欢迎访问《草业学报》官方网站,今天是 分享到:

草业学报 ›› 2022, Vol. 31 ›› Issue (8): 13-23.DOI: 10.11686/cyxb2021481

• 研究论文 • 上一篇    下一篇

基于机器学习的阿勒泰地区草地地下生物量估测与数字制图

厉方桢1,2,3(), 钟华平2, 欧阳克蕙1, 赵小敏1,3(), 李愈哲2()   

  1. 1.江西农业大学,江西 南昌 330045
    2.中国科学院地理科学与资源研究所,北京 100101
    3.江西省鄱阳湖流域农业资源与生态重点实验室,江西 南昌 330045
  • 收稿日期:2021-12-23 修回日期:2022-02-28 出版日期:2022-08-20 发布日期:2022-07-01
  • 通讯作者: 赵小敏,李愈哲
  • 作者简介:E-mail: liyuzhe@igsnrr.ac.cn
    E-mail: zhaoxm889@126.com
    厉方桢(1993-),男,浙江嵊州人,在读博士。E-mail: lifzhen@163.com
  • 基金资助:
    国家重点研发计划(2020YFD1100603-02);国家科技基础性工作专项(2013FY110900);国家自然科学基金项目(41971276);江西省研究生创新专项资金(YC2020-B089)

Estimation and digital mapping of grassland belowground biomass in the Altay region, China, based on machine learning

Fang-zhen LI1,2,3(), Hua-ping ZHONG2, Ke-hui OUYANG1, Xiao-min ZHAO1,3(), Yu-zhe LI2()   

  1. 1.Jiangxi Agricultural University,Nanchang 330045,China
    2.Institute of Geographic Sciences and Natural Resources Research,Chinese Academy of Sciences,Beijing 100101,China
    3.Key Laboratory of Poyang Lake Watershed Agricultural Resources and Ecology of Jiangxi Province,Nanchang 330045,China
  • Received:2021-12-23 Revised:2022-02-28 Online:2022-08-20 Published:2022-07-01
  • Contact: Xiao-min ZHAO,Yu-zhe LI

摘要:

为精准估算草地地下生物量,分析其水平及垂直空间格局,实现草地地下生物量(BGB)的数字制图。调查了2015年阿勒泰地区草地生长季节(6-8月)的生态要素和地下生物量。以地理位置、地形、气候、土壤和植被中的代表性信息为基础,基于机器学习算法估测研究区0~30 cm的草地地下生物量,并根据估测结果利用空间插值法得到地下生物量的空间分布格局,最终实现草地地下生物量的数字制图。结果表明:1)相比偏最小二乘回归(PLS)和随机森林模型(RF),支持向量机模型(SVM)在0~10 cm、10~20 cm和20~30 cm土层地下生物量的估测中表现出最高的精度,验证集数据的精度(R2)依次为0.77、0.67和0.69,相应的RMSE为245.56、98.81和63.58 g·m-2。从空间插值的效果看,反距离权重插值(IDW)优于径向基函数插值(RBF)和张力样条插值(SPL)。2)进一步比较了不同估测模型与空间插值方法间的组合能力,结果显示,在阿勒泰地区的草地地下生物量研究中,SVM+IDW是可靠的估测模型和空间化方法的组合。在0~10 cm、10~20 cm和20~30 cm土层的草地地下生物量数字制图的R2为0.73、0.64和0.60,RMSE为269.73、108.14和73.01 g·m-2。3)阿勒泰地区草地地下生物量均值为1265 g·m-2,是全国平均值的两倍,与全球平均水平相当。其中,高寒草甸的地下生物量最大,为2908.50 g·m-2,温性荒漠的地下生物量最小,为776.84 g·m-2,全区草地地下生物量共计1.27×108 t(≈0.13 Pg)。全区草地地下生物量的空间变化明显,整体上自北向南,由山地向平原呈递减的趋势。

关键词: 草地地下生物量, 阿勒泰地区, 机器学习, 空间插值, SVM模型, IDW插值, 数字制图

Abstract:

This research investigated the spatial pattern of grassland belowground biomass (BGB) in the Altay region. Ecological indicators and BGB during the peak season (June-August) in 2015 were surveyed. Based on representative information for specific locations, terrain types, climate, and soil and plant factors, a forecasting model involving machine learning was developed to estimate BGB in the 0-30 cm soil layers in the study area. Based on this eatimation model, the spatial pattern was then obtained by the spatial interpolation method for digital mapping of BGB. It was found that: 1) The support vector machine model (SVM) had a higher estimation accuracy than an alternative partial least squares regression model and random forest model, also tested. The R2 values from the SVM model for verification data were 0.77, 0.67 and 0.69, respectively in the 0-10 cm, 10-20 cm and 20-30 cm soil layers, with accompanying root mean square error (RMSE) values of 246.56, 98.81 and 63.58 g·m-2, respectively. The outputs of three spatial interpolations indicated that an inverse distance weighting (IDW) method performed better than radial basis function and tension spline methods. 2) We further compared the output accuracy of different combinations of estimation model and spatial interpolation method. This comparison showed that SVM+IDW was a dependable combination of estimation model and spatial interpolation method for estimating the spatial pattern of BGB in the Altay region. The R2 values of the digital map obtained from this combination in the 0-10 cm, 10-20 cm, and 20-30 cm soil layers were, respectively, 0.73, 0.64, and 0.60, and the RMSE values were 269.73, 108.14 and 73.01 g·m-2. 3) The mean value of BGB in Altay region was 1265 g·m-2, which was twice that of the average for China and comparable to the global mean value. Temperate meadow steppe had the largest BGB, with a mean value of 2908.50 g·m-2, the temperate deserts hold the least biomass, estimated at 776.84 g·m-2. The BGB in the whole region is 1.27×108 t (≈0.13 Pg). Our modelling identified a strong spatial variability that decreased from the north to the south and from the mountains to plains.

Key words: grassland belowground biomass, Altay region, machine learning, spatial interpolation, SVM model, IDW interpolation, digital mapping