<div>2.3. Construction of the species distribution model</div><div>In this research, the Maxent model was used to predict the species
suitability distribution of <i>L. japonicus</i> in three periods. A
total of 489 distribution sample data points and 12 environmental
factors were imported into Maxent 3.4.4 software for modeling. We used
75% of the distribution samples randomly chosen as training data for
modeling, and 25% of the distribution samples were used as test data to
evaluate the model’s ability. Bootstrap replicates were set to 10, and
the distribution value was output in the logistic form. The weight of
each environmental factor was evaluated by the knife-cut analysis
method, and the dominant environmental factor was determined by
combining the percentage contribution of each environmental factor and
the replacement importance value.</div><div>The knife-cut analysis method represents the importance of the
explanatory variables based on the arrangement, reveals the importance
of the variables, and is a highly reliable way for assessing a model.
The accuracy value of the AUC-ROC of the test data represents the model
fitting degree and indicates the prediction reliability of the model
(Khan et al., 2022). The evaluation result of the AUC is a value in the
range of 0.5–1.0; the greater the AUC value, the more precise the
prediction. A result in the range 0.5–0.6 indicates that the simulation
result is unqualified; 0.6–0.7 indicates poor simulation results;
0.7–0.8 indicates the general simulation results; 0.8–0.9 indicates
that the simulation results are good; 0.9–1.0 indicates the good
prediction effect (Ouyang et al., 2019; Wang et al., 2018; Wang et al.,
2021).</div><div>2.4 Optimization of the model</div><div>Referring to Robert Muscarella’s latest optimization method, the
Checkerboard2 method was used to divide the study area into four bins.
This masked geographic structure method can better adjust the model
regularization level. The Maxent model regularization level consists of
two parameters, the regularization multiplication value (RM) and the
feature combination (FC) optimized by calling the ENMeval packet in the
R language. The Maxent model provides five features: a linear feature
(L), a quadratic feature (Q), a fragmented feature (H), a multiplicative
feature (P), and a threshold feature (T). In this research, the default
parameters of Maxent software were RM = 1.5 and FC = H; to optimize the
Maxent model, the RM was set to 0.5–4; and each increase was by 0.5.
There were a total of eight control frequency doublings, and six
combinations with one or more features were used: L; L and Q; H; L, Q
and H; L, Q, H and P; and L, Q, H, P and T. According to the
permutations and combinations, 48
parameter combinations were calculated. The ENMeval data package uses
the above 48 combinations of parameters to test the complexity of the
model based on the delta AICc value and a 10% test miss rate, where the
lower the value, the more accurate the model prediction (Phillips et
al., 2017; Zhao et al., 2021).</div><div>2.5. Data processing</div><div>ArcGis10.4.1 software was used to divide and visualize the suitability
of <i>Leonurus japonicus</i> . The suitability threshold of<i>Leonurus japonicus</i> was predicted based on the Maxent model. The
natural breakpoint method was used to derive the habitat suitability
index of <i>L. japonicus</i> . The minimum level of threshold
establishment was 0.47, and the distribution below this level was
excluded. Therefore, the habitat suitability grade of <i>L.
japonicus</i> was divided into unsuitable area (0–0.47), low suitable area
(0.47–0.55), moderate suitable area (0.55–0.65), and most suitable
area (0.65–1). We compared the differences in the suitable areas of<i>Leonurus japonicus</i> in different periods to obtain the change map
of the spatial distribution pattern of <i>Leonurus japonicus</i> under
future climate change scenarios. The SDMTool data package in the R
language was used to calculate the centroid position of the suitable
area of <i>L. japonicus</i> under six different economic paths in the
current and future periods, and the migration direction of the spatial
distribution of the suitable area of <i>L. japonicus</i> was reflected
by the change in the centroid position. The geosphere data package in
the R language was used to calculate the centroid migration distance of<i>Leonurus japonicus</i> under different economic scenarios. The ArcGis
overlay tool was used to overlay the current and future suitable area
distribution data layers. The tool can merge different raster data into
one output to define the specified set, reclassify the new layer, and
divide the suitable level to obtain the final suitable zoning map of<i>Leonurus japonicus</i> . In this study, three suitable grades of low
suitable area, medium suitable area, and highly suitable area were taken
as the total suitable area; the values of dominant environmental
variables and the suitability of different periods were extracted from
489 distribution points to analyze the relationship between
environmental factors and potential distribution areas.</div><div>2.6. Niche differentiation analysis and priority reserve analysis</div><div>This study quantitatively analyzed niche differentiation in current and
future periods. Based on the natural distribution points and
environmental factor layers of the current period, the ecospat package
for the R language was used to analyze and visualize niche
differentiation, and the niche parameter D that ranges from 0 to 1 was
calculated, indicating niches that are non-overlapping to fully
overlapping (Warren et al., 2008; Yan et al., 2021). The software
ENMTools v1.4 was used to calculate the niche breadth in the
geographical and environmental space of each period as the average
Levins B1 (inverse concentration) and B2 (uncertainty) values of the
habitat suitability map for each period (Warren et al., 2008; Levins et
al., 1968). Levins B1 and B2 values range from 0 to 1, with values close
to 0 representing a narrow niche breadth and values close to 1
representing a wide niche breadth (Yuan et al., 2021).</div><div>The priority reserve is a protected area with a relatively large
designated suitability index value. The main feature of this area is
that the number of national key protected plant species is relatively
large and less disturbed by human activities. Marxan shows wonderful
performance in identifying priority protected areas, and it has been
applied to system protection planning (Da Luz Fernandes et al., 2018;
Carrasco et al., 2020; He et al., 2021). Based on the analysis results
of the Maxent model, the main environmental factors selected were used
for Marxan modeling. According to the distribution of the target
species, the locations of the protected areas were determined to achieve
the purpose of biodiversity conservation (Zhang et al., 2021).</div><div>3. Results and analysis</div><div>3.1. Model optimization and accuracy evaluation</div><div>Based on 489 distribution points, 12 layers of environmental variables,
and the AIC information criterion, the Maxent model was used to simulate
and predict the potential distribution area of <i>L. japonicus</i> .
Under the default parameter settings of Maxent, the magnification RM =
1, the feature combination FC = LQHPT, and delta AICc = 6.99. When RM =
1.5, FC = H, and delta AICc = 0, the model was optimal,
and the 10% training omission
rate was 13.48% lower than that of the model under the default
parameters (Table 2).</div><div>Therefore, the control frequency doubling RM = 1.5 and the feature
combination FC = H were selected as the final parameters of the model.
The AUC value of the simulation training under these parameters was
0.830. The standard deviation was 0.006 (Figure 2), indicating that the
prediction results were accurate.</div><div>Table 2 Evaluation results of the Maxent model under different parameter
settings</div>