基于机器学习和经验知识的青藏高原多时段植被制图  

Multi temporal vegetation mapping of the Tibetan Plateau via machine learning model simulation and experiential knowledge

作　　者：周继华[1,2] 来利明 陈巧娥[1,3] 宋长青 高培超 叶思菁[4] 沈石 杨刚刚 郝海霞 王贵豪 熊喆 郑元润 Jihua Zhou;Liming Lai;Qiaoe Chen;Changqing Song*;Peichao Gao;Sijing Ye;Shi Shen;Ganggang Yang;Haixia Hao;Guihao Wang;Zhe Xiong;Yuanrun Zheng(State Key Laboratory of Plant Diversity and Specialty Crops,Institute of Botany,Chinese Academy of Sciences,Beijing 100093,China;China National Botanical Garden,Beijing 100093,China;University of Chinese Academy of Sciences,Beijing 100049,China;Faculty of Geographical Science,Beijing Normal University,Beijing 100875,China;Institute of Atmospheric Physics,Chinese Academy of Sciences,Beijing 100029,China)

机构地区：[1]中国科学院植物研究所,植物多样性与特色经济作物全国重点实验室,北京100093 [2]国家植物园,北京100093 [3]中国科学院大学,北京100049 [4]北京师范大学地理学部,北京100875 [5]中国科学院大气物理研究所,北京100029

出　　处：《科学通报》2025年第1期134-144,共11页Chinese Science Bulletin

基　　金：第二次青藏高原综合科学考察研究(2019QZKK0608)资助。

摘　　要：理解植被长期演变趋势及其对气候变化与人类活动的响应需要多时段植被分布资料.目前青藏高原植被图主要为2000、2020年前后两个时段,以及1990~2020年时段;主要是基于地面调查数据的人工制图以及基于地面调查数据、遥感数据、地形与气候资料,采用机器学习模型分类的自动制图,存在部分植被类型混淆、制图精度有待提高等问题,同时,缺乏更早时段的植被分布资料.本文发展了整合地面调查数据、遥感数据、地形数据、气候数据等,基于随机森林机器学习模型的人工制图技术,获得了1980s、2020年青藏高原植被分布图;基于植被地带性分析和1980s植被图,获得青藏高原近似复原植被图;以2020年青藏高原植被图为基础,采用随机森林模型模拟了2035、2050年青藏高原植被分布图.近似复原植被图、1980S和2020年青藏高原植被图的空间分辨率可达到100 m,与公认较为可靠的《中华人民共和国植被图(1:100万)》相比,本研究获得的1980s和2020年植被图对植被类型边界的刻画更为准确,植被斑块数目增加较多,有利于更加准确理解和模拟青藏高原植被对气候变化的响应.Understanding long-term trends in vegetation distribution and its responses to climate change and human activities requires vegetation distribution data from multiple periods.At present,the vegetation maps of the Tibetan Plateau mainly exist for periods of time surrounding 2000 and 2020,during which time they were based on field survey data,and 10-year periods from 1990 to2020,during which time they were based on random forest classifiers and Landsat imagery.These maps were created though manual mapping based on ground observation data and automatic mapping based on ground observation data,remote sensing data,terrain data and climate data,with machine learning models used for classification.There are problems,such as confusion among some vegetation types,and map accuracy needs to be improved.Moreover,there is a lack of vegetation distribution data from earlier periods.This study developed artificial mapping techniques based on random forest machine learning models,integrating observation data,remote sensing data,terrain data,climate data,etc.,and obtained vegetation distribution maps of the Tibetan Plateau in the 1980s and 2020.On the basis of zonal vegetation analysis and the 1980s vegetation map,an approximated original vegetation map of the Tibetan Plateau was obtained.On the basis of the vegetation map of the Tibetan Plateau from 2020,a random forest model was used to simulate the vegetation distribution maps of the Tibetan Plateau in 2035 and 2050.The spatial resolution of the approximated original vegetation map,the 1980s vegetation map and the 2020 vegetation map was approximately 100 m.The approximated original vegetation map,the 1980s vegetation map and the 2020 vegetation map contain11 vegetation groups,43 vegetation types,and 2 vegetation subtypes.Owing to the influences of climate,terrain,soil and other conditions,the vegetation distribution patterns of forests,shrubs,grasslands,and deserts on the Tibetan Plateau from southeast to northwest did not change across the five periods,but the areas and dist

关 键 词：青藏高原 植被制图 机器学习模型 多源数据 植被分类 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]