Spatio-temporal evolution and driving forces of habitat quality in Guizhou Province

This study aimed to analyze spatio-temporal changes in habitat quality in Guizhou Province during the 1990–2018 period and identify factors influencing habitat quality. Land-use data for the period were used to evaluate spatio-temporal variations in habitat quality using the InVEST model, and factors influencing habitat quality were analyzed using GeoDetector. According to the results, cultivated land and forestland decreased by 0.48% and 0.88%, respectively, during the study period. Grassland, water, and construction land areas increased, with construction land increasing the most (0.92%) followed by water area (0.37%). The main land-use changes included conversion of cultivated land to forestland, grassland, and construction land. The average habitat quality index for Guizhou Province changed from 0.633 to 0.627 over the 1990–2018 period, showing an overall downward trend. The distribution pattern of habitat quality was spatially "high in the north, south, and, east, and low in the west". The most significant improvement of habitat quality was in the western region, whereas the most significant decline of habitat quality was in the central region. Land-use was the major factor influencing the spatio-temporal variations in habitat quality, and the interactive effect between any two factors was stronger than that of a single factor. Natural factors and human factors co-dominated the temporal-spatial changes in habitat quality.


Materials and methods
Study area. Guizhou Province, which is located in the hinterland of southwest China (103°360-109°350 E; 24°37-29°130 N) (Fig. 1), and covers a total area of 1.76 × 10 5 km 2 . Guizhou Province belongs to the low-latitude zone and experiences a subtropical humid monsoon climate with abundant annual precipitation of 1000-1400 mm and relative humidity is above 70%. The annual average temperature is 15 °C. Guizhou is a unique mountainous province in China without plains, more than 90% of the total area is occupied by mountains or hills with the altitudes ranging from 143 to 2889 m. The complex topography and climate contribute to the diversity of land-use types. The study area has nine municipalities and is quite underdeveloped. The population pressure is generally high, with the province having a population of more than 36 million population in 2019. Majority of the residents are poor and live in poverty in remote mountain villages (Data obtained from Guizhou Statistical Yearbook, http:// stjj. Guizh ou. gov. cn). Guizhou Province is a relatively unique regional environment www.nature.com/scientificreports/ unit, where over 60% of the total area is covered by an outcrop carbonate karst. The unique geological structure makes the area a vulnerable ecosystem.
Data sources. Land-use data for 1990Land-use data for , 2000Land-use data for , 2010, and 2018 were obtained from the Data Center for Resources and Environmental Sciences of the Chinese Academy of Sciences (http:// www. resdc. cn) and the spatial resolution was 30 m × 30 m 35 . The land-use data included six primary land-use types and 20 secondary landuse types ( Table 1). The accuracy of interpretation of the data was more than 90%, which satisfied the requirements of the study 36 . Digital elevation model (DEM) data was acquired from the Geospatial Data Cloud with a resolution of 30 m (http:// www. gsclo ud. cn/) using ASTER global DEM (v2). Slope and elevation were acquired from DEM data. Temperature, precipitation, net primary productivity (NPP), normalized difference vegetation index (NDVI), population density, and gross domestic product (GDP) data were obtained from the Data Center for Resources and Environmental Sciences of the Chinese Academy of Sciences at a resolution of 1 km × 1 km. Specifically, temperature and precipitation data were generated using the ANUSPLIN interpolation software. The spatial distribution dataset (NDVI) comprises a monthly vegetation index dataset generated since 1998 based on the SPOT-VEGETATION NDVI satellite remote sensing data of continuous time series and the maximum synthesis method. GDP dataset, which is based on the national GDP data of a county, describes the spatial relationships among land-use type, night lighting brightness, settlement density data closely associated with human activities and GDP, and generates spatial grid data through spatial interpolation. The spatial distribution data for population density is basically consistent with GDP. NPP is the fixed energy or organic matter produced per unit area and time left after a green plant respires, and is based on simulations of the light energy utilization model, GLO-PEM. In addition, due to the availability of data, NPP, temperature, precipitation, and population density data were from 2015; while GDP and NDVI data were from 2018.

