Research on Vegetation Dynamics and Driving Mechanisms in Karst Desertified Areas Integrating Remote Sensing and Multi-Source Data

Abstract

Rocky desertification severely restricts socio-economic development in the karst regions. However, assessments linking karst rocky desertification and NPP changes over the long term and at high resolution are limited. This study aims to reveal the spatiotemporal patterns and driving mechanisms of NPP changes in Wenshan Prefecture, addressing the scientific gap in quantitative process research and mechanism identification in karst desertification areas. We estimated vegetation NPP from 2000 to 2020 using remote sensing data and the CASA model. The Theil–Sen trend analysis and Mann–Kendall test were applied to assess temporal variation, while a Geographical Detector identified the dominant natural and human factors and their interactions shaping NPP spatial patterns. Our results showed that NPP increased overall by 4.07 gC m⁻² a⁻¹, alongside a general decline in rocky desertification. The most significant improvement occurred between 2010 and 2015, when rocky desertification shrank by 2224 km² and the dynamic rate reached 1.42%. Mean NPP reached 1057 gC m⁻² a⁻¹, with a “northwest high–southeast low” spatial pattern, and 77% of the region showed significant increases. Rocky desertification was most severe at elevations between 1000 and 2000 m. In the karst region, NPP is mainly controlled by natural factors, with soil depth and slope being the strongest influences. Human activity had the largest negative impact, and most factors interacted synergistically, where hydrothermal gradients and human disturbances more strongly suppressed NPP on steep, thin slopes than individually expected. These findings provide robust scientific evidence and practical decision-making support for ecological restoration, rocky desertification control and long-term sustainable development in Wenshan and other karst regions, highlighting the importance of continuous monitoring and adaptive management strategies to consolidate restoration achievements and guide future land-use planning and regional ecological policy.

1. Introduction

The world’s largest contiguous karst region, with an area of approximately 3.44 × 106 km², is located in southwest China, where well-developed tower and cone karst serve as the archetype of subtropical karst landforms [1,2]. However, average soil thickness of <30 cm and concentrated rainfall render the landscape extremely fragile, so that slight perturbations rapidly escalate into widespread rocky desertification [1]. Intensification of rocky desertification exerts substantial ecological and socio-economic impacts: it accelerates soil erosion and nutrient depletion in thin karst soils, weakens water storage capacity, and, due to high infiltration and limited retention, elevates surface runoff and disaster risk. The resulting declines in vegetation cover and degradation of karst aquifers lead to reduced ecosystem productivity and carbon sequestration, impacting agricultural yields and threatening rural livelihoods in mountainous areas [3,4]. To curb the spread of rocky desertification and associated ecological degradation, China has implemented a series of integrated management measures with notable success. Among these, vegetation-based restoration measures contribute most significantly to enhancing ecosystem functions. During rocky desertification management, vegetation restoration boosts regional carbon storage capacity, exhibiting dynamic growth patterns influenced by both management duration and the severity of rocky desertification [5]. Therefore, understanding the spatiotemporal dynamics of vegetation and its response to environmental changes is crucial for the protection and restoration of ecosystems.

Vegetation net primary productivity (NPP)—the living biomass increment—offers a quantitative proxy for both carbon cycling and restoration potential, making it a primarily employed indicator in assessing karst vegetation recovery [6,7,8]. Moreover, NPP dynamics are closely tied to rocky desertification processes in rocky desertification regions, and understanding their spatiotemporal relationship is essential for elucidating ecological changes and identifying the driving forces [9]. Since the launch of the International Biological Program in the 1960s, global research on NPP has advanced steadily, driving the development of climate-based models and quantitative studies of the NPP–biomass relationship [10]. Among these, the CASA model, grounded in the principle of photosynthetic energy use efficiency, integrates remote sensing, geospatial analysis, and ecological process simulation, and is currently one of the most mature parameterized models for dynamic NPP assessment [11]. CASA model has been widely applied in China and East Asia. Park Se-ryong et al. first used it to estimate terrestrial NPP in China from 1982 to 1998, systematically analyzing national spatiotemporal patterns [12]. Xue P et al. applied CASA model to six coastal provinces from 1982 to 2015, finding overall increases in NPP, though declines occurred in some cities due to agricultural expansion and urban construction [13]. Yu D Yet al. further modeled NPP in East Asia using land cover data, NDVI datasets, and monthly meteorological data, showing close agreement with observations [14].

Despite its wide application, the CASA model remains underutilized in typical karst rocky desertification regions. Existing studies largely focus on the effects of climate and land-use change on vegetation productivity, but robust long-term analyses and characterizations of regional heterogeneity specific to karst rocky desertification areas are lacking [15,16]. This gap is scientifically important because karst ecosystems exhibit pronounced spatial heterogeneity, and the response mechanisms of vegetation productivity in rocky desertification zones may differ substantially from those in non-karst regions [17]. Moreover, numerous studies have assessed NPP trends in karst regions, two knowledge gaps persist. First, robust, long-term assessments that align the progression of karst rocky desertification with NPP changes at high spatial detail remain insufficient. Second, driver factors that jointly consider natural gradients and human pressures are often incomplete.

