Developing a method for assessing environmental sustainability based on the Google Earth Engine platform

Environmental sustainability is the foundation and of great significance for the sustainable development of urban agglomerations. Taking the Beijing-Tianjin-Hebei urban agglomeration as an example, we developed a method to effectively assess long-term regional environmental sustainability based on the Google Earth Engine (GEE) platform. We used the GEE to obtain 5206 Landsat remote sensing images in the region from 1983 to 2016 and developed the comprehensive environmental index (CEI) to assess regional environmental sustainability based on the theme-oriented framework proposed by the United Nations Commission on Sustainable Development. We found that the environmental sustainability of the urban agglomeration showed a trend of first rising, then falling, and then rising again in the past 30 years. The average CEI increased from 0.621 to 0.631 from 1985 to 1990, dropped to the lowest value of 0.618 in 2000, and then rose to the highest value of 0.672 in 2015. In particular, the extent of areas in which environmental sustainability improved (56% of the region) was greater than the extent of areas in which environmental deterioration occurred. The environmental sustainability of Hengshui, Xingtai, and Cangzhou in the southeast of the region has been significantly improved. The method proposed in this study provides an automatic, rapid, and extensible way to assess regional environmental sustainability and provides a scientific reference for improving the sustainability of the regional environment.


Introduction
Sustainability is related to human survival and development and is a hot topic in the twenty-first century. Regional sustainability refers to the societal, economic, and environmental development that meet the developmental needs of contemporary people but do not damage the development of future generations and other regions (Brundtland et al., 1987). In addition, environmental sustainability is ensuring the development needs of future generations without destroying the environment (Goodland, 1995), and is fundamental to regional sustainability (Goodland and Daly, 1996). Among the three dimensions of sustainability (i.e., societal, economic, and environmental sustainability), environmental sustainability is the basis of regional sustainability and guarantees social sustainability and economic sustainability (Wu, 2013;Olafsson et al., 2014). Therefore, assessing regional environmental sustainability and its dynamic characteristics in time and space is important for achieving regional sustainable development.
Communicated by Philippe Garrigues.
In recent years, remote sensing has played an increasingly important role in regional environmental sustainability assessment studies. A recent review found that approximately 18% of more than 230 indicators used to assess global sustainable development make direct or indirect use of remote sensing monitoring data (Estoque, 2020). Remote sensing can provide a macroscopic and scientific data basis for environmental sustainability monitoring because it can acquire satellite monitoring data over a large area quickly and with high accuracy (Ustin, 2004;Wang, 2021). Early studies tended to focus on a single dimension of environment, such as atmosphere, water, soil, land use, or natural disasters. Many research institutions have evaluated the environmental conditions of an area using various indicators. Then, the evaluation indicator has also developed from a single dimension index to a comprehensive index which considers multiple factors of the environment. For example, the Environmental Sustainability Index (ESI) was developed by American scholars to evaluate the sustainability of countries around the world (Sands & Podmore, 2000). The Ministry of Environmental Protection of the People's Republic of China (2015) issued the revised Technical Criterion for Ecosystem Status Evaluation and proposed a comprehensive indicator, ecological index, to measure environmental condition from six dimensions (biodiversity, vegetation cover, water network density, land stress, pollution loading, and environmental restriction). As an example of a comprehensive index based on remote sensing data, Sutton (2003) developed the environmental sustainability index based on nighttime lighting data and land cover data to assess global environmental sustainability in 2001. Similarly, Xu (2013a) developed the remote sensing ecological index (RSEI) using Landsat images to assess urban environmental sustainability in Fuzhou, China, from 2001 to 2009. More recently, Shruti (2021) developed the smart city environmental sustainability index based on 24 environmental indicators to assess environmental sustainability in the construction of new smart cities. Although remote sensing-based means provide effective methods for monitoring and assessing environmental sustainability, limitations such as large data volume, heavy preprocessing work, and low efficiency of indicator calculation still exist, making them difficult to implement at large spatial scales or over a long-term scale.
The Google Earth Engine (GEE) platform provides a convenient way to process massive and complex remote sensing data. First, the GEE provides online access to remote sensing data from different satellites and scales around the world, including over 30 years of historical images and datasets, breaking the limitation of data downloading and local data storage. Second, the GEE has various algorithm interfaces, such as cloud masks, radiometric correction, and mountain shadow correction, so that remote sensing data can be easily and quickly preprocessed. In addition, the GEE is a cloud service platform that is able to perform parallel computing on the Google Cloud, which greatly improves the efficiency of data processing and indicator calculation. With the Google Cloud, the GEE has a great advantage in the integration and computation of big data (Kumar and Mutanga, 2018). For example, Hansen (2013) used 654,178 Landsat 7 remote sensing images (approximately 707 TB in size) to monitor forest cover change at the global scale. This data would take 10 6 h to preprocess on an ordinary personal computer and only 100 h to preprocess through accessing and computing online using the GEE. Therefore, it is expected that studies can monitor regional environment sustainability on a large scale and long-term scale using the strengths of the GEE.
The purpose of our study is to develop the comprehensive environmental index (CEI) based on the GEE for the rapid assessment of regional environmental sustainability. Taking the Beijing-Tianjin-Hebei urban agglomeration as an example, we assessed the dynamic characteristics of its environmental sustainability over the last 30 years. First, we used the GEE to acquire Landsat images in the study area for the summers of 1983-2016. Then, based upon the theme-oriented framework proposed by the United Nations Commission on Sustainable Development (UNCSD), we developed the CEI after calculating four indices representing three themes in the framework (UN, 2007). Finally, we analyzed the spatial and temporal dynamic characteristics of environmental sustainability in the Beijing-Tianjin-Hebei urban agglomeration from 1985 to 2015, which can provide a scientific reference for regional sustainable development.

