Evaluation and prediction of ecological environment of karst world heritage sites based on google earth engine: a case study of Libo–Huanjiang karst

The assessment and prediction of ecological quality can help to quickly and systematically understand the ecological status of World Natural Heritage and assist with developing appropriate strategies to ensure the sustainable and healthy development of that heritage. Using the google earth engine platform, the remote sensing ecological index (RSEI) is rapidly constructed based on principal component analysis to evaluate the spatial and temporal distribution characteristics and the main impact indicators of the ecological environment, and the cellular automata (CA)-Markov model are used to simulate and predict the ecological quality, taking the Libo–Huanjiang karst world natural heritage site as the study area. The results show that: (a) the contribution rate of the four ecological indicators on the first principal component (PC1) are more than 90%. The construction of RSEI based on PC1 are applicable in karst world nature heritage sites and can be used to monitor and evaluate the spatial and temporal variation characteristics of the ecological environment. (b) The mean values of RSEI of the Libo–Huanjiang heritage site during 2000, 2007, 2013 and 2020 were 0.5435, 0.5465, 0.6009 and 0.5101, The overall ecological quality is mainly moderate and good, but in the eastern part of the heritage site the ecological quality is poor. (c) The evolution of the ecological environment quality during 20 years is mainly divided into the development trend of rapidly getting better, slowly getting better, and maintaining stability. (d) In the analysis of the relationship between RSEI and altitude, it is found that the ecological environment quality is mainly inferior, less favorable and moderate in areas with altitudes below 600 m, and there is a positive relationship between ecological quality level and altitude. (e) By analyzing the apparent spatial aggregation between the ecological environment quality, and then simulating the ecological grades in 2027 and 2033 using CA-Markov model, it is predicted that the area of medium and excellent ecological grades will increase in the future, but the ecological environment quality still needs to be improved in the eastern region due to the development of the tourism industry. Overall, the remote sensing ecological index is an effective model for evaluating and monitoring the ecological environment quality of karst heritage sites; the ecological environment quality of the Libo–Huanjiang heritage site is in a steady state of improvement, and the conservation measures of relevant departments are beginning to bear fruit; further coordination between conservation and development is needed to promote the sustainable development of heritage sites and to provide effective solutions for monitoring other karst-like heritage sites.


Introduction
As a valuable treasure for all mankind [1], world natural heritage sites (WNHS) have been receiving attention from people worldwide for their unique scenic landscapes, rich cultural heritage, and long historicity [2]. While WNHS bring great benefits to the heritage sites, they are facing serious threats, and some have been destroyed [3]. Nowadays, 52 WNHS have been placed on the danger list, and conservation and longterm sustainable development of WNHS is urgent. The 'South China Karst' is a WNHS project proposed by the Chinese government to the UNESCO World Heritage Committee in batches. Its unique geomorphology, ecosystems, biodiversity, natural beauty, and evolutionary processes have significant global value and significance [4]. Karst is one of the most remarkable landscapes in the world [5], at the same time, karst areas are characterized by soil vulnerability, hydrological vulnerability, vegetation vulnerability, and human vulnerability [6]. Therefore, it is completely necessary to monitor them regularly and regularly to protect the value attributes of karst WNHS.
The world heritage (WH) convention states that monitoring is a fundamental requirement for world natural reserves [7]. In the 1970s the WH center developed a program of regular monitoring reports, conservation status, etc, to investigate and track the health of natural WH properties [8]. Since the signing of the heritage convention, WH conservation has become a major concern, and heritage conservation and monitoring have become a worldwide movement and trend [9]. However, in monitoring WNHS, it is inevitable to consider the constraints of human and financial resources and choose the appropriate strategy [10,11]. Different scholars have different approaches to the way heritage is monitored, and various monitoring efforts are basically carried out with individual indicators as monitoring factors [12,13]. Ecological environment quality measurement methods mainly include two types: one is single-factor change analysis, such as analysis of changes in factors closely related to ecological environment such as land use change [14,15], net primary production (NPP) [16] and ecoefficiency change [17]. The other type is multi-factor change integrated analysis, which is more comprehensive and accurate compared with single-factor analysis, and scholars have proposed various evaluation index systems for this purpose [18,19]. Xu [20] proposed a purely remote sensing-driven remote sensing based ecological index (RSEI) to reflect ecological status comprehensively, with normalization and principal component analysis from four aspects of greenness (NDVI), heat (LST), humidity (WET) and dryness (NDSI) to achieve remote sensing ecological status evaluation; subsequently several scholars have conducted practical studies in typical regions such as Bayinbrook WNHS and Karajun-Kurdening WNHS in Xinjiang, China [21,22] Although this method has been successfully applied in different regions, there is no remote sensing ecological evaluation study of karst heritage sites. Compared with traditional remote sensing monitoring means, google earth engine (GEE), as a cloud-based platform, can more easily access high-performance data resources and calculate huge data sets [23,24], The trends of ecological quality changes in the Xishuangbanna and Yunnan Erhai watersheds in China has been quantitatively assessed by the GEE platform, providing an effective reference for the subsequent ecological environment in the study area [25][26][27][28]. Compared with traditional RSEI modeling, using the GEE platform allows researchers to focus more on the research purpose itself rather than some repetitive technical work [29,30]. The GEE platform provides many built-in codes and functions that are easy to call [31,32], such as, built-in codes and functions that ensure researchers can obtain timely and accurate changes in regional RSEI. In this paper, the principal component analysis utilizes the GEE platform, which greatly increases the efficiency of the study. Therefore, it is necessary and feasible for us to choose to use the GEE platform as a long time series ecological environment quality monitoring platform.
To meet the needs for higher spatial resolution, longer time series and high performance computing, and moreover for the assessment of ecological environment quality of karst WNHS, we choose the GEE platform with Landsat as the data source to study the ecological environment condition of Libo-Huanjiang karst WNHS, in order to prove the applicability of RSEI to the ecological environment quality assessment of karst WNHS, and to reveal the change pattern of the ecological quality of the Libo-Huanjiang, and finally tried to simulate the future ecological environment quality condition of the WNHS using metacellular automata as a prediction model to provide scientific basis for the conservation and monitoring of other karst WNHS.

