地球信息科学学报 ›› 2017, Vol. 19 ›› Issue (7): 915-923.doi: 10.3724/SP.J.1047.2017.00915

• 地理空间分析综合应用 • 上一篇    下一篇

基于深度学习的区域生态安全时空模拟与预测

陈工1(), 李琦1,2,*(), 金玲艳1, 梁贺明1, 莫玉琴1   

  1. 1. 北京大学 数字地球工作室,北京 100871
    2. 北京大学 智慧城市研究中心,北京 100871
  • 收稿日期:2016-09-28 修回日期:2017-04-14 出版日期:2017-07-10 发布日期:2017-07-10
  • 通讯作者: 李琦 E-mail:gong.chen@pku.edu.cn;qi.lee009@gmail.com
  • 作者简介:

    作者简介:陈 工(1989-),男,湖北荆门人,博士生,研究方向为遥感、GIS和人工智能。E-mail:gong.chen@pku.edu.cn

  • 基金资助:
    国家科技支撑计划项目(2012BAC20B06)

Spatio-temporal Simulation and Prediction of Regional Ecological Security Based on Deep Learning

CHEN Gong1(), LI Qi1,2,*(), JIN Lingyan1, LIANG Heming1, Hamed Karimian1, MO Yuqin1   

  1. 1. Digital Earth Studio, Peking University, Beijing 100871, China
    2. Smart City Research Center, Peking University, Beijing 100871, China
  • Received:2016-09-28 Revised:2017-04-14 Online:2017-07-10 Published:2017-07-10
  • Contact: LI Qi E-mail:gong.chen@pku.edu.cn;qi.lee009@gmail.com

摘要:

区域作为人类、自然、社会共同作用和互相影响的复杂系统,对区域进行生态量化建模与模拟仿真,是实现区域可持续发展战略的关键。传统机器学习方法对区域生态系统建模取得了一定的成果,但难以确定学习特征和实现时空模拟。深度学习不需事先确定训练特征,具有优异的特征学习能力,能够提高模型预测精度,因此利用深度学习进行建模具有显著优势。本文使用植被净初级生产力(NPP)、气溶胶光学厚度(AOD)和人口格网数据,充分利用深度学习的优点,采用最优深度神经网络时空模拟,得到了河南省2007-2014年3 km分辨率的生态赤字空间分布图和河南省2015-2020年的生态赤字时间预测结果并进行分析,为区域生态的科学管理和建设供科学依据和参考。

关键词: 深度学习, 机器学习, 生态承载力, 生态赤字, 时空模拟

Abstract:

A region is a complex system of human, nature and society. The quantitative modeling and simulation of the ecology of the region are the key to realize the strategy of regional sustainable development. Traditional methods of machine learning have made some achievements in the modeling of regional ecosystems, but it is difficult to determine the learning characteristics and realize the simulation of time and space. Deep learning does not need to determine the training characteristics and has excellent feature learning ability and higher accuracy of model prediction. In this paper, we used the net primary productivity (NPP), aerosol optical thickness (AOD) and population grid data to make full use of the advantages of depth learning. The optimal deep neural network is used to simulate the spatial and temporal patterns of Henan Province. The spatial distribution map of ecological deficit and the forecast of ecological deficit in Henan province from 2015 to 2020 are generated and analyzed. Our work provides relevant basic scientific support and reference for the scientific management and construction of regional ecology.

Key words: deep learning, machine learning, ecological capacity, ecological deficit, spatio-temporal simulation