Study area
The Beijing-Tianjin-Hebei urban agglomeration is located at 113° 27′ E-119° 50′ E and 35° 03′ N-42° 40′ N in the northern part of the North China Plain. The agglomeration borders the Bohai Sea to the east, the Huang-Huai Plain to the south, the Taihang Mountains to the west, and the Mongolian Plateau to the north. The agglomeration has a typical temperate continental monsoon climate zone. The average annual temperature rises gradually from north to south, while precipitation is unevenly distributed and mostly concentrated in summer. The Beijing-Tianjin-Hebei urban agglomeration covers an area of approximately 210,000 km 2 . The agglomeration includes the two municipalities of Beijing and Tianjin and 11 prefecture-level cities in Hebei Province: Zhangjiakou, Chengde, Qinhuangdao, Tangshan, Cangzhou, Hengshui, Langfang, Baoding, Shijiazhuang, Xingtai, and Handan (Fig. 1). Since China's reform and opening up, the study area has experienced rapid economic growth and urbanization, and the conflict between environmental protection and urban construction has also intensified. Therefore, the agglomeration has been a key area for environmental construction .

Data sources
In this study, we used remote sensing monitoring data from 1985 to 2015 and vector administrative boundary data for the Beijing-Tianjin-Hebei urban agglomeration. The remote sensing data included Landsat satellite images, atmospheric water content data, and surface emissivity data (Table 1). Landsat satellite data were obtained from the United States Geological Survey (USGS), with a spatial resolution of 30 m (https:// devel opers. google. com/ earth-engine/ datas ets/ catal og). Specifically, the satellite data used in this study were images from two  . Therefore, we did not use Landsat 7 satellite data in this study. Atmospheric water content data were obtained from the NCEP/NCAR Atmospheric Reanalysis Data jointly produced by the National Centers for Environmental Prediction (NCEP) and the National Center for Atmospheric Research (NCAR). The surface emissivity data are derived from ASTER-GED, a global surface emissivity dataset with a spatial resolution of 100 m published by NASA.
In addition, the administrative boundary vector data of the Beijing-Tianjin-Hebei urban agglomeration were obtained from the Resource Environment Science and Data Center, Institute of Geographical Sciences and Natural Resources Research, Chinese Academy of Sciences (www. resdc. cn/ Defau lt. aspx).

