基于无人机多光谱影像的夏玉米SPAD估算模型研究

快速获取作物叶片叶绿素含量对及时诊断作物健康状况、指导田间管理具有重要意义。本研究以关中地区2020年夏玉米为研究对象，获取试验区无人机多光谱影像，提取植被指数，分析所选植被指数与SPAD的相关性，筛选得到模型的输入变量，利用偏最小二乘法（PLS）、随机森林回归（RF）和分层线性模型（HLM）分别构建拔节期、抽雄期、灌浆期以及全生育期的SPAD估算模型，最终选出最优估算模型，以期为快速获取夏玉米SPAD提供参考。研究发现：除NRI之外，NDVI、OSAVI、GNDVI、RVI、MCARI、MSR、CIre与SPAD均显著相关，其中，OSAVI、NDVI与SPAD呈现出较强且稳定的相关性；各个生育期的最优模型均是RF模型，在拔节期、抽雄期、灌浆期和全生育期，验证集R2分别为0.81、0.81、0.73、0.61，RMSE分别为1.24、2.32、3.13、3.20；对于SPAD估算模型，将降雨量、最高气温这两个气象因子与植被指数耦合的HLM模型可以一定程度提升线性模型的估算精度，但其精度低于RF模型。因此，基于无人机多光谱影像的RF模型可以实现夏玉米SPAD的快速准确估算。

Estimation of Summer Maize SPAD Based on UAV Multispectral Images

Obtaining chlorophyll content of crops rapidly is of great significance for timely diagnosing the health status of crops and guiding field management. In recent years, the development of unmanned aerial vehicle (UAV) has made it possible to quickly and accurately obtain information at the farm scale. The purpose was to estimate SPAD of summer maize based on UAV multispectral images, especially focusing on whether the hierarchical linear model with meteorological data had high accuracy. SPAD in the jointing stage, the tasseling stage and the filling stage were measured by SPAD-502Plus chlorophyll meter, and the multispectral images were captured by RedEdge mounted on DJI M600 Pro. Firstly, the vegetation indices of 21 experimental plots were extracted by band math and establishing the region of interest. Then, the correlation between the vegetation indices and SPAD was analyzed, and the vegetation indices with high correlation coefficient were selected as the input variables of SPAD estimation model. At last, the SPAD estimation models for the jointing stage, the tasseling stage, the filling stage and the whole growth stage were constructed by using partial least squares (PLS), random forest regression (RF) and hierarchical linear model (HLM), respectively. The results were compared to select the best model, which could provide support for SPAD estimation. It was found that except NRI, other vegetation indices (NDVI, OSAVI, GNDVI, RVI, MCARI, MSR, CIre) were significantly correlated with SPAD, furthermore, OSAVI and NDVI had strong and stable correlation with SPAD. The best model for each growth period was established by RF. For the jointing stage, the tasseling stage, the filling stage and the whole growth period, the R2 of the test set was 0.81, 0.81, 0.73 and 0.61, and the RMSE was 1.24, 2.32, 3.13 and 3.20, respectively. The HLM model, which coupled rainfall and maximum temperature with vegetation index, could improve the accuracy of linear model for estimating SPAD, but its accuracy was lower than that of RF. Therefore, the RF model based on UAV multispectral images could realize the estimation of SPAD of summer maize timely and accurately.