中国农业气象 ›› 2021, Vol. 42 ›› Issue (09): 775-787.doi: 10.3969/j.issn.1000-6362.2021.09.005

• 农业气象信息技术 栏目 • 上一篇    下一篇

基于深度学习算法的京津冀地区综合干旱评估模型构建

胡小枫,王冬利,赵安周,刘宪锋,王金杰   

  1. 1.河北工程大学矿业与测绘工程学院,邯郸 056038;2.邯郸市自然资源空间信息重点实验室,邯郸 056038;3. 中国科学院地理科学与资源研究所资源与环境信息系统国家重点实验室,北京 100101;4. 陕西师范大学地理科学与旅游学院,西安 710119
  • 收稿日期:2021-01-07 出版日期:2021-09-20 发布日期:2021-09-11
  • 通讯作者: 赵安周,博士,副教授,硕士生导师,主要从事城市扩张对生态环境影响等方面研究,E-mail: zhaoanzhou@126.com E-mail: zhaoanzhou@126.com
  • 作者简介:胡小枫,E-mail: 729420832@qq.com
  • 基金资助:
    河北省普通高等学校青年拔尖人才计划资助项目(BJ2018043);国家自然科学基金资助项目(42071246;42171212);2021年河北省硕士研究生创新资助项目(CXZZSS2021089)

A Comprehensive Drought Evaluation Model in Beijing-Tianjin-Hebei Region Based on Deep Learning Algorithm

HU Xiao-feng, WANG Dong-li, ZHAO An-zhou, LIU Xian-feng, WANG Jin-jie   

  1. 1. School of Mining and Geomatics, Hebei University of Engineering, Handan 056038, China; 2. Key Laboratory of Natural Resources and Spatial Information, Handan 056038; 3. Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Science, Beijing 100101; 4. School of Geography and Tourism, Shaanxi Normal University, Xi,an 710119
  • Received:2021-01-07 Online:2021-09-20 Published:2021-09-11

摘要: 及时准确的干旱评估对社会经济发展和农业生产具有重要的指导意义,当前的干旱评估指标通常仅考虑植被或降水等单方面影响因素,在实际干旱评估中存在一定的局限性。本研究综合考虑降水、温度、地形等多个干旱致灾因子,以主要产粮基地京津冀地区为例,基于2007-2017年地表温度(Land Surface Temperature,LST)、归一化植被指数(Normalized Difference Vegetation Index,NDVI)以及降水等多源数据,利用深度学习框架Tensorflow构建以标准化降水蒸散发指数(Standardized Precipitation Evapotranspiration Index,SPEI)为目标值的综合干旱评估模型。利用决定系数(R2)和均方根误差(RMSE)对模型进行测试;利用站点标准化降水指数(Standardized Precipitation Index,SPI)、土壤相对湿度数据以及2016年京津冀地区的气象灾害数据,从时间和空间上对模型的可靠性进行验证。结果表明:模型的训练集和测试集在不同月份上均表现较好(R2均大于0.5而RMSE均小于0.55)。模型输出的综合干旱指数(Comprehensive Drought Index,CDI)在密云站上与SPI和SPEI接近,变化趋势基本一致,并且与站点SPI和土壤相对湿度的相关系数分别大于0.7和0.4,均通过了0.01水平的显著性检验。空间上,相较于SPEI,CDI计算的2016年3-7月京津冀地区干旱事件结果与实际情况符合度更高,表明该模型适用于京津冀地区干旱评估。

关键词: 综合干旱评估模型, 深度学习, 多源数据, 京津冀地区, 综合干旱指数

Abstract:

Drought is the most common natural hazard in the Beijing-Tianjin-Hebei region. Timely and accurate drought evaluation is crucial for socio-economic development and agricultural production. Unilateral factors such as vegetation or precipitation are usually only considered in current drought assessment, which have some limitations in actual drought evaluation. In this study, multiple drought-causing factors such as precipitation, temperature, soil and terrain were considered comprehensively. Land surface temperature (LST), normalized difference vegetation index (NDVI), precipitation and other multi-source data from 2007 to 2017 were used to construct a comprehensive drought evaluation model with standardized precipitation evapotranspiration index (SPEI) as the target value under Tensorflow frame in the Beijing-Tianjin-Hebei region, Chinese main grain production base. Determination coefficient (R2) and root mean square error (RMSE) were used to evaluate model accuracy. The station standardized precipitation index (SPI), soil relative moisture data and meteorological disaster data for the Beijing-Tianjin-Hebei region in 2016 were used to validate the accuracy of model in time and space. The results showed that model in training and test sets had different performance in various months, with R2 both greater than 0.5 and RMSE less than 0.55. Comprehensive drought evaluation model had the best performance in November. Comprehensive drought index (CDI) output from the model was close to SPI and SPEI at Miyun station, and the change trend was basically consistent. The correlation coefficients between the CDI of the model and SPI, relative soil moisture at a 10cm depth were greater than 0.7 and 0.4 respectively. Both of them passed 0.01 significance level test. Spatially, compared with SPEI, the results of drought events in Beijing-Tianjin-Hebei region from March to July in 2016 calculated by CDI were more consistent with actual situation, which indicated that the comprehensive evaluation model was applicable for drought monitoring in Beijing-Tianjin-Hebei region. 

Key words: Comprehensive drought evaluation model, Deep learning, Multi-source data, Beijing-Tianjin-Hebei region, Comprehensive drought index (CDI)