Methods
In this study, we quantified and analyzed the environmental sustainability of the Beijing-Tianjin-Hebei urban agglomeration over the past 30 years using the GEE platform (Fig. 2). First, we accessed Landsat 5/8 surface reflectance images online using the GEE for the corresponding study years and preprocessed the images with cloud masking, stitching, and integration. Then, we developed the CEI based on the sustainability assessment framework proposed by the UNCSD. Specifically, we selected and calculated four indices: NDVI (normalized difference vegetation index), NDBSI (normalized difference build and soil index), WET (wetness index), and LST (land surface temperatures), to calculate the CEI. Meanwhile, we calculated the modified normalized difference water index (MNDWI) for water masking. After water masking and normalization of the four indicators, we analyzed the characteristics of the variation in environmental sustainability in the Beijing-Tianjin-Hebei urban agglomeration from 1985 to 2015 based on the CEI.

Selecting images
We selected seven time points from 1985 to 2015 including 1985, 1990, 1995, 2000, 2005, 2010, and 2015. The integrated images for each time point, except for 1985, are synthesized from remote sensing images of the summer (from June 1 to September 30) of 3 years, including the current year, the previous year, and the year after. Because of the lack of images from 1984 to 1986, we used remote sensing images in the summers from 1983 to 1987 to synthesize the target images for 1985. Finally, we used 5206 remote sensing images of Landsat 5 TM and Landsat 8 OLI between 1983 and 2016 (Fig. 3). In GEE, we first filtered the remote sensing images in the time range of the target year and then used the cloud mask function provided by the GEE to mask the clouds of images from different satellites. Because of the frequent cloud cover in remote sensing images during summer, we used median pixel-by-pixel synthesis in image integration to avoid low-quality images (Midekisa et al., 2017). According to Fig. 3, there are 25 image tiles covering the Beijing-Tianjin-Hebei region, the number of images used for each Landsat tile in this study is not less than 118, and the number of images for each study year is not less than 539.

Developing the comprehensive environmental index
We developed the CEI based on the sustainability assessment framework proposed by the UNCSD. The framework identifies five major environmental themes to ensure that environmental sustainability is assessed in multiple dimensions: freshwater (water quality and water quantity); land (cropland, vegetation, desertification, and urbanization); atmosphere (air quality, ozone layer depletion, and climate change); biodiversity (ecosystems and species); and oceans, seas, and coasts (coastal zone and Flowchart for quantifying and analyzing regional environmental sustainability fisheries) (UN, 2007). It ensures a holistic assessment of the environmental conditions at the national or regional scale on multiple dimensions and is one of the most commonly used frameworks for assessing environmental sustainability worldwide (Wu and Wu, 2012;Brent et al., 2005). According to this framework, previous studies have used remote sensing data, e.g., MODIS data, to examine the change in environmental sustainability in China (He et al., 2017) and megacities in Eurasia (Lu et al., 2019).
The five themes in the UNCSD proposed framework can be represented, at least, partially by Landsat image-based indices (Fig. 4). Here, we selected four indices and covered three themes. The four indicators are originally selected by Xu (2013a) to form the RSEI-based Landsat images and The connection between the five themes of the UNCSD Environmental Assessment Framework and the selected four indices in this study principal component analysis. First, the NDVI is selected because it is closely related to vegetation biomass and the leaf area index (Goward et al., 2002). Therefore, it covers two themes (land and biodiversity) and can be a good indicator of the cover, diversity, and growth of surface vegetation, as well as an indicator of changes in surface land use and land cover (Gandhi et al., 2015;Fairbanks et al., 2004;Senay et al., 2000). Second, the NDBSI is selected because it characterizes the land in the region that is not covered by vegetation nor has an excavated surface. It is strongly associated with human urban construction activities (Hu and Xu, 2018). Therefore, we believe that it covers the desertification and urbanization in the theme of land. Third, the WET represents the moisture status of the observed surface and is closely related to surface water and moisture levels of atmosphere, soil, and vegetation (Xu, 2013b). It is closely related to the evapotranspiration of forests, grassland, and croplands in the theme of land, as well as the hydrological cycle between land surface and atmosphere in the theme of atmosphere. Fourth, we chose LST because heat island effect is pronounced in urban areas and threatens the health of urban residents (Tan et al., 2010). This index covers two themes, i.e., land and atmosphere, as it is highly related to the composition and configuration of vegetation cover and impervious surface, and heat exchange between land surface and atmosphere. Two themes, freshwater and oceans (seas and coasts), are not considered in this study due to a lack of effective remote sensing-based indicators. Nevertheless, we discussed the potentials in the future to represent the two remaining themes in the "Discussion" section. Different from Xu (2013a) which used the principal component analysis (PCA) to form the composite index, we used the geometric mean of these four indicators following previous studies (He et al., 2017). Using the geometric mean of the four indicators can guarantee that our results are comparable across regions, while the PCA-based results may vary across regions and cannot be comparable across studies. We also discussed other advantages of the CEI index in the "Discussion" section (see "Features of the developed method" section). The formula for the CEI is as below: where NDVI, NDBSI, WET, and LST are their value after normalization. It is worth noting that, when one of the four indicators is 0, the value of CEI would be 0, no matter how large the other three indicators are. To avoid this issue, we add 0.01 to each of the four metrics after normalization. Thus, the calculated CEI ranged from 0.01 to 1.01, which directly represents regional environmental sustainability. When the CEI is closer to 1.01, the environmental sustainability of the study area is better.

