农作物物候遥感反演方法的适用性研究

Crop phenology is critical to many fields, such as agricultural management, crop growth monitoring, crop planting area extraction, crop growth simulation. The retrieval of vegetation phenology from remotely sensed data includes two major procedures: firstly, smoothing the time-series vegetation index data; and then, retrieving vegetation phenology from the smoothed data. However, in current research and applications, the characteristics of crops have been neglected, leading to the empirical selection of retrieval methods and the thresholds for key parameters. Although some researchers have conducted preliminary explorations on the evaluation of some methods, research on this problem is still relatively weak. In this research, we choose the crop phenology under five typical cropping patterns in China as research subject. Firstly, we develop the smoothing method evaluation model and the retrieval method evaluation model with evaluation indices from different perspectives. And then based on these two evaluation models, we systematically and qualitatively analyze six frequently-used smoothing methods and seven frequently-used phenology retrieval methods, to reveal the applicability of smoothing methods and phenology retrieval methods in crop phenology retrieving. Finally, the retrieved crop phonological parameters in three typical farming areas are used to produce thematic products. This research will deepen the understanding of crop phenology retrieval methods, improve the evaluation model for phenology retrieving methods and enhance the practice of crop phenology retrieval from remotely sensed data.

农作物物候对农业生产管理、作物长势监测、种植面积提取和作物生长过程模拟等都具有重要意义。植被物候遥感反演包括两个主要步骤：遥感植被指数时间序列数据滤波处理和物候反演。目前农作物物候遥感反演研究和应用中，研究者忽视了农作物物候的自身特点，在滤波方法和物候反演方法的选择及其参数阈值的确定上都存在一定的盲目性和经验性。虽然有学者对部分方法的评价进行了初步探索，但是这方面的研究还比较薄弱。本研究以我国5种典型种植模式下农作物的物候为对象，从多个角度、多个层面构建评价指标，建立滤波方法评价模型和物候反演方法评价模型，对6种常用的滤波方法和7种常用的物候反演方法进行定量评价和系统性研究，揭示各方法的区域、种植模式和农作物适用性，并在三个典型农区进行物候参数的应用研究。本研究对于深化对农作物物候反演方法的认识，完善物候遥感反演定量评价方法、增强农作物物候遥感反演的实践应用都具有重要意义。

准确反演农作物物候信息有助于提高相关研究的准确性和应用的有效性。本项目以基于我国农业气象观测数据，从中选出典型种植模式站点的返青期（SOS）和成熟期（EOS）记录，结合MODIS EVI时间序列数据，从多个角度、多个层面构建评价指标，建立滤波方法综合评价模型和物候反演方法评价模型，进而对常用的滤波方法和物候反演方法适用性进行分析。得到的主要结论如下：（1）各种滤波方法应用于不同区域、不同种植模式时存在最优阈值，应用最优阈值进行滤波能提高滤波效果；Savitzky-Golay滤波法的普适性最强，在各种熟制区和种植模式下均能获得很好的滤波效果。（2）各种物候反演方法在反演农作物SOS/EOS时，反演效果与其阈值有较大的关系，采用合适的阈值能够提高SOS/EOS反演的可行性和准确性，在5分析的物候反演方法中，动态阈值法反演物候的准确性最高，并且在不同熟制区的均能取得较好的反演效果。（3）应用物候参数可有效提取耕地的复种指数、有效识别一些主要农作物的类型。