Wenshan Prefecture, located in the extensive karst belt of Southwest China, exemplifies a subtropical karst landscape, where fragile soils, concentrated rainfall, and human disturbance combine to drive pronounced rocky desertification. This study selects Wenshan as a representative karst area because it features extensive Triassic carbonate outcrops and a full suite of classic karst landforms, a terrain with strong relief and high fragmentation shaped by prolonged dissolution and erosion, and marked elevation and bioclimatic gradients from temperate to subtropical zones [18,19,20]. These conditions make it well suited to examining the linkages between rocky desertification and vegetation productivity. This not only helps refine the theoretical framework of global NPP modeling under special ecological settings but also provides scientific support for understanding the mechanisms of rocky desertification and for evaluating ecological restoration.

To address these gaps, this study selects Wenshan Prefecture, Yunnan, as a representative karst landscape. Leveraging Landsat and NPP datasets from 2000 to 2020, we set out to: (1) characterize the spatiotemporal dynamics of rocky desertification and NPP; (2) elucidate how NPP evolves along altitude gradients under varying desertification levels; and (3) disentangle the dominant natural and anthropogenic drivers of NPP. The findings are expected to inform precision ecological restoration and sustainable land management in karst areas.

2. Materials and Methods

2.1. Research Area Overview

Wenshan Prefecture is located in the karst belt of the Yunnan Plateau and is characterized mainly by medium-altitude plateaus as well as mountainous and semi-mountainous terrain, with elevations ranging from approximately 107 to 2991 m and an overall slope from northwest to southeast. Vegetation is dominated by evergreen broadleaf forests in the humid mid- to high-mountain belts, forming a key component of regional water conservation and ecological barriers [21,22]. Rocky desertification in Wenshan Prefecture is mainly caused by the inherently fragile natural conditions of karst areas—characterized by abundant exposed rock, little soil, and steep slopes with intense rainfall—combined with long-term, intense human disturbances, which have exacerbated soil erosion and vegetation degradation. Since the late 1990s, a series of ecological projects—such as Grain-for-Green and comprehensive rocky-desertification control—have been implemented in Wenshan, substantially altering land-use patterns and vegetation cover [23,24,25]. At present, rocky desertification in Wenshan exhibits a contrasting pattern in which severely degraded patches continue to expand while plantations and secondary forests are recovering rapidly, making the area a typical case study for investigating NPP dynamics, restoration thresholds, and driving mechanisms in karst rocky-desertification regions [20,21,22].

Based on the 1:200,000 regional geological maps released by the National Geological Database, we extracted information on geological structures and lithological distributions in Wenshan Prefecture. Carbonate rocks are mainly represented by assemblages of limestone, dolomite, and their interbeds, showing pronounced regional clustering. By identifying and delineating areas with exposed carbonate rocks, we estimated the karst area of Wenshan Prefecture to be 19,870.19 km², accounting for 63.17% of the prefecture’s total area (Figure 1).

2.2. Data Collection

This study used Landsat time-series imagery provided by the USGS, from which vegetation cover (FVC) and bedrock exposure (RE) were derived (https://code.earthengine.google.com/, accessed on 23 November 2025). Terrain data came from the Geospatial Data Cloud (SRTMGL1_003), and elevation (ELE) and slope (SLO) were extracted using ArcGIS 10.8 spatial analysis, at 30 m resolution (https://www.gscloud.cn/, accessed on 23 November 2025). Temperature (Temp) and precipitation (Pre) were obtained from the Qinghai–Tibet Plateau Data Center at 1 km resolution (https://data.tpdc.ac.cn/, accessed on 23 November 2025). Solar radiation (SR) data were obtained from the Geospatial Data Cloud, at 1 km resolution (http://www.gis5g.com/data/, accessed on 23 November 2025) NPP was estimated from the NASA MOD17A3HGF annual product at 5 km resolution (https://code.earthengine.google.com/, accessed on 23 November 2025). Soil moisture (SM) and the Palmer Drought Severity Index (PDSI) were sourced from the TerraClimate dataset at 5 km resolution (https://code.earthengine.google.com/, accessed on 23 November 2025). GDP and nighttime-lights (NL) data were obtained from the Resource and Environment Science Data Center (RESDC), Chinese Academy of Sciences, at 1 km resolution (https://www.climatologylab.org/, accessed on 23 November 2025). Total population (PD) data were obtained from the WorldPop dataset. This gridded product is constructed from national census and population registry data, with an approximate spatial resolution of 1 km (https://hub.worldpop.org, accessed on 23 November 2025). Human footprint intensity (HFP) was obtained from the open datasets provided by the College of Land Science and Technology, China Agricultural University, at 1 km resolution (https://www.x-mol.com/groups/, accessed on 23 November 2025). Road-density (RD) data were produced from OpenStreetMap (https://www.openstreetmap.org/, accessed on 23 November 2025) by extracting transportation features (e.g., highways and railways) and deriving a 100 m road-density raster for Wenshan in ArcGIS 10.8. Soil depth (SD) data were downloaded from the ISRIC data center at 250 m resolution (https://data.isric.org/, accessed on 23 November 2025). Geological data were obtained from the 1:200,000 maps of the National Geological Archives of China (https://www.ngac.cn/, accessed on 23 November 2025); lithological information was extracted after converting MapGIS files to Shapefile format. For datasets with different spatial resolutions, we applied resampling using the nearest-neighbor method to standardize the spatial scale and thereby minimize the impact of resolution differences on the analytical results [26].