Calculating the four indices
The NDVI is calculated as follows: The NDBSI is the average of the soil index (SI) and indexbased built-up index (IBI) used to characterize the "dryness" of the ground surface (Xu, 2008;Rikimaru et al., 2002).
The SI and IBI are calculated using the following two formulas: WET is the wet component obtained from a tasseled cap transformation. For Landsat 5 TM and Landsat 8 OLI, different calculation formulas are to be used, and here are the formulas (Crist, 1985;Baig et al., 2014): In the above formulas, B , G , R , NIR , SWIR1 and SWIR2 are the reflectances in the blue, green, red, near-infrared, shortwave infrared 1, and shortwave infrared 2 bands from Landsat TM and Landsat 8 OLI satellites, respectively.
We used the LST obtained from the inversion of remote sensing images as an index representing the heat of land and atmosphere. Based on the mono-window algorithm for surface temperature proposed by Duguay-Tetzlaff et al. (2015), Ermida et al. (2020) achieved automated surface temperature, which can automatically invert the surface temperature of a single remote sensing image, using the GEE platform. The following is the basic formula, consisting of a linearization of the radiative transfer equation: where Tb is the top-of-atmosphere (TOA) brightness temperature in the Landsat's thermal infrared (TIR) channel, and ε is the surface emissivity for the same channel, which can be derived using the vegetation-cover method Caselles et al., 1997). The algorithm coefficients A i , B i , and C i are determined from linear regressions of radiative transfer simulations performed for 10 classes of total column water vapor (TCWV) (Kalnay et al., 1996). The TCWV data is available on GEE from NCEP/NCAR reanalysis data, with a sixhourly temporal resolution. But unlike a single remote sensing image, in our study, summer images from the year before and the year after the target year were used to integrate the target image. Hence, we used the average TCWV of the target year to represent the atmospheric water content value of the integrated image. Using this method, we achieved the calculation of land surface temperature of integrated images.
To check whether the NDVI and LST based on the GEE platform are able to represent regional environmental conditions, we compared the results in 2015 with the corresponding MODIS product in 2015 after resampling our results to a spatial resolution of 250 m. The correlation coefficients of NDVI and LST calculated based on these two data types were 0.79 and 0.76, which proved the reliability of the indices we calculated based on GEE.

Masking water
Because the NDVI values above water surface would be contradictory to our general assumption of environmental sustainability, we used the mask to exclude water for assessing sustainability. In specific, we assumed that NDVI has a positive relationship with environmental sustainability. However, for large water surface in the study area, NDVI is low, which suggests that the environmental sustainability on water surface is low. To avoid this issue, we used the MNDWI to identify the water. And based on the MNDWI, we ran the water mask algorithm for the NDVI, NDBSI, WET, and LST. The MNDWI is calculated using the following formula (Xu, 2005):

Normalizing the four indices
In order to eliminate the differences in indicators and enhance comparability between different years, it is necessary to normalize each indicator. According previous studies (Xu, 2013c;Xu et al., 2020), we adopted the maximum-minimum method to normalize the indicators: where x i is the value of the image pixel i. x max is the maximum value of x i , and x min is the minimum value of x i . y i1 is the normalization formula for positive indicators, and y i2 is the normalization method for negative indicators (Liu and Hao, 2017). According to Xu (2013a), NDVI and WET play positive roles and NDBSI and LST play negative roles in environmental sustainability. Therefore, the NDVI and WET used formula y i1 and the NDBSI and LST used formula y i2 to calculate the normalized results.