Introduction to the study area
The Libo-Huanjiang karst WNHS is located at the junction of Guizhou Province and Guangxi Province of China. The heritage Site is located at the border of Guizhou and Guangxi provinces in China (figure 1). The heritage site is home to the Shui, Buyi, Yao, Zhuang and Miao ethnic minorities, which account for 84.20% of the total population. The livelihood of the inhabitants has a more obvious impact on the ecological environment of the heritage site and the buffer zone. The Libo-Huanjiang karst WNHS is dominated by karst landscape with the inherent characteristics of the karst binary hydrogeological structure, facing the vulnerability of karst structure-function-habitat and the urgency of heritage conservation.

Data sources and pre-processing
The remote sensing image data used for this study were derived from the T1-rated (highest quality) Landsat 5(TM), Landsat 7(TM), and Landsat 8(OLI) time series image sets provided by the GEE platform database, which were geometrically corrected, radiometrically corrected, and atmospherically corrected, with a spatial resolution of 30 m and a temporal resolution of 16 d. The digital elevation model (DEM) data were obtained from the Geospatial Data Cloud (www.gscloud.cn).

RSEI model
The remote sensing ecological index is a new remote sensing ecological index that uses multi-source remote sensing data and integrates several ecological factors as the main drivers to monitor and evaluate the ecological environment in a certain area. Compared with the Ecological Index (EI) of the code, it has the characteristics of a short cycle time and wide application, which is a good complement to the EI index [20]. Regarding the four indicators, the formula is shown in the supplementary materials.
The four standardized images were then synthesized and the four indicators were coupled using the principal component analysis (PCA) method, a multidimensional data compression technique that selects a few important variables by orthogonal linear transformation of multiple variables. With the greatest advantage of integrating the weights of each indicator, avoiding artificial determination and automatically and objectively based on the nature of the data itself and the contribution of each indicator to each principal component [33]. The calculation of the initial RSEI was performed using the following formula, Similarly, the calculated RSEI values were standardized to obtain the final remotely sensed ecological index.