2.3. Extraction of Karst Rocky Desertification Information

Carbonate rocks are mainly distributed in lithological assemblages of limestone, dolomite, and their interbedded sequences, showing pronounced regional clustering [27]. Based on the identification and extraction of exposed carbonate rock, the karst area was estimated at 19,870.19 km², representing 63.17% of the prefecture’s total area. This confirms that Wenshan is a typical karst-dominated region in Southwest China.

To support the GEE platform, multi-period remote sensing images were used to calculate the Normalized Difference Vegetation Index (NDVI) and Normalized Difference Rock Index (NDRI), which characterize vegetation cover and surface exposure in the study area [28,29]. On this basis, pixel-wise binary classification was applied to estimate vegetation cover (FVC) and bedrock exposure (RE) separately. FVC reflects the proportion of vegetation cover and serves as a key parameter for evaluating ecosystem health, whereas RE denotes the degree of bedrock exposure and is a critical indicator of rocky desertification processes [30]. The calculation formula is as follows [29,30].

$$NDVI={\displaystyle \frac{{Band}_{NIR}-{Band}_{RED}}{{Band}_{NIR}+{Band}_{RED}}}$$

(1)

In the formula: Band_NIR refers to the near-infrared band, and Band_RED refers to the red light band.

$$NDRI={\displaystyle \frac{{Band}_{SWIR}-{Band}_{NIR}}{{Band}_{SWIR}+{Band}_{NIR}}}$$

(2)

In the formula: Band_SWIR refers to the short-wave infrared band.

$$FVC={\displaystyle \frac{NDVI-{NDVI}_{MIN}}{{NDVI}_{MAX}-{NDVI}_{MIN}}}$$

(3)

In the formula: NDVI_MIN denotes the value where the cumulative contribution rate of NDVI reaches 5%; NDVI_MAX denotes the value where the cumulative contribution rate of NDVI reaches 95%.

$$RE={\displaystyle \frac{NDRI-{NDRI}_{MIN}}{{NDRI}_{MAX}-{NDRI}_{MIN}}}$$

(4)

In the formula: NDRI_MIN denotes the value at which the cumulative contribution rate of NDRI reaches 5%; NDRI_MAX denotes the value at which the cumulative contribution rate of NDRI reaches 95%.

Based on a comprehensive consideration of other scholars’ research findings, ecological and environmental standards, and the specific conditions of the study area, the degree of rocky desertification was ultimately classified into six levels: no rocky desertification, potential rocky desertification, mild, moderate, severe, and extremely severe rocky desertification [31]. The classification criteria are shown in

Table 1. Classification Criteria for Rocky Desertification Levels.

Rocky Desertification Level	Code	FVC	Rock Exposure Rate	Slope
No rocky desertification	N-KBRD	0.8–1.0	0–0.1	0–5
Potential rocky desertification	P-KBRD	0.6–0.8	0.1–0.3	5–8
Light rocky desertification	L-KBRD	0.4–0.6	0.3–0.5	8–10
Moderate rocky desertification	M-KBRD	0.2–0.4	0.5–0.7	10–20
Severe rocky desertification	S-KBRD	0.1–0.2	0.7–0.9	20–30
Extremely revere rocky desertification	ES-KBRD	0–0.1	0.9–1.0	30–90

.

2.4. Dynamic Analysis of Karst Rocky Desertification

This study applied both the single dynamic degree (K) and the composite dynamic degree (P) to quantitatively analyze the spatiotemporal dynamics of rocky desertification in Wenshan Zhuang and Miao Autonomous Prefecture. The aim was to evaluate the rates of change across different rocky desertification levels during distinct periods and to characterize their regional evolutionary patterns. The single dynamic degree measures the average annual rate of change for a specific rocky desertification level, providing an intuitive indication of its expansion or retreat. In contrast, the composite dynamic degree integrates changes across all levels, offering an overall understanding of the regional trajectory of rocky desertification and the effectiveness of its management [32]. The calculation method is as follows [33].

$$K={\displaystyle \frac{U_b-U_a}{U_a}}\times 100\%$$

(5)

$$P=\sum _{i=1}^n{\displaystyle \frac{\left(\left|U_{bi}-U_{ai}\right|\right)}{S}}\times {\displaystyle \frac{1}{T}}\times 100\%$$

(6)

where: $K$ denotes the dynamic degree of a specific rocky desertification type during the study period; U_a and U_b are the areas of that type at the beginning and end of the study period, respectively; $T$ is the number of monitoring years; P represents the comprehensive dynamic degree of all rocky desertification types; $U_{ai}$ and $U_{bi}$ are the areas of the level i rocky desertification type at the beginning and end of the study period, respectively; S is the total area of the study region; and n is the number of rocky desertification types.