Static characteristics
We divided the CEI values into five quality grades considering the distribution of the values between 1985 and 2015 (Chen et al., 2019a, b). The values followed a normal distribution with most values lie within a range between 0.5 and 0.7. Then, we divided the CEI into five classes of 0-0.52, 0.52-0.59, 0.59-0.65, 0.65-0.71, and 0.71-1, according to the Natural Breaks method. These five quality grades, from low to high, represent environmental sustainability classes of poor, fair, moderate, good, and excellent, respectively. Based on this, we qualitatively and quantitatively analyzed the environmental sustainability of the Beijing-Tianjin-Hebei region in 2015. At the city scale, we quantitatively analyzed the distribution of the CEI of 13 cities in the Beijing-Tianjin-Hebei urban agglomeration in 2015 using methods such as zonal statistics.

Spatiotemporal dynamic characteristics
We calculated the CEI at seven time points from 1985 to 2015 based on the remote sensing images synthesized by the GEE and obtained the spatial distribution characteristics and dynamic change patterns of the CEI in the Beijing-Tianjin-Hebei region. Then, we used the abovementioned five quality grades to analyze the dynamic changes of each quality grade for the entire Beijing-Tianjin-Hebei region and the city scale. Finally, we analyzed the changes in the environmental sustainability of the Beijing-Tianjin-Hebei urban agglomeration over the past 30 years by detecting the changes in the quality grades for the results from 1985 to 2015.

Environmental Sustainability of the urban agglomeration in 2015
The average CEI of the Beijing-Tianjin-Hebei urban agglomeration in 2015 was 0.672, indicating that the agglomeration had a good environmental sustainability. As for the spatial distribution, the CEI was higher in the southeast and northeast and lower in the northwest of the region (Fig. 5a). Approximately 65% of the areas had excellent and good environmental sustainability quality grades and were mainly in Hengshui, Handan, and Xingtai in the southeast and northeast of Beijing-Tianjin-Hebei. The areas that had poor and fair sustainability quality grades accounted for 19% and were mainly concentrated in Zhangjiakou in the northwestern region (Fig. 5b). At the city scale, the CEI of more than half of the cities in the Beijing-Tianjin-Hebei region was higher than the average of the entire region. Among these cities, the CEI of 7 cities, namely, Chengde, Hengshui, Qinhuangdao, Beijing, Handan, Xingtai, and Tangshan, were higher than the regional average of 0.672. The highest value was 0.715 in Chengde, which was 6.5% higher than the regional average. However, the CEI of the other six cities were below 0.672. The lowest value was 0.613 in Zhangjiakou, which was only 91.2% of the average value of the entire region (Fig. 6).

