and NDVI, EVI2, and MSR_RE were sensitive to Leaf Area Index (LAI). Soil sampling (n = 195) at three layers (0~10 cm, 10~20 cm, and 20~40 cm) were implemented evenly across the study area. In total, five soil attributes were measured, including soil salinity (SALT, g/kg), pH, organic matter content (C, g/kg), total nitrogen content (N, g/kg), and available nitrogen content (SN, mg/kg). In our study, we first conducted an interpolation method using Inverse Distance Weighted (IDW) to map the spatial heterogeneity of soil attributes. Our interpolation results show that all the soil attributes showed obvious horizontal spatial heterogeneity, while pH and SALT showed remarkable vertical spatial heterogeneity. Second, we conducted the Pearson Correlation Analysis (PCA) between different soil attributes at each soil layer. The results of PCA showed that SALT and pH had a significantly negative correlation, and these two attributes were not related to SN, N, and C. While SN, N, and C had significantly positive relationships with each other. Finally, the influences of soil attributes on the growth status of maize and sorghum were assessed separately using the Recursive Feature Elimination (RFE) method along with the random forest model based on 3-fold cross validation and 100 times iteration. According to the importance values of soil attributes to VIs, the influence of soil attributes on crop growth from high to low was that SN>N, C>pH>SALT for maize, and SALT>pH>SN, N, C for sorghum. However, the dominant soil attributes that influenced crop growth were SN2 (i.e., SN at 10~20 cm soil layer) and SALT at 0~40 cm soil layer for maize and sorghum, respectively. This study proposes a 'soil-crop growth-VIs' framework for monitoring crop growth status based combining field sampling and UAV remote sensing observations, which is essential for agronomic management in saline-alkali land and contributes to the development of precision agriculture.