Interactive comment on “ Characterization and interaction of driving factors in karst rocky desertification : a case study from Changshun , China ” by

The geographical information system techniques and geographical-detector model can effectively explore the relationship between driving factors and the evolution of karst rocky desertification (KRD) at spatial dimension. The paper found some interesting results based on the quantified indicator (PD value) from the geographical-detector model). It concluded that there was no significant difference between the impacts of natural and anthropogenic factors. As we know, human influence is an important factor on the KRD evolution. However, from the finding in the paper, the impact of human influence cannot be over-emphasized and specific karst environment would have a great impact. Also the enhanced interaction of factors should be taken into consideration in the planning of combating KRD. The findings of MS can help effectively


Introduction
China has approximately 3.44 × 10 6 km 2 of karst areas, about 36% of its total land, and 15.6 % of all the 22 × 10 6 km 2 karst areas in the world (Jiang et al., 2014).The continuously distributed karst region, which is mostly located in eight provinces of southwestern China (Guizhou, Yunnan, Guangxi, Chongqing, Sichuan, Hunan, Hubei, and Guangdong) is one of the world's most extensive and welldeveloped karst landscapes (Wang et al., 2004b;Xu et al., 2013).Karst rocky desertification (KRD) has been identified as the most severe ecological issue threatening and constraining southwestern China (Bai et al., 2013;Wang et al., 2004b).The KRD is a process of land degradation involving serious soil erosion, extensive exposure of basement rocks, drastic decrease in soil productivity, and the appearance of a desertlike landscape (Wang et al., 2004b).Therefore, the National Reform and Development Commission commenced implementation of a pilot KRD restoration project for 100 counties in the region in 2008 and expanded the project to 200 counties in 2011.
The restoration and reconstruction projects have been effective (Qi et al., 2013;Zeng et al., 2007), but KRD improvement can easily reverse in local areas (Sheng et al., 2013).The mutual transformation (improvement or deterioration) of different types of KRD land has been remarkable (Bai et al., 2013), perhaps because KRD occurs as a result of multiple factors including the inherent natural background and intensive human activities in the complex karst environment (Jiang et al., 2014;Liu et al., 2008b;Wang et al., 2004b;Yang et al., 2011).The lack of insight into the joint impacts of driving factors on KRD evolution can affect restoration efforts.To control, manage and restore KRD, analyses of how relative driving factors influence the evolution of KRD and determinations of their corresponding contributions are essential.
Previous studies have undertaken analyses of different driving factors on KRD based on measurable physical data and socioeconomic census data.The evolution of KRD is associated with a considerable number of natural factors, e.g., meteorological factors including temperature and precipitation (Peng and Wang, 2012;Xiong et al., 2009), topographic factors consisting of the elevation, slope (Huang and Cai, 2007;Jiang et al., 2009) and aspect (Zhou et al., 2007), and lithological factors (i.e., types of carbonate rock assemblages) (Li et al., 2009b;Wang et al., 2004a).Human activities, such as farming on hilly land, overgrazing, and felling of forest and shrubs for cooking (Jiang et al., 2014;Li et al., 2009bLi et al., , 2008a;;Wu et al., 2011;Yan and Cai, 2013), have also recently become pivotal factors in KRD.The natural and anthropogenic factors above jointly influence the evolution of KRD in southwestern China.
Few previous studies have discussed the relative importance of driving factors contributing to the evolution of KRD (Huang and Cai, 2007;Jiang et al., 2009Jiang et al., , 2014;;Li et al.,  2009b; Peng and Wang, 2012;Xiong et al., 2009;Yan and Cai, 2013).Impacts of driving factors have been analyzed separately, and the leading driving factors have been qualitatively or semi-quantitatively evaluated.A limited number of studies have used regression modeling (Liu et al., 2008a), factor analysis (Li et al., 2009a), or redundancy analysis (Yang et al., 2011) to study such relationships.These studies were constrained at the scale of a county or regions (e.g., the third or fourth administrative unit in China in terms of scale), and were missing any spatially consistent information on KRD evolution as a result of driving factors.Additionally, consideration of the interaction of different factors is lacking in previous work.All of these factors are necessary for a deeper understanding of the problem.Fortunately, a geographical detector model (Li et al., 2013;Wang et al., 2010b), that can calculate the relative importance of various factors, provides an opportunity to consistently quantify more spatial information on the driving factors and their interactions with regard to the evolution of KRD.
The objective of this study is to investigate the relationships between the spatial evolution of KRD and its driving factors at a fine scale.Based on a geographical detector model, we have quantified the relative importance of different natural and anthropogenic factors.Furthermore, we have identified the leading factors and analyzed their interactive impacts on both the improvement and deterioration of KRD.

