Research on an Ecological Sensitivity Evaluation of Mountain-Type National Parks Under Multi-Modal Optimization: A Case Study of Shennongjia, China

Abstract

As ecologically sensitive zones within natural ecosystems, national parks demand more precise evaluation models for ecological sensitivity assessment. This study takes Shennongjia Forestry District, a pioneer among China’s first-batch national parks, as an study object to optimize the ecological sensitivity evaluation framework. In this study, we developed an integrated methodology incorporating high-precision ASTER GDEM elevation data, Landsat8 TM vegetation density inversion, SWAT-based flash flood simulation, and SVM-LSM landslide prediction while introducing dynamic protection elements including species migration corridors and human activity risks. The results demonstrate that the refined data structure enhances terrain coupling accuracy by transitioning from “Vegetation Type—Runoff Coefficient” to “Vegetation Density—Runoff Coefficient” conversions, with the optimized model exhibiting superior sensitivity in spatial element identification. This approach provides scientifically grounded technical support for balancing ecological conservation and visitor management in protected areas.

1. Introduction

The concept of national parks has evolved from protecting single animal/plant species or scenic views to safeguarding entire interconnected ecosystems. In 2016, the U.S. National Park Service (NPS) formulated detailed scientific management plans for each national park to ensure the effective preservation of natural resources [1]. In 2020, China entered into partnership with the International Union for Conservation of Nature (IUCN) in advancing the construction and management of national parks (IUCN, 2020) [2]. Then, in 2019, China proposed to construct a system of protected areas centered on national parks by prioritizing the protection of the most important areas within the national natural ecosystem, which feature the most unique landscapes, richest natural heritage, and most concentrated biodiversity. Involving a more efficient identification and evaluation of the objects and contents that deserve protection [3], these measures promote the scientific and comprehensive ecological protection of national parks.

Ecological sensitivity evaluation represents a key focus of research on ecological protection [4,5]. As technology advances, especially computers and geographic information systems (GISs), there is a shift in ecological sensitivity evaluation from single-factor to multi-factor assessments. As a result, more elements are introduced, such as water conservation, biodiversity evaluation, and soil conservation, which provides more comprehensive scientific guidance on preserving sensitive ecological areas [6,7,8]. Additionally, a variety of contents are incorporated into ecological protection evaluations, such as carbon sequestration and glacier protection [9,10].

However, three critical limitations persist in existing methodologies: (1) inadequate model adaptability across diverse ecosystems and geographical scales [11], (2) the insufficient integration of dynamic protection elements (e.g., species migration corridors, human activity risks) with static sensitivity metrics [12], and (3) the lack of standardized protocols for multi-source data fusion [13]. While scholars have acknowledged accuracy limitations in conventional models and datasets, current solutions remain fragmented—focusing predominantly on parameter optimization rather than systemic framework development. This methodological fragmentation hinders the operationalization of evaluation outcomes for adaptive management strategies, particularly in protected areas requiring dynamic monitoring capacities [14].

Specifically, allowing for the value assessment requirements for soil conservation and water conservation in national parks, it is imperative to enhance the accuracy of annual average rainfall and surface runoff data (including surface cover and vegetation classification) as the key data [15,16,17]. Three persistent methodological gaps undermine current practices: (1) The precision paradox—while city/county-level studies propose 20–30 m resolution standards [18], statistical yearbook data remain predominantly at the >1:100,000 scale due to meteorological station placement constraints, particularly inadequate in remote parks with complex terrain demanding hydrological precision. (2) Standardization void—no operational guidance exists for upgrading substandard datasets, exacerbated by sparse sampling density (avg. <0.3 meteorological stations/100 km² in mountain parks), far below the required precision, which is especially true in those remote and sparsely populated national parks where detailed rainfall data are particularly needed due to complex terrain and climate [18,19,20]. (3) Functional disconnect—surface runoff analyses rely on generic vegetation categories (e.g., “forested land”), neglecting the surface cover data, which contain ecophysiological parameters that are critical for rainwater interception (e.g., coniferous vs. broadleaf forest differentiation) [20,21,22,23]. As a matter of fact, general information about surface cover data is used more extensively, which allows for a relatively simplistic surface runoff data analysis.

