The GBGEZ is situated in the coastal areas of Southwest China and consists of the administrative regions of Chongzuo, Yulin, Qinzhou, Fangchenggang, Beihai, and Nanning (Figure 1). The land area is 42,500 square kilometers, and it had a population of 24.56 million at the end of 2019. In January 2008, China proposed building the GBGEZ into an important regional and international economic cooperation zone. The GBGEZ is China's first regional and international economic cooperation zone and is intended to become China's fourth economic growth pole. The construction goal of the GBGEZ is to become the China-ASEAN business, cooperation logistics, manufacturing and processing bases, and information exchange center to drive and support the development of the western region of strategically important open highland with economic prosperity, vibrant culture, healthy ecology, social harmony, and an important international economic cooperation area.

The RSEI index data in this study were Landsat 7 and MODIS data for the GBGEZ, all of which were downloaded from the Cloud Computing Website (https://code.earthngine. google.com/), and the effective data time was 2001.01.01-2020.12.31. The greenness, wetness, and dryness were computed from the Landsat 7 dataset. The LST data were downloaded from the QC_Day band of MOD11A2.006 Terra Land Surface Temperature dataset. The LST is a stable product in the MODIS dataset that can be reduced and calculated directly. The greenness, wetness, and dryness index data were reduced with GEE, which took the average pixel value for each index every year. The administrative boundary, population, and GDP spatial data of the study region are from the Resource and Environmental Science and Data Center of the Institute of Geographical Sciences and Natural Resources Research. The detailed data sources and descriptions are shown in Table 1.

The four indices used in the RSEI were wetness, heat, dryness, and greenness, which are often used in evaluating ecological status, are typically associated with eco-environmental status, and can be instantly understood by person (Mishra et al., 2015; Coutts et al., 2016). RSEI indicates that the adjustment of the four indices will cause the transformation of the land surface ecological environment. Accordingly, we can use a four-index function to define the RSEI:

The greenness index is represented by the normalized difference vegetation indicator (NDVI), which is typically calculated using satellite remote sensing data to evaluate the growth status of green vegetation in the study region (De Araujo et al., 2015). The index method is calculated based on the reflection of red and near-infrared light, which can describe productivity, ecosystem vitality, plant growth, and other information. The larger the value, the greater the plant growth. The formula of NDVI is as follows (Rouse et al., 1973):

where Red is the pixel value of the red light reflection band and Nir is the pixel value of the near-infrared light reflection band. There is a maximum value of 1 when Red = 0. In contrast, there is a minimum value of -1 when Nir = 0; thus, the output value of the index is between -1.0 and 1.0.

The wetness index is represented by the humidity obtained by remote sensing tasseled cap transformation. The wetness index reflects the humidity of soil, water and vegetation, which is closely related to ecology (Kauth and Thomas, 1976; Huang et al., 2002). For Landsat TM images, the formula of wetness is as follows (Crist and Cicone, 1984):

where Blue, Green, Swir1 and Swir2 are the blue light reflection, green light reflection, short infrared band 1 reflection, and short infrared band 2 reflection band values of pixels, respectively. The greater the value of wetness is, the greater the contribution to the EEQ.

Dryness was expressed by the NDBI index composed of the index-based built-up index (IBI) (Xu, 2008) and bare soil index (BI) (Rikimaru et al., 2002), which are defined as follows:

The larger the NDBI value is, the smaller the contribution to the EEQ.

Thermal sensor data of MODIS LST product data represented heat in this study. LST is an important index that is typically used to examine climate change and ecological processes, as well as to study evapotranspiration, drought, surface energy balance, and vegetation density (Rhee et al., 2010; Weng et al., 2014; Xu, 2014).