Study area
Changshun County is located in central Guizhou Province in China, roughly between 106 • 13 06 and 106 • 38 48 E, and 25 • 38 48 and 26 • 17 30 N (Fig. 1).The county covers an area of 1552 km 2 within a karst peak-cluster depression landscape that contains a combination of karst cones and depressions between cones.It lies within the subtropical monsoon climate zone with a mean annual temperature of 14 • to 14.5 • C and a mean annual precipitation of 1160 to 1355 mm (Xu et al., 2013).Its elevation is between 661 and 1572 m.Changshun is one of the areas most affected by KRD in Guizhou, with nearly 30 % of its area impacted (Wang, 2010).ETM+ image was preprocessed by applying geometric and atmospheric corrections.A digital elevation model (DEM) with 30 ground control points, which were taken from 1 : 50 000 topographic maps (provided by the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences), was incorporated in the geometric corrections.This resulted in a root-mean-square spatial positioning error of less than 0.5 pixels for each image.The FLAASH module from the Environment for Visualizing Images (ENVI) software (Solutions, 2009) was used for atmospheric corrections.

Categorizing and quantifying karst rocky desertification
Based on mapping of bedrock exposures, vegetation and soil coverage (Li et al., 2009b), karst areas can be classified as no KRD, potential KRD, slight KRD, moderate KRD, severe KRD and extremely severe KRD (Table 2).The distribution of KRD in Changshun County in 2000 and 2010 was mapped by visual interpretation.In 2010, 86 field measurements were made to assess the accuracy of the method in Changshun (Fig. 1).The accuracy was 90.7 % (Xu et al., 2013).
Then evolution index of KRD for each raster can be calculated as follows: where t 0 and t denote the period, i.e., 2000 and 2010, respectively in this study; i and j range from 1 to 6, representing the six levels of KRD intensity as above; KRD indicates that the KRD land improves as the bedrock rate of an area decreases; a negative E-KRD (deterioration index) indicates that the KRD land deteriorates.A larger absolute value of E-KRD signifies a more significant intensity of KRD transformation.

Driving factors data sets
We chose nine relevant driving factors to study their relationship with the evolution (improvement or deterioration) of KRD based on previous studies (Jiang et al., 2009;Li et al., 2009b;Wang et al., 2004a;Yang et al., 2011).Natural factors include soil type classified by the genetic soil classification of China (Shi et al., 2004), lithology, vegetation type, elevation and slope; human factors include road influences (buffer of roads), settlement influences (buffer of settlements), gross domestic product (GDP) density, and population density.It is difficult to quantify human activities as a detailed spatial unit.Therefore, we used geographic information system (GIS) techniques to quantify such land use information, including farming on hilly lands, overgrazing, felling of forest or shrubs for cooking, and KRD restoration projects.The buffer methods calculate the Euclidean distance to roads and settlements (i.e., road and settlement influence) as proxies of the distribution of human activities (Simpson and Christensen, 1997).A shorter distance denotes a greater human influence.The presence of roads corresponds to the utilization of natural resources (Yang et al., 2013), and the distribution of settlements is related to the range of residents' daily lives and the implementation of restoration projects (Yang et al., 2011).The detailed data sources and processing of the nine sets of driving data are listed in Table 3.
In the geographical detector model, continuous data should be transformed into discrete intervals (Cao et al., 2013a;Wang et al., 2010b).The discretization method depends on optimal classifications or prior knowledge (Li et al., 2008;Wang et al., 2010a).With the exception of the soil, lithology and vegetation data, the driving data can be considered to be continuous variables (Table 3).Based on the distribution and prior knowledge of data, we majorly used the method of natural breaks coupled with the professional knowledge and rounded the intervals as the integer format (Fig. 2).As continuous variables have local characteristics, natural breaks can seek to minimize each interval's average deviation from the interval mean while maximizing each interval's deviation from the means of the other groups.In other words, the method seeks to reduce the variance within intervals and maximize the variance between intervals.The discretization standards for the corresponding data are also shown in Fig. 2.