In terms of ecological sensitivity evaluation models, the current recognized ecological sensitivity assessment models include InVEST for ecosystem service quantification, DPSIR frameworks for human–environment interaction analysis, and AHP-based fuzzy evaluation for multi-criteria decision making. While InVEST excels in spatial explicitness and policy scenario simulations, its reliance on high-resolution input data limits its applications in data-scarce regions [24,25]. The DPSIR model effectively links socio-economic drivers with ecological impacts but struggles with quantitative dynamic feedbacks [26,27]. Machine learning approaches (e.g., Random Forest) demonstrate superior predictive accuracy in complex systems yet suffer from interpretability challenges [28,29]. Meanwhile, most research relies on single models to assess and optimize the ecological environment. However, some models are subjected to significant limitations and are incapable of meeting the needs of multiple ecological protection aspects in various specific environments such as mountainous areas and national parks [30,31]. For example, in biodiversity protection evaluations, which are highly valued in national parks, most models are limited to predicting species distribution (source habitats), and there is a lack of analysis on species migration corridors. Consequently, the conflict zones between wildlife movement paths and human activities are overlooked [32,33]. Regarding soil conservation evaluations, which are commonly required in national parks, the modified Universal Soil Loss Equation (RUSLE) is applied in most studies to identify vulnerable areas. However, the RUSLE, which is only suitable for slope-based models, is unfit to estimate runoff, gully erosion, or deposition and related disasters, which renders it ineffective in mountainous areas where soil erosion hazards are severe [34,35,36]. Therefore, hybrid methodologies integrating geostatistical techniques (e.g., EBK regression–Kriging) with process-based models are emerging as robust solutions for multi-scale assessments. Recent advances prioritize coupled model architectures to balance precision and generalizability, particularly in protected areas requiring adaptive management protocols.

In summary, existing ecological sensitivity assessments face critical limitations in data quality and model adaptability for complex mountain–forest ecosystems. Challenges include insufficient data resolution, fragmented coverage, and oversimplified classifications that inadequately represent heterogeneous environments, coupled with an overreliance on single-model frameworks requiring unrealistic data granularity. This study addresses these gaps through an integrated approach combining ecological refinement with hybrid modeling, systematically incorporating dynamic conservation elements (e.g., species corridors, human activity risks) into sensitivity metrics. By establishing standardized protocols for multi-source data fusion and scenario-adaptive validation, the proposed framework enhances ecological assessment precision and operational utility in data-scarce protected areas. This study emphasizes innovations in assessment frameworks and data refinement techniques, rather than asserting the factual occurrence of disaster-induced disturbances in the Shennongjia Forestry District. Consequently, the research focus will center on vulnerability analysis within the ecological sensitivity evaluation paradigm.

2. Materials and Methods

2.1. Study Area and Research Framwork

In 2021, China announced the first batch of 10 national park pilots, more than half of which are characterized by mountainous terrains and forest landscapes, each with rich and diverse ecological environments, characteristic of their settings. Shennongjia National Park is a typical representative among them. Therefore, Shennongjia National Park will be used in this study as a case to conduct research on the ecological sensitivity evaluation of “mountain-type” parks, with the aim of developing the optimized evaluation techniques that can be applied to more similar areas. This is conducive to creating more refined optimization schemes for the ecological environment of mountain-type national parks. Research demonstrates that portions of critical habitats for key protected species in Shennongjia lie beyond the national park boundaries, necessitating ecological considerations to expand the study’s scope. In alignment with local conservation management practices, the entire Shennongjia forestry district encompassing the national park has been included in the research framework [37].

Located in the northwest of Hubei Province, Shennongjia Forestry District covers an area of 3253 square kilometers. Situated in its southwestern part, the national park accounts for about 36% of the total area of the forestry district. It represents the only administrative division in China named “forestry district” (see Figure 1), hosting a globally representative biodiversity system. Included in the “World Network of Biosphere Reserves”, it is regarded as the “Water Tower of Central China”, playing a crucial role in water conservation and soil preservation. In this sense, it is imperative to conduct targeted ecological sensitivity evaluations on water conservation, biodiversity, and soil preservation. The International Union for the Conservation of Nature (IUCN) report identifies Shennongjia as “the most intact subtropical forest ecosystem at its latitude in the Northern Hemisphere”, hosting exceptionally rich biodiversity through its “unique transitional zone between subtropical and warm-temperate flora” [38].

