基于无人机热红外时序遥感的玉米冠层温度响应模式解析

Analysis of Maize Canopy Temperature Response Patterns Using UAV Thermal Infrared Time-series Remote Sensing

摘要:

高温胁迫严重制约玉米的生长发育与产量形成,但目前对冠层温度动态响应及其遗传基础的认识仍不充分。本研究利用无人机热红外遥感技术获取了V6-VT期800份玉米育种材料在连续3d、共15个时相的高时间分辨率冠层温度时序数据,构建了涵盖统计特征、偏差特征、昼夜节律、胁迫响应、峰值特征、DTW 形态、小波多尺度及拓扑持久性的多维度时序解析特征集,并提出融合农学先验知识的自适应加权特征选择框架以降低特征冗余。在此基础上,采用5 种基聚类器集成的聚类方法识别冠层温度响应模式,同时建立多指标加权融合的最优聚类数判定体系,有效缓解了单一聚类方法和评价指标的局限性。基于Teixeira热胁迫模型量化累积热胁迫强度,揭示了2种显著不同的温度响应模式:低温平稳型模式(聚类1)显著富集于DH和TST亚群(分别占82.7%和71.0%),且在冠层覆盖度(Cohen's d=1.172)和冠层体积指数(d=1.009)等生长指标上显著优于聚类0,而聚类0呈现更高的峰值温度与更强的昼夜波动。遗传背景与温度响应模式呈极显著强关联(φc=0.304,p<0.001),与热胁迫分布呈极显著中等强度关联(φc=0.204,p<0.0001)。热胁迫因子与所有生长指标呈显著负相关,其中与冠层覆盖度(r=-0.598)和冠层体积指数(r=-0.510)的相关性尤为突出。研究结果表明,冠层温度响应模式与遗传背景密切相关,可作为耐热性快速高通量筛选指标。本研究所提出的特征选择与聚类优化框架为时序表型分析提供了新途径。

Abstract:

Heat stress critically constrains maize growth, development and yield formation. Nevertheless, the dynamic responses of canopy temperature and their underlying genetic architecture remain poorly characterized. UAV-borne thermal infrared remote sensing was employed to acquire high-temporal-resolution time-series canopy temperature data from 800 maize breeding accessions at the V6-VT stage, spanning 15 temporal phases across three consecutive days. A multi-dimensional feature set for timeseries dissection was constructed, encompassing statistical attributes, deviation characteristics, diurnal rhythms, stress responses, peak features, DTW-based morphological similarity, wavelet multiscale patterns and topological persistence. To mitigate feature redundancy, an adaptive weighted feature selection framework incorporating agronomic prior knowledge was developed. Subsequently, canopy temperature response patterns were identified by using an ensemble clustering approach that integrated five base clusters, and a multi-criteria weighted fusion system was established for determining the optimal number of clusters, thereby overcoming the limitations inherent in any single clustering algorithm or evaluation metric. The cumulative heat stress intensity was quantified by using the Teixeira heat stress model, which revealed two markedly distinct canopy temperature response modes. The low-temperature stable mode (Mode 1) was significantly enriched in the DH and TST subgroups (accounting for 82.7% and 71.0%, respectively) and exhibited substantially superior growth traits, including canopy cover (Cohen's d=1.172) and canopy volume (d=1.009), relative to Mode 0. Conversely, Mode 0 displayed higher peak temperatures and more pronounced diurnal fluctuations. Genetic background showed a highly significant association with the temperature response mode (φc=0.304, p<0.001) and a statistically significant but moderate association with heat stress distribution (Cramér's V=0.204, p<0.0001). Heat stress intensity was significantly negatively correlated with all growth indicators, with particularly strong correlations observed for canopy cover (r=-0.598 ) and canopy volume (r=-0.510). Collectively, these findings indicated that canopy temperature response patterns were closely linked to genetic background and can serve as rapid, high-throughput phenotypic indicators of heat tolerance. The proposed feature selection and clustering optimization framework can provide a novel methodological tool for time-series phenotyping analysis.