资源科学 ›› 2018, Vol. 40 ›› Issue (12): 2414-2424.doi: 10.18402/resci.2018.12.09

• 土地资源 • 上一篇    下一篇

基于BP神经网络的作物Cd含量预测及安全种植分区

侯艺璇1(), 赵华甫1,2(), 吴克宁1,2, 李凯1   

  1. 1. 中国地质大学(北京)土地科学技术学院,北京 100083
    2. 国土资源部土地整治重点实验室,北京 100035
  • 收稿日期:2017-11-29 修回日期:2018-10-19 出版日期:2018-12-20 发布日期:2018-12-10
  • 作者简介:

    作者简介:侯艺璇,女,河北邢台市人,硕士生,研究方向为土地资源利用与评价。E-mail: yxhou@cugb.edu.cn

  • 基金资助:
    国家重点研发计划(2017YFD0800305);国土资源部公益性行业专项经费项目(201511082)

Prediction of crop Cd content and zoning of safety planting based on BP neural network

Yixuan HOU1(), Huafu ZHAO1,2(), Kening WU1,2, Kai LI1   

  1. 1. School of Land Science and Technology, China University of Geosciences (Beijing), Beijing 100083, China
    2. Key Laboratory of Land Regulation Ministry of Land and Resources, Beijing 100035, China
  • Received:2017-11-29 Revised:2018-10-19 Online:2018-12-20 Published:2018-12-10

摘要:

为科学预测作物重金属含量,实现重金属污染农用地的安全利用,本研究利用地理探测器选取对作物Cd含量影响较大的土壤Cd含量、土壤pH值、与交通主干线的距离等10个因素为输入因子,农作物可食部分Cd含量作为输出因子,分别建立小麦、水稻、油菜籽、蔬菜可食部分Cd含量的BP神经网络预测模型,对作物种植污染情况进行预测分析,划分作物安全利用种植区。研究结果表明:① BP神经网络模型预测精度明显优于多元回归预测模型;② 对小麦、水稻和油菜籽的可食部分Cd含量预测结果进行评价,得到作物污染可能的空间分布及特征;③ 依据评价结果,对3种作物进行配置,划分得到4种作物适宜种植区,并提出管控策略。研究可为污染农用地的安全利用及作物种植调整提供思路和依据,兼具理论和现实意义。

关键词: 作物Cd含量, BP神经网络, 预测模型, 重金属污染, 农用地安全利用, 小麦, 水稻, 油菜籽, 蔬菜, 作物适宜种植区

Abstract:

In order to predict the heavy metal content of crops and realize the safe utilization of heavy metal contaminated agricultural land, 10 factors including soil Cd content, soil pH value, soil organic matter, cation exchange capacity, distance to mining land, and distance to traffic lines etc. which affect Cd content in crops were selected by GeoDetector model, being used as input factors. The Cd content in edible parts of crops was used as output factor. The BP neural network prediction models of Cd content in edible parts of wheat, rice, rapeseed, and vegetables were established respectively. The pollution of crop planting was predicted and analyzed. The safety zones of planting was categorized. The results illustrate that: (1) The BP neural network exhibit a good predictive effect, which has a nice applicability to the prediction of Cd content of edible parts of crops. The prediction accuracy of BP neural network models was better than that of the multiple regression prediction models according to the accuracy evaluation indexes. (2) Due to the low prediction accuracy of vegetables prediction model, the support for prediction and management partitions was limited. Only the prediction results of Cd content in edible parts of wheat, rice, and rapeseed were evaluated. The crop pollution prediction distribution map and distribution characteristics were obtained. (3) With the goal of safe use of agricultural land, the three crops in the study area were re-allocated based on the evaluation results, which was divided into four crop suitable planting safety zones. The findings of this study can provide ideas and basis for safe utilization of contaminated agricultural land and crop planting adjustment, which is of both theoretical and practical significance.

Key words: Cd content of crops, BP neural network, prediction models, heavy metal pollution, safety use of agricultural land, wheat, rice, rapeseed, vegetables, zoning of suitable planting for crops