| Under the background of climate change,Ratoon rice has become an important planting pattern for actively adapting to climate change,making full use of water and heat resources,ensuring food security,and mitigating agro-ecological environmental problems.Remote sensing monitoring research for ratoon rice is still in its infancy.Southwest China is cloudy and rainy,the validity of optical remote sensing data is limited,and a single sensor cannot meet the high spatial and temporal resolution data required for the study.Scientific questions such as reconstructing high quality,high spatiotemporal resolution remote sensing datasets in cloudy and rainy areas,developing remote sensing monitoring models for ratoon rice,and scientifically identifying the spatial and temporal patterns of ratoon rice need answering.We took the Yongchuan District of Chongqing as the study area,and constructed models to realize the extraction of ratoon rice planting areas using Sentinel-2,Landsat8,MODIS optical remote sensing data and Sentinel-1 microwave remote sensing data respectively,and compared and analyzed the similarities and differences between the two results from qualitative and quantitative perspectives.The main work and conclusions are as follows:(1)Given the limitations of optical remote sensing data in cloudy and rainy areas,we proposed to combine the cloud removal interpolation method with a spatiotemporal fusion model to construct a high spatiotemporal series dataset,and then combined it with the phenological information to construct a threshold model for ratoon rice planting extraction.We used the modified neighborhood similar pixel interpolator(MNSPI)method and the flexible spatiotemporal data fusion(FSDAF)model to remove cloud interpolation and fuse Sentinel-2,Landsat8,and MODIS data to generate a fused time series dataset with a spatial resolution of 10 m and a temporal resolution of 5～16 days.On this basis,we calculated the normalized difference vegetation index(NDVI)and land surface water index(LSWI),and used the Savitzky-Golay filter(S-G)method for smoothing.We used growth characteristics and curve changes in ratoon rice to construct a threshold model to monitor ratoon rice.We verified the accuracy based on field data,the overall accuracy was 90.73%,and the Kappa coefficient was 0.81.The results showed that the planting area of ratoon rice in the study area was 194.17 km~2.(2)Microwave data has the benefit of penetrating clouds and fog,and has great potential for application in cloudy and rainy regions.Based on Sentinel-1 radar data,we obtained the change characteristics of the backscattering coefficient during the growth period of ratoon rice and combined it with the phenological information to construct a threshold model for microwave remote sensing monitoring of ratoon rice in the study area.We first selected the VV and VH polarization methods of Sentinel-1 radar data.Then used the Lee sigma filter for speckle noise processing and the S-G filter for smoothing.Next,depending on the variation characteristics of backscattering coefficients of ratoon rice,water bodies and building areas established a threshold model.We obtained the result by determining the thresholds based on the training samples and the phenological information.We verified the accuracy based on field data,the overall accuracy was 90.24%,and the Kappa coefficient was 0.80.The results showed that the planting area of ratoon rice in the study area was 199.79 km~2.(3)We have compared the obtained optical and microwave remote sensing monitoring results,including quantitative and qualitative analyses.The quantitative analysis includes a comparison of the overall distribution and detailed characteristics of ratoon rice planting areas.The qualitative analysis includes the characteristics of the growth period curves of ratoon rice under the two types of data,and the comparison of the differences and correlations of the ratoon rice area in each township.The two types of results were consistent in terms of the overall distribution,detailed characteristics and curve characteristics,and the area correlation coefficient of each township was 0.8894(P<0.01),which was significantly correlated.The differences were mainly reflected in the area of ratoon rice planted in several towns around 29°20’00"N,which is mainly related to the original data,potential noise interference and other factors.Based on the above analysis,if there are more optically valid images of the study area,we suggest using a spatiotemporal fusion model to reconstruct the optical dataset.If the optical remote sensing images cannot meet the time series requirements,we suggest using microwave data.If we wish to improve the accuracy of the results,we think we can fuse the optical and microwave data features for classification extraction.Aiming at the monitoring of ratoon rice in cloudy and rainy areas,we constructed two threshold models from both optical remote sensing data sources and microwave remote sensing data sources,and integrated methods such as cloud interpolation,spatiotemporal fusion,and phenological information to improve the accuracy of ratoon rice.Our model could effectively distinguish ratoon rice from other cropping patterns,enriching the technical system for remote sensing monitoring of rice,and has certain implications for agricultural remote sensing applications in cloudy and rainy areas.It also provides more scientific and detailed technical support for food security and sustainable management of the ecological environment,which is of scientific and innovative significance. |