Granular computing is an intelligent research method that divides and simplifies complex problems into several simple problems. The key assignment of granular computing is to construct, represent and process information granules (Artiemjew, 2020). Information granulation refers to the decomposition of an object space into many subspaces or the aggregation of individuals in the space into different classes based on useful information and knowledge. These classes are called granules, and the elements in the granules can be described as instances of corresponding concepts (Liang et al., 2015). Remote sensing images are an object space that conforms to the study and computing category of granular computing. Combined with the definition of granular computing, this study defines remote sensing information particles. In a remote sensing image, there are equivalence relations, similarity relations, and close relations among remote sensing pixels, and the set of remote sensing pixels that exhibit a certain relationship is called a remote sensing information particle. The remote sensing image decomposition that creates different spatial unit aggregates according to a certain relationship is called remote sensing image information granulation. For example, in Figure 3, the value of an index (e.g., a land type) of the image composed of 12 remote sensing pixels is {1, 2, 3}. According to the index value equality relationship, the remote sensing image information is granulated into remote sensing pixel information granule 1, remote sensing pixel information granule 2, and remote sensing pixel information granule 3. Using remote sensing information granulation, each index of the remote sensing image can be granulated into different spatial information particles.

Each index of the RSEI is a remote sensing image that is composed of continuous values. Different index values have the corresponding spatial distribution of remote sensing pixels, and all index values and the corresponding remote sensing pixels form the index histogram. For example, Figure 4a shows a spatial distribution map of the greenness index that is composed of continuous values; Figure 4b shows the corresponding histogram of the greenness index, where the ordinate represents the number of spatial pixels of the greenness index value; and Figure 4c shows the spatial remote sensing pixel information granules of several values of the greenness index. Each index of the RSEI can be granulated into different spatial units aggregated through pixel value-histogram-equal relation-remote sensing pixel information granules so that the complex calculation of continuous remote sensing image values can be transformed into a simple granular computing problem.

The weight reflects the position of each index in the eco-environmental evaluation index system and represents the contribution of the index itself to the RSEI evaluation. Most existing studies of the areal environmental quality evaluation of the RSEI use principal component analysis to determine the weights (Xu, 2013; Xu et al., 2019; Yue et al., 2019). Principal component analysis does not produce clear results when interpreted as a weighting method but can produce good results compared to high-dimensional evaluation indices. The distribution of remote sensing pixel values is random and uncertain. The greater the uncertainty, the greater the information entropy of the remote sensing index, the greater the amount of information contained, and the smaller the uncertainty, the smaller the information entropy of the remote sensing index, and the smaller the amount of information contained. The amount of remote sensing index information is a measure to judge the randomness and dispersion of the index spatial distribution. The greater the dispersion of the index is, the greater the impact of the index on the comprehensive evaluation of the RSEI. This study used the information entropy weight method to calculate each index weight and uses the spatial distribution of all pixel values of each RSEI index to determine the index weight. The relevant formula is as follows:

where Er is the value of information entropy for one index, n is the number of distinct numbers in each RSEI index, and p_ij is the probability value of each pixel value of remote sensing pixel information particles in all remote sensing pixel information particles in all indices. The pixels with different pixel values in each index constitute remote sensing information granules. In this study, combined with the concept of knowledge granularity and knowledge granularity entropy in granular computing, a method to establish RSEI weights based on remote sensing pixel values and remote sensing spatial information particles was proposed. In the evaluation of the RSEI, evaluation indices are refined into different hierarchical partitions according to different pixel values. Different zones and levels were subdivided by different pixel values for the evaluation of the RSEI. The knowledge granularity is a type of value of this type of zone variation, which can be calculated by the following formula (Liang et al., 2004):

where Gk is the knowledge granularity, $\left| {{X}_{i}} \right|\text{ }\!\!~\!\!\text{ }$is the number of samples of remote sensing pixel information particles for each pixel value, $\left| U \right|\text{ }\!\!~\!\!\text{ }$is the total number of samples in the area, and n is the number of pixel values for each index in an area (i.e., the total number of remote sensing pixel information particles). The smaller the value of knowledge granularity is, the stronger the resolution of the index; otherwise, the larger the value of knowledge granularity is, the weaker the resolving ability of the index. The knowledge granularity entropy based on knowledge granularity is defined as follows:

where k = 1/log_{2 $\left| U \right|\text{ }\!\!~\!\!\text{ }$}, Er(R) is the knowledge granularity entropy of the index. The smaller the value of index knowledge granularity entropy, the finer it divides the evaluation area, the greater its contribution to RSEI, and the stronger its resolution. According to the law of the entropy weight method and the knowledge granularity entropy of every index, the weight formula of every index in the RSEI is defined as follows:

where m is the number of indices and W(C_i) is the final weight value of each index.

