探索盐碱地青贮玉米产量的快速估测方法，实现饲料作物长势无损监测与精准评估。基于无人机多光谱遥感技术，以河西走廊不同改良剂处理下的青贮玉米金岭 67为研究对象，开展植被指数筛选、产量模型构建与空间反演研究。获取收获期多光谱影像，通过构建植被指数和相关性分析，优选敏感指数作为模型输入，采用偏最小二乘回归( PLSR)、岭回归( RR)、支持向量机回归( SVR)和随机森林回归( RFR) 4种算法构建估产模型。结果表明，归一化差异植被指数( NDVI)、归一化红边指数( NDRE)、面叶绿素指数( LCI)与产量相关系数最高，均达 0.76；4种模型中 SVR表现最优，决定系数( R2)为 0.77，均方根误差( RMSE)为 139.47 kg/hm2，非线性机器学习算法估产精度高于传统线性方法；基于 SVR模型的空间反演显示，盐碱地改良剂杰效利 3 kg/hm2处理青贮玉米平均产量最高，为 72 769 kg/hm2，较常规施肥处理[N 480kg/hm2、P2O5 225kg/hm2、 K2O 105 kg/hm2)]增产 5.9%。无人机多光谱结合机器学习可实现盐碱地青贮玉米产量精准估测，为改良剂效果评价与饲草精细化管理提供技术支撑。

This study aimed to explore a rapid method for estimating silage maize yield in saline-alkali land and to achievenondestructive monitoring and precise assessment of forage crop growth. Based on UAV multispectral remote sensing technology, si.lage maize variety Jinling 67 grown under different soil amendment treatments in the Hexi Corridor was used as the research object toconduct vegetation index screening, yield model construction, and spatial inversion. Multispectral imagery was acquired during the har.vest period, and sensitive vegetation indices were selected as model inputs through vegetation index construction and correlation analy.sis. 4 algorithms, namely Partial Least Squares Regression (PLSR), Ridge Regression (RR), Support Vector Regression (SVR), andRandom Forest Regression (RFR), were employed to develop yield estimation models. The results showed that the normalized differ.ence vegetation index (NDVI), normalized difference red-edge index (NDRE), and leaf chlorophyll index (LCI) exhibited the highestcorrelation coefficients with yield (all reaching 0.76). Among the 4 models, SVR performed best, with a coefficient of determination(R2) of 0.77 and a root mean square error (RMSE) of 139.47 kg/ha, indicating that non-linear machine learning algorithms achievedhigher estimation accuracy than traditional linear methods. Spatial inversion based on the optimal SVR model revealed that the treat.ment Jiexiaoli at 3 kg/ha achieved the highest average yield (72 769 kg/ha) in saline-alkali land, representing a 5.9% increase com.pared to that under conventional fertilization (N 480 kg/ha, P2O5 225 kg/ha and K2O 105 kg/ha). This study demonstrates that UAVmultispectral remote sensing combined with machine learning enables accurate yield estimation of silage maize in saline-alkali soils,providing technical support for evaluating amendment effectiveness and precision management of forage crops.

F127；S513

国家重点研发计划项目( 2024YFD2201105)；中央引导地方科技发展资金( 24ZYQA053)；甘肃省农业农村厅农业科技支撑项目( KJZC-2026-31)。