Generating MODIS-consistent High-resolution Leaf Area Index for Landsat and Sentinel-2 with a Data-driven Approach

Abstract:

Long-term records of Leaf Area Index (LAI) are conventionally available at coarse spatial resolutions (e.g., 0.5 – 4 km) and have been widely applied to Earth system modeling. Recently, the advancements in decametric-resolution sensing systems provide new opportunities to develop global LAI products with higher resolution and revisit frequency, which is essential to boost agricultural and natural resources applications. Here, we present a robust, data-driven approach to estimate LAI from Landsat and Sentinel-2 images. This approach is based on a comprehensive dataset composed of Landsat/Sentinel-2 surface reflectance and MODIS LAI across the globe over multiple years. We adopt advanced data cleaning and balancing techniques to ensure global representativeness across diverse biomes, vegetation growth stages, and different sensors. Machine learning algorithms are trained to model the underlying relationships between surface reflectance and LAI. We also perform deep analysis to understand the impact of land cover information on LAI modeling at a global scale. The combined use of Landsat and Sentinel-2 data provides frequent LAI observation at a 5-day time step on average. Implemented on Google Earth Engine, our algorithm also enables the rapid generation of long-term high-resolution LAI estimations over the entire globe. In this presentation, we will describe the general methodology and discuss preliminary evaluation results against MODIS LAI products and ground measurements.