Geographical detector model
The geographical detector model is a new spatial analysis method used to assess the relationship between driving factors and relevant resultant outcomes (Hu et al., 2011;Li et al., 2013;Wang et al., 2010b).The model compares the spatial consistency of a resultant outcome distribution (e.g., the KRD in our study) to strata of driving factors based on a spatial variance analysis.It can extract information without any assumptions or restrictions with respect to the vari-ables.Also, it can be used for both quantitative and nominal data.The model can resolve the following four questions.
(1) What is the geographical domain of the outcome?( 2) What are the driving factors responsible for the outcome?
(3) What is the relative importance of each driving factor?(4) Do driving factors operate independently or are they interconnected?(Wang et al., 2010b).
The power of determinant (PD) value and the interactive PD value are two novel indicators used to assess the relationship between outcomes and their driving factors in the geographical detector model (Li et al., 2013;Wang et al., 2010b).The geographical detector model overlays the distribution of K (e.g., E-KRD in our study) over several strata of driving factors of D (i.e., one of driving data).D i , where i = 1, 2,. ..n, and n is the number of categorical types of D, are the discrete attributes associated with a stratum of driving factors of D, which is denoted as D = {D i } (Li et al., 2013;Wang et al., 2010b), then the study regions were divided to sub-regions (D 1 , D 2 , . ..D n ).The mean value and the dispersion variance of K (denoted as σ 2 K D,i ) can be calculated by the model.Next, PD is calculated as follows: where N denotes the number of samples over the entire region of interest; N D,i is the number of samples in the subregion of i category of D; σ 2 K denotes the variance of K over the entire region; the dispersion variance of K over the subregions of the attributes D i is denoted as σ 2 K D,i .When the σ 2 K D,i of each subregion is small, the variances between subregions is large and the PD is large (which means that such a division explains most or even all of the spatial K variation).Therefore, a higher PD indicates that the driving factor (D) has a larger impact on K.If the driving factor (D) completely controls the total K, then PD = 1.
The package of geographical detector model used in our study is a version of Excel-GeoDetector (http://www.sssampling.org/Excel-GeoDetector/).All of the driving data and the E-KRD index were resampled in the WGS84 projection constrained by the same boundary of Changshun County in GIS.A pixel size of 100 × 100 m was used to extract rel-evant data.Finally, all the data were extracted as input data for the Excel-GeoDetector.

Importance of driving factors influencing KRD improvement
Using the geographical detector model, PDs of the nine aforementioned driving factors influencing KRD improvement index were calculated.Their order is as follows (the number in parentheses is the corresponding PD value): lithology (0.The lithology, road influence, soil and population density (ranking first to fourth, respectively) are considered to be the major driving factors in Changshun.In contrast, the slope and GDP density have low relative importance when it comes to explaining the spatial distribution of KRD improvement.Considering the order of PDs, both natural and anthropogenic factors influence the improvement of KRD.This implies that natural factors (with ranks of first, third, fifth, seventh, and eighth) and anthropogenic factors (with ranks of second, fourth, sixth, and ninth) have similar overall impact.This information confirms that the effect of restoration project implementations would improve if the natural impact factors on the evolution of KRD were considered.

Variations in KRD improvement corresponding to different driving factor levels
The geographical detector model calculated mean values for the KRD improvement index for various levels of each driving factor, and indicated significant differences for all levels.The model helps to analyze how a particular driving factor influences KRD improvement.For example, the PD of lithology shows that it has the greatest impact.According to the model, the mean values of improvement indices for three lithological types show significant differences, and their ranks are as follows (with the number in brackets as the corresponding index value): clastic rock mixed with limestone/dolomite (32.6) > limestone interbedded with clastic rock (27.7) > homogenous limestone/dolomite (22.3).These three lithological types show an increasing proportion of limestone or dolomite, which indicates that the proportion of limestone or dolomite is positively associated with improved intensities of KRD.
Our results also indicate that the distance to a road, as a proxy of human activity levels, influences the spatial distribution of KRD improvement.The order of mean index values for six levels of road influence (Table 4) suggests that the distance to a road is positively correlated to improved intensities of KRD.This means that a greater road influence (i.e., better access to roads) enhances the successful implementation of KRD restoration projects and accelerates the improvement of KRD. ."Y" denotes that the difference of influence between the two factor levels is significant with the confidence of 95 %; and "N" denotes that it is not.The order of mean values for six road influence levels of improvement index is as follows: level 2 (27.1) > level 1(26.3)> level3 (24.0) ≈ level4 (22.6) > level 5 (20.6) ≈ level6 (20.1)."≈" denotes that there is no significant difference between two levels (the same meaning is the following content).