In this study, the spatiotemporal variations in dryness, wetness, and greenness were calculated from 2001 to 2020. From 2001 to 2020, the maximum greenness mean value of each year in the GBGEZ was 0.3568, which occurred in 2013, and the minimum greenness mean value of each year was 0.257, which occurred in 2003. From 2001 to 2020, the greenness mean value of each year showed a growth trend, and the greenness contribution to EEQ increased year by year, as shown in Figure 5. The maximum standard deviation of greenness was 0.0949, which occurred in 2011, and the minimum standard deviation was 0.0502, which occurred in 2002. The standard deviation showed an increasing trend, and the spatial difference in greenness in the study region increased year by year, as detailed in Figure 6. The GBGEZ considered ecological and environmental protection during the study period; the vegetation coverage in GBGEZ showed an increasing trend; and the increased area of vegetation in all local regions levels is greater than the reduced area of vegetation, indicating that in the process of economic development, local municipal governments paid attention to environmental protection, put the ecological environment in the first place, and encourage people to perform afforestation and returning farmland to forest activities. Also, the process had effectively implemented the national policy of ecological civilization construction.

From 2001 to 2020, the maximum dryness mean value of each year in the GBGEZ was -0.0987, which occurred in 2020, and the minimum dryness mean value of each year was -0.1625, which occurred in 2003. From 2001 to 2020, the dryness mean value of every year in the GBGEZ showed a decreasing trend, the surface dryness decreased year by year, and the contribution of dryness to the EEQ increased year by year, as shown in Figure 5. The maximum standard deviation of dryness was 0.0632, which occurred in 2019, and the minimum standard deviation was 0.0322, which occurred in 2001. The standard deviation showed an increasing trend, and the spatial difference in dryness in the study region increased year by year (Figure 6). In the early study period of GBGEZ, the rural residential area occupied a lot of land, and construction land was extensive, which posed a certain threat to the maintenance of the regional ecological environment, particularly the coastal ecological environment. With the continuous promotion of urban-rural integration, the intensity of construction in the study area increased. During urban construction, increasing attention has been given to the restoration of the ecological environment.

From 2001 to 2020, the maximum wetness mean value of each year in the GBGEZ was -0.0847, which occurred in 2013, and the minimum wetness mean value of each year was -0.1132, which occurred in 2003. From 2001 to 2020, the wetness mean value of each year in the GBGEZ showed a decreasing trend, while the land surface wetness increased from year to year, and the contribution of wetness to the EEQ increased, as shown in Figure 5. The maximum standard deviation of wetness was 0.0412, which occurred in 2012, and the minimum standard deviation was 0.0267, which occurred in 2002. The standard deviation showed an increasing trend, and the spatial difference in wetness in the study region increased year by year (Figure 6). Compared with greenness and dryness, the change in wetness in the GBGEZ slowed. The GBGEZ once encouraged the planting of eucalyptus and other fast-growing forests. Sparse forestland and high coverage grassland were planted into timber forest and economic forest, which had a certain impact on soil water conservation. With the gradual cancellation of the subsidy policy for fast-growing forests, the soil water conservation capacity has been gradually improved.

From 2001 to 2020, the maximum heat mean value of each year in the GBGEZ was 21.0143, which occurred in 2009, and the minimum heat mean value of each year was 15.1025, which occurred in 2002. From 2001 to 2020, the mean heat value every year in the GBGEZ showed a marginal decreasing trend. The maximum standard deviation of heat was 6.5516, which occurred in 2003, and the minimum standard deviation was 4.9479, which occurred in 2002. The standard deviation showed an increasing trend, and the spatial difference in heat in the study region increased year by year, as shown in Figure 7.

