Evaluation of model prediction results
We evaluated the MaxEnt model by measuring the area under the receiving operator curve (ROC) (receiver operating characteristic curve) of the model AUC (area under the receiving operator curve)39. The AUC is one of the statistics built in MaxEnt that provides good statistics to judge model performance40. AUC values are generally between 0.5 and 1; the higher the value is, the better the model prediction41. When AUC < 0.7, it indicates poor model performance, 0.7–0.9 indicates moderate model performance, and > 0.9 indicates good model performance42. The model results showed a mean AUC of 0.985 for 10 repeated runs with a standard deviation of 0.011 (Fig. 2). This result indicates that the model has good performance and high prediction accuracy. The species distribution data and environmental variable data used in the modeling could effectively predict the potential suitable distribution areas of Emeia pseudosauteri in Zhejiang Province.
Contribution of environmental variables
We analyzed the importance and contribution of each environmental variable in the model prediction by the jackknife method in the MaxEnt prediction model. The results showed that the contributions of bio_4 (temperature seasonal variation) and Alt (altitude) were 33.5% and 30%, respectively, and the regularized training gain values were both greater than 0.75, which can provide a greater benefit to the prediction model. Thus, bio_4 and Alt had more useful information than the other environmental variables and had the greatest effect on the distribution of Emeia pseudosauteri in Zhejiang Province. Additionally, riv_dis (distance to rivers), bio_12 (annual precipitation) and LUCC (land use) were the three environmental variables with the next highest contribution and regularized training gain values and had some influence on the distribution of Emeia pseudosauteri in Zhejiang Province. The remaining environmental variables, including the NDVI, slope direction, slope, light, distance from roads, and distance from settlements, had a small or even negligible effect on the distribution of Emeia pseudosauteri in Zhejiang Province (Table 3, Fig. 3).
Table 3
Percent contribution of environmental variables in the MaxEnt model
Variable abb
|
Description
|
Contribution
|
Bio_4
|
Temperature seasonality
|
33.5%
|
Alt
|
Altitude
|
30%
|
Riv_dis
|
Distance to river
|
15.9%
|
LUCC
|
Land Use
|
7.2%
|
Bio_12
|
Annual precipitation
|
5.9%
|
NDVI
|
Normalized difference vegetation index
|
3.3%
|
Asp
|
Slope direction
|
1.9%
|
Slope
|
Slope
|
1.1%
|
light
|
Lighting
|
0.8%
|
Res_dis
Roa_dis
|
Distance to settlement
Distance to road
|
0.2%
0.2%
|
Response curve analysis of environmental variables
To explain the effect of each environmental variable on the distribution of suitable habitat, the MaxEnt model gives response curves between the probability of species presence and environmental variables43. The response curves for the main environmental variables affecting the distribution of Emeia pseudosauteri are shown below (Fig. 4). The seasonal change in the air temperature response curve showed an overall trend of increasing and then decreasing, with the probability of Emeia pseudosauteri reaching a maximum of 93% at 7.7°C. When the amount of seasonal change in air temperature was less than 7.7°C and more than 8°C, the presence probability decreased rapidly, while the curve changed slowly between 7.7°C and 8°C. Thus, Emeia pseudosauteri is more sensitive to seasonal temperature changes. In terms of altitude, the presence probability of Emeia pseudosauteri increased rapidly with increasing altitude, reaching an extreme value at 100 m, after which the presence probability gradually decreased with increasing altitude. This result suggests that a specific altitude can provide a unique and suitable habitat for Emeia pseudosauteri. The presence probability of Emeia pseudosauteri was negatively correlated with the distance from the river, with the presence probability decreasing as the distance from the river increased. The probability of Emeia pseudosauteri survival was most suitable when the annual precipitation was approximately 1500 mm, and the distribution probability reached an extreme value of 95%. Between 1000 mm and 1500 mm, the probability of presence gradually increased and then gradually decreased. The response curves of land use variables indicated that Emeia pseudosauteri habitat selection preferred mudflats, river canals, and shrubby woodlands. Collectively, Emeia pseudosauteri is highly dependent on wetland environments near rivers with high annual precipitation, and the dense vegetation of shrubby woodlands can provide a sheltered, safe, and stable habitat environment for Emeia pseudosauteri.
Prediction of suitable habitat for Emeia pseudosauteri in Zhejiang Province
According to the fitness index P output from MaxEnt software, the predicted fitness zones were classified into four levels: 0-0.08, nonsuitable zone; 0.08–0.32, low suitability zone; 0.32–0.64, medium suitability zone; and 0.64-1, high suitability zone. Thus, the potential suitable distribution areas of Emeia pseudosauteri in Zhejiang Province were obtained. The suitable area of Emeia pseudosauteri was mainly distributed in southern Zhejiang Province, and the area of highly suitable area was approximately 775.91 km², accounting for approximately 0.74% of the land area of Zhejiang Province; the area of moderately suitable area was approximately 1458.09 km², accounting for approximately 1.38%; and the area of less suitable area was approximately 3996.96 km², accounting for approximately 3.79%. The distribution of medium- and high-fitness zones of Emeia pseudosauteri was more scattered, and a continuous larger area of high-fitness zones was found only in Lishui city (Fig. 5).
Highly suitable areas: Daxi watershed in Lishui city, Longquan Creek and Jingshuitan watershed in Yunhe County, Songyin Creek watershed in Songyang County, parts of Longquan City, Xiaoxi and Qianxia Lake watersheds in Qingtian County and parts of the Oujiang River watershed from Qingtian County to Wenzhou City; Dannan Creek watershed in Yongjia County, Wenzhou; around Xianxia Lake watershed in Hunan Town, Quzhou city; around Qiandao Lake watershed in Chun'an County, Hangzhou city; around Changtan Reservoir watershed in Huangyan District, Taizhou city.
Middle fitness zone: The distribution of the middle fitness zone was basically the same as that of the high fitness zone, mainly extending outward to a certain extent with the high fitness zone at the center, including the area around Xianxia Lake waters in Lishui city, Dongdu Township in Jinyun County, Lanju Township in Longquan city, Xiakou Township in Quzhou city, Baita Township in Xianju County in Taizhou city, around Niutoushan Reservoir in Linhai city, and some areas in Wenling city, in addition to those in the high fitness zone mentioned above.