Data Fusion of MODIS, HLS, and Environmental-Climate Data Using Bayesian Neural Network for Daily 30m Resolution NDVI Time Series

Keywords: Data Fusion, HLS, MODIS, Bayesian Neural Network, NDVI
Abstract Type: Poster Abstract

Abstract

The rapid advancement of satellite and airborne remote sensing technology provides abundant
data for monitoring vegetation. However, a significant challenge remains in balancing swath
width and revisiting time of remote sensing platforms. Integrating data from multiple remote
sensing data with different spatial and temporal characteristics offers a practical solution for
producing high-resolution synthetic observations more frequently.

The novel fusion strategy proposed in the research not only fuses the complementary
characteristics of the Moderate Resolution Imaging Spectroradiometer (MODIS) and
Harmonized Landsat Sentinel-2 (HLS) in terms of spatial and temporal resolutions, but also
integrates environmental-climate data, that shape and influence vegetation growth in the
Bayesian Neural Network (BNN) framework. BNN provides a probability distribution rather
than a single value for each contribution weight for each variable and uncovers hidden non-linear
relationships among variables. The research’s outcome, focused on daily 30-m surface
reflectance data and Normalized Difference Vegetation Index (NDVI) analysis, has critical
applications in monitoring vegetation phenology, predicting crop yields, and modeling biomass.

The proposed work addresses an urgent need for reliable, high-resolution vegetation data over
space and time by utilizing a cross-disciplinary data fusion approach to integrate multi-source
cross-scale datasets. The resulting framework offers a powerful tool for scientists in agriculture,
land use, and ecology, promoting better-informed decision-making and contributing to
advancements in agronomic research and practices.