# Cloud gap-filling with deep learning for improved grassland monitoring

arXiv: https://arxiv.org/abs/2403.09554 · HTML: https://mlsanity.com/paper/2403.09554/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 5 of its 69 references that are in the map.
- First posted on arXiv 14 Mar 2024.

## Builds on (most cited first, 5 of 5)

- [Multi-Temporal Land Cover Classification with Sequential Recurrent Encoders](https://mlsanity.com/paper/1802.02080/) (2018)
- [Fusing Optical and SAR time series for LAI gap filling with multioutput Gaussian processes](https://mlsanity.com/paper/2012.02998/) (2020)
- [Long time-series NDVI reconstruction in cloud-prone regions via spatio-temporal tensor completion](https://mlsanity.com/paper/2102.02603/) (2021)
- [Fuzzy clustering for the within-season estimation of cotton phenology](https://mlsanity.com/paper/2211.14099/) (2022)
- [A Data Cube of Big Satellite Image Time-Series for Agriculture Monitoring](https://mlsanity.com/paper/2205.07752/) (2022)

## Built on by (most cited first, 2 of 2)

- [Mapping of Weed Management Methods in Orchards using Sentinel-2 and PlanetScope Data](https://mlsanity.com/paper/2504.19991/) (2025)
- [Deep Learning Super Resolution for Satellite Cloud Mask Downscaling](https://mlsanity.com/paper/2608.24715/) (2026)

## Abstract (arXiv)

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 employs a hybrid architecture combining Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) to generate continuous Normalized Difference Vegetation Index (NDVI) time series, highlighting the role of NDVI in the synergy between SAR and optical data. 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 improvement in the performance of the mowing event detection, with F1-score up to 84% using two widely applied mowing detection methodologies. Our method also effectively mitigated sudden shifts and noise originating from cloudy observations, which are often missed by conventional cloud masks and adversely affect mowing detection precision.