Interactions of driving factors influencing KRD improvement
The interactive impact of two factors (A and B) on KRD improvement generally differs from the simple linear summation of the separate impacts of A and B, i.e., PD (A ∩ B) = PD (A) + PD (B).PDs of their interactive impacts were calculated by the geographical detector model.The top 10 PDs of interactive impacts, which range from 0.205 to 0.283, are greater than the highest PD of a single factor (i.e., PD (lithology) = 0.150) (Table 5).This shows that the combined impact of factors on KRD improvement would be greater than single factors, i.e., PD (A ∩ B) > PD (A) or PD (B).Moreover, the road influence associated with vegetation and the elevation associated with population density reflect a non-linearly enhanced impact on the KRD improvement, i.e., PD (A ∩ B) > PD (A) + PD (B).

Importance of driving factors influencing KRD deterioration
The PDs of the nine driving factors influencing the KRD deterioration index were calculated.
"Y" denotes that the difference of influence between the two factor levels is significant with the confidence of 95 %; and "N" denotes that it is not.
respectively) are still detected as the major driving factors of KRD deterioration in Changshun.The GDP and population density have low impacts on the spatial distribution of KRD deterioration.It seems that natural factors (with ranks of first, second, fifth, sixth, and seventh) have a relatively greater impact on KRD deterioration than anthropogenic factors (ranked third, fourth, eighth, and ninth).

Variations of KRD deterioration at different driving factor levels
The PD of soil, which has the highest ranking, shows that it is the most important of the driving factors influencing KRD deterioration.Table 6 shows significant differences in the mean deterioration KRD index between the eight soil types.The order of mean values of deterioration index for the eight soil types is as follows: acidic lithosol (−57.0)≈ purplish soil (−53.1)> terra rossa (−36.6)≈ yellow soil (−35.3)> paddy soil (−33.3)> calcareous soil (−28.7)> skeletal soil (−23.8)> rendzina (−15) (where "≈" denotes that there is no significant difference between deterioration index of the two variables).This confirms that soil types are associated with their different hydrological proper ties and susceptibil-ity to erosion, which influence KRD transformation.Results suggest that the acidic lithosol and purplish soil significantly worsened KRD deterioration.The influence of roads was again identified as being the most important of the anthropogenic factors influencing KRD deterioration.The order of mean values of the deterioration index for six road influence levels is as follows: level 2 (−39.8)> level 6 (−36.3)> level 1 (−34.1)> level 4 (−32.3)> level 5 (−28.6)> level 3 (−27.0).This shows that road influence is not monotonously correlated to the KRD deterioration intensities.Areas of levels 1 and 2, which are the nearest to roads, have severe KRD deterioration intensities.This indicates that human activities would enhance KRD deterioration.With distances to roads increasing, the areal deterioration intensity of levels 3, 4 and 5 decreases.However, the areal deterioration intensity at level 6 actually increases.An area with few human activities easily deteriorates, which may be attributed to other driving factors.

Interactions of driving factors influencing KRD deterioration
The top 10 PDs corresponding to interactive impacts on KRD deterioration are listed in  0.477.Except for the interactive impact of the lithology associated with elevation, the other interactive impacts of driving factors on KRD deterioration show non-linear enhancement (Table 7).For example, the PD of the soil associated with the lithology is 0.477, far greater than their linear summation result (i.e., 0.345).Furthermore, the deterioration index of homogenous limestone/dolomite is minimized (i.e., −35.3) along with the deterioration of the acidic lithosol (i.e., −57.0).If the area corresponds to a region of homogenous limestone and acidic lithosol, then the interactive impact of high-permeability carbonate rocks, in terms of lithology and a low soil formation rate in soil types, would exacerbate the KRD deterioration.

