Abstract:<p>Multi-sensor remote sensing data fusion technologies have been developed and widely applied in recent years, providing a feasible and economical solution to increase the availability of high spatial and temporal resolution data. These methods, however, have been challenging to apply in highly heterogeneous areas, especially in complex agricultural landscapes where there are rapid changes at small scales, while features at larger scales change more slowly. In this study, we developed a novel method to reconstruct daily 30 m Normalized Difference Vegetation Index (NDVI) using imagery from the Moderate Resolution Imaging Spectroradiometer (MODIS), Landsat and Landsat-like platforms, and the Cropland Data Layer (CDL). This method utilizes a crop reference curve (CRC) approach, in which a set of NDVI time series are extracted from pure MODIS pixels (250 m resolution) identified using the CDL, and then used to fit Landsat-like observations (30 m). The CRC based method was applied over a complex agricultural landscape in the Choptank River watershed on the Eastern Shore of Maryland. Landsat data from 2013 and 2014 and Harmonized Landsat and Sentinel-2 (HLS) data from 2018 were used to reconstruct 30 m daily NDVI maps for major crop types. Results show that the relative error (RE) in reconstructed NDVI is around 6–8% during periods of rapid crop growth, and 3–5% during peak periods when growth is slow. The accuracy of the CRC method outperforms a standard image pair-based data fusion algorithm (Spatial and Temporal Adaptive Reflectance Fusion Model; STARFM), which yields RE of 4–9% in slow-growth periods and 10–16% in fast-growth periods when clear Landsat images are scarce. The CRC method was also compared with time-series data fusion methods, including a harmonic fitting model and the SaTellite dAta IntegRation (STAIR) model. The results show that CRC gives similar results when the Landsat-like image availability is high (around 27 images per year), but outperforms other methods when availability is limited (less than 15 images per year). The reconstructed NDVI time series for corn, soybean, winter wheat/soybean and forest at 30-m resolution show clear phenological patterns at the sub-field scale. The resulting 30-m NDVI timeseries data provide useful information for mapping crop phenology and monitoring crop condition in complex agricultural landscapes, especially for complex double-cropping areas. However, the input requirement of an accurate 30-m crop classification map constrains its application to areas and periods where classifications are available.</p>

NDVI Reconstruction by Synthesizing Optical and Synthetic Aperture Radar Images Based on CNN

Fusion of optical and SAR images based on deep learning to reconstruct vegetation NDVI time series in cloud-prone regions

RDNRnet: A Reconstruction Solution of NDVI Based on SAR and Optical Images by Residual-in-Residual Dense Blocks

A new deep learning-based model for reconstructing high-quality NDVI time-series data in heavily cloudy areas: fusion of Sentinel 1 and 2 data

Generative Adversarial Network Based NDVI Super-resolution Reconstruction

SAR Image Generation by Integrating Differentiable SAR Renderer with Neural Networks

Superpixel-based time-series reconstruction for optical images incorporating SAR data using autoencoder networks

Multi-Temporal Sentinel-1 and -2 Data Fusion for Optical Image Simulation

Reconstructing daily 30 m NDVI over complex agricultural landscapes using a crop reference curve approach

A Deep Learning-Based Spatio-Temporal NDVI Data Fusion Model

Efficient and Effective NDVI Time-Series Reconstruction by Combining Deep Learning and Tensor Completion

SAR-NeRF: Neural Radiance Fields for Synthetic Aperture Radar Multi-View Representation

Comparison of convolutional neural networks for cloudy optical images reconstruction from single or multitemporal joint SAR and optical images

A Novel Strategy to Reconstruct NDVI Time-Series with High Temporal Resolution from MODIS Multi-Temporal Composite Products

Dense NDVI Time Series by Fusion of Optical and SAR-derived Data

Nonsparse SAR Scene Imaging Network Based on Sparse Representation and Approximate Observations

A Deep Learning Approach for SAR Tomographic Imaging of Forested Areas

SAR2NDVI: a pix2pix generative adversarial network for reconstructing field-level normalized difference vegetation index time series using Sentinel-1 synthetic aperture radar data

GAN-based SAR and optical image translation for wildfire impact assessment using multi-source remote sensing data

Crop Mapping from Sentinel-1 Polarimetric Time-Series with a Deep Neural Network

Soil Moisture Retrieval in Farmland Areas with Sentinel Multi-Source Data Based on Regression Convolutional Neural Networks