CEI dynamics from 1985 to 2015
Over the past 30 years, the environmental sustainability of the Beijing-Tianjin-Hebei urban agglomeration showed a trend of rising, falling, and rising again (Fig. 7). From 1985 to 1990, the CEI rose from 0.621 to 0.631 with an increased rate of 1.6%. However, from 1990 to 2000, the CEI decreased by 2.1%, from 0.631 to the lowest value of 0.618. Then, from 2000 to 2015, the CEI rose to the highest value of 0.672, increasing by 8.7% compared to 2000. As for the distribution of the CEI, the differences in environmental sustainability were significant in the study area. The maximum values of the CEI in the seven target years ranged from 0.80 to 0.90, the minimum values ranged from 0.35 to 0.47, and the mean and median values fluctuated between 0.60 and 0.70. In addition, the median and mean values of the CEI for the 7 years did not exceed 0.02, and the upper and lower quartiles were almost symmetrical around the means and medians, which means that the statistical distribution of the CEI in the Beijing-Tianjin-Hebei region were approximately normal. Therefore, the mean and standard deviation can be used as references when grading the environmental sustainability quality.
Based on the results of the environmental sustainability quality grading, the trend of change in the areas of "excellent" quality grades was the same as the trend of the CEI in Beijing-Tianjin-Hebei urban agglomeration: rising, then falling, then rising again. The area with "good" quality fluctuated between 22.6 and 32.8%, and the area with "moderate" quality decreased from 27.6 to 15.8%. The areas of the "poor" and "fair" quality classes showed the same trend: first increasing and then decreasing (Fig. 8). In addition, the extent of areas in which environmental sustainability improved was much greater than the extent of areas in which it worsened in the Beijing-Tianjin-Hebei region. The areas where environmental sustainability improved comprised more than half of the region (Fig. 9). Approximately 56% of the regions experienced an increase in environmental sustainability quality, 12% experienced a decrease, and 32% remained stable. The areas with considerable improvements in quality grades were mainly located in Xingtai, Hengshui, and Cangzhou in the southeastern Beijing-Tianjin-Hebei urban agglomeration. The northwestern and northeastern areas experienced slight improvements. The areas that worsened were mainly concentrated in Beijing, Tianjin, and Shijiazhuang.
At the city scale, except for Tangshan and Shijiazhuang, the changes in the CEI in the remaining 11 cities first increased, then decreased, and then increased again. The lowest CEI for each city generally occurred in 1995 or 2000 while the highest CEI occurred in 2015. The CEI in Shijiazhuang first decreased and then increased. The overall trend of the CEI in Tangshan was a gradual increase. Considering the magnitudes of the changes, we sorted the CEI of 13 cities according to the extreme difference. We found that 3 cities, Beijing, Tianjin and Tangshan, had low magnitudes of change and high mean CEI, which meant that they had better environmental sustainability and less drastic changes. In contrast, Xingtai, Zhangjiakou and Hengshui had high fluctuations and low mean CEI, which meant that their environmental sustainability was low and dramatically changed (Fig. 10).
From the sustainability grading results of the 13 cities, we found that the 5 cities in the northeast Beijing-Tianjin-Hebei region (Beijing, Chengde, Qinhuangdao, Tangshan, and Tianjin) had higher environmental sustainability grades than the other cities in the past 30 years overall. They had higher percentages of "excellent" and "good" quality grades and lower percentages of "poor" and "fair" grades compared to other cities. In addition, after grading the environmental sustainability of the 13 cities, the results showed that the trend and magnitude of the changes in each quality grade showed significant differences between the cities (Fig. 11).