Discussion
Using GIS techniques and the geographical detector model, our novel investigation has explored the available information at a much finer scale than previous studies (Huang and Cai, 2007;Jiang et al., 2009Jiang et al., , 2014;;Li et al., 2009a, b;Liu et al., 2008a;Xiong et al., 2009;Yan and Cai, 2013;Yang et al., 2013).Such studies omitted information about the spatial consistency of KRD evolution with respect to driving factors and also lacked consideration of interactions between various factors.We calculated the PDs of driving factors, which relate to the spatial consistency of driving factors compared with the E-KRD index.The order of the PDs is indicative of their relative importance to the evolution of KRD.Interactive PDs also enable the investigation of the combined impact of driving factors.The influences of lithology, soil and roads are identified as the leading factors for KRD transformation in Changshun.Lithological types associated with the permeability of carbonate rocks and soil types associated with soil erosion and formation rates constitute the basic geographical environment that significantly affects the evolution of KRD (Jiang et al., 2014;Peng et al., 2013).Such influences suggest that in Changshun the design of restoration projects should consider the impacts of lithology and soil, and even more so their interactions.Our results confirm the important impact that human factors have on KRD.In particular, roads, as a proxy for human activity, have a significant influence as noted by their association with KRD improvement or deterioration.However, KRD deterioration in areas with a low road influence may be attributed to other driving factors.On one hand, better access to roads does enhance successful implementation of KRD restoration projects (Deng et al., 2011;Xu et al., 2013;Yang et al., 2011).On the other hand, however, when roads penetrate into (or are improved in) areas, they may intensify efforts to exploit resources (Deng et al., 2011) and therefore cause KRD deterioration (Mick, 2010;Yang et al., 2013).Road construction can create numerous roadcuts and embankments that can contribute to increased sediment production and further cause land degradation (Lee et al., 2013).Serious soil erosion can occur on bare road embankments (especially those under construction) (Cerdà, 2007) and on unpaved roads (Cao et al., 2013b).Restoration efforts should be directed to guarantee key ecological processes and support soil formation for roadcuts and embankments (Jimenez et al., 2013;Lee et al., 2013).New strategies should emphasize the need to decrease the impact of road construction on plant and animal habitats and the benefits of using local species for the re-vegetation of land surrounding the roads (Cheng et al., 2013).
Comparing the relative importance of natural and anthropogenic impact factors reveals some interesting findings.Previous studies have considered anthropogenic activities as being more significant in KRD than natural factors (Lan and Xiong, 2001;Li et al., 2009a;Yan and Cai, 2013;Yang et al., 2011).However, other studies have claimed that natural factors were the major factor in KRD evolution (Gu et al., 2011;Hu et al., 2004;Shan, 2006).In our case study, there is no significant difference observed between the impacts of natural and anthropogenic factors influencing the KRD improvement based on the order of their PDs (Table 8).Even natural factors have a higher impact on KRD deterioration.In fact it could  (Yang et al., 2009), and may cautiously support the importance of anthropogenic driving factors in KRD.However, there is little unassailable evidence to support whether or not natural or anthropogenic factors have the greater impact.In the field survey, a significant contrast in adjacent regions with carbonate rock and non-carbonate rocks (granite) can be found, but under nearly the same conditions of anthropogenic factors.There is an obvious KRD in areas with the carbonate rock outcropping.In contrast, areas with noncarbonate are covered with lush vegetation.Furthermore, a historical study showed that the spatial distribution of suspect KRD in Guizhou in 1940 was similar to the situation in 2005 (Han et al., 2011).Within this stage of rapid population growth, the spatial distribution stays relatively stable.
Above evidence indicated that we cannot overemphasize the impact of anthropogenic factors.Karst is an integrated and unique geo-ecosystem, and its evolution is caused by natural and human factors (Febles-González et al., 2012).KRD occurs under a specific karst background, and anthropogenic factors can aggravate or reverse the KRD in a relatively short period, but may not be the essential driving factor (Zhang and Zhou, 2001).Without the specific environments, informal human activities cannot cause KRD (Shan, 2006).
Although slope angles have a significant influence on KRD (Huang and Cai, 2007;Jiang et al., 2009), this impact factor was ranked quite low for both KRD improvement and deterioration (the PDs rank eighth and seventh respectively) in our study.Soil on steepened slopes is often susceptible to erosion, which is one of the reasons leading to KRD (Ying et al., 2014).Therefore, areas of KRD are mainly located in sloping regions in Changshun.Also, the mean KRD improvement index with small slope angles is greater than that with large slope angles.However, local areas with steepened slopes can also be highly vegetated in our field survey, which may be attributed to the fact that high slope angles limit the range of possible human activities and thereby act to reduce anthropogenic disturbances (Xu et al., 2013).Areas of low-slope angles (especially < 5 • ) may be suitable for agricultural use which then can lead to KRD (Huang and Cai, 2007).Therefore, slope angle might not play a significant role in soil erosion of karst regions (Peng et al., 2013).The complex effects of slope cause an inconsistent spatial distribution of the KRD transformation versus slope angles.
Until now, few studies have investigated the interactive effects of driving factors on KRD transformations.Our results show that those interactive factors lead to enhanced impacts.Major interactive results of the KRD improvement conclude that "PD (A ∩ B) < PD (A) + PD (B)" (Table 5), whereas the major results of the KRD deterioration conclude that "PD (A ∩ B) > PD (A) + PD (B)" (Table 7).The enhanced effect on KRD deterioration is more significant than for KRD improvement.Compared with the impact of a single factor, the nonlinearly enhanced effects of interactive factors on KRD transformations should be explored in the further studies to effectively characterize KRD evolution.For example, the interactive impact of lithology associated with soil on KRD deterioration and road influence associated with vegetation on KRD improvement should be considered in the design of restoration projects.
There are some uncertainties in our study.Human activities, such as farming on hilly lands, overgrazing, and felling of forest and shrubs for cooking (Jiang et al., 2014;Li et al., 2009b;Liu et al., 2008a;Wu et al., 2011;Yan and Cai, 2013), are rather difficult to measure in a straightforward manner and would not be depicted at a detailed spatial distribution.Instead, they were represented by proxies of distances to roads and settlements in our study, which have proven to be acceptable in recent studies.Moreover, summing PDs of driving data (0.779 for improvement and 0.957 for KRD deterioration), it indicates that our selected driving data have a relatively larger impact on KRD deterioration than improvement.Indeed, we could not include all the driving data in the study.For example, meteorological factors are relatively similar at the county scale and were neglected in this study.However, the impact of frequent droughts and extreme floods induced by global climate change on the KRD transformation should be investigated further (Gu et al., 2011;Huang et al., 2009;Jiang et al., 2014).It would be interesting to explore the significance of meteorological factors and their interactions with other factors in a macro scale study.

