基于LAI时间序列重构数据的冬小麦物候监测

农作物物候信息对农作物长势监测和估产具有重要意义。该文以河北省中南部冬小麦为研究对象,以叶面积指数(LAI,leaf area index)为同化量,采用重采样粒子滤波算法同化WOFOST(world food studies)作物生长模型和遥感观测LAI,重构LAI时间序列数据,基于重构数据提取冬小麦返青期、抽穗期和成熟期等关键物候期。重构结果表明,重构的LAI具有良好的时间连续性和空间连续性,可减缓WOFOST作物模型LAI变化剧烈程度,峰值出现时间与遥感LAI曲线基本同步,且可一定程度上解决遥感观测LAI数值整体偏低和数据缺失的问题。物候期监测结果表明,在空间分布上与冬小麦实际生长状况基本相符,时间上也较为合理,但因在返青期存在LAI高初始值、成熟期存在LAI下限不确定性等问题致使在具体日期存在偏差。

Monitoring of phenology by reconstructing LAI time series data for winter wheat

Abstract: A method to monitoring winter wheat phenology based on reconstructing Leaf area index (LAI) by assimilating crop model and remotely sensed LAI product is introduced in this paper. WOFOST is a crop model developed by the Center for World Food Studies in cooperation with the Wageningen Agricultural University to simulate and estimate the growth of winter wheat. LAI is one of the important input and output parameters in the model, and the next day's LAI output is derived from the previous day's LAI input. NASA's Moderate Resolution Imaging Spectroradiometer (MODIS) LAI time series has a good time resolution and can reflect the growth trend of crops, but the LAI values is generally lower than actual value for the impact from mixed pixels and cloud noises. Resampling particle filter (SIP) is a sequential Monte Carlo state estimation method for Non-Linear and Non-Gaussian system, and is introduced to reconstruct the winter wheat LAI time series by assimilating the WOFOST model and MODIS LAI. In the SIP assimilating algorithm, the WOFOST model is imported as the state transition equation of LAI for its ability to simulating the winter wheat growth process, and the MODIS LAI time series is treated as the observations from the winter wheat. To get the trend as MODIS LAI, the weights of particles are calculated from the first derivative of the MODIS LAI time series. The particles that have the similar trend as the MODIS LAI time series can achieve a higher weight. The algorithm was applied on the winter wheat in Hebei province, China. First, a WOFOST model was calibrated to get applicable to local region, and the actual meteorological data were preprocessed to meet the format requirement, then the algorithm was run and the reconstructed LAI time series are achieved. The results show that the reconstructed LAI curve has good temporal continuity. The reconstruct algorithm can decrease the LAI change scope and avoid the problem of low value of MODIS LAI to a certain degree, and the day corresponding to the reconstructed LAI curve peak is basically the same as that of the MODIS LAI curve. At the regional scale, the reconstructed LAI images have higher LAI values and good spatial continuity, and reduce the impact from the low-value and missing data of MODIS LAI. Based on the reconstructed LAI time series, the key stages of the winter wheat are monitored including its green-returning stage, heading stage, and ripen stage. The reconstructed LAI curve is a substantially horizontal line in over-wintering stage, and has an initial value of approximately 0.4, which is higher than the actual value in the early green-returning stage. Therefore the start day of the green-returning stage is identified by the upslope point of the reconstructed LAI curve. In the heading stage, the winter wheat LAI reached the maximum, so the day corresponding to the curve peak was identified as the start day of the heading stage. The identification of the ripen stage is made by the threshold method. The threshold is set to 20% of the spacing between maximum and minimum from the minimum of the right side of the LAI curve. The day corresponding to the threshold is identified as the start day of the ripen stage. Comparing the monitoring results with the actual situation, basically both have the same spatial distribution, and may be the bias for the high initial value in the early green-returning stage and the uncertainty of the minimum in the ripen stage of the LAI curve.