Moran's index
The first law of geography states that the correlation between features is related to distance, and in general, the closer the distance, the greater the correlation between features; the farther the distance, the greater the dissimilarity between features [34]. Therefore, the law of spatial correlation is often used in spatio-temporal evolution studies, and the global Moran's I index are used to express the global spatial autocorrelation, in the formula, I is the global Moran's I index with values ranging from [−1,1], I < 0 indicates negative correlation, and I > 0 indicates positive correlation; ω ij is the weighting coefficient, X i and X j are the remotely sensed ecological indices at i and j in the study area. Local spatial autocorrelation in terms of local Moran's I index,

Cellular automata (CA)-Markov model prediction
CA-Markov is a model that can simulate and predict spatial and temporal evolution based on image elements [35], which integrates the spatial simulation advantages of CA models and the temporal prediction advantages of Markov models. It is widely used in the simulation and prediction of land-use and landcover change. We used the Markov model to predict the number of future ecological quality classes, and the number of conversions of each class will be counted during the model run. Therefore, we try to use the CA-Markov model to predict the ecological quality of karst WNHS. The CA model is a local network dynamics model with discrete time, space and state, which can simulate the spatio-temporal evolution process of complex systems [36]. Its formulation is as follows, in the formula, S is the metacell state; is the metacell transformation rule in local space; t and t + 1 are the respective metacell moments; n is the interval time of two moments. The Markov model considers that the change of things obeys the Markov process, which considers that the latter state of things is only related to the previous moment state. The transfer matrix is formulated as follows, (6) in the formula, X (t+1) and X (t) are the states at moments t + 1 and t. P ij are the transfer matrices.

RSEI model test
As can be seen from  [20]. The mean RSEI values and data distribution for the four years in the study area are statistically presented in figure 2, which shows that the RSEI values in the study area increased from 0.543 in 2000 to 0.6 in 2017, with an increasing trend of 0.003/a in average, but then decreased to 0.51 in 2020. After a visit to the local heritage authority, it was learned that during this period, a study on the policy of migration to the buffer zone for the local community residents, and the small-scale development and construction in the buffer zone, the maintenance of facilities in the two tourist areas, and the construction of the Guinan Railway due to the needs of the community residents, did not affect the value attributes of the heritage site. Figure 3 reflects the spatial distribution of RSEI values in the study area (2000-2020). As a whole, it can be divided into two stages. From 2000 to 2013, the areas with poor (I) and less favorable (II) and medium (III) ecological environment quality were mainly distributed in the northwestern part of the study area as well as in the central scattered areas. The areas with good (IV) and excellent (V) ecological environments are located in the eastern and southeastern parts of the study area. From 2013 to 2020, the areas with poor (I) and less favorable (II) and medium (III) ecological environment quality are mainly located in the central part of the study area, mainly in the middle buffer area of the two heritage sites, while other areas are basically in good (IV) and excellent (V). And from figure 3, we can also get that from 2000 to 2013, the ecological quality grade of the study area is mainly medium (III) and good (IV), and the area accounts for more than 90%, From 2013 to 2020, the ecological quality grade of the study area is mainly medium (III) and good (IV), and the area accounts for more than 95%.

Spatial and temporal patterns of ecological quality
Overall, it can be seen that the ecological quality of the study area from 2000 to 2020 decreased in the middle years due to the policy influence, but then increased again, and the ecological quality remains good.

Trends in the evolution of ecological environment quality
Based on the above analysis, the evolution of the ecological quality of the study area was divided into two periods (2000-2013 and 2013-2020). From figures 4 and 5, it can be seen that the ecological environment quality of the study area shows a general trend of 'rapid improvement-slow improvement-stable' during this period. The specific analysis is as follows.
From figure 4, it can be seen that during 2000-2007, the study area was basically in the trend of significant improvement and significant improvement. The area with improved ecological grade was 484.90 km 2 , accounting for 60% of the total area of the study area, while the area with degraded ecological quality was 199.56 km 2 , accounting for 24.51% of the total area. The improved area exceeds the degraded area by 285.34 km 2 . It can be seen from figure 5 that the ecological environment quality improved mainly due to the change from good to excellent (116.77 km 2 ) and from medium to excellent (343.20 km 2 ). During the period 2007-2013, the ecological environment quality started slowly getting better, and the area of ecological grade improvement was 128.79 km 2 , accounting for 15% of the total area of the study area. 2013-2020, the ecological environment quality tends to be in a stable state, and the area of grade improvement is 96.1 km 2 , while the area of decrease is 45.56 km 2 , the ecological quality of other parts of the study area remained stable and unchanged.