2.5. CASA Model Estimates NPP

In this study, we adopted the CASA model version developed by Wang [34] and Wu [35] as the methodological foundation. This version incorporates regionally adaptive optimization of solar energy utilization parameters relative to the traditional model. By integrating multiple inputs—including NDVI, land-use type, air temperature, precipitation, and solar radiation—it improves the model’s applicability and accuracy in the complex ecological settings of southern China’s karst regions. Using this enhanced framework, vegetation NPP in Wenshan Zhuang and Miao Autonomous Prefecture was estimated for the period 2000–2020 year. The calculation formula is as follows [36].

$$KNPP\left(x,t\right)=APAR\left(x,t\right)\times \epsilon \left(x,t\right)$$

(7)

$$FPAR\left(x,t\right)=min\left[{\displaystyle \frac{SR-{SR}_{min}}{{SR}_{max}-{SR}_{min}}}\right],0.95$$

(8)

$$SR\left(x,t\right)=\left[{\displaystyle \frac{1+NDVI\left(x,t\right)}{1-NDVI\left(x,t\right)}}\right]$$

(9)

$$\epsilon \left(x,t\right)=T_{\epsilon 1}\left(x,t\right)\times T_{\epsilon 2}\left(x,t\right)\times W_{\epsilon }\left(x,t\right)\times {\epsilon }_{max}$$

(10)

$$T_{\epsilon 1}\left(x,t\right)=0.8+0.02\times T_{opt}\left(x\right)0.0005\times {\left[T_{opt}\left(x\right)\right]}^2$$

(11)

$$T_{\epsilon 2}\left(x,t\right){\displaystyle \frac{1.184}{(1+e^{0.2\times \left(T_{opt}\left(x\right)-10-T\left(x,t\right)\right)})(1+e^{0.3\times (T\left(x,t\right)-T_{opt}\left(x\right)-10)})}}$$

(12)

$$W_{\epsilon }\left(x,t\right)=0.5+0.5\times {\displaystyle \frac{E\left(x,t\right)}{E_p\left(x,t\right)}}$$

(13)

where:, APAR(x,t) denotes the photosynthetically active radiation; ε(x,t) represents the actual photosynthetic efficiency; SOL(x,t) refers to the total solar radiation received by pixel x in month t; FPAR(x,t) is the fraction of incident photosynthetically active radiation absorbed by the vegetation layer, reflecting the effective utilization of solar radiation by pixel x during month t; T_ε1(x,t) captures the low-temperature stress on photosynthetic efficiency; T_ε2(x,t) captures the high-temperature stress on photosynthetic efficiency; Wε(x,t) is the water stress coefficient; E(x,t) represents the actual regional evapotranspiration; and E_p(x,t) the potential regional evapotranspiration. Finally, εmax indicates the maximum photosynthetic efficiency attainable under ideal conditions.

To assess the reliability and accuracy of the CASA model developed in this study for estimating NPP in Wenshan Prefecture, we used the MODIS MOD17A3H annual NPP product as reference data [34]. Corresponding values were extracted from both the CASA simulations and the MOD17A3H dataset for the same time periods, and a linear regression model was established between the two. The coefficient of determination (R²) was then calculated to quantify the strength of the relationship between the model outputs and the reference data. An R² value closer to 1 indicates stronger agreement between the two datasets, thereby demonstrating higher reliability and accuracy of the model estimates [33]. The calculation formula is given as follows [37].

$$R^2=1-{\displaystyle \frac{\sum \begin{array}{c}n\\ i=n\end{array}{\left(y_i-{\widehat{y}}_i\right)}^2}{\sum \begin{array}{c}n\\ i=1\end{array}{\left(y_i-\overline{y}\right)}^2}}$$

(14)

where: $y_i$ represents the vegetation NPP inverted by the CASA model; ${\widehat{y}}_i$ denotes the fitted value calculated by the regression model; $\overline{y}$ is the mean of the fitted values; n is the sample size.

2.6. Spatiotemporal Trend Analysis of NPP

This study applied the Theil–Sen median trend analysis to assess changes in vegetation net primary productivity (NPP) in Wenshan Prefecture from 2000 to 2020. Compared with conventional least-squares regression, the Theil–Sen approach offers greater robustness and adaptability. It does not require data to follow a normal distribution and is applicable to cases with non-normality, heteroscedasticity, or extreme values. Moreover, it is insensitive to outliers and effectively reduces their influence on trend detection, making it particularly suitable for analyzing remote sensing time series in ecological and environmental studies [38]. The calculation formula is as follows [39].

$$S_{enij}=Median\left({\displaystyle \frac{{x_j}_{x_i}}{j_i}}\right),{\forall }_j>i$$

(15)

where: S_enij denotes the Theil–Sen trend slope; x_j and x_i represent the sequence values at time points j and i, respectively, where 1 < i < j < n; n denotes the sequence length.