Features of the developed method
We developed a method for regional sustainability quantitative assessment using the GEE based on the environmental sustainability assessment framework proposed by the UNCSD. This method has at least four advantages, as well as several shortcomings.
First, it is intuitive, simple, and easy to interpret. We developed the CEI by calculating the geometric mean of the four indicators (NDVI, NDBSI, WET, and LST) after normalization. In other words, we set equal weights for the four indicators, rather than using principal component analysis (PCA), e.g., RSEI (Xu, 2013b) or varied weights based on entropy method (Cheng et al., 2021). It is intuitive to compare the changes in the CEI with the changes in the four indicators. By contrast, it is difficult to interpret the contribution of each indicator for the composite index using PCA or entropy weight method. In addition, equal weight setting also echoes the United Nation's principle of "leaving no one behind" of the 2030 Agenda for Sustainable Development Goals, and such setting was widely adopted previous studies (Schmidt-Traub et al., 2017;Xu et al., 2020). On the contrary, PCA cannot guarantee a high contribution rate (previous studies with a varying level of contribution rates between 60 and 90%), and the contribution rate of the first principal component gradually decreases as the number of indicators increases (Xu, 2013a;Shan et al., 2019;Xu et al., 2019).
Second, the developed CEI can provide a robust assessment of environmental sustainability, and the results are consistent with other sustainability assessment outcomes. To evaluate the robustness of the CEI, we compared the environmental assessments between two additional indicators, the RSEI and RSEIwi, and the CEI. The RSEI also used the four indicators (NDVI, NDBSI, WET, and LST) as the input. Different from CEI, the RSEI used the first component of the principal component analysis (PCA) to represent environmental condition (Xu, 2013a), and the RSEIwi was calculated using the entropy weight method (Cheng et al., 2021). The comparison among the three indicators showed that the CEI had a significant correlation with the results calculated by the classical RSEI model or RSEIwi (Fig. 12). The correlation coefficient between the CEI and RSEI was 0.86, and the correlation coefficient between the CEI and RSEIwi was 0.99, which proved that our method was reliable. The CEI can reflect the condition of environmental sustainability in urban agglomerations. In addition, to evaluate the reliability of the CEI, we calculated the correlation coefficients between the CEI and statistical indicators, which were derived from censuses.
As the data in the census are mainly indicators of anthropogenic activities (e.g., industrial emission) rather than environmental conditions at the city scale, we only The results showed that CEI was significantly correlated with some aspects of environmental conditions (e.g., NPP, urban green space per capita, and sewage treatment rate) and the level of economic development (e.g., GDP). It is also worth noting that CEI has a weak association with PM2.5 concentration, which suggested that CEI has limited ability to represent air quality. We expected that air condition indicators (e.g., Air Quality Index (AQI) based on statistical data) could be used in the future to represent such dimension.
Third, CEI can be feasibly applied to other regions to assess environmental sustainability, and the results of the two regions are comparable. The geometric means of the four indicators are more transparent than the PCA and entropy method to assign the weights. Therefore, if we define an upper and a lower bound to normalize the indicators, we can compare the results across regions and times. For example, the NDVI values vary between − 1 and 1. If we used this range to normalize the NDVI to a range of 0 to 1, we can guarantee the comparability. By contrast, the first component of PCA would vary across regions, which makes the results in different regions not comparable with each other.
Fourth, in our study, we developed the CEI with the GEE platform, which has great advantages in the online access, preprocessing, and indicator computing of remote sensing big data. The GEE is able to access 5206 online pieces of data and over 1200 GB of remote sensing data from the USGS between 1983 and 2016, and the GEE is free from the limitation of local computer data storage. Moreover, using the GEE to conduct preprocessing operations, including cloud mask and image integration, eliminates complicated offline manual processing. In addition, the GEE uses Google's high-performance servers for parallel cloud computing, which allows fast calculations and an efficient data processing.
However, there are some shortcomings of the method that could be improved. First, there are five themes included in the environmental sustainability assessment framework of the UNCSD, but NDVI, NDBSI, WET, and LST could only represent three of them that was because we lacked valid remote sensing indicators to represent the conditions in freshwater (water quality and water quantity) and oceans, seas, and coasts (coastal zone and fisheries). Therefore, it is expected that in the future valid remote sensing indicators representing these two themes can be developed to cover all five themes in the framework. In addition, the linkages between indicators and the themes are indirect. As an example, NDVI could not directly represent the biodiversity of an ecosystem; it only established an indirect correlation through the vegetation status. Also, WET and LST only indirectly represented the moisture or temperature condition of the atmosphere. In the future, it is expected that other indicators can be developed by combining remote sensing data with field data to represent biodiversity (Coops et al., 2008), water quality (Singh et al., 2013), and air quality (Wan et al., 2021) directly.
Second, we developed the CEI by calculating the geometric means of four indicators: the NDVI, NDBSI, WET, and LST. However, the elements of environment are not independent and juxtaposed with each other, so there must be interactions among the indicators of the four dimensions of the CEI. For example, when emissions of toxic gases from human industrial activities increase, it does not only affect the atmosphere, but it also affects surface vegetation, surface water quality, and ecosystem biodiversity, but such effect cannot be clearly defined. It is expected that new methods that integrate the indicators of various dimensions will be developed in the future.
Third, the method we adopted in grading the CEI determined the grading limits based on the distribution characteristics of the data. The grading results were not examined and verified with the actual situations, leading to possible errors. In the future, it is expected that we can access the environmental sustainability data based on filed investigations. By combining the investigation data and the theoretical data, we can form policies for the environment sustainable development.
Finally, we have not yet analyzed the driving factors of the CEI. The CEI covers four indices on three themes. Therefore, any factor that affects a change in one of the indices may cause a change in the CEI. But this effect becomes quite complicated when multiple factors change at the same time, because their contributions to CEI may be different. It is expected that in the future, the physical data (temperature and precipitation) and socioeconomic data (population, GDP, pollutant emissions, and land use) of the Beijing-Tianjin-Hebei urban agglomeration in a long time series can be used to analyze the driving factors of ecological and environmental quality changes.

