基于MODIS-EVI数据和Symlet11小波识别东北地区水稻主要物候期

作物物候信号能够反映温度和降水等变化对植被生长的影响,是进行农作物动态分析和田间管理的重要依据。基于2008年EOS-MODIS多时相卫星遥感数据,研究了我国东北地区水稻的主要物候期的识别方法。首先提取研究区24个农业气象观测站所在位置的MODIS-EVI(Enhanced Vegetation Index,增强型植被指数)指数的时间序列;同时利用小波滤波消除时间序列上的噪音,小波滤波选用函数包含Daubechies(7-20),Coiflet(3-5)和Symlet(7-15)共26种类型。然后根据水稻移栽期、抽穗期和成熟期在EVI时间序列上的表现特征来识别水稻主要物候期。最后与东北地区24个站点水稻物候观测资料对比并分析误差。结果表明,Symlet11小波滤波的效果最好,其移栽期识别结果的误差绝大部分在±16 d,抽穗期和成熟期识别结果的误差在±8 d。表明通过此方法可以较好地识别东北水稻主要物候期,并可进一步应用到整个东北地区水稻的物候空间分布和时间变化特征研究上。

Detecting major phenological stages of rice using MODIS-EVI data and Symlet11 wavelet in Northeast China

Plant phenology refers to the emergence of an annual cycle of natural phenomena in plants affected by climate and other environmental factors. Information of crop phenology can directly reflect the effect of temperature and precipitation on the crop, which is essential for evaluating crop growth, productivity and crop management. Remote sensing technique is an important method to detect vegetation phenology with high spatial-temporal scales. Vegetation index generated by infrared and near-infrared band based on satellite remotely sensed data can reflect the status of vegetation growth and coverage more accurately. The Northeast China includes Liaoning, Jilin and Heilongjiang provinces, which rice area accounts for about 10% of total rice area and its rice yield accounts for 11% of total rice yield in China. Rice planting in Northeast China plays an important role in China's food safety. In this study, we developed a method for detecting phenological stages of rice in Northeast China based on the EOS-MODIS multi-temporal remote sensing data, including MOD09A1 and MCD12Q1 in 2008. The rice crop phenological stages were detected by using EOS-MODIS Enhanced Vegetation Index (EVI) data, and compared with the observed rice phenological stages in 24 selected agro-meteorological sites in Northeast China. The method consists of four procedures: (1) calculating the EVI value and extracting time profile from the location where 24 selected sites through MODIS-EVI data; (2) filtering the noise in EVI time profile by wavelet transform with twenty six types of wavelets; (3) identifying the rice planting date, heading date and ripening date by the variation characteristics in the smoothed EVI time profile; (4) comparing the result calculated by this method with the observed data from the 24 agro-meteorological sites in study area in 2008 and calculating the root mean square error (RMSE), then choosing the best type of wavelets. Due to the temporal resolution of the MODIS/Terra is 8 days, there will be missing data in the EVI time profile. The cubic spline interpolated (CSI) method (more smoother and stabler) was applied to repair the missing data, which can reflect the missing data more authentic. The twenty six types of wavelet: Daubechies(7-20), Coiflet(3-5) and Symlet(7-15) were used when filtering EVI time profile. The results showed that, the case using Symlet11 shows a remarkably good result in determining phenological stages, which is compared with the observed data. Most of the RMSE in planting date were less than 16 days. Most of the RMSE in heading date and ripening date were less than 8 days. It was shown that the Symlet11 filtering is the best method, which can be used to detect rice phenology in Northeast China. Furthermore, the Symlet11 filtering can be used to analyze variations and distribution of rice phenology with high-spatial scale in the whole Northeast China.