基于无人机高光谱数据的小麦生物量估测

Wheat Biomass Estimation Based on UAV Hyperspectral Data

高光谱因其通道多、数据量大、信息丰富等特点,在小麦农学参数估测方面被广泛应用。对小麦生物量和植被指数进行相关性分析,结果表明17种植被指数中在拔节期、孕穗期和全生育期与生物量达到显著相关水平的植被指数各有16种;拔节期DVI和RDVI指数相关性最高,r均为0.784;孕穗期GNDVI指数相关性最高,r为0.766;全生育期WI指数相关性最高,r为-0.799;与开花期生物量达到显著相关的植被指数有8种,WBI指数相关性最高,r为-0.642。分别利用各时期与生物量达到显著相关的植被指数构建生物量PLSR估测模型,模型的验证R2和建模R2均是全生育期最高,分别为0.85和0.93,其次是孕穗期、拔节期、开花期。建模RMSE最低的是孕穗期,为461.74 kg/hm^(2),验证RMSE最低的是拔节期为354.92 kg/hm^(2)。建模和验证R2提升最大的是全生育期,提升了0.11;RMSE下降最多的同样是全生育期,下降了298.93 kg/hm^(2)。总体来看,利用全生育期数据构建生物量估测模型精度最优。该研究所构建的小麦生物量预测模型可为田间作物长势监测以及农业管理决策提供有效参考。

Hyperspectral is widely used in the estimation of agronomic parameters of wheat because of its many channels,large amount of data and rich information.We analyzed the correlation between wheat biomass and vegetation indices,and the analysis showed that 16 vegetation indices were significantly correlated with biomass at each of the 17 vegetation indices,including the pulling,gestation and full-fertility stages;the highest correlation between DVI and RDVI indices was 0.784 at the pulling stage;the highest correlation between GNDVI indices was 0.766 at the gestation stage;the highest correlation between WI indices was-0.799 at the full-fertility stage;and the highest correlation with biomass at the flowering stage.The highest correlations were found for the vegetation indices at flowering,and the highest correlations were found for the WBI indices,with an r of-0.642.The vegetation indices that were significantly correlated with biomass at each period were used to construct the biomass PLSR estimation model.It gradually decreased at the gestation,nodulation and flowering stages.The lowest modeled RMSE was 461.74 kg/hm^(2)at gestation and the lowest validated RMSE was 354.92 kg/hm^(2)at nodulation.The largest increase in modeled and validated R 2 was at full-gestation,with an increase of 0.11,and the largest decrease in RMSE was also at full-gestation,with a decrease of 298.93 kg/hm^(2).The biomass estimation model using full-gestation data had the best accuracy.Overall,the accuracy of the biomass estimation model using full fertility data was the optimal.The wheat biomass prediction model constructed could provide an effective reference for crop growth monitoring in the field as well as agricultural management decisions.