The Mann–Kendall test was applied to assess the significance of the Sen trend analysis results [40]. As a nonparametric statistical method, the Mann–Kendall test is widely used for detecting trends in climate, hydrological, and ecological time series. It does not assume normality and is suitable for sequential data with missing values or outliers. Owing to its robustness and computational simplicity, it is particularly appropriate for long-term analyses of remote sensing-derived data [41]. The test statistic S is calculated as follows [40,41].

$$S=\sum _{i=1}^{n-1}\sum _{j=i+1}^{n}sgn(x_j-x_i)$$

(16)

where: x_j − x_i denote the item j and i terms of the time series, respectively.

$$Sgn\left(\theta \right)=\left\{\begin{array}{c}1,\theta >0\\ 0,\theta =0\\ -1,\theta <0\end{array}\right.$$

(17)

When n > 10, the standardized test statistic for S is calculated as follows.

$$Z=\left\{\begin{array}{c}{\displaystyle \frac{S-1}{\sqrt{V_{ar}\left(S\right)}}},S>0\\ 0, S=0\\ {\displaystyle \frac{S+1}{\sqrt{V_{ar}\left(S\right)}}},S<0\end{array}\right.$$

(18)

$$Var\left(S\right)={\displaystyle \frac{n\left(n-1\right)\left(2n+5\right)}{18}}$$

(19)

2.7. Analysis of Primary Driving Factors for NPP

Vegetation productivity (NPP) in karst desertification areas is driven by both natural environmental factors and human activities. Natural factors determine the hydrotrophic resources, topographic patterns, and soil conditions for vegetation growth, while human activity factors reflect regional economic development intensity, spatial development patterns, and disturbance pressures [34,36]. Therefore, this study selected 7 ecologically significant natural factors and 5 human activity-related factors with available data, using a geographic detector to quantify their explanatory power for spatial variations in NPP.

The Optimal Parameters-Based Geographical Detector (OPGD), implemented in R 4.3.1, was used to identify the optimal discretization method by calculating the maximum interpretability (q-value) [42]. Using the GD package in R 4.3.1, several classification methods—including equal intervals, natural breaks, quantiles, geometric intervals, and standard deviation—were tested. Classification schemes ranging from four to seven categories were evaluated, and the one yielding the q-value was selected for analysis [43]. The ranking of q-values provides an intuitive indicator for assessing the relative contributions of different factors to net primary productivity formation, thereby establishing a theoretical foundation for understanding spatial variations in regional net primary productivity. The formula for calculating the q-value is as follows [44].

$$q=1-{\displaystyle \frac{S_{SW}}{S_{ST}}}=1-{\displaystyle \frac{{\sum }_{h=1}^L{N}_h{\sigma }_h^2}{N{\sigma }^2}}$$

(20)

$$S_{SW}=\sum _{h=1}^L{N}_h{\sigma }_h^2,S_{ST}=N{\sigma }^2$$

(21)

In the equation: S_SW and S_ST represent the variance within each layer and the total variance across the region, respectively; q denotes the explanatory power of each relevant primary driver factor for NPP, with a value range of [0, 1]; h indicates the category or stratum of different influencing factors; N is the number of sample units across the entire region; N_h is the number of sample units within a subdomain; L denotes the number of subdomains; ${\sigma }_h^2$ and ${\sigma }^2$ represent the variance within the hth layer and the variance of Y values across the entire region, respectively.

Interaction detectors can be used to determine whether the explanatory power of multiple factors on the dependent variable is enhanced or weakened under their combined influence. Their core function lies in analyzing whether significant synergistic effects or independence exist among variables [45]. The results of the assessment are shown in

Table 2. Interaction Detection Type Classification Rules.

Interaction Effect	Discrimination Method
Synergy	q(X1∩X2) > q(X1) or q(X2)
Double synergy	q(X1∩X2) > q(X1) and q(X2)
Nonlinear synergy	q(X1∩X2) > q(X1) + q(X2)
Antagonism	q(X1∩X2) < q(X1) + q(X2)
Single antagonism	q(X1∩X2) < q(X1) or q(X2)
Nonlinear antagonism	q(X1∩X2) < q(X1) and q(X2)

[44].

3. Results and Analysis

3.1. Spatiotemporal Evolution Analysis of Karst Rocky Desertification

3.1.1. General Characteristics of Spacetime

As shown in Figure 2 and Figure 3, From 2000 to 2020, rocky desertification was relatively severe in the central part of Wenshan Prefecture, whereas the southern, northern, and eastern sectors exhibited weaker degrees of rocky desertification. Rocky desertification was mainly concentrated in the karst belt where Qiubei country, Yanshan country, Guangnan country, and Wenshan city converge, radiating outward along a north–south axis. Over the past two decades, areas at risk of P-KBRD expanded markedly, whereas L-KBRD and ES-KBRD decreased, producing a more fragmented overall pattern. This trend reflects continuous ecological improvement and the effectiveness of governance measures. The area of P-KBRD increased substantially, with the largest expansion between 2010 and 2015 (657.37 km²). M-KBRD showed the most pronounced decline, shrinking from 2784.34 km² in 2000 to 1883.83 km² in 2020 (a reduction of 900.51 km²), with the steepest decrease occurring between 2010 and 2015 (762.51 km²). S-KBRD fell from 315.47 km² to 194.27 km² (>38% reduction), while ES-KBRD declined extremely from 17.81 km² to 8.33 km² (>50% reduction), with the most rapid decrease recorded between 2015 and 2020. Overall, rocky desertification in Wenshan Zhuang and Miao Autonomous Prefecture exhibited a positive trajectory with fluctuations, with particularly notable improvements in the southwestern and central regions.