This study performed a comprehensive evaluation of the GBGEZ over five years: 2001, 2005, 2010, 2015, and 2020. In these five evaluations, the granularity entropy of the heat index is the largest, the spatial differentiation of the heat index in the study region is relatively small, its contribution to the EEQ is the smallest, and its weight is relatively small. The overall dryness granularity entropy is the smallest, and the spatial differentiation of the dryness index in the study region is relatively large, which contributes the most to the EEQ and has a relatively large weight. The “dryness” caused by economic activities in the study region has a strong effect on the EEQ. From 2001 to 2020, the contribution order of the RSEI indices for the eco-environment in the GBGEZ was dryness, wetness, greenness, and heat (Table 2). The detailed values of knowledge granulation entropy and the weight value of every index in 2001, 2005, 2010, 2015, and 2020 are listed in Table 2. The change in land use mode and the change in soil water conservation brought by economic forest had the strongest impact on the eco-environmental quality of the GBGEZ, and the vegetation coverage in the study area was high.

During the 2001-2020 period, the mean ecological remote sensing quality in the GBGEZ did not change much but showed a marginal upward trend. The maximum value is 0.8120, which occurred in 2001, and the minimum value is 0.1151, which occurred in 2001. The difference between each year was low, and the spatial trend fluctuates. The maximum mean was 0.4110, which occurred in 2005, and the minimum mean was 0.3937, which occurred in 2020. The mean value slowly increased and declined slowly, as shown in Table 3. The overall fluctuation of ecological remote sensing quality in the GBGEZ is thus weak. During the study period, ecological damage behaviors such as soil erosion, desertification, desertification, sharp decline of forests, land degradation and reduction of biodiversity were rare.

From 2001 to 2020, the minimum and maximum means were considered to be the upper and lower limits of the eco-environmental zoning, and the GBGEZ was divided into five ecological zones. The GBGEZ is primarily composed of second- and third-level ecological zones. The first-level ecological zones with extremely poor ecological conditions and the fourth- and fifth-level ecological zones with excellent ecological conditions account for a small proportion, and these areas typically are at higher altitudes, have higher temperatures, and exhibit slow soil development and fragile ecological environment. The overall proportion of second- and third-level ecological zones remained relatively constant, and the spatial difference was not apparent overall, as shown in Table 4. The first-level ecological areas are scattered in some areas in northwestern Nanning and northeastern Chongzuo. The second-level ecological areas are distributed in the middle of the GBGEZ, which primarily consists of cities, farmland, and other economically active areas. The third-level ecological areas are distributed in the east and west of the GBGEZ, and the west and east are primarily mountainous areas. In terms of spatial distribution, the EEQ of the GBGEZ in the west and east was better than that in the middle, as shown in Figure 8.

From 2001 to 2020, the transformation of the eco-environment in the GBGEZ was primarily between levels 2 and 3, and 3 and 4. In general, there are more areas of improved remote sensing EEQ than areas of degraded EEQ. Remote sensing eco-environment degradation areas are distributed in the east and west, while remote sensing ecological environment improvement areas are distributed in the middle of the study region. The remote sensing EEQ in the GBGEZ improved markedly from 2010-1015, as shown in Figure 9.

From 2001 to 2005, 27.446% of the EEQ results of the GBGEZ were converted from level 2 to level 3; 15.269% of the regions were converted from level 3 to level 2; and at this stage, the EEQ was converted to the direction of ecological environment improvement. From 2005 to 2010, 12.412% of the EEQ results of the GBGEZ were converted from level 2 to level 3; 28.471% of the regions were converted from level 3 to level 2; and the EEQ was converted to the direction of ecological environment degradation. From 2010 to 2015, 25.677% of the EEQ results of the GBGEZ were converted from level 2 to level 3; 9.93% of the regions were converted from level 3 to level 2; and the EEQ was converted to the direction of ecological environment improvement. From 2015 to 2020, 11.104% of the EEQ results of the GBGEZ were converted from level 2 to level 3; 26.549% of the regions were converted from level 3 to level 2; and the EEQ was converted to the direction of ecological environment degradation. Detailed values are listed in Table 5.

