Mapping the ecological vulnerability status quo area is one of the most useful methods for making appropriate ecological management plans and mitigating ecological damage decisions (Xu K. et al. 2020). It is indeed necessary to assess the ecological vulnerability of karst areas, because it is used as a spatial data to determine the location of relatively high ecological vulnerability, and can effectively and specifically plan, design and protect the area. Using IGR to screen the impact factors before model training can maximize the efficiency of vulnerability assessment model (Yi Y. et al. 2020). Finally, IGR was applied to select 16 impact factors, and these impact factors were reasonably considered in several studies above (Eakin et al., Yu et al. 2020, Wang F, et al. 2021, Berrouet et al. 2018). Using these influencing factors, this study successfully explored the possibility of using a variety of machine learning models to evaluate the ecological vulnerability of Chongqing. The model accuracy verification results AUC (DNN) = 0.89, AUC (CNN) = 0.93, indicating that the accuracy of the two models is better, and the CNN model is better than the DNN. The results of MAE and RSME can also prove that the evaluation accuracy of the two models is comparable and can accurately assess the vulnerability of the study area. In addition, as a recent research, machine learning technology has higher accuracy than traditional method models in the field of vulnerability modeling (Han J. et al. 2019, Fan J. et al. 2019), and in nonlinear models, massive data processing capabilities are superior to other data-driven models (Dodangeh et al. 2019, Vakhshoori et al. 2019). In this study, CNN is undoubtedly the best model to evaluate the results. Its advantage is that it takes into account all domain information and can determine the manifold stage of representation from the input data (Zhang L. et al. 2019). It uses multiple factors and identifies internal elements to maintain pixel association (Wiegand et al. 2013).
In addition to the exploration of model methods, the evaluation results also expressed the relative spatial distribution of ecological vulnerability in the study area. A considerable number of areas in the study area suffer from high-vulnerability compared to other study areas. From the spatial distribution, it can be seen that the high-vulnerability areas are generally located in the northeast and southeast corners of the study area, the mountain valleys and urban areas in the west of the central region.
The reason for the high vulnerability of the northeastern corner of the study area is mainly determined by its own geological and geomorphological conditions and human activities. The area is mainly karst landform, which is affected by geological processes. The overall dip angle of the stratum is large, forming peak clusters, isolated peaks and other landforms, multi-gorges, cliffs and steep slopes. In the field survey, it can be found that the vegetation in this area can only grow in hard limestone fissures. The soil development on the surface is low, and the soil is mostly concentrated between rock fissures (Fig. 11), so the growth environment of vegetation is poor. In the past, the area has been reclaimed on a large scale (Wang Y et al. 2022), and trees have been cut down until the government began to control and protect the ecological environment of the area. In the next 20 years, trees and habitats have been restored, but the degree of restoration is far less than before, which can also be used as evidence of the degree of ecological vulnerability in the area. Although the forest area in this area is now large, according to recent forest survey literature, the quality of forests in the region is not high, and forest savings per kilometer are below the national average. Its vulnerability therefore persists and, together with the effects of human activities on it, must be higher than in other parts of the study area.
The reason for the high vulnerability in the southeast corner of the study area is the same as that in the northeast corner, and it is also karst landform, but the difference is that the elevation fluctuation in the area is not as large as that in the northeast corner, and it is greatly affected by human activities. The farming methods of the main residents in this area are relatively backward. The result of adopting backward farming methods in special landforms is that the local ecological environment has been seriously damaged, and soil erosion is serious. After years of governance and protection, the ecological environment has been fundamentally improved. In the field survey, artificial cultivation has led to a large number of trees being cut down. Although the state has implemented the policy of returning farmland to forest in recent years, the trees in the area of returning farmland to forest grow slowly or even do not grow. In Fig. 12a, the pine trees planted by aerial seeding grow in the limestone gap when the trees are ecological restoration. In Fig. 12b, local residents cultivate and plant crops in the soil between rocks on the slopes of karst landforms. Serious rocky desertification and soil erosion problems have always existed in such areas, and it is difficult to restore vegetation once it is destroyed, so there is also a high ecological vulnerability.
In addition, the limestone in the karst area can be used as the basic material for building use. Therefore, there are many quarries in the northeastern and southeastern corners of the study area, some are being mined and some are stopped and abandoned. Human activities for direct mining of rocks have greatly increased the pressure on the ecological environment in these areas, and the damage to the ecological environment is irreparable.
The intermountain valleys in the western central part of the study area are also areas with relatively high ecological vulnerability in Chongqing. The spatial characteristics of the area are distributed in valleys between two parallel mountains, which are mainly affected by human activities. Chongqing is a major grain producing area in China, and this area is an excellent place for farming, so the mountain valleys in the central and western Chongqing are full of cultivated farmland Fig. 13. In the past, the farmland on the slope was directly cultivated on the slope. After the ecological management of the government, it was changed to terraced fields and stopped reclamation, which greatly reduced the soil erosion problem and the destruction of the ecosystem in the area. Although a series of ecological restoration and protection measures such as slope to terrace and returning farmland to forest have been taken, but high ecological vulnerability still exists.
Currently, the pressure on ecosystems worldwide is increasing, and the main influencing factors of this phenomenon are attributed to climate change and human activities (Milton et al. 2018). For the study area, in addition to the two main influencing factors, geological formation factors also play a major role, which can be seen from the results of RF machine learning. In this paper, the two machine learning models for the judgment of the impact factors are the most important impact factors are geological formation, indicating that the karst landform is the main ecological vulnerability impact factor in the study area. The second is vegetation coverage, which is also in line with general cognition. Land use type also occupies a major position, it can reflect the transformation and impact of human activities on the environment, and land use types in Chongqing, the largest proportion of arable land, also reflects the above status of vulnerability. In short, the evaluation effect of the machine learning model established for the study area is excellent and reliable, which is consistent with the cognitive situation of our field investigation. However, the machine learning model in this paper can only be applied to the study area, and cannot be used to evaluate other areas, and the universality is not high.