Methods
Geological atlas analysis. Geographical Atlas analysis can simultaneously express the spatial dynamic changes in landscape structure over time, as well as synthesizing land-use changes and HQ patterns by mapping land-use and HQ 37,38 . To determine land-use and HQ changes in Guizhou Province over the 1980-2018 period, growth map, decrease map, and HQ grade transfer map of land-use changes for the period were obtained using the following formula 39,40 : where N is the code for land-use change type; F is the code for land-use type before conversion; L is the code for land-use type after conversion.
InVEST-HQ assessment. The InVEST-HQ model was used to assess HQ of Guizhou Province, which was estimated based on information regarding LULC changes and biodiversity threats 25 . The assessment depended on the distance and intensity of habitat and non-habitat land-use. HQ was reflected by the HQ index that was calculated as follows 22 : where Q xj denotes the HQ of pixel x in land-use type j, D xj is the threat level of pixel x in land-use type j, H j is the habitat suitability of land-use type j, and k is half the saturation constant (half of the maximum value of D xj ).
where R denotes the number of stress factors, Y r is the total number of grid cells of stress factors, ω r is weight of the stress factor, r y is the number of stress factors in a grid cell, β x is the accessibility level of grid x, S jr is the sensitivity of land-use type j to stress factors, with a value range of 0-1, and i rxy is the maximum distance of influence of the stress factors. www.nature.com/scientificreports/ where d xy denotes the distance between grid x and y, and d rmax is the maximum impact scope of the threat factor r. The data inputs (spatial and nonspatial), required to run the InVEST-HQ model include multiple-date LULC maps, threat sources and impacts, habitat types, habitat sensitivity to the threats, and the half-saturation constant. InVEST-HQ model finally generates an HQ map with values ranging from 0 to 1 41 , where 1 represents the highest suitability. Construction land reflects the threat to biodiversity threat posed by human activities. As a semi-natural and semi-artificial ecosystem, cultivated land has a negative impact on the natural environment 42 . Therefore, construction and cultivated lands were used as threat factors. Considering that the study area is a typical karst landform, predominantly a mountainous and hilly plateau with scattered rural settlements, construction land was converted to urban construction land, rural settlements, and other types of land-use. Based on the model manual 22 and literature review 43,44 , the relative weight of threat factors, the maximum distance between habitats and each threat source, and the habitat sensitivity of each threat were determined (Tables 2 and 3). Global Moran's I is a measure of spatial autocorrelation proposed by Patrick Moran 45 .

Global Moran's I.
Spatial autocorrelation refers to the degree to which an attribute of a geographical location is related to different spatial locations, and can be used to describe whether HQ has a clustering effect over the entire area, with a value range of [−1, 1] 46 . Global Moran's I was calculated using the following formula:  www.nature.com/scientificreports/ where I is the global Moran's index; n is the total number of study units; y is the average habitat quality; y i and y j are the habitat quality indices for i and j study areas; w ij is the spatial weight coefficient matrix of regions i and j, which reflects the spatial relationship between regions i and j, and is defined as w ij = 1 or otherwise w ij = 0; and E(I) and VAR(I) represent the expected value and variance of the global Moran's index y, respectively.

Hot-spot analysis (Getis-Ord Gi*).
Hot spot analysis is a method of evaluating the distribution of clusters in local areas and can be used to indicate whether there are statistically significant high and low values in the spatial distribution pattern of HQ 47 . A significantly positive G* value shows HQ is a clustering of high values, which is a hot spot, whereas a significantly negative G* value shows HQ is a clustering of low values, which is a cold spot 48 . The G*value was calculated using the following formula: where xj is the habitat quality of grid j; Wij is the spatial weight matrix of grid i and grid j, if i is adjacent to j, its spatial weight is 1, otherwise it is 0; is the average habitat quality; S is the standard deviation of habitat quality; n is the total number of rasters.