Conclusions
A comprehensive exploration of driving factors influencing KRD evolution can provide the information needed to effectively combat further deterioration in a fragile karst environment.We use GIS techniques to quantify the information on human activity farming (hilly lands, overgrazing, felling and restoration projects) by proxies of distances to roads and settlements for it is difficult to measure it directly.We used a geographical detector model to incorporate fine-scale spatial information in our investigation of the relationships between driving factors and KRD transformation in Changshun.The orders of PDs for natural and anthropogenic factors show the variable impact that they have on the spatial distribution of KRD improvement or deterioration.Lithology, soil and road influence are identified as the major driving factors associated with KRD evolution.Our results imply that there is no significant difference between natural and anthropogenic factors when it comes to influencing KRD improvement, but natural factors have a higher impact on KRD deterioration.The specifics of the karst background create a fragile and vulnerable environment that is susceptible to human activities, and then anthropogenic factors significantly influence KRD transformation further.There is little unassailable evidence to support whether or not the natural or anthropogenic factors have a greater impact.However, we did detect the impact of interactive factors and found that non-linearly enhanced impacts significantly aggravated the KRD.Our results have explored new information concerning the relevant influence on KRD evolution, which can help to effectively restore KRD.The limitations and improvements mentioned above should be explored in future studies.

Figure 1 .
Figure 1.The location of Changshun County.

Figure 3
Figure 3 shows improvement and deterioration indices from 2000 to 2010 for Changshun County.Areas of KRD transformation are widely detected.The transformed area is 410 km 2 , including 301 km 2 where the class improved and 109 km 2 where it deteriorated.The improvement index ranges from 15 to 70 and the deterioration index ranges from −15 to −85.Areas of KRD improvement and deterioration and their intensities show different spatial distribution characteristics.In general, significantly transformed areas (i.e., large absolute values of E-KRDs) are relatively small and concentrated in local areas.Areas of significant deterioration are mainly located in the townships of Weiyuan and Kaizuo.In comparison, KRD lands in western Changshun, including the townships of Baisuo and Yingpan, significantly improved from 2000 to 2010.Meanwhile, areas with small absolute values of E-KRDs, which indicate a relatively low intensity of KRD transformation, are widely distributed through the whole study area.

Figure 3 .
Figure 3.The spatial evolution index of karst rocky desertification from 2000 to 2010 (a) improvement index; (b) deterioration index.

Table 2 .
Classification and quantification standard for karst rocky desertification.

Table 3 .
Driving data sources and processing.

Table 4 .
Difference of mean improvement KRD index between levels of road influence.

Zhang: Characterization and interaction of driving factorsTable 5 .
Interaction between pairs of driving factors influencing the KRD improvement.

Table 6 .
Difference of mean deterioration KRD index between soil types.

Table 7 .
Interaction between pairs of driving factors influencing the KRD deterioration.↑" denotes A and B enhance each other; "↑↑" denotes a nonlinear enhancement of A and B. "

Table 8 .
PDs of driving data and their order for both the KRD improvement and deterioration.