基于无人机多源遥感与气象参数的大田玉米地上生物量估算研究

Estimation of Aboveground Biomass of Maize Based on Multisource Remote Sensing and Meteorological Parameters

摘要:

快速、准确地监测作物地上生物量的时空分布情况对于作物长势状况评估及农田精准灌溉管理具有重要意义。光谱指数已广泛应用于地上生物量估算，但高覆被下光谱饱和效应严重影响模型性能，另外，结合无人机遥感光谱、温度和纹理信息之间的互补与协同效应进行农田尺度不同生育期和不同灌溉量处理下玉米地上生物量估算的适用性以及气象因素对于遥感模型估算年际间地上生物量的影响仍需深入探究。本文以2018、2019、2021年内蒙古不同实验地数据为基础，使用无人机多源遥感系统获取的光谱指数、温度指数和纹理信息以及气象参数（参考蒸散量和饱和水汽压差）作为输入特征参数，采用逐步回归、随机森林回归、自适应增强回归、支持向量回归和一维卷积神经网络回归5种建模方法，建立多源特征融合的大田玉米地上生物量遥感估算模型。研究结果表明：与单一类型遥感信息相比，融合光谱、温度和纹理信息的地上生物量估算模型精度较好，其中一维卷积神经网络回归模型的精度相对较好（验证集R2=0.87，RMSE为295.77g/m2）；引入气象参数后，地上生物量无人机多源估算模型精度无明显提升，表明气象参数与无人机多源遥感信息存在冗余；考虑到无人机机载设备的成本因素及作业效率，本研究中基于光谱指数和气象参数的地上生物量随机森林估算模型的精度也较高（验证集R2=0.85，RMSE为296.74g/m2），说明气象参数可作为光谱指数的补充，推荐其进行年际间大田玉米地上生物量估算；地上生物量一维卷积神经网络模型的性能在不同生育期存在差异，因此需根据作物生育期来综合评估模型性能。本研究为促进无人机多源遥感技术在农田精准灌溉管理上的应用提供了技术支持。

Abstract:

Accurately and rapidly monitoring the spatio-temporal distribution of aboveground biomass in crops is crucial for assessing growth conditions and managing precision irrigation in agricultural fields. Spectral indices have been widely utilized for estimating aboveground biomass;however, the spectral saturation effect under high coverage significantly impacts model performance. Furthermore, the effectiveness of integrating complementary and synergistic effects of spectral, temperature, and texture information from UAV remote sensing for estimating maize aboveground biomass across different growth stages and irrigation treatments at the farmland scale remains unclear. Additionally, the influence of meteorological factors on the interannual performance of remote sensing models for aboveground biomass estimation requires further investigation. Based on data from experimental sites in Inner Mongolia in 2018, 2019, and 2021, the spectral indices, temperature indices, and texture information obtained from UAV-based multisource remote sensing system, along with meteorological parameters such as reference evapotranspiration and vapor pressure deficit, were used as input features. Five modeling methods—stepwise regression, random forest regression, adaptive boosting regression, support vector regression, and one-dimensional convolutional neural network regression— were employed to establish a multisource feature fusion remote sensing estimation model for aboveground biomass of maize in large fields. The results showed that compared with single-type remote sensing information, the aboveground biomass estimation model integrating spectral, temperature, and texture information had better accuracy. Among them, the one-dimensional convolutional neural network regression model had relatively better accuracy (R2=0.87, RMSE=295.77g/m2). After introducing meteorological parameters, there was no significant improvement in the accuracy of the aboveground biomass estimation model by using drone multisource remote sensing information, indicating redundancy between meteorological parameters and drone multisource remote sensing information. Considering the cost of drone-mounted equipment and the operation efficiency, the aboveground biomass estimation model based on spectral indices and meteorological parameters using random forest regression also had satisfactory accuracy (R2=0.85, RMSE=296.74g/m2), suggesting that meteorological parameters can serve as a supplement to spectral indices. The performance of the one-dimensional convolutional neural network model for aboveground biomass estimation varied across different growth stages, indicating that the accuracy and reliability of the model need to be comprehensively assessed based on the crop growth stage. The research result can provide technical support for promoting the application of drone multisource remote sensing technology in precision irrigation management in farmlands.