Through the integration of current research experiences and innovation, this paper aims to construct a more scientific and comprehensive evaluation technology system, producing more precise and effective results for mountain national parks (see Figure 2).

2.2. Research Data

2.2.1. Rainfall Data Optimization

In general, traditional spatial rainfall data are generated through the spatial interpolation of station data, such as the Kriging method [39,40]. However, there are only two meteorological stations in Shennongjia Forestry District, as a result of which data density is insufficient for high-precision interpolation. To address the insufficiency of sampling points, it is proposed in this paper that the quality of the predicted data can be enhanced by combining Kriging with regression models to form an EBK Regression Prediction (Esri) and incorporating high-precision known spatial independent variables, given a close spatial relationship between local rainfall and the terrain. As documented in ESRI’s ArcGIS10.8 Pro software documentation and relevant research papers, this approach integrates Empirical Bayesian Kriging (EBK) with known explanatory variable rasters to influence interpolated data values. By combining Kriging with regression analysis, the method produces more accurate predictions than those achieved using either regression analysis or Kriging alone [41,42,43].

As suggested by scholars, the tertiary uplift terrain of western Hubei, where Shennongjia is located, causes intense water vapor lifting and increases rainfall by blocking the low vortex mountain-climbing air flows from the southwest Sichuan direction [44], which illustrates the correlation between rainfall and local terrain around Shennongjia and its neighboring areas. Additionally, scholars collected data from 58 local rainfall stations and 30 m elevation data from ASTER GDEM in Wushan County to the south of Shennongjia Forestry District making EBK regression predictions to obtain more precise local rainfall data [45,46]. For this reason, meteorological station data are comprehensively collected in this paper from around Shennongjia Forestry District, including 58 township weather stations in Wushan County, 31 in Xingshan County, and scattered sites in Baokang, Fang County, Zhushan, Zhuxi, and Wuxi. Additionally, 30 m DEM data are used to obtain more accurate rainfall data (see Figure 3).

2.2.2. Surface Runoff Coefficient Optimization

The evaluation results are of low accuracy as traditional surface runoff coefficients are limited due to missing surface vegetation data (hence the use of surface cover data) [47,48]. Therefore, multiple studies are integrated in this paper to construct a process of “Vegetation Density—Surface Vegetation Runoff Coefficient” analysis, with data granularity improved by using local and surrounding sampling point data and regression models (see Figure 4).

The procedures are detailed as follows:

• Forest vegetation density is inverted using the remote sensing Kirchhoff Transform formula. Specifically, the sampling point data include 25 existing local sampling points collected by scholars, 20 additional sampling points supplemented by the author through street view photos on Baidu Maps and the well-known domestic outdoor sports website “2bulu” (https://www.2bulu.com/) (accessed on 3 April 2023), and a variety of factors such as the ratio vegetation index, moisture, brightness, and greenness generated from Landsat8 TM’s 30 m data, which are treated as independent variables for the regression and prediction groups.
• Forest density is converted to surface runoff coefficient via a multivariate regression model, where the independent variables include multiple measured results of forest vegetation runoff from the surrounding areas with similar climate and terrain conditions (Chongqing) [49,50].

2.3. Evaluation Model

2.3.1. Biodiversity Protection Evaluation Model

Regarding biodiversity protection evaluations, many rare animals are prioritized due to their extensive activity and migration ranges, which render them susceptible to unintentional encroachment or disturbance [51,52]. However, the existing models ignore migration corridors and human activity disturbances since they are often limited to analyzing species distribution only. In this paper, the biodiversity evaluation model is optimized for Shennongjia Forestry District by integrating assessments of species distribution, migration corridors, and human activity disturbances (Figure 5).

Firstly, the maximum entropy model (MaxEnt) is used in this paper to conduct a more comprehensive and detailed assessment of how 69 rare protected species are distributed within the Shennongjia ecological zone. Distinct from traditional research, which focuses narrowly on a few representative species, this study focuses on improving the accuracy of predicting species distribution through an in-depth analysis on the ecological habits, environmental needs, and spatial distribution characteristics of different species, with a variety of influencing factors taken into consideration.