Relationship between DEM and ecological environment quality
Since the study area belongs to the Southern China Karst WNHS, it reflects unique physical geographic features. It demonstrates the topography, landforms, and other natural aspects of the karst region with unique physical geographic features of the same latitude region as in the northern hemisphere. In particular, the cone-shaped karst landforms in the study area are distributed 250-1250 m above sea level, and the pyramidal conical hills are spectacular, which are not only typical representatives of karst landforms in the world but also outstanding types of continental landforms. Therefore, based on the spatial statistical   function of the geographic information system (GIS) to analyze the relationship between RSEI grades and elevation, it is found that the area of grades of poor, less favorable and moderate ecological environment quality in the area below 600 m in elevation accounts for more than 60%. In contrast, while the area of good grades accounts for nearly 40% and the percentage of excellent grades is almost zero ( figure 6). However, the proportion of medium and good grades increased with increasing altitude, indicating that the ecological grades increased with increasing altitude in the study area and showed a positive correlation.

Spatial correlation analysis and ecological prediction 3.5.1. Moran's index analysis
Calculation of the global Moran's I index for the study area from 2000 to 2020 (figure 7) reveals that the global Moran's I index shows a decreasing trend. Still, most of it is located in quadrants one and three, indicating that there is a significant spatial aggregation among ecological environment quality. Its spatial characteristics show that areas with higher ecological environment quality are clustered with each other, and areas with lower ecological environment quality are the spatial characteristics show that areas  with higher ecological quality are clustered together. Areas with lower ecological environment quality are the spatial characteristics show that areas with higher ecological quality are clustered together and areas with lower ecological quality are adjacent to each other. Generally, the spatial clustering of ecological environment quality in the study area is significantly enhanced.
The local autocorrelation analysis of the ecological environment quality in the study area, (figure 8) showed that the study area mainly showed five types of spatial characteristics: high-high aggregation, high-low aggregation, low-low aggregation, bottom-high aggregation and non-significant. The high and high aggregation types were mainly concentrated in the eastern part of the study area and stabilized within the heritage site. In contrast, the low and low aggregation types moved from the eastern part of the study area to the central and central-western parts of the study area, and gradually moved to the buffer zone of the study area.

Correlation analysis
We discuss the spatial dependence of RSEI by using regression models and, based on previous studies, find that spatial error models are more suitable for explaining the distribution patterns and interactions of RSEI [21]. The standardized regression results of  were significantly and positively correlated with the RSEI index, while LST and NDSI were significantly and negatively correlated with the RSEI index, and from the results, it was mainly NDVI that influenced the RSEI index, indicating that the distribution rank characteristics of the RSEI index were significantly related to the condition of vegetation growth, which was also consistent with the vegetation cover of the WNHS.

CA-Markov model prediction
Based on the calculated results of RSEI levels in 2000, 2007, 2013, and 2020, the CA-Markov model was applied to simulate the RSEI levels in 2013 and 2020. After calculation, the Kappa coefficients between the simulated and calculated results in 2013 and 2020 were 0.68 and 0.70, respectively, which belonged to a high degree of agreement, indicating the model's suitability in the prediction of ecological quality in the study area. Therefore, this study applied the model to simulate the ecological quality condition of the study area, and based on the results, simulations were conducted for 2027 and 2033 ( figure 9). After the GIS spatial statistics calculation, the percentage of areas with grades of poor, less favorable, moderate, good and excellent in 2027 was 0.05%, 3.71%, 55.72%, 40.12%, and 0.39%, respectively. Compared to 2020, the area of moderate and excellent has increased, but the good area has slightly decreased and the other grades have not changed. By 2033, the ecological grade status of most areas in the study area is good and medium. Still, none of them reaches excellent, mainly because the study area has the brand effect of a WNHS, and the tourism industry is developing rapidly. The establishment of some tourism facilities and the life of local community residents will affect the ecological grade status of the heritage site. Therefore, focus future ecological management and regulation of heritage sites should focus on the coordination between the development of the local tourism industry and heritage conservation.