3.1.2. Dynamic Degree Analysis

Figure 4 shows that changes in the degree of rocky desertification in Wenshan Zhuang and Miao Autonomous Prefecture (2000–2020 year) exhibit distinct phased characteristics. Overall, areas affected by rocky desertification and those at potential risk expanded, whereas L-KBRD and ES-KBRD categories declined. Notably, areas classified as moderate or worse showed a sustained downward trend. From 2000 to 2005, rocky desertification expanded most rapidly, with an increase of 606.43 km² (dynamic rate 2.17%). P-KBRD also grew by 326.37 km², while L-KBRD and M-KBRD decreased by 541.96 km² and 413.64 km², respectively. Between 2005 and 2010, a brief rebound occurred: the total affected area declined, yet M-KBRD expanded by 352.38 km² (dynamic rate 2.97%). M-KBRD decreased sharply by 762.51 km² (–5.60%), and S-KBRD recorded its first notable reduction (76.51 km²). From 2015 to 2020, the rate of change slowed. L-KBRD, M-KBRD, and S-KBRD all declined further, and ES-KBRD contracted by 10.99 km² (–11.38%). Overall, the period 2000–2015 represented the most significant phase of rocky desertification control, with dynamic rates consistently above 0.7%. The largest improvement occurred between 2010 and 2015, with a net reduction of 2224.62 km² and a dynamic rate of 1.42%, marking the most prominent achievements in governance efforts.

3.1.3. Subsubsection

Analysis of rocky desertification across elevation zones (2000–2020) reveals distinct vertical patterns (Figure 5). At 0–500 m, the affected area increased annually from 26.55 to 38.73 km², dominated by P-KBRD and L-KBRD, with only small patches of moderate to severe degradation. In the 500–1000 m range, the total affected area was 608.00–693.50 km², including 316.09–355.33 km² of P-KBRD, 134.54–195.80 km² of L-KBRD, 28.44–81.69 km² of M-KBRD, and 1.40–11.98 km² of S-KBRD, indicating progressively worsening conditions. The largest extent of rock desertification occurred at 1000–2000 m. Within 1000–1500 m, the affected area spanned 3227.95–4220.12 km², with notable increases in M-KBRD and S-KBRD. At 1500–2000 m, rocky desertification covered 3119.21–4418.12 km², including 885.17–1519.80 km² of M-KBRD and 96.91–217.87 km² of S-KBRD, marking this range as the highest-risk zone. At 2000–3000 m, rocky desertification declined sharply, consisting mainly of P-KBRD and L-KBRD. M-KBRD and S-KBRD were minimal, and areas above 2500 m showed almost N-KBRD. Overall, rocky desertification was most severe between 1000 and 2000 m, while low (<500 m) and high (>2000 m) elevations were dominated by P-KBRD and L-KBRD, posing lower ecological risks. Between 2000 and 2020, moderately to severely degraded areas expanded significantly at mid-elevations, whereas total affected areas at low and high elevations remained relatively stable. These results confirm that rocky desertification follows a distinct elevation-dependent zonation pattern.

3.2. Analysis of NPP Trend Patterns

Based on the 20-year mean NPP simulated by the CASA model for 2000–2020 and the corresponding NPP values extracted from the MOD17A3HGF product for the same period, we constructed a simple linear regression model (Figure 6), with the regression line in Figure 5b fitted using ordinary least squares (OLS). The two datasets exhibited a highly significant correlation (p < 0.01), with a correlation coefficient of 0.87 and an R² of 0.75. These results indicate that the CASA-simulated NPP is highly consistent with the MOD17A3HGF NPP, thereby supporting the applicability of the model for simulating vegetation productivity under the complex topographic conditions of the Wenshan Zhuang and Miao Autonomous Prefecture.

From 2000 to 2020 year, the average NPP of vegetation in Wenshan Zhuang and Miao Autonomous Prefecture exhibited a fluctuating but overall upward trend (Figure 6). NPP values ranged from 951.17 to 1138.97 gC m⁻² a⁻¹, with a mean of 1057.13 gC m⁻² a⁻¹ and an average annual increase of 4.07 gC m⁻² a⁻¹. The highest value occurred in 2019 year and the lowest in 2001 year. The steepest rise was observed between 2017 and 2018 year (+108.81 gC m⁻² a⁻¹), while the sharpest decline occurred between 2016 and 2017 year (–147.83 gC m⁻² a⁻¹), though the long-term upward trajectory remained unchanged.

As shown in Figure 6 and Figure 7, NPP change rates in Wenshan Prefecture ranged from –9.433 to 24.406 gC m⁻² a⁻¹. Most areas exhibited increasing trends, although localized declines were also observed. Spatially, the pattern was characterized by “higher in the northwest and lower in the southeast”. Counties such as Qiubei, Guangnan, and Yanshan showed relatively high rates of increase, whereas southeastern regions including Funing and Maguan displayed lower or even negative changes.