Under the maximum entropy model, it is assumed that the most reasonable probability distribution is the one with the maximum entropy in the absence of any other validating information. In this paper, known environmental condition data are used to determine the optimal probability distribution of species distribution in line with the principle of maximum entropy [53,54]. By maximizing entropy, the model is applicable to obtain the most objective and unbiased distribution estimates as it avoids unnecessary assumptions about species distribution while satisfying known constraints. This method maintains the highest uncertainty in predicting species distribution under limited information and ensures the robust capture of the complex relationships between environmental variables and species distribution.

Given the widespread utilization of the maximum entropy model in such assessment studies, its equations are not elaborated in this paper to maintain conciseness and focus on methodological application.

For the challenges in collecting sufficient occurrence data for all 69 protected animal species in Shennongjia National Park, this study adopted a taxonomic grouping strategy based on shared ecological requirements within the same orders and families. Species were systematically categorized into 24 families, with at least one representative species selected per family (totaling 28 species after merging taxonomically similar groups lacking sufficient data). This framework enabled the compilation of 370 georeferenced occurrence points from local surveys and the literature [55], coupled with five critical habitat variables: suitable terrain, vegetation, climate, water proximity, and human activity disturbance (detailed in Appendix A).

Inputting these parameters into MaxEnt generated spatially explicit predictions of optimal habitat suitability across the Shennongjia ecoregion. Model evaluation demonstrated robust performance, with mean AUC values of 0.902 (training) and 0.836 (validation), indicating strong discriminatory capacity between suitable and unsuitable habitats. Environmental variable contribution analysis revealed the following hierarchy: suitable vegetation (33.4%), terrain (27.6%), climate (17.7%), human disturbance (12.2%), and water proximity (9.1%). Vegetation metrics, particularly canopy density validated through Kirchhoff Transform inversion, emerged as the dominant predictor. This finding underscores the critical role of vegetation structure—specifically canopy cover dynamics—in shaping habitat selection patterns for focal species.

Secondly, the connectivity corridors linking “source” habitats are constructed in this paper by integrating the circuit model (incorporated into the Linkage Mapper tool on the ArcGIS platform), the minimum cumulative resistance (MCR) model, and graph theory. Thus, the analytical precision of species migration corridors is enhanced [56,57]. Different to the traditional methods of constructing ecological corridors, the circuit model not only takes into consideration the impact of landscape heterogeneity on species migration but also provides a more accurate corridor structure by simulating actual ecological resistance and connectivity.

The core idea of the circuit model is to view the ecological landscape as a circuit, where different geographical units (such as habitats, forests, rivers, etc.) act as resistance elements, and the paths of species migration are similar to the flow of electrical current, with the current flowing along the path of least resistance [58]. Mathematically, it is expressed as follows:

$$\mathrm{P}(\mathrm{x})={\displaystyle \frac{1}{\mathrm{R}\left(\mathrm{x}\right)}}$$

(1)

where P(x) represents the migration potential of a species at location x, and R(x) denotes the “ecological resistance” at that location. In this model, the R(x) value is a measure of the ecological suitability of the geographic unit. A higher resistance value indicates greater difficulty in migration for species. Conversely, it is easier for species to migrate. The flow of the current (species migration paths) conforms to the principle of minimum cumulative resistance, meaning that species will choose the path with the least resistance from several possible paths for migration.

To simulate this process more specifically, the shortest path algorithm from graph theory is further applied by the model to determine the optimal migration paths for species. With the landscape grid converted into a graph structure, the circuit model can be adopted to calculate not only the optimal path from the “source” to the “destination” for species, but also the impact of different landscape elements on species migration flow, such as forests, grasslands, water bodies, etc.

The ecological migration resistance mentioned in this paper is obtained from the aforementioned five habitat data based on their contribution rates. For instance, suitable forest types and warm, moist climatic conditions are usually associated with lower migration resistance, while the areas significantly affected by a variety of human activities, such as urbanization and agricultural lands, are associated with higher resistance values. Specifically, the formula used to calculate ecological resistance values is expressed as follows:

$$\mathrm{R}(\mathrm{x})={\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathsf{\omega }}_{\mathrm{i}}{\mathrm{f}}_{\mathrm{i}}\left(\mathrm{x}\right)$$

(2)

where R(x) represents the total ecological resistance at location x, fi(x) indicates the value of the ith environmental variable (such as land use type, vegetation type, etc.) at that location, and ωi denotes the weight of that environmental variable, signifying the relative contribution of different factors to the species migration resistance.

