Cloud gap-filling with deep learning for improved grassland monitoring

Uninterrupted optical image time series are crucial for the timely monitoring of agricultural land changes, particularly in grasslands. However, the continuity of such time series is often disrupted by clouds. In response to this challenge, we propose an innovative deep learning method that integrates cloud-free optical (Sentinel-2) observations and weather-independent (Sentinel-1) Synthetic Aperture Radar (SAR) data. Our approach utilizes a combined Convolutional Neural Network (CNN)-Recurrent Neural Network (RNN) architecture to generate continuous Normalized Difference Vegetation Index (NDVI) time series, emphasizing the contribution of NDVI input component in the SAR-NDVI synergy. We demonstrate the significance of observation continuity by assessing the impact of the generated NDVI time series on the downstream task of grassland mowing event detection. We conducted our study in Lithuania, a country characterized by extensive cloud coverage, and compared our approach with alternative interpolation techniques (i.e., linear, Akima, quadratic). Our method outperformed these techniques, achieving an average Mean Absolute Error (MAE) of 0.024 and a coefficient of determination (R^2) of 0.92. Additionally, our analysis revealed enhanced performance in the subsequent mowing event detection. Evaluation based on widely applied mowing detection methodologies demonstrated a significant improvement in detection accuracy, with F1-score values of up to 84%. Furthermore, our method effectively mitigated sudden shifts and noise originating from cloudy observations, which are often missed by conventional cloud masks and adversely affect mowing detection precision.

This H5 file contains the feature space data used for training and evaluating the SF model.

grassland_geometries.geojson schema