Layout Optimization Of Rural Residential Land In Mountainous Areas Based On Machine Learning Evaluation And Function Zoning Guidance

China is covered with a wide range of mountains,which account for a high proportion of land area.Traditionally,mountainous counties in China have been in a relative weak position of the economic and social development,which is due to the limited resources and environment conditions.In recent years,upgrading of economic development mode have promoted the inter-regional flow of industries and production capacities,with the additional stimulus of policy support and fiscal transfer payment,rural economic and social development in mountainous areas accelerated step by step,patterns of population and resource between urban and rural areas have been continuously changing,a large number of rural labor forces have been transferred to non-agricultural sectors,it has caused the continuous outflow of rural population,leaving the rural residential land abandoned.Industrialization and urbanization in mountainous areas has produced a strong demand for construction land,making the allocation of mountainous land resources between urban and rural areas tend to be unreasonable.Restricted by the terrain and other factors in mountainous areas,the spatial layout of rural residential land is scattered and the coverage of infrastructure is limited,which makes the rural living environment worse.At the same time,the development of residential land is extensive,per capita land area is too large,bringing serious waste of land and low efficiency of its utilization.Under this background,the scale and layout of rural residential land needs to be further adjusted.Therefore,the study on the layout optimization of rural residential land should be carried out to improve the intensive and economical land use and the rural living environment,which is of great practical significance for the coordinated development of urban and rural areas,Rural Revitalization and ecological civilization in mountainous areas.Based on this,this study focuses on optimizing the layout of rural residential land in mountainous areas.It first combs the studies on the layout optimization of rural residential land and its related areas in the domestic or foreign publications,summarizes the research progress and shortcomings,and introduces the main concepts and basic theories involved.Then,it analyses the characteristics of the development mountainous residential land,compares the existing layout optimization models,defines the core ideas of layout optimization of rural residential land,and probes into the supporting ways of machine learning and function zoning in this optimization.Finally,it takes Fang County as an example to carry out empirical research and put forward some related policy recommendations.The main conclusions of this study are as follows:(1)The connotation of the layout optimization of rural residential land in mountainous areas was enriched.The new optimization model is developed.Due to the limitations of natural environment,resource endowment and economic development in mountainous areas,the existing optimization model is facing geographical,economic and management constraints.Considering the reality of land use,the optimization work should fit for Rural Revitalization and other development goals,follow the guidance of the main function zoning,and implement the reuse or transition of all the residential lands involved.Based on these new connotations,this study proposes a new model for optimizing the layout of rural residential land,which mainly includes stabilizing the long-term use of suitable one,scientifically converting the non-suitable one to another use,and adjusting the rights and interests between people and the land involved in the optimization process.(2)The suitability evaluation of rural residential land use in mountainous area is carried out based on machine learning techniques.The recognition of residential land with reservation and conversion expectations is achieved.Using the object-based analysis,the fusion of evaluation process and machine learning techniques is realized.The evaluation index system is constructed from three aspects: land endowment,agricultural production conditions,and living and development conditions.Three typical machine learning algorithms: Random Forest,Support Vector Machine and Naive Bayes are used.Through parameter optimization,cross-validation and other methods,the spatial positions of suitable and non-suitable rural residential land parcels in Fang County are comprehensively obtained,and the factor importance function is used to analyze the influence degree of the main factors.The results show that Random Forest performs the best;accessibility,altitude and size of residential land are the most important factors affecting the suitability of residential land;the majority of non-suitable residential lands are concentrated in the northeastern and the central-northern and central-eastern areas of Fang County,with a large number;the change of residential land use is expected to be more intense.(3)The township-level land use function zoning of Shiyan City is carried out to guide the development of rural residential land.Considering the factors of natural/ecological carrying capacity,economic development competitiveness and future development potential,the evaluation system of land use function zoning at the township level is established.Taking Shiyan city as a whole,the synthesis of different zoning methods is mainly supported by manual revision,and four types of zones are distinguished: key development zone,optimized development zone,restricted development zone and prohibited development zone.The results show that the key development zones are concentrated in the north,while the optimized development zones are concentrated in the central part that Chengguan locates on and the surrounding towns that spreads from East to the west.A small number of towns in the south are classified as restricted development zones,while Shangkan and Wu Taishan are classified into prohibited development zones.Combining with the main function orientation of different townships,guiding suggestions on the use or conversion of regional rural residential areas are put forward.(4)Scheme of layout optimization is made and its contributions to the expected adjustment of land use planning are estimatedReferring the results of machine learning evaluation and functional zoning,and the predictions of land use structure in the year 2020,556.32 hectares of suitable rural residential lands(4535.59 hectares in total)are designated as reserve lands for further conversion.The remaining ones are for the long-term use.2012.23 hectares of residential lands are identified as the non-suitable type that to be converted.Combining the reserve part and the to-be-converted part can accomplish 100% of the total amount of residential lands needed to be converted in the target year.The to-be-converted land index can solve 86.71% of the source of new construction land,and provides a total of 1155.34 mu of new-built land for cultivation or gardening.Then the policy recommendations and implementation paths for this optimization are discussed.The main innovations of the study are:(1)It puts forward a new mode for optimizing the layout of rural residential land in mountainous areas.The connotations of optimizing the layout of rural residential land in mountainous areas are expounded: it should fit for Rural Revitalization and other development goals,follow the guidance of land spatial planning,and implement the use or conversion of all the residential lands involved.Based on this,a new optimization model is proposed,the core ideas are: stabilizing the long-term use of suitable one,scientifically converting the non-suitable one to another use,and adjusting the rights and interests between people and the land involved in the optimization process.It enriches the research system of layout optimization of rural residential land,and provides a good reference for carrying out the layout optimization in mountainous areas.(2)It introduces machine learning techniques into the suitability evaluation of rural residential land in mountainous areas.Machine learning technique can analyze a large number of unknown data through learning a small number of sampled data.Sample’s feature-driven analysis makes land use suitability evaluation implemented at the land parcel level,which is conducive to the more elaborate layout optimization of land use.The study proves that the application of machine learning method can make up for the shortcomings of traditional methods and index system in the field of evaluation of rural residential land use,the combination of different methods can improve the credibility of the evaluation results and bring the rich layerings of it.This study enriches the method system of land use evaluation and provides a new idea for optimizing the process of land use evaluation.(3)It estimates the contributions of the optimization results to the adjustments of planned land use structure,and expands the practical significance of the optimization.The connection between the results of layout optimizing and regional land use planning is discussed.The land use structure in the target year is forecasted.According to the use or conversion arrangement of the reserved and to-be-converted residential land in the final layout optimization scheme,their contributions to the expected adjustments of various types of land use are estimated.It expands the practical significance of optimizing the layout of rural residential land to the intensive and economical use of land resources in mountainous areas and the integration of urban and rural development.