Files

No attached file found for this publication.

Details

Authors
Abstract
The green area index (GAI) is a key biophysical variable for crop monitoring, widely used to assess crop health, growth, and productivity. Most large-scale and cost-effective methods for estimating the GAI rely on optical remote sensing data; consequently, frequent cloud cover can severely limit their reliability. This challenge is ever more pressing in tropical regions, where timely vegetation monitoring is essential for food security and where cloud cover often overlaps with key phases of vegetation growth. In these instances, synthetic aperture radars (SARs) offer a valuable alternative, as their cloud-penetrating capabilities enable the generation of dense time series that can enhance the spatial and temporal coverage of optical data. In recent years, various methods have been proposed to address gaps in time series of biophysical variables retrieved from optical data, including fusion techniques that aim to reproduce or enhance optical imagery using ancillary Earth observation data to compensate for cloud cover. However, there have been relatively few efforts to systematically leverage dense SAR time series to directly fill gaps in GAI time series, despite the potential for reducing modeling errors and production time. In this study, a method is proposed to fill these gaps as they occur along the crop growing season with current and past SAR data as well as past GAI values. The focus on near real-time gap-filling ensures enhanced temporal resolution and timeliness, addressing critical needs in operational crop monitoring. The approach involves the use of a transformer encoder, a deep learning architecture that exploits the sequential nature of the values of the target variable and its complex relationship with SAR backscatter and interferometric coherence. Sentinel-1 and Sentinel-2 data acquired from 2018 to 2021 over the Hesbaye region of Belgium are used for cross-validation. The results demonstrate the robustness of the method. The model can successfully retrieve the GAI at the parcel level on an unseen growing season with a mean R2 of 0,88 and RMSE of 0,71. External validation with in situ data collected from 10 maize fields in 2018 in Belgium further confirms its accuracy, outperforming traditional approaches based on Water Cloud model inversion. These promising results not only highlight the immediate applicability of this approach but also its potential for broader impact. While this study focused on maize—a high-biomass crop that has been shown to be challenging to monitor using C-band SAR—the method shows promise for extension to other major temporary crops. Additionally, future advancements incorporating multi-frequency SAR data, using the L-band data from forthcoming NISAR and ROSE-L missions, are anticipated to further enhance its performance. In the end, by enabling the generation of accurate and dense GAI time series throughout the crop growing season, this method has the potential to significantly advance the capability of operational crop monitoring systems in cloud-prone regions, where timely delivery of information on crop condition is critical for informed decision-making.
Affiliations

Citations

  • APA
  • Chicago
  • FWB

Bouchat, J., Deffense, Q., De Maet, T., & Defourny, P. (2025). Synergistic use of optical and SAR imagery for near real-time green area index retrieval in maize. Living Planet Symposium 2025, Vienna, Austria.