谱系聚类法在小区域粮食安全预测中的应用——以昆山市为例

The Application of Hierarchical Cluster Analysis to the Prediction of Grain Security of Small Research Areas—A Case Study of Kunshan

粮食安全对国民经济的可持续发展起着不可替代的基础性作用,小区域由于受政策因素的影响较大,粮食安全相关指标的变化呈一定阶段性,长时间序列的数学规律并不突出,不利于规划工作的展开。论文基于昆山市的研究,提出谱系聚类与数学模型相结合的基本思路,在此基础上推出了聚类结果有效性的量化判定标准并对聚类法运用准则做了深入的探讨。实际数据分析结果表明:昆山的粮食安全相关的社会经济指标变化确实呈明显阶段性;与利用全部时间序列数据建立的模型相比,运用谱系聚类的模型拟合和预测效果都有明显优势;至2015年,昆山市粮食自给率将下降至6%,最小人均耕地面积降低至0.022 hm2。通过进一步的分析、对比及讨论,文章认为,谱系聚类法运用于小区域粮食安全预测,方法可操作性强,结论科学性显著。

Grain security is fundamental to the sustainable development of our society and national economy. As research regions with small area are vulnerable to the impacts of policy changes, indexes related to grain security of these areas often change in the form of stages, which means that the mathematical regularity of long-term datasets is not significant. As a result, it is difficult to implement grain security programming for the future. We put forward a new method of combining hierarchical cluster analysis with traditional mathematical models, and established a quantification standard for the validity judgment of the clustering results. Meanwhile, a criterion for the using of hierarchical cluster analysis was also proposed, but we strongly recommended that mass data from other research areas are needed to calibrate and perfect it. Kunshan （1985 -2007） was chosen as a study region to prove the new method, because it is small in area but with rapid economic development. The results of analysis showed that ： the indexes related to grain seeurity did change in the form of stages, which were stage 1 from 1985 to 2001 and stage 2 from 2002 to 2006; new models based on hierarchical cluster analysis showed better results in simulating the grain security situation from 1985 to 2006 and smaller deviations in predicting it in 2007 than traditional models-the prediction deviations of total population, farmland acreage and grain productivity in 2007 were 3.4%, - 10. 9% and 5.6% with clustering and -25.7% , 96. 1% and 122.5% without clustering, while the determination coefficients of three different models were 0. 977, 0. 981 and 0. 914 with clustering and 0. 801, 0. 518 and 0. 426 without it; in 2015, Kunshan would have a decreased grain self-sufficient ratio of 6% and a decreased minimum fal-mland acreage per capita of 0. 022 hm2. After a comparison with works of other researchers（ on average, the minimum farmland acreage per capita of Zhejiang Province and East China were 0. 044 hm2 and 0. 075 hm2 respectively） and a subsequent analysis, we concluded that the new method of combining hierarchical cluster analysis with traditional mathematical models was both feasible and scientific. In the discussion, we thought that although such models with hierarchical cluster analysis can be applied to predict the grain security in the future, the predicting precision of the model is dependent on the duration of the recent stage to a great degree. However, the deficiency does not harm its function as a decision support implement for future grain security programming as well as town planning.