Population and GDP are closely related to EEQ. For data reasons, this study used the Pearson correlation coefficient to study the relationship between population, GDP, and the RSEI in the GBGEZ in 2001, 2005, 2010, and 2015. The population density and GDP of the GBGEZ increased during the study period. The population and GDP in the city and surrounding areas are larger than those in the mountainous and coastal areas, and the central part is larger than the eastern, western, and northern areas. In 2001-2015, there was a negative correlation coefficient value between population, GDP, and RSEI (Table 6). Population and GDP inhibited the improvement of the EEQ in the GBGEZ to a certain extent.

Because the economic growth of the GBGEZ still belonged to the resource-dependent growth mode, the improvement of resource use efficiency cannot offset the increase in the demand for natural resources in economic development. Therefore, the ecological environment was not well protected during social and economic development. The GBGEZ should rely on advanced technical means to continuously promote the resource-saving production mode, improve regional green competitiveness and practice the concept of sustainable development. The correlation coefficient value first increased and then decreased, indicating that the government of the study region is paying increasing attention to the management of eco-environment quality while developing the economy.

In existing studies of spatiotemporal evaluation of RSEI, most weights were determined by principal component analysis (Xiong et al., 2021; Ning et al., 2020). Principal component analysis is a type of orthogonal transformation calculation based on the continuous value of the index, which requires a lot of computational resources and is difficult to explain. The RSEI spatiotemporal evaluation based on remote sensing pixel information granulation proposed in this paper granulates the remote sensing image information into different remote sensing spatial information granules according to the index values. This method preserves the information of all the index values, avoids the information profit and loss caused by information granulation (Liao and Jiang, 2020), makes calculations easier, and exhibits strong interpretability. Through the granules and granulation of remote sensing information, complex spatial computing problems can be abstracted and divided into several simpler problems to simplify the spatial computing problems of remote sensing and improve the efficiency of solving the problems. As a form of remote sensing spatial information, remote sensing information granules have become a basic concept for researchers to process and perceive complex spatial information. Remote sensing EEQ is the result that is directly driven by each index. Based on the evaluation result of the RSEI, existing studies randomly sampled a certain number of spatial sample points in the study region to analyze the impact of each index on the remote sensing eco-environment (Cheng and He, 2019; Song et al., 2019). The evaluation framework of this study, which considers all remote sensing spatial pixels and index values, can directly measure the magnitude and ranking of the impact of evaluation indices on the remote sensing eco-environment.

The GBGEZ government must establish ecological perspectives and ecological values and construct a land and space planning system from the perspective of ecological civilization epistemology. Ecological environmental monitoring must be expanded from past monitoring primarily on construction land to cultivated land, forest, wetland, water area, mangroves, and other categories, and a satellite remote sensing monitoring system with all elements, all-weather, all-time, and all-scale landscapes, forests, lakes, and grass should be established. Ecological environment monitoring should be converted from quantity monitoring to quality monitoring. Concurrently, time-series observations should be performed, and the change in ecological status and the analysis of ecosystem function should be investigated and monitored by integrating the factors of geography, climate, economy and humanity. The GBGEZ should strengthen the construction of the concept of regional integrated development of the ecological environment and promote the construction of the guarantee system of coordinated governance of ecological environment.

Due to the limitation of remote sensing data, this study only calculated the spatiotemporal evaluation of the RSEI in the GBGEZ from 2001 to 2020 and did not calculate the RSEI of the study region with longer time series. Although the RSEI based on the RSEI can quickly and accurately evaluate the regional EEQ, more testing is required to determine whether the four indices can fully describe the EEQ of the study region. This study only calculated the correlation between GDP, population density, and the RSEI and did not calculate more related factors to find the driving factors of eco-environmental change in the GBGEZ. In future studies, the discovery of an association between land-use change and RSEI change should be found in combination with land-use change.