Policy implication and future perspectives
The results show that the environmental sustainability measured by CEI in the Beijing-Tianjin-Hebei region has a trend of rising, then falling, then rising again from 1985 to 2015. The overall rising trends of CEI in the studied urban agglomeration suggested that the implementation of various conservation and restoration projects were effective for improving environmental sustainability at the regional scale. During this period, a series of conservation and restoration projects were implemented in this region to improve local environmental conditions. For instance, the National Soil and Water Conservation Project since 1983 and Grain for Green Project (returning farmland to forest) from 1999 to 2013 have increased greatly the forest cover through reforestation and afforestation (planting forests where no forests had been planted before). These large areas of trees have been planted to effectively control sandstorms, soil erosion, and the expansion of desertified land. They not only relieved the ecological risks of desertification and rapid urbanization, but also protected biodiversity and increased the carbon sequestration services of the regions (Bryan et al., 2018;Li et al., 2021). In addition, China's sustainable development program has implemented a number of water management projects over the past few decades. They have improved water quality, reduced river siltation, and increased soil water retention capacity. And the increase in surface moisture has contributed to stabilizing land surface temperatures in the region. Moreover, it is worth noting that from 1998 to 2015, the Chinese government's total annual investment in sustainable development projects steadily increased as the Chinese economy grew (Bryan et al., 2018). In the studied Beijing-Tianjin-Hebei urban agglomeration, our results also confirmed that the increase in environment sustainability measured by CEI had a significant positive association with regional GDP (Table 2).
Our finding that the environmental sustainability in this region has an overall rising trend and the turning point from decline to incline was approximately in the year of 2000 was also consistent with previous studies Chen et al., 2019a, b;Xu et al., 2021). In addition, our result also showed that Zhangjiakou city's CEI has increased more greatly than other cities in the urban agglomeration, especially after the year of 2000 (Lu et al., 2020). As the critical region for water conservation, wind prevention, and sand fixation, as well as the hosting city of the 2022 Winter Olympics Games, the environmental conditions in Zhangjiakou have received substantial improvement in light of ecological and environmental projects (Yeerkenbieke et al., 2021). Nevertheless, the CEI in 2015 in the cities of Cangzhou, Langfang, and Shijiazhuang were relatively lower than other cities. It suggests that their environmental quality should be further improved in the future.

Conclusions
In this study, we developed the Comprehensive Environmental Index (CEI) using Google Earth Engine to provide an assessment of environmental sustainability. The developed method has four advantages. First, the CEI is simple, intuitive, and easy to interpret. Second, it provides a reliable assessment of environmental sustainability. Third, it can guarantee comparability across space and time. Fourth, using GEE, it overcomes the difficulty in accessing, storing, and processing remote sensing big data, and achieved dynamic assessment of environmental sustainability on a large spatial scale and over a long period of time.
From 1985 to 2015, the overall environmental sustainability of the Beijing-Tianjin-Hebei urban agglomeration generally experienced a rising trend. Among the 13 cities, a few cities in the northeast (e.g., Beijing, Chengde, Qinhuangdao, and Tangshan) had generally better environmental sustainability than others cities in the agglomeration. In the future, the improvement of environmental quality in cities of Cangzhou, Langfang, and Zhangjiakou should be continued to promote the environmental construction of the Beijing-Tianjin-Hebei urban agglomeration.
Author contribution Xianwang Xia: data processing and analysis, writing-original draft preparation, and visualization.
Chentai Jiao: data processing and analysis. Shixiong Song: data processing and analysis, conceptualization, and methodology.
Funding This research was supported in part by the National Natural Science Foundation of China (Grant No. 41971225), the Beijing Municipal Natural Science Fund (Grant No. 8192027), and the Beijing Normal University Tang Scholar.

Data Availability
The datasets generated and/or analyzed during the current study are not publicly available due to the sensitive nature of the raw data but are available from the corresponding author on reasonable request.

Declarations
Ethics approval Not applicable.

Consent to participate Not applicable.
Consent for publication Not applicable.

Competing interests
The authors declare no competing interests.