Improving Leaf Area Index estimation with satellite imagery

Abstract

The Leaf Area Index (LAI) is an important parameter characterising vegetation and knowledge of LAI is crucial for describing the activities within an ecosystem. It is widely used as a basic input parameter in hydrological and bio-chemical models for the estimation of the water-cycle, agricultural primary production and other parameters. This study attempted to improve the spatial