Combining Theil–Sen slope estimation with the Mann–Kendall test, the spatiotemporal trends of NPP in Wenshan Prefecture revealed a pattern of “increase in the west, decrease in the east, with core-area degradation”. Overall, 77.32% of the prefecture showed increasing trends (mainly in the western and central regions), 22.66% showed decreases (concentrated in the east and south), and only 0.02% remained stable. The southeastern part of Qiubei County and the central part of Guangnan County emerged as core zones of NPP growth. Within the increasing regions, extremely significant and significant increases accounted for 40.07% of the total area. Conversely, extremely significant and significant declines in eastern and southern Wenshan Prefecture exhibited strong spatial clustering, forming distinct degradation cores. These areas, covering only 1.58% of the prefecture, were mainly associated with karst terrain and intensive human activities, such as mineral extraction and agricultural expansion.

3.3. Coupled Analysis of Karst Rocky Desertification and NPP

As shown in Figure 8, both the average and total NPP across different rocky desertification types in Wenshan Prefecture exhibited an overall upward trend from 2000 to 2020 year. Areas N-KBRD and those with P-KBRD consistently maintained higher NPP levels, with marked increases over time. These regions were the core contributors to regional NPP, reflecting the ecosystem’s overall positive trajectory. In contrast, areas with M-KBRD and S-KBRD generally declined in extent and NPP, indicating that degraded zones are gradually shrinking under management interventions and ecological functions are being restored. Specifically, average NPP in N-KBRD areas increased from 958.45 to 1050.14 gC m⁻² a⁻¹, and in P-KBRD areas from 960.41 to 1070.68 gC m⁻² a⁻¹, reflecting notable improvements in vegetation growth. Although fluctuations occurred in L-KBRD and ES-KBRD areas, overall NPP in these zones also rose. Among them, ES-KBRD areas showed the largest increase, from 852.10 to 1006.59 gC m⁻² a⁻¹. Overall, NPP across rocky desertification gradients in Wenshan Prefecture has recovered effectively; however, S-KBRD and ES-KBRD still require sustained management and attention.

3.4. Analysis of Factors Influencing NPP Driving Forces

Figure 9a shows clear differences in the linear correlations between NPP and the various factors. Among natural factors, slope has the highest correlation with NPP (r = 0.42), followed by a moderate positive correlation between mean annual precipitation and NPP (r = 0.26). Among anthropogenic factors, socio-economic variables are all negatively correlated with NPP, with the human footprint showing the strongest correlation in magnitude (r = −0.34). Other factors exhibit weaker correlations with NPP and may influence its distribution through combined effects with other variables.

Figure 9b (factor detector) indicates that all factors pass the significance test, but their explanatory powers differ markedly. Among natural factors, soil depth has the highest q value (q = 0.76), far exceeding the others, and is the primary control on the spatial variation in NPP in karst rocky-desertification areas. Slope ranks second (q = 0.31), underscoring the key role of soil and terrain conditions in regulating soil and water conservation, soil development, and vegetation growth. Among anthropogenic factors, the human footprint has the highest q (q = 0.12); by contrast, other human-related indicators have lower q values, suggesting their effects on NPP are more indirect, typically manifesting through changes in land-use patterns or when compounded with terrain and soil conditions.

Figure 9c shows that most pairwise interactions have q values greater than those of the corresponding single factors, indicating bivariate or nonlinear enhancement effects. Interactions between slope and other factors are consistently higher than the single-factor effects (with the exception of soil depth), implying that on steep, rocky-desertified slopes, the superposition of heavy precipitation or intense human disturbance more readily triggers soil loss and vegetation degradation. In addition, the interaction q values between soil depth and precipitation, soil moisture, and elevation (≈0.17–0.23) are generally higher than the single-factor effects of those variables, indicating that under thin-soil conditions the influence of natural hydrothermal gradients on NPP is further amplified. Overall, natural and anthropogenic factors exhibit pronounced synergies rather than simple linear additivity.

4. Discussion

4.1. Classification Analysis of Karst Rocky Desertification

Using a decision-tree classification method based on remote sensing data, a multi-tiered grading system was established with three core indicators—FVC, RE, and slope gradient—to enable automated identification and classification of rocky desertification severity. This approach offers clear logic and transparent classification while effectively integrating multi-source remote sensing factors to capture rocky desertification patterns across complex terrain. Previous studies have demonstrated that decision trees achieve high accuracy and efficiency in rocky desertification classification. For example, DF Wei et al. propose estimating vegetation cover using Landsat 8 OLI imagery, based on Fractional Vegetation Cover and Fractional Rock Cover; the classification is refined with local rocky desertification conditions and previous standards [46]. Their results showed strong agreement with field surveys, with accuracy sufficient for regional-scale analysis. Compared with traditional maximum likelihood methods or support vector machines, decision trees require fewer training samples and are more adaptable to nonlinear relationships and threshold-based divisions. They are particularly effective for remote sensing tasks involving multiple interacting indicators and pronounced hierarchical variation [47].

