Monitoring crop growth using a canopy structure dynamic model and time series of synthetic aperture radar (SAR) data

Normalized Difference Vegetation Index (NDVI) time series data are used by agricultural agencies for many essential operational crop monitoring programmes. But optical sensors often miss key growth stages due to cloud cover interference, impacting the performance of operational activities. Although the synergistic use of optical and Synthetic Aperture Radar (SAR) imagery can provide time series data, any SAR-optical integration necessitates building a relationship between these two data sources. The objective of this study was to use a semi-empirical Canopy Structure Dynamics Model (CSDM), Growing Degree Days (GDD), and SAR parameters calibrated to optical NDVI to derive daily estimates of canola crop condition over an entire growing season. RADARSAT-2 Fine Quad-pol and RapidEye images were collected over three years for a study site in western Canada. Object-based image analysis was applied to study the relationship between the optical and SAR time series data. Significant correlations were documented between a number of SAR parameters and optical NDVI, specifically a ratio of backscatter intensities (HH-HV)/(HH+HV), a ratio of volume to surface scattering extracted from the Freeman Durden decomposition, and Entropy from the Cloude-Pottier decomposition. Correlations (r-values) between these SAR parameters and optical NDVI ranged from 0.63 to 0.84 for the three years of data. Based on this analysis, a simple statistical model was used to relate SAR parameters to optical NDVI, creating a SAR-calibrated NDVI (SAR(cal)-NDVI). A CSDM was fit to the SAR(cal)-NDVI for each canola field, constructing a temporal vegetation index curve which captured canopy development from emergence to senescence. Coefficients of determination (R-2) were 0.87 0.86 and 0.82 for entropy, the volume-surface scattering ratio, and the ratio of backscatter intensities (HH-HV)/(HH+HV), respectively, demonstrating a good model fit. The CSDM describes well the temporal evolution of SAR(cal)-NDVI. Using the CSDM, SAR(cal)-NDVI and GDD, the canola condition can be estimated for any given day in the growing season. In fact when the CSDM was used to estimate SAR(cal)-NDVI for the exact days of RapidEye acquisitions, correlations with optically derived NDVI were high. The strongest correlations with RapidEye NDVI were reported for the volume-surface scattering ratio (R-2 of 0.69 and RMSE of 0.15). The SAR(cal)-NDVI estimated from the CSDM was also physically meaningful. Field-based biomass was significantly correlated (R-2 of 0.79) with the SAR(cal)-NDVI calculated using the volume-surface scattering ratio. Although further research is needed to extend this method to other crops, these results demonstrate that SAR data can be used to estimate vegetation conditions and when coupled with a CSDM, integrated into current monitoring operations based on optical NDVI. As a next step, the research team will be assessing SAR(cal)-NDVI in a national operational programme which reports on crop yields using modelling with optical-based NDVI.