GeoDetector analysis.
GeoDetector is a statistical tool used for detecting spatially stratified heterogeneity and revealing the factors associated with it. The GeoDetector has four main components: divergence and factor detection, interaction detection, risk zone detection, and ecological detection, which are mainly used to analyze the factors influencing various phenomena and the interactions among multiple factors 36,49 . We analyzed the extent to which factors influence spatial variations in HQ using factor and interaction detectors. The factor detector is typically used to measure the extent to which each factor is associated with spatial variation in HQ while the interaction detector is used to measure the interaction between two factors. The magnitude of the divergence is determined by the q-value of the GeoDetector, which is calculated using the following formula 50 : where q is the degree of explanation of the spatial distribution of HQ by an influencing factor with a value ranging from 0 to 1, where larger values indicate a greater effect of the factor on the spatial variation of HQ and smaller values indicate a small effect of the factor; L is the sample size of the influencing factor; N h and N are the numbers of cells in stratum h and the entire area, respectively; σn 2 and σ 2 are the variances of HQ in stratum h and the entire area, respectively. Interaction detection focuses on determining the interaction between various factors and assessing whether multiple factors acting together increase or decrease the driving power of HQ or whether there is no interaction between the factors (Table 4) 50 .
The study area was sampled based on a 5 km × 5 km grid using ArcGIS 10.7 and a total of 7036 evaluation units were obtained. The factors influencing HQ were divided into natural and human factors, where natural factors included elevation (X1), slope (X2), temperature (X3), rainfall (X4), NPP (X5), NDVI (X6), whereas human factors included population density (X7), GDP (X8), and land-use type (X9). The HQ index was the dependent variable and the factors influencing HQ were the independent variables. In addition, the natural breakpoint method of ArcGIS 10.7 was used to discretize and classify each factor, and correlation results were calculated using the GeoDetector.

Results
InVEST-HQ assessment. Land-use change analysis. From 1990 to 2018, substantial changes in land-use patterns occurred in Guizhou Province and forestland, cultivated land, and grassland were the main land-use types (Table 5). Grassland, water, and construction land areas increased, whereas cultivated land, forestland, and unused land areas decreased during the study period. Specifically, cultivated and construction land areas increased significantly from 1990 to 2018 by − 98.37% and 398.71%, respectively, followed by grassland and water Table 4. Types of interactions between two Covariates to HQ.
Maps showing increases and decreases in land-use changes in Guizhou Province from 1980 to 2018 were obtained from a geographic atlas (Fig. 2). Land-use transformations in Guizhou Province during the 1990-2018 period mainly involved cultivated land, forestland, and grassland, as well as the conversion of forestland, cultivated land, and grassland to construction land. Cultivated land was mainly transformed into forestland 3557.52 km 2 , which was sporadically distributed across the province; forestland was transformed into grassland 4707.10 km 2 , which was largely distributed in the central and western regions; grassland was transformed into forestland 3810.20 km 2 , which was distributed in the western region; other land-use types were converted to construction land 1664.16 km 2 , which was distributed in the central region of the study area. Specifically, the conversion of cultivated land during the 1990-2018 period was the most notable land-use change, with 1357.50, 218.67, and 968.39 km 2 of cultivated land area being converted to grassland, water, and construction land, respectively. A total of 3526.27km 2 , 337.85km 2 , and 371.26 km 2 of forestland were transformed into cultivated land, water and construction land, respectively. A total of 1689.22, 130.14, and 321.57 km 2 were transformed into cultivated land, water, and construction land, respectively (Table 6). Therefore, the increase in construction land in Guizhou   www.nature.com/scientificreports/ Province is primarily attributed to the loss of cultivated land, forestland, and grassland, while the decrease in forestland is attributed to the increase in grassland.

