融合无人机遥感与气象数据的水稻地上生物量估计模型

Aboveground Biomass Estimation Model of Rice Using UAV Remote Sensing and Meteorological Data

摘要:

及时准确地估计水稻地上生物量对田间精准管理具有重要意义，而现有研究集中在利用单一无人机遥感数据，因光谱量饱和效应难以实现作物生育后期地上生物量精准估计。为此，本文于2023年和2024水稻生长季进行了无人机多光谱遥感图像、气象数据及水稻地上干生物量数据采集，构建多源特征融合的地上生物量估计模型，实现全生育期及跨生长季的精准有效估计。结果表明：分别以植被指数、植被指数和纹理特征、植被指数和纹理特征及有效积温作为输入变量，利用多元线性回归（Multiple linear regression，MLR）、随机森林（Random forest，RF）、偏最小二乘（Partial least squares，PLS）和支持向量机（Support vector machine，SVM）4种机器学习算法建立水稻地上生物量估测模型时，精度逐渐提升且RF建立的模型精度均为最高。以植被指数作为模型输入变量，在开花前期、开花后期和全生育期RF调整决定系数（Adjusted coefficient of determination，调整R2）分别为0.71、0.67、0.7，均方根误差（Root mean square error，RMSE）分别为268.62、300.29、249.43 g/m2;以植被指数和纹理特征作为模型输入变量，对应调整R2分别为0.75、0.72和0.74，RMSE分别为213.79、239.81、289.46 g/m2;以植被指数和纹理特征及有效积温作为输入变量，对应调整R2分别为0.84、0.87和0.87，RMSE分别为176.9、162.81、163.08 g/m2。以2024年相关数据作为验证集时，在全生育期RF调整R2为0.60，RMSE为288.19 g/m2，可实现水稻地上干生物量的跨生长季精准估计。本文融合无人机遥感及气象数据的估计方法可实现水稻地上生物量的全生育期及跨生长季精准估计，可为智慧农业背景下的水稻精准管理提供技术支撑。

Abstract:

It is of great significance to estimate the aboveground biomass of rice accurately and timely for precision management of rice field. However, the existing researches focus on using single UAV remote sensing data, which is difficult to achieve accurate estimation of aboveground biomass in the late growth stage of rice due to the spectral saturation effect. To this end, the drone multispectral remote sensing images, meteorological data, and aboveground dry biomass data of rice during the 2023 and 2024 growing seasons were collected. A multi?source feature fusion model for aboveground biomass estimation was constructed to achieve accurate and effective estimation throughout the entire growth period and across multiple growing seasons.The results showed that the vegetation index，vegetation index and texture characteristics，vegetation index and texture characteristics and effective product temperature as the input variable，using multiple linear regression（MLR）, random forest（RF）, partial least squares（PLS）and support vector machine（SVM） to establish the rice ground biomass estimation model. The accuracy was gradually improved and the model accuracy established by the RF algorithm was the highest. With the vegetation index as the model input variable，the adjusted coefficient of determination (adjusted R2) during flowering，late flowering and all reproductive stages were 0.71,0.67 and 0.7，respectively，root mean square error (RMSE) were 268.62 g/m2，300.29 g/m2 and 249.43 g/m2，respectively. With the vegetation index and texture features as the model input variables，the corresponding adjustment R2 were respectively 0.75, 0.72 and 0.74，RMSE were 213.79 g/m2，239.81 g/m2 and 289.46 g/m2，respectively. With vegetation index and texture characteristics and effective product temperature as input variables，the corresponding adjustment R2 were respectively 0.84, 0.87 and 0.87，and RMSE were 176.9 g/m2，162.81 g/m2 and 163.08 g/m2. Using 2024 data as validation, RF achieved an adjusted R2 of 0.60 and RMSE of 288.19 g/m2 across the entire growth cycle, enabling precise estimation of aboveground dry biomass in rice across growing seasons. The proposed integrated approach combining UAV remote sensing and meteorological data provided a robust method for accurate aboveground biomass estimation throughout the growth cycle and across seasons, offering technical support for precision rice management in smart agriculture.