基于物候特征的时序SAR水稻指数构建及验证

Construction and validation of paddy rice index using phenologicalfeatures of SAR time series

准确获取水稻的种植分布情况对于农业生产管理具有重要意义。针对多云雨地区缺乏有效光学影像影响水稻识别的问题,该研究以多时相Sentinel-1 SAR影像为主要数据源,基于后向散射系数的时序变化提取多个特征构建遥感指数以实现水稻提取。首先采用SG(savitzky-golay)滤波算法对不同地物的后向散射系数时序曲线进行平滑处理;然后计算像元与水稻样本点在水稻生育期内时序曲线的动态时间规整距离、像元时序曲线最小值与水体的差异、像元时序曲线最大值与植被的差异这3个值以量化水稻生育期的不同特征,并将三者相乘作为归一化差异水稻遥感指数(normalized difference rice index, NDRI)进行阈值分类后得出水稻空间分布情况。为验证该模型的有效性,该研究选取云南省盈江县为研究区。结果表明:1)基于时序SAR特征能够实现对水稻实现准确提取,模型总体精度达到89.42%,Kappa系数为0.82;2)盈江县2023年水稻种植面积为199.83 km^(2),且大盈江沿岸种植面积占全县的90.76%,水稻种植存在明显的集聚性。该研究为多云雨地区的水稻空间分布的准确提取提供一种方法,并可为相关农业政策制定提供参考。

Rice cropping has been the great contribution to the water consumer and greenhouse gas emitter.It is of great significance to accurately monitor the water resources and rice cropping area for the food security under climate change.Among them,Yingjiang County is located in the central area of Yunnan Province,indicating the better representation of typical agriculture.However,the optical remote sensing cannot fully meet the accurate identification of some crops,due to the pixel heterogeneity,mixed pixels,and spectral similarity.Moreover,it is still lacking on the optical images for crop monitoring under the cloudy and rainy climates during the paddy rice growing period in tropical regions.Some difficulties are also remained to extract the rice using multi-spectral remote sensing.Current techniques are focused mainly on the optics and radar image fusion for mapping paddy rice.The applicability has been limited in the absence of optical images.The purpose of this study is to monitor the rice planting areas with frequent clouds and rain in the potential application of Sentinel-1 SAR data.Several features were extracted from the temporal curve of backscattering coefficient using SAR data in the entire growth period of rice.Multiple features were also extracted from the temporal changes of backscattering coefficient in the SAR images.The large-scale water rice maps were obtained to combine the new remote sensing index,named NDRI(normalized difference vegetation index).The Sentinel-1 satellite was provided all-weather synthetic aperture radar imaging in the C-band.The Sentinel-1 SAR GRD dataset was accessed in the GEE cloud computing platform.Firstly,the SG filtering was used to smooth the temporal curves of backscattering coefficients in the different land covers.Then,the dynamic time warping distances among different temporal curves were calculated using rice samples as a reference.Meanwhile,two feature values were calculated to quantify the different characteristics of rice growth,including the difference between the minimum temporal curves and water bodies,and the difference between the maximum temporal curves and vegetation.Their multiplication was taken as the remote sensing index for the classification of rice threshold.Finally,the spatial distribution of rice was further corrected to generate the slope data from NASADEM images using overlay analysis.The effectiveness of the improved model was validated,taking Yingjiang County in Yunnan Province as the study area.The optical images were almost unusable during the period from June to September in the whole growing season of paddy rice,due to the extensive cloud cover.The results showed that:1)The temporal SAR features of rice were accurately extracted with an overall accuracy of 89.42%and a Kappa coefficient of 0.82;2)The planting area of rice in Yingjiang County in 2023 was 199.83 km^(2),where the planting area along the Dayingjiang River was accounted for 90.76%of the total county.There was the significantly spatial aggregation of rice planting.This finding can provide the strong reference to accurately extract the spatial distribution of rice in the cloudy and rainy areas for the decision-making on agricultural policies.