Temporal and spatial variations in HQ.
The InVEST-HQ model was used to obtain a HQ map of Guizhou Province and HQ was graded into five levels using the natural breakpoint method as follows: low (0-0.2), moderately low (0.2-0.4), medium (0.4-0.6), moderately high (0.6-0.8), and high (0.8-1.0). The habitat areas and percentages of each grade over the four periods are summarized in Table 7. From a spatial scale (Fig. 3). HQ for construction and rural residential land was low, while that for natural forest landscape and grassland was high. The main habitat types in Guizhou Province had medium and high HQ (Table 7). High HQ areas were mainly distributed in the central and western parts of Zunyi, central part of Tongren, and southwestern and eastern parts of Qianxinan. However, low HQ areas were predominantly distributed in the southern part of Guiyang, eastern part of Bijie, and northern parts of Anshun and Qianxinan.
Based on temporal changes, the proportion of low HQ areas decreased by 0.66% (from 6.88 to 6.22%) over the 1990-2018 period; the proportions of both moderately low and medium HQ areas increased by 0.95% (from 10.73% to 11.32%) and 0.28% (34.88% to 35.16%), respectively. The proportions of moderately high and high quality habitat areas decreased by 0.18% (from 21.75 to 21.57%) and 0.03% (from 25.75 to 25.72%), respectively, during the study period (Table 6). www.nature.com/scientificreports/ Overall, the average HQ indices for Guizhou Province in 1990, 2000, 2010, and 2018 were 0.633, 0.629, 0.630, and 0.627, respectively, indicating that the HQ declined slightly over the period. HQ initially decreased, then increased, and eventually decreased. The overall habitat change in Guizhou Province was negligible due to the relatively stable proportions of forestland, grassland, and cultivated land that make up the landscape patch of the province. The percentage decreases in HQ over the 1990-2000, 2000-2010, and 2010-2018 periods were 0.63%, 0.15%, and 0.47%, respectively.
To gain insight into the spatial and temporal characteristics of HQ, ArcGIS was used to generate a map showing the spatial variations in HQ in Guizhou Province from 1900 to 2018 (Fig. 4). HQ remained stable in most parts of Guizhou Province from 1990 to 2000, with significant decline in HQ being observed in the central and western regions and increased HQ being observed in the southwestern region. HQ of Guizhou Province improved from 2000 to 2010, although a decrease in HQ was observed in the central region. HQ changed significantly from 2000 to 2010, with a decrease in HQ being observed in areas located in the central-western and northern regions of Guizhou Province, and improved HQ being observed in the southwestern region. Overall, no significant differences were observed in HQ of 84% of the total land area from 1990 to 2018. HQ has decreased in Guizhou Province, especially in the central, western, and northern regions due to the rapid increase in construction land caused by accelerated urbanization, which has in turn, stressed the surrounding habitats and increased habitat fragmentation and loss of connectivity, in addition to decreasing HQ in the areas. The habitats with improved quality were mainly located in the southwestern region, which is attributed to the implementation of projects, such as "returning cultivated land to forestland" and "closing mountains for afforestation and grass cultivation".
Spatial  (Fig. 5). The overall geographical distribution of HQ in Guizhou Province was "high in the south and east, low in the central and west, and high and low in the north". The hot spots were mainly concentrated in the southern and eastern regions, where grasses are interspersed in forests, and a few hot spots were distributed in the northern and central regions of the woodland and water areas during the study period. Cold spots were mainly distributed in construction land and some portions of cultivated land.