These resistance values are inputted into the circuit model for the construction of different species migration paths and ecological corridors. When the model is applied, the Linkage Mapper tool is used to simulate the flow of species as a current in the landscape for the calculation of suitable ecological corridor areas. Effectively connecting the “source” and “destination”, these areas minimize ecological resistance during species migration.

Finally, a line density model is used to identify human activity hotspots through over 1000 pedestrian outdoor activity tracks obtained from the “2bulu” website within the Shennongjia ecological area. They are overlaid with the species migration corridor network to determine the potential human activity conflict areas. Then, the Zonation5 software is applied to categorize these conflict areas by invasion risk, revealing that the top 50% of high-risk areas cover 70% of low-resistance areas in the potential conflict zones. Through the combination of potential distribution probability maps and sampling point distributions, the top 60% of areas are identified as the zones where “reinforcement” is required. According to the different requirements on biological migration corridor width, these areas are then expanded to varying widths of 200–1200 m [59], which leads to comprehensive results of the biodiversity protection evaluation.

2.3.2. Soil Conservation Evaluation Model

In terms of soil conservation evaluation, the RUSLE model is not applicable to mountainous terrains. Thus, this paper adopts the risk levels of soil conservation as determined by predicting the natural disasters triggered by short-duration heavy rainfall, such as flash floods and landslides.

Firstly, flash flood prediction is made in this paper by introducing the hydrological component of the SWAT model, and the SCS (Soil Conservation Service) curve number method is used to calculate surface runoff under storm conditions. Also, the precision of the CN (Curve Number) parameter is optimized to enhance the accuracy of model prediction, which allows the identification and simulation of potential flood areas in the Shennongjia ecological zone during storm conditions [60,61]. According to research comparing 1:2000 scale topographic maps and field surveys, a 300 threshold effectively identifies dry gullies in the Shennongjia area and delineates Hydrologic Response Units (HRUs). Therefore, ASTER GDEM elevation data are used to divide the four major watersheds in Shennongjia into approximately 1500 sub-watersheds. Then, surface flood volume prediction is made using the water balance equation (Equation (5)) in the SWAT model, where runoff occurs when surface precipitation exceeds loss amounts. The critical Qsurf uses the SCS Curve Equation (Equation (6)), and the CN value is required to calculate key parameters Ia and S (Equations (7) and (8)), such as the surface runoff curve number data optimized in this study. After the surface runoff volume is calculated, the drainage amount of the local channel is subtracted by allowing for an estimation of flood volume in each sub-watershed. Finally, the potential flood area is back-calculated for each sub-watershed by comparing the flood volume with the water capacity measured at different local elevations (Figure 6).

$$\mathrm{S}{\mathrm{W}}_{\mathrm{t}}=\mathrm{S}{\mathrm{W}}_{\mathrm{o}}+\sum _{\mathrm{i}=1}^{\mathrm{t}}({\mathrm{R}}_{\mathrm{d}\mathrm{a}\mathrm{y}}-{\mathrm{Q}}_{\mathrm{s}\mathrm{u}\mathrm{r}\mathrm{f}}-{\mathrm{E}}_{\mathrm{a}}-{\mathrm{W}}_{\mathrm{s}\mathrm{e}\mathrm{e}\mathrm{p}}-{\mathrm{Q}}_{\mathrm{g}\mathrm{w}})$$

(3)

In Formula (5), SWt represents the final soil moisture content transferred, SWo represents the soil moisture content in the early stage, t is the length of time, Rday is the precipitation on the i-th day, Qsurf is the surface runoff on the i-th day, Ea is the evaporation on the i-th day, Wseep is the infiltration and measured flow at the bottom of the soil profile on the i-th day, and Qgw is the groundwater content on the i-th day.

$$\mathrm{Q}={\displaystyle \frac{{(\mathrm{P}-{\mathrm{I}}_{\mathrm{a}})}^2}{\mathrm{P}-{\mathrm{I}}_{\mathrm{a}}+\mathrm{S}}\hspace{1em}\hspace{1em}(\mathrm{P}\geqq \mathrm{Ia})}$$

(4)