4.2. Trends in Karst Rocky Desertification and NPP

The overall trend of rocky desertification in Wenshan Prefecture showed fluctuating improvement, characterized by the continuous expansion of P-KBRD areas, while zones classified as moderate or more severe gradually declined. This pattern is consistent with the evolutionary characteristics of karst landscapes in southeastern Yunnan. Lu T [48] reported a decline in the degree of rocky desertification in eastern Yunnan from 1987 to 2020, with particularly notable progress after 2010, when large areas of L-KBRD and P-KBRD transitioned to non-desertified status. Dynamic degree analysis indicated that governance outcomes in Wenshan Prefecture exhibited phased variation, with the most pronounced improvement occurring between 2010 and 2015, when the dynamic degree reached 1.42%. This corroborates that, during the implementation of the Natural Forest Protection Program and the Grain-for-Green Program in Yunnan, natural forest stewardship and logging restrictions were strengthened, consolidating existing stands and curbing new disturbances, while re/afforestation scales and subsidy levels were expanded, thereby accelerating vegetation recovery on sloping farmland and other fragile sites [23,24]. In addition, during the 12th Five-Year Plan period, karst rocky desertification in Southwest China was prioritized, and a pathway and evaluation framework centered on enclosure, re/afforestation, and soil-and-water conservation was established to support local implementation [25]. This suggests a gradual shift in rocky desertification control from policy-driven initiatives to measurable spatial improvements [49]. Vertical zonation analysis further showed that rocky desertification was less severe and improved markedly at low elevations, more pronounced at mid-elevations, and predominantly potential at high elevations. From 2000 to 2020, NPP in Wenshan Prefecture exhibited a fluctuating but overall upward trend, with an average annual increase of 4.07 gC m⁻² a⁻¹. This suggests that regional ecosystem productivity has gradually improved under the combined influence of favorable climate conditions and ecological engineering initiatives [50,51]. Spatially, NPP followed a northwest–southeast gradient, with high values concentrated in northern Qiubei and northwestern Guangnan, where forest cover is extensive and solar energy utilization is strong.

4.3. Main Factors Influencing NPP

A synthesis of existing studies indicates that soil depth sets the carrying capacity for vegetation productivity, while slope modulates it primarily by controlling soil erosion, runoff convergence, and the transfer of energy and water [52,53,54]. At the same time, human activity footprints such as mining, road construction, and urban expansion suppress vegetation NPP in rocky desertification areas [55]. Our results are consistent with these findings, identifying soil depth and slope as the key constraints on NPP recovery in rocky desertification regions. It is also important to note that the effects of human activities on NPP are not purely inhibitory: ecological restoration generally enhances productivity and carbon sequestration, whereas on thin, steep slopes at mid to high elevations, development-oriented disturbances alter local hydrothermal conditions and couple with terrain and soil constraints, thereby indirectly weakening NPP recovery [56].

4.4. Limits and Outlook

The CASA model relies on multiple input datasets, which are subject to spatiotemporal inconsistencies and inversion uncertainties that can be propagated and even amplified throughout the NPP simulation process. The Theil–Sen trend analysis and Mann–Kendall test are mainly suitable for detecting monotonic changes and have limited ability to characterize time series with autocorrelation, nonlinear variations, or abrupt shifts. The results of the Geographical Detector are sensitive to factor discretization schemes and spatial scale, and they only reflect the explanatory power in terms of spatial variance rather than strict causal relationships, while multicollinearity among multiple factors further complicates the interpretation of interaction effects. Overall, the combined limitations of data accuracy, model assumptions, and the analytical methods themselves affect the reliability of NPP estimation and the identification of its driving factors. In analyzing the relationship between NPP and rocky desertification, the focus has largely been on changes in area and total extent at the regional scale, without fully addressing the mechanisms underlying soil erosion, vegetation restoration, and water cycling [57]. Future research should integrate ecological parameters such as soil moisture and nutrients to enhance the applicability of the CASA model in karst environments [58]. Long-term observation sites, field sampling, and ground monitoring will enable direct measurement and calibration of key parameters (e.g., photosynthetically active radiation use efficiency, and temperature and moisture stress coefficients), thereby establishing a parameterization framework better aligned with regional characteristics. Combining ground observations with remote-sensing inversion results will further validate the model’s reliability and practical value [59].

5. Conclusions

This study analyzed karst rocky desertification in Wenshan Zhuang and Miao Autonomous Prefecture using remote sensing, an improved CASA model, and optimized geographic detectors to explore NPP dynamics and driving forces. From 2000 to 2020, rocky desertification generally improved, with moderate-to-severe areas reduced and potential areas expanding. It was concentrated in karst depressions and valleys, most severe in the north and center. The overall dynamic rate was ~0.9%, peaking between 2010 and 2015 with a 2224.62 km² reduction. Mean NPP was 1057.13 gC m⁻² a⁻¹, showing an overall increase; gains were in northern Qiubei and central-west Guangnan, while declines occurred near mining and cropland. Rocky desertification displayed clear vertical zonation, most pronounced at 1000–2000 m. Factor analysis indicates that slope, soil depth, and human footprint are the primary determinants of NPP distribution, and they exhibit synergistic interaction effects with other variables. Future management should adopt a combined strategy of rebuilding thin soils, stabilizing steep slopes, limiting disturbances, and implementing zoned control: prioritize soil and slope engineering restoration in moderately to severely degraded areas, while emphasizing protection and low-intensity management in potential and mildly affected zones to reduce the risk of progression to higher levels of rocky desertification at its source. Future research should incorporate additional ecological parameters, such as soil moisture and soil nutrients, and combine them with long-term ground observations to further constrain the uncertainties in NPP estimation and enhance the applicability and transferability of this framework to other karst regions in Southwest China and beyond.