Variations in HQ characteristics of different LULC.
Land-use has a significant impact on regional HQ and changes in land-use lead to changes in HQ 51 . Human activities have altered the spatial patterns of land-use, thereby resulting in habitat change and biodiversity loss. To better understand the impact of land-use change on  (Fig. 6). HQ indices for the different land-use types varied over the four study periods. During the 1990-2018 period, forestland had the highest average HQ index, followed by grassland and water, and the lowest average HQ value was observed in construction land. HQ indices of forestland, cultivated land, and unused land did not exhibit significant variations; however, the HQ index of water initially increased and then decreased. The HQ indices of grassland, cultivated land, and water remained unchanged from 1990 to 2000, while those of forestland and cultivated land exhibited a decreasing trend from 2000 to 2010. The HQ indices of cultivated land and forestland improved and exhibited an increasing trend from 2010 to 2018, whereas those of grassland and water decreased. With the increase in population, the demand for land is likely to increase accordingly, and more forestland and grassland will be cleared for cultivation, as well as the conversion of large areas of cultivated land to construction land. Therefore, HQ indices of various LULC are affected by human activities. However, the implementation of active interventions, such as restoration of forestland and grassland and relevant environmental conversation activities, has led to a decline in habitat degradation of land-use types disturbed by anthropogenic factors  www.nature.com/scientificreports/ has declined and improved HQ. The HQ index of forestland over the 1990-2018 period remained above 0.73, which could be because the forestland in Guizhou Province is mainly distributed in high altitude and sloping areas. Furthermore, mountainous areas are not conducive to activities, such as cultivation and construction land development. Although the HQ index of forestland remained high during the 1990-2018 period, the total forestland area decreased, which is attributed to the considerable degradation of forestland through its conversion to cultivated and construction land. However, topographic conditions of forestland limit human activities and the high vegetation cover makes it more resistant to disturbances. HQ has been maintained at a high level due to the implementation of various ecological conservation activities and restoration practices.

Factors influencing spatial variations in HQ.
Single factor analysis. GeoDetector analysis results (Table 8) revealed differences in the interpretation abilities (q-values) of the various factors influencing variations in HQ. The order of the q-values was land-use type (X9) > NDVI (X6) > slope (X2) > GDP > NPP (X5) > temperature (X3) > rainfall (X4) > elevation (X1) > population density (X7). Therefore, land-use type was identified as a key factor influencing variations in HQ (q-value of 0.264), followed by NDVI (q-value of 0.106). Slope, temperature, rainfall, population density, elevation, and GDP had weak interpretation ability for spatial variations in HQ; however, the factors should be taken into consideration. Overall, natural factors (land-use type, NDVI, and NPP) significantly influenced the spatial distribution patterns of HQ and their q-values were greater than those of topographic (slope, elevation), socioeconomic (GDP, population density), and meteorological (precipitation, temperature) factors.
Analysis of interactions between two factors. The interaction detector of the GeoDetector was used to test the explanatory power of two-factor interactions on HQ and the results (Table 9) indicated that the q-values of two-factor interactions were greater than the q-value of a single factor. The interactions exhibited a two-factor enhancement or non-linear enhancement. The most significant interactive effect on the spatial variation in HQ was that of land-use type ∩ NDVI (0.309), followed by that of land-use type ∩ slope (0.300). The most dominant factor interaction was observed when a dominant single factor (land-use type) was combined with another factor, suggesting that different land-use types determine the distribution patterns of ecosystem types. The interaction of NDVI with precipitation, temperature, and slope exhibited considerable interpretation ability, indicating that natural factors, such as temperature, precipitation, and slope affect changes in land-use type and in turn, HQ to a certain extent. Although the q-values of single factors, GDP and population density were low, their interaction with other factors exceeded the interpretation ability of single factors influencing the spatial differentiation of HQ, which highlights the need to take such factors into consideration. Urbanization and population growth  Table 9. The results of interactive detection. Enhance, bi-: means that the interaction between the two factors is a two-factor enhancement, Enhance nonlinear: means a non-linear enhancement. www.nature.com/scientificreports/ will lead to rapid expansion of construction land and consequently increase ecological pressure on the surrounding habitats. Therefore, the spatial patterns of towns and cities should be optimized to improve the local HQ through the implementation of ecological restoration projects.

