基于2种数据源的糖料蔗固碳量定量估算模型研究——以上思县为例OA
Research on Prediction Method of Carbon Sequestration in Sugarcane Based on Two Data Sources:A Case Study of Shangsi County
为提升蔗区糖料蔗固碳量智能估算水平,应用2022-2024年上思县糖料蔗固碳量观测数据、地面气象观测数据和风云卫星FY3D-NDVI遥感数据,分别构建了基于随机森林的气象要素估算模型和基于多函数拟合的遥感植被指数估算模型,通过对比拟合效果评价了 2种模型的适用性.研究结果表明:①随机森林特征因子重要性分析发现,散射辐射与光合有效辐射对糖料蔗固碳量的贡献权重高于降水量、日照等传统气象因子;②糖料蔗固碳量与NDVI指数关系密切,2023年同期NDVI值与糖料蔗总、茎、叶的固碳量均呈最优S型关系,S型生长规律同样适用于NDVI对糖料蔗固碳量的定量表征;③随机森林模型在预测误差控制方面表现更优,总、茎和叶固碳量的均方根误差RMSE和平均绝对误差MAE均小于遥感估算模型,经特征筛选与加权决策树优化后,R2提高至0.655,绝对误差为2.820t/hm2,可作为基于气象数据的固碳量估算优选模型;而遥感反演模型拟合优度(R2)更高,能更好地捕捉固碳量的整体变化趋势.本研究为区域糖料蔗固碳能力评估与农业碳汇计量提供了科学依据.
To improve the intelligent estimation level of sugarcane carbon sequestration in the area,this study applied observational data of sugarcane carbon sequestration,surface meteorological observation data,and FY3D-NDVI remote sensing data in Shangsi County from 2022 to 2024,a meteorological factor estimation model based on random forest and a remote sensing vegetation index estimation model based on multi-function fitting were constructed,respectively.Two estimation models were respectively constructed:a meteorological factor-based model using random forest(RF)algorithm,and a remote sensing vegetation index-based model employing multi-function fitting.The applicability of the two models was evaluated through comparative analysis of their fitting performance.The results indicate that:(1)Importance analysis of RF feature variables revealed that diffuse radiation and photosynthetically active radiation contributed more significantly to sugarcane carbon sequestration than conventional meteorological factors such as precipitation and sunshine duration.(2)A close relationship was observed between sugarcane carbon sequestration and NDVI;the NDVI values of the corresponding period in 2023 exhibited an optimal S-shaped relationship with the carbon sequestration of total aboveground biomass,stems,and leaves,respectively.This S-shaped growth pattern also proved applicable to the quantitative characterization of sugarcane carbon sequestration using NDVI.(3)The RF model demonstrated superior performance in prediction error control,with both root mean square error(RMSE)and mean absolute error(MAE)for total,stem,and leaf carbon sequestration being lower than those of the remote sensing-based estimation model.Following feature selection and weighted decision tree optimization,the RF model achieved an R2 of 0.655 and an absolute error of 2.820 t/hm2 making it the preferred model for carbon sequestration estimation based on meteorological data.In contrast,the remote sensing inversion model exhibited a higher coefficient of determination(R2),indicating its enhanced capability in capturing the overall trend of carbon sequestration dynamics.This study provides a scientific foundation for regional assessment of sugarcane carbon sequestration capacity and agricultural carbon accounting.
陆永顺;丁美花;潘春江;陆思宇;谭宗琨;孙明;方建芳;许彦灵;文徐能
上思县气象局,广西防城港 535500||北部湾国家气候观象台开放实验室,广西南宁 530022广西壮族自治区气候中心,广西南宁 530022贺州市气象局,广西贺州 542899广西壮族自治区气象科学研究所,广西南宁 530022广西壮族自治区气候中心,广西南宁 530022广西壮族自治区气候中心,广西南宁 530022上思县气象局,广西防城港 535500||北部湾国家气候观象台开放实验室,广西南宁 530022上思县气象局,广西防城港 535500||北部湾国家气候观象台开放实验室,广西南宁 530022上思县气象局,广西防城港 535500||北部湾国家气候观象台开放实验室,广西南宁 530022
农业科技
糖料蔗固碳量随机森林气象因子植被指数
SugarcaneCarbon sequestrationRandom forestMeteorological factorsVegetation index
《甘蔗糖业》 2026 (4)
76-88,13
2025年度广西气象科研计划青年人才培养项目"基于多元数据的上思县糖料蔗生物固碳估算模型研究"(桂气科2025QN20)广西重点研发项目"广西糖料蔗碳汇气象评估关键技术与应用"(桂科AB22035069)广西气象科研计划项目"甘蔗关键生育期土壤水分垂向响应机制的初步研究"(桂气科2025QN06)
评论