基于高光谱和激光雷达遥感的水稻产量监测研究

【背景】快速、准确地估算水稻产量对于肥水精确管理及国家粮食政策的制定至关重要。高光谱与激光雷达遥感作为2种不同的主被动监测技术,为水稻长势信息获取提供了多样化手段。【目的】对比2种遥感监测手段在不同生态点的独立数据集中的验证精度,寻求可移植性强的产量估算模型,对水稻长势监测提供理论与技术支撑,及为精确农业提供科学指导具有重要意义。【方法】本研究通过实施3年（2016—2018年）包含不同地点、不同品种与不同氮素水平的水稻田间试验,在抽穗后各时期同步获取点云数据和光谱数据,结合线性回归与随机森林回归来估算产量,探究抽穗后点云数据与光谱数据估算水稻产量的差异;同时评估产量模型在不同数据集的时空可移植性,寻求可移植性强的产量估算模型。【结果】利用点云数据估算产量的精度（R² = 0.64—0.69）优于光谱数据的估算精度（R² = 0.20—0.58）;基于线性回归的产量估算模型,其验证精度明显优于基于随机森林回归的产量模型;产量模型在同一生态点的可移植性更强（不同生态点：RRMSE 16.69%—17.85%;同一生态点：RRMSE 11.37%—12.41%）。【结论】本研究为抽穗后水稻产量监测提供了新的方法和不同遥感手段的性能比较,为收获前作物产量的实时估算提供重要支撑。激光雷达技术凭借其全天候工作的特点,在长江中下游水稻产量实时监测中有着较好的应用前景。

Assessment of Terrestrial Laser Scanning and Hyperspectral Remote Sensing for the Estimation of Rice Grain Yield

【Background】Non-destructive and accurate estimation of crop biomass and yield is crucial for the quantitative diagnosis of growth status and planning of national food policies. Hyperspectral and Terrestrial Laser Scanning (TLS) remote sensing provide convenient and effective ways to monitor crop condition.【Objective】The aim of this study was to examine the feasibility of developing models with various independent datasets to build a universal yield monitoring model. The expected results can provide theoretical and technical support for rice growth monitoring and scientific guidance of precision agriculture.【Method】Field plot experiments were conducted in 2016, 2017 and 2018 and involved different study sites, nitrogen (N) rates, planting techniques and rice varieties. Linear regression (LR) and random forest (RF) were evaluated in estimating yield with TLS and spectral data collected since the heading stage, and the feasibility of developing models with various independent datasets was examined to build a universal yield monitoring model.【Result】 The results showed that TLS models exhibited higher estimation accuracies for yield in the heading stage, early-filling stage and late-filling stage (R²: 0.64-0.69) than hyperspectral models (R²: 0.20-0.58). Compared to RF, LR modeling yielded significantly higher validation accuracies for growth stages after heading. While the predictive model was transferred to other datasets, the validation accuracies from the same site (RRMSE: 11.37%-12.41%) were higher than those from a different site (RRMSE: 16.69%-17.85%).【Conclusion】The results suggested that TLS was a promising technique to monitor yield at post-heading stages with high accuracy and to overcome the saturation of canopy reflectance signals encountered in optical remote sensing.