Discussion
The present study assessed the spatio-temporal changes in HQ of Guizhou Province from 1990 to 2018 and factors influencing HQ using the HQ module of the InVEST model and GeoDetector. The research has significant implications for biodiversity conservation and the construction of ecological civilization in karst regions.
The essence of land-use transformation is to change the form of land-use during socio-economic development. The primary cause of land-use transformation is the change in land-use types leaded by human activities on natural ecosystems 52 . The results of the present study have shown that land-use has changed considerably over the last few decades, with cultivated land, grassland, and forestland being the predominant land-use types in Guizhou Province. The results of the present study revealed that cultivated land and forestland areas decreased during the 1990-2018 period, while construction land area increased with increasing economic development and urban expansion, which explains the decline in HQ in Guizhou Province. Moreover, the development of rural infrastructure has accelerated due to the revitalization of the countryside and tourism, in turn, promoting sustainable development of the regions in Guizhou Province. These projects have provided favorable conditions for addressing poverty in rural regions, although they have also had considerable negative impacts on the ecosystem, leading to a reduction in HQ. The average HQ index of Guizhou Province during the 1990-2018 period decreased slightly and was characterized by an initial decrease, followed by an increase, and then a decrease. However, the implementation of a range of restoration and protection measures, such as returning cultivated land to forestland and grassland and rocky desertification treatment, have substantially enhanced the ecological environment in Guizhou Province. The increase in population, economic development, and urbanization during the 1990-2018 period has led to an increase in construction land area by nearly four times. The expansion of construction land has created new sources of threats, disturbed and destroyed the surrounding habitats, leading to a decline in HQ.
The dominant factor influencing HQ during the study period was land-use type, as determined by single factor detection. Land-use type is a key factor determining regional HQ, and land-use types determine the distribution patterns of various ecosystems, especially the occupation of ecological spaces by construction land, which accelerates landscape fragmentation and leads to a decline in HQ. Followed by NDVI index, which exhibits a strong influence on HQ, and is a key parameter that reflects vegetation growth status and cover. NDVI is closely associated with HQ 53 . According to the interaction detection results, n atural and socio-economic factors had a substantial effect on the spatio-temporal variations in HQ of the study area. The interaction between factors enhanced the effect on HQ, indicating that two-factor interaction effect was greater than that of a single factor. The interactive effect between land-use and other factors was considerably stronger than that between other factors, indicating that land-use is a major factor influencing HQ changes.
The InVEST-HQ model was used in this study to provide a feasible method for the quantification of HQ and visualization of the calculated results. However, our study had some limitations. First, the results of regionalscale analysis can only be used as a reference for the relationship between LULC changes and variations in HQ in Guizhou Province. However, specific conservation measures need to be studied on a more comprehensive scale. Second, the InVEST-HQ model requires several parameters, and the relevant parameters used in this study were obtained from the model manual, previous studies, and expert experience that are associated with a certain degree of subjectivity in the parameter settings, which could lead to uncertainty in the evaluated results. Final, inconsistencies in data accuracy, which result in certain deviations in factor extraction when using fishnet sampling, may have led to deviations in the driving force results obtained using the GeoDetector model.

Conclusions
The present study investigated the spatio-temporal variations in HQ and factors influencing HQ changes in Guizhou Province, using the InVEST-HQ model and GeoDetector. The main conclusions are as follows: (1) The main land-use types in Guizhou Province during the 1990-2018 period were forestland, cultivated land, and grassland. Cultivated land and forestland decreased by − 98.27% and − 98.37%, respectively, while grassland, construction land, and water increased by 100.41%, 398.71%, and 264.21%, respectively; unused land was relatively stable. (2) The average HQ index of Guizhou Province during the 1990-2018 period initially decreased, then increased, and eventually decreased. The spatial distribution of hot and cold spots associated with HQ showed a pattern of "high in the south and east, low in the central and west, and high and low in the north". (3) Single and interactive factor detection using the GeoDetector showed that the major factor influencing HQ changes was LULC, followed by natural factors, such as NDVI and slope. Socio-economic factors such as population density and GDP were minor factors influencing HQ changes (Supplementary Information).

Data availability
The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.