Discussion
There are many methods for monitoring and evaluating the quality of the ecological environment. From the 1970s, when NASA began monitoring land use, vegetation cover, and water accumulation in Appalachia using Earth resources satellites [37], remote sensing has gradually developed into an important tool for monitoring the ecological environment. The factors affecting the quality of the ecological environment are also complex and diverse, and the evaluation of a single indicator cannot fully explain the combined effect of many factors in the ecological environment [38,39]. As can be seen from table 1, the contribution rate of all four indicators in PC1 reaches more than 90%, indicating that PC1 has concentrated most of the information of the four indicators; therefore, the RSEI index constructed by choosing PC1 has the advantages of comprehensive, objective and efficient access to the changing state of ecological environment quality as well as easy visualization, spatial and temporal analysis and prediction compared with other indices [20]. This is also consistent with the results of other scholars who have evaluated the ecological quality of WNHS [21]. Although RSEI cannot completely reflect the ecological environment quality of a watershed, it is the most comprehensively considered of the existing ecological indices, so it is also the most widely used at present.
With the increasing concern of UNESCO about the environmental threat status of WNHS [40]. In this paper, we constructed the RSEI model through the GEE platform to objectively, comprehensively, and deeply analyze the changing trends and spatial and temporal patterns of ecological environment quality in the Libo-Huanjiang karst WNHS in the past 20 years, and the study is representative and reference value. Through the comprehensive analysis results of this paper, the areas with poor and worse ecological grades of the Libo-Huanjiang heritage site are mainly concentrated in the centraleastern part of the site (figure 3), which is mainly since the area was formerly a scenic area with a relatively well-developed tourism industry [41]. The buffer zone in the area has aboriginal people who have lived here for generations. The life and livelihood of the aboriginal people will have an impact on the ecological environment. Therefore, this area is a priority area for the heritage authorities. In the trend of ecological environment evolution in the past 20 years, it is obvious that the ecological environment quality in general shows 'rapid improvement-slow improvement-stable' (figures 4 and 5). Xiong et al found that since the Guinan railroad opened in 2017 crossed the buffer zone of the WNHS, it did not affect its Outstanding Universal Value (OUV) value, however, it had the influence of surface reflection spectral errors in the evaluation of the ecological and environmental quality of the buffer zone [42]. Overall, the trend of better ecological environment quality in the Libo-Huanjiang karst WNHS is more obvious, which is inextricably related to the guidance of UNESCO, the policy management of local government, and the practical actions of management departments, and the conscious protection of the heritage by indigenous people.
In this paper, we have completed the calculation and analysis of RSEI for the full spatial and temporal coverage of the Libo-Huanjiang WNHS and the buffer zone since 2000, revealing the macroscopic pattern of ecological condition changes, but many aspects still need to be improved. Future research should consider the establishment of a synergistic monitoring system between the sky and the earth and the actual situation of the study area to conduct microscopic environmental evaluation, such as acoustic environment, water environment, and atmospheric environment. Meanwhile, when the applicability of the CA-Markov model was examined, the Kappa coefficients of the simulation results were highly consistent, but there were some errors, which were related to the uncertainty of the images themselves and the comprehensiveness of the selection of factors, and were expected to be improved subsequently.

Conclusion
Ecological and environmental quality assessment is important for the sustainable development of WNHS. This study attempts to use the GEE platform to rapidly assess the ecological environmental quality, which helps to understand the dynamic changes of environmental quality of the WNHS in a long time series. The results show that the contribution of PC1 principal component eigenvalues of the four-phase images reaches more than 90%, and RSEI is applicable to the ecological environment quality assessment of the Libo-Huanjiang karst WNHS; the ecological environment quality shows a distribution pattern that the eastern part is higher than the western part, and the central buffer zone has the lowest ecological grade, and the ecological environment quality of the heritage site shows a general pattern of. The ecological quality of the heritage sites shows an evolutionary trend of 'rapidly improving-slowly improving-maintaining stable' , and the ecological grade is maintained from low to high; the ecological quality grade increases with the elevation and is positively correlated; there is an obvious spatial aggregation among the ecological quality and a significant strengthening trend. Based on this, the CA-Markov model is used to simulate the future ecological quality. The area of medium and excellent ecological grade will increase in 2027 and 2033, but the ecological quality of the eastern region still cannot reach good and excellent.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https:// earthexplorer.usgs.gov/.