Construction and validation of paddy rice index using phenological features of SAR time series

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 km², 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.