In Formula (6), Q represents actual runoff, P represents total rainfall, Ia represents initial rainfall loss, and S represents maximum potential infiltration before the rainfall in the region.

$$\mathrm{S}={\displaystyle \frac{25400}{\mathrm{C}\mathrm{N}}}-254$$

(5)

$$I_a=\lambda ({\displaystyle \frac{25400}{CN}}-254)$$

(6)

Secondly, landslide-prone areas are predicted, as landslides are the complex geological disasters triggered by sustained rainfall causing water to infiltrate soil layers and rock masses, ultimately leading to slip layer movement. In this paper, landslide susceptibility risk levels are mapped using the support vector machine–landslide susceptibility assessment model (SVM–LSM) within the ArcGIS tool, with sampling point data provided and factor selected [62,63,64]. The SVM–LSM model is particularly applicable to deal with multifactorial, nonlinear, and high-dimensional spatial data.

In this paper, 10 years’ worth of historical landslide information is collected from the Shennongjia area, with 405 risk areas identified where landslides occurred in previous years and the residential areas exceeding 4 square kilometers marked as 200 non-risk zones. Generated as the dependent variable dataset (risk areas as 1 and non-risk areas as 0), a 30 × 30 m grid point dataset is inputted into the SVM model for training. The SVM model searches for the optimal hyperplane to maximally separate the positive and negative samples (i.e., landslide risk points and non-risk points). Thus, the accuracy of the underlying classification is maintained. Subsequently, the relevant influencing factors and SVM model parameters are filtered using ROC (Receiver Operating Characteristic) curves and AUC (Area Under the Curve) values. When the model is trained, AUC values are analyzed to determine the most effective 11 influencing factors and optimal SVM classification parameters. Ultimately, the weight parameters of each factor are determined through these filtering processes, enabling the SVM-LSM model to identify landslide risk areas more precisely.

Mathematically, the SVM-LSM model is expressed as follows:

$$f\left(\mathrm{x}\right)={\mathsf{\omega }}^{\mathrm{T}}+\mathrm{b}$$

(7)

where f(x) is the decision function, representing the model’s output; x is the input data (including slope aspect, slope gradient, terrain curvature, soil type, precipitation, vegetation density, etc.); ω is the weight vector, determining the direction of the hyperplane; and b is the bias term, determining the position of the hyperplane.

According to the results of model training, the AUC value for the vegetation density factor performs as well as the basic terrain factors (such as slope gradient, slope aspect, and curvature), geological factors (such as soil type and rock type), and climatic factors (such as precipitation and temperature).

The final evaluation of soil conservation is an overlay of assessments for potential flood-prone areas and landslide-prone areas.

Lastly, the water yield module of the commonly used InVEST model is used to evaluate the ecological sensitivity of water conservation [65,66].

3. Results

In this paper, the optimized evaluation results are found to be more scientific and reliable by interpreting, comparing, and validating the data and evaluation model results before and after optimization.

3.1. Data Densification Results

As indicated by comparing the annual average rainfall data before and after optimization (Figure 7), the latter are more precise, fitting better with the terrain. In line with the division standards for low-mountain (500–1000 m), mid-mountain (1000–2000 m), and subalpine (>2000 m) rainfall zones, the statistics of the corresponding elevation rainfall data in both figures (see

Table 1. Comparison of annual average rainfall data before and after optimization.

Shennongjia Climate Zone Names and Elevation Ranges	Original Result		Optimized Result
	Average Rainfall Within the Interval	Range Value of Rainfall Within the Interval	Average Rainfall Within the Interval	Range Value of Rainfall Within the Interval
Low-mountain climate zone (500–1000 m)	918 mm	220.50 mm	699 mm	64.83 mm
Mid-mountain climate zone (1000–2000 m)	1091 mm	264.26 mm	1017 mm	62.46 mm
Subalpine climate zone (>2000 m)	1463 mm	193.51 mm	1888 mm	59.15 mm

) reveal that the optimized data vary more significantly in rainfall between different zones, with the average value changing to a greater extent, and data distribution is more concentrated (a smaller range of variation) and the data are more consistent with the actual rainfall conditions in Shennongjia.

