地球信息科学学报 ›› 2023, Vol. 25 ›› Issue (9): 1908-1922.doi: 10.12082/dqxxkx.2023.220623

• 遥感科学与应用技术 • 上一篇    

基于CatBoost的植被总初级生产力遥感模拟方法及在福建省的应用

李玉洁1,2,3(), 江洪1,2,3,*(), 刘宣广1,2,3   

  1. 1.福州大学空间数据挖掘与信息共享教育部重点实验室,福州 350108
    2.福州大学地理空间信息技术国家地方联合工程研究中心,福州 350108
    3.数字中国研究院(福建),福州 350108
  • 收稿日期:2022-08-24 修回日期:2022-11-03 出版日期:2023-09-25 发布日期:2023-09-05
  • 通讯作者: * 江 洪(1975— ),男,福建永安人,副研究员,研究方向为环境遥感、信息管理研究。E-mail: jh910@fzu.edu.cn
  • 作者简介:李玉洁(1997— ),女,山西吕梁人,硕士生,研究方向为遥感技术与应用、资源环境遥感。E-mail: 205527018@fzu.edu.cn
  • 基金资助:
    福建省科技计划引导性项目(2021Y0005);福建省水利科技项目(MSK202301)

Simulation of Vegetation Gross Primary Productivity and Its Application in Fujian Province of China Using Remote Sensing and CatBoost Algorithm

LI Yujie1,2,3(), JIANG Hong1,2,3,*(), LIU Xuanguang1,2,3   

  1. 1. Key Laboratory of Spatial Data Mining and Information Sharing of Ministry of Education, Fuzhou University, Fuzhou 350108, China
    2. National Engineering Research Center of Geospatial Information Technology, Fuzhou University, Fuzhou 350108, China
    3. The Academy of Digital China (Fujian), Fuzhou 350108, China
  • Received:2022-08-24 Revised:2022-11-03 Online:2023-09-25 Published:2023-09-05
  • Contact: * JIANG Hong, E-mail: jh910@fzu.edu.cn
  • Supported by:
    Science and Technology Plan Leading Project of Fujian Province, China(2021Y0005);Water Conservancy Science and Technology Project of Fujian Province, China(MSK202301)

摘要:

植被总初级生产力(GPP)作为衡量陆地生态系统健康的重要指标,可直接反映区域环境状况和改善情况,因此准确估算植被GPP变化对区域可持续发展具有重要意义。本文利用中国及日本涡度通量观测数据,构建了基于CatBoost算法融合地形特征的GPP估算模型;并将模型应用于具有复杂地形特征的福建省,实现了该省GPP长时序模拟。研究结果表明:① 地形特征是GPP机器学习估算的重要参数,融合地形特征建模的GPP模拟结果精度明显提高,均方根误差(RMSE)下降16%。② CatBoost GPP估算模型有效降低了传统GPP估算模型和常用机器学习(随机森林和支持向量机)GPP估算模型中存在的高估和低估现象,模型拥有更高的精度和更强的鲁棒性。本文GPP模拟精度:决定系数(R2)为0.888,RMSE为1.164 gC·m-2·day-1,平均绝对误差(MAE)为0.773 gC·m-2·day-1。③ 基于CatBoost GPP估算模型模拟的福建省多年GPP变化与GOSIF GPP估算结果高度一致,且其对福建省GPP空间分布表达更准确。福建省2002—2020年GPP均值1697 gC·m-2·a-1,空间变化整体呈现出“由东南向西北递减”的分布特征,多年GPP变化呈“不显著波动增加”趋势。本研究可为实现区域GPP精确估算和生态环境有效治理提供新方法和可靠科学依据。

关键词: 植被总初级生产力, 地形特征, CatBoost, 随机森林, 支持向量机

Abstract:

As an important indicator to measure the health of terrestrial ecosystems, the Gross Primary Productivity (GPP) of vegetation can directly reflect the improvement of regional environment. Therefore, accurate estimation of vegetation GPP changes is of great significance to regional sustainable development. In this paper, a GPP estimation model using the CatBoost algorithm integrating topographic data was developed. Using the vorticity flux observation data from China and Japan, this model was applied to simulate the long term GPP of Fujian Province where the topographic effect is significant. The results show that: (1) Terrain features are important parameters for the estimation of GPP using machine learning methods. The accuracy of GPP simulation results with terrain features included is significantly improved, and the Root Mean Square Error (RMSE) is decreased by 16%; (2) The GPP estimation model based on CatBoost has higher accuracy and stronger robustness and effectively reduces the overestimation and underestimation phenomena existing in traditional GPP estimation models and commonly used machine learning models (e.g., random forest and support vector machine). The coefficient of determination (R2) is 0.888, the RMSE is 1.164 gC·m-2·day-1, and the Mean Absolute Error (MAE) is 0.773 gC·m-2·day-1; (3) The multi-year GPP changes in Fujian Province simulated by the CatBoost GPP estimation model are highly consistent with the GOSIF GPP estimation results, indicating a more accurate GPP spatial distribution in Fujian Province. It is found that the mean GPP of Fujian Province from 2002 to 2020 was 1 697 gC·m-2·a-1. The overall spatial distribution is characterized by "decreasing from southeast to northwest", and the multi-year GPP variation shows a trend of "non-significant fluctuation increase". This study provides a new method and useful data for regional GPP estimation and ecological environment management.

Key words: Gross Primary Production, topographic features, CatBoost, Random Forest, Support Vector Machine