Land Suitability Assessment Based On Machine Learning And Its Application

Research Objective: Land resources are the foundation of socio-economic development and human development,the most basic guarantee for human survival activities,and the foundation for realizing modernization.In the land spatial planning system,land suitability evaluation is the basis for the optimal allocation of land resources.Therefore,in the process of establishing and developing the land space system,the land suitability assessment should also keep up with the pace and update and adjust in time.The traditional land suitability assessment method is expensive,cumbersome and inefficient,so the land suitability assessment method needs to be innovated to improve efficiency and combine with practice.Research method: Machine learning algorithms can obtain classification results of new data after entering new data by computational modeling of past data,learning the classification experience in past data.Land suitability evaluation conforms to the definition of classification problems in machine learning,so machine learning can have strong adaptability in land suitability evaluation.Machine learning models can be established through the model method to learn from the past land suitability evaluation experience,so as to obtain the latest land suitability evaluation according to the data in the new period.Results: In this paper,Wuxianqi,Ordos City was selected as the research area,and a variety of machine learning land suitability evaluation models were constructed.The land suitability evaluation work is carried out by machine learning model,and the performance data of 32 machine learning classifier models under different sampling methods and classification strategies are obtained.The data of the new period are entered into the three models with the best performance,so as to obtain the urban suitability evaluation results and the quantitative area statistics of the new period.Conclusions:(1)For the suitability of urban land in the new era,the distance from other grassland,rural roads and railway land is an important indicator affecting the suitability evaluation of Wuxianqi town;From the perspective of the change trend,the suitability land range of Wuxianqi level 3 towns showed an increasing trend.From the perspective of spatial distribution,the distribution of urban suitability land is uneven,with high land suitability evaluation grades in the north and central parts of Wuxianqi and low land suitability evaluation grades in the south.(2)It is suggested that when constructing a new land suitability evaluation model based on machine learning,the sampling method should be random sampling,the classification strategy should be selected as the ovr strategy,and the Random Forest Classifier model,Gradient Boosting Classifier model and Nu SVC model should be preferred.(3)For the selection of land suitability evaluation results obtained by different machine learning models,in addition to the performance as the selection criterion in previous research,a new selection criterion is proposed,and it is suggested that in the model with little difference in performance,the selection is made by analyzing the feature importance of the machine model combined with the requirements of normative standards or expert experience,so as to ensure the efficiency of land suitability evaluation,and can also better select land adaptation evaluation results based on the actual situation.