As revealed by comparing the surface runoff coefficient data based on original land cover with the optimized data, the latter feature more varied spatial layers, which aligns more with local terrain and natural environmental characteristics. With forest density converted to surface runoff coefficients according to Niwat Ruangpanit’s table, the optimized data are tested for average values by intervals (

Table 2. Verification of optimized surface runoff coefficient data.

Vegetation Density Intervals	Research by Niwat Ruangpanit [67]	Optimized Data Study	Absolute Difference Ratio
	Corresponding Surface Runoff Coefficient Standards	Surface Runoff Coefficient Statistical Average
20–30%	19.33%	22.14%	14.5%
40–50%	15.16%	16.03%	5.7%
50–60%	11.35%	13.12%	15.6%
60–70%	7.36%	7.86%	6.8%
70–80%	4.88%	5.13%	5.1%
80–90%	2.22%	2.15%	3.2%

). The absolute difference ratio indicates a good overall fit between the two, particularly in the intervals with vegetation density above 0.7, confirming the reliability of the optimized data.

3.2. Model Improvement Results

As revealed by comparing the biodiversity protection model evaluation results, optimized in this paper, with the forest above-ground biomass data estimated by Yang Qiuli for the 30 m spatial resolution in China [68], the “source” prediction results obtained in this paper cover 65.5% of the local land area (approximately 1998 km²), encompassing 82.28% of the biomass estimated by the former (including the highest 30% of unit biomass areas), which illustrates the validity of this research (see Figure 8). The thresholds used to reclassify the multiple zones were derived from a peer-reviewed study [69].

According to the spatial verification of the SVM-LSM prediction made in the landslide analysis conducted for the soil conservation evaluation, over 85% of historical landslide disaster points are concentrated within high-risk areas, which suggests consistence between the evaluation results and local actual conditions.

3.3. Comprehensive Improvement Results

As clearly indicated by comparing the ecological sensitivity evaluation results before and after optimization (Figure 9), there are noticeable large “mosaic” blocks in the original evaluation images, which are difficult to couple with local terrain and features. In contrast, the optimized images exhibit higher data granularity and smoother transitions, which are more conducive to subsequent planning efforts. According to the analytical results, the original biodiversity protection and soil conservation evaluation models are incapable of identifying some ecologically sensitive spatial elements, which results in an excessively large number of low-protection spaces (red patches). This is not in alignment with the needs of national parks for more detailed and precise protection. The optimized models are effective at significantly improving these issues.

4. Discussion

In this paper, the ecological sensitivity evaluation of mountain-type national parks is conducted to propose an optimized evaluation scheme through data densification and model improvement. With the Shennongjia Forestry District as a case study, a detailed empirical analysis is conducted to provide a more scientific guidance on ecological protection and management.

According to the optimized evaluation results, there are extensive biodiversity conservation patches above 1000 m in elevation in the Shennongjia Forestry District from the perspective of biodiversity. A considerable proportion (about 40%) of them are distributed within the Shennongjia National Park, serving as habitats for numerous rare flora and fauna [70,71]. Therefore, it is crucial to further increase ecological management efforts in the national park and other high-altitude areas within the forestry district and improve protection mechanisms to maintain the health and stability of its ecosystem. Biodiversity conservation patches are relatively fragmented in the low mountain valley areas to the northeast and southwest, where human and agricultural activities are performed frequently [72,73]. A recommendation is that the species migration corridors identified in this paper can be constructed in these areas to enhance their interconnections, which not only reduces the risk of biodiversity loss but also enhances ecosystem stability [74,75].

In terms of soil conservation, the southwestern part of the Shennongjia Forestry District is the area with the highest risk of landslides, which is attributed to its complex geological structure and high rainfall [76,77]. In contrast, the landslide risks are relatively lower in the downstream valleys of the northeast, including Songbai, Yangri, and Xinhua. As emphasized in some research, the risk of flash floods must be considered due to the terrain and rainfall impacts despite the lower landslide risks of these valleys [78,79].

Upon the comprehensive overlay of ecological sensitivity evaluation results for the Shennongjia Forestry District, it can be found out that the areas with high, relatively high, and medium ecological sensitivity account for 25.09%, 44.51%, and 19.42% of the area, respectively, which means 89.01% in total. This demonstrates a high ecological protection value in most of the land in the Shennongjia area. The areas with high protection deserve strict protection, with any form of development activity prohibited. The areas of medium protection should be developed under limited conditions with strict environmental monitoring. The areas with low protection can be planned for urban and tourism development, but the basic principles of ecological protection must be adhered to. What is essential for the sustainable development of the local ecology and economy is to integrate ecological protection and economic development in the Shennongjia area in a harmonious way [12,80].

In the low-protection areas available for development, Songbai Town as the administrative center of the Shennongjia area is relatively large in scale but has limited space for expansion. Therefore, it is recommended that Yangri Town should be prioritized for future urban development. This development strategy can meet the needs of local economic development while offering protection for the ecological environment [81]. Additionally, it is noteworthy that Muyu Town and Jiuhu Township, located in the southern part of the Shennongjia Forestry District, are both popular tourist destinations that have been evaluated as having high or medium ecological protection. Thus, it is necessary to restrict the pace of development in these areas as appropriate. In actual development and construction, it is also necessary to protect the multispecies environment from irreversible damage by carefully selecting development sites through the combination of various ecological sensitivity assessments [82,83].

5. Conclusions

With its vast territory and complex natural conditions, China’s national parks are of much significance ecologically, exhibiting diversity in ecological conditions. Therefore, a significant precondition for the effective and high-quality protection of national park ecosystems is to explore more scientific, comprehensive, and precise ecological sensitivity evaluation data and models.

In this paper, Shennongjia Forestry District is taken as a case study to explore the national parks featuring mountainous terrain and forest landscapes. As revealed by optimizing both the data and models in the ecological sensitivity evaluation scheme, there is a significant improvement in the objectivity and precision of the final evaluation results. This provides a valuable reference for the evaluations of ecological sensitivity for other natural environments with similar characteristics.

In terms of data, the granularity of annual rainfall and surface runoff data is enhanced in this paper by introducing additional sampling points, optimizing data sources (such as using 30 m data types like ASTER GDEM and Landsat TM), incorporating regression models (including spatial Empirical Bayesian Kriging and non-spatial multivariate linear regression), and adjusting the corresponding processing procedures. Through these improvements, more refined data support is provided for subsequent model evaluations. In terms of models, species migration corridors are evaluated using the Linkage Mapper model and human conflict risks are evaluated using such models as Zonation. Moreover, more accurate and comprehensive evaluation results are obtained through a shift from focusing on soil conservation effects (the RUSLE) to simulating and predicting soil loss risks (using the SWAT model for flash floods and the SVM-LSM model for landslides).

The optimized evaluation framework provides a decision support tool for spatially adaptive management in mountain protected areas. By integrating multi-dimensional sensitivity zoning (e.g., biodiversity corridors, landslide risk gradients) with administrative boundaries, it enables dynamic zoning adjustments that balance conservation priorities with regional development needs. For instance, the identification of critical connectivity corridors between fragmented habitats could inform the design of transboundary ecological networks, extending protection beyond formal park boundaries. In low-sensitivity zones designated for limited development, the methodology supports site-specific tourism infrastructure planning—such as elevated walkways or seasonal access restrictions—to minimize habitat disruption while fostering eco-tourism.

Through the above research, this paper aims to provide technical support for the refinement of ecological sensitivity evaluation systems for mountain-type national parks and other critical ecological regions, which would promote the scientific management and utilization of ecological resources in the Shennongjia area. However, there remains room for further research. As mentioned earlier in the water conservation evaluation, an integrated three-dimensional hydrological model of both surface and sub-surface environments can be constructed if more detailed geological data are obtainable. Thus, the interactions between hydrological processes and ecosystems can be better elucidated. So, there are indeed implementation barriers that we may encounter while assessing this, including the following: Data Accessibility: High-resolution terrain and microclimate data required for model calibration are often restricted or inconsistently available across jurisdictions, particularly in transboundary mountain regions. Stakeholder Conflicts: Tourism developers and local communities may resist strict zoning in medium-sensitivity areas due to perceived economic constraints, necessitating complex negotiation. Climate Uncertainty: Increasing rainfall variability under climate change may alter landslide and flood risk patterns, requiring continuous model recalibration beyond initial implementation and cross-sectoral collaboration. The approach we provide transcends conventional static zoning by enabling adaptive governance, yet its effectiveness hinges on resolving multiscale data-policy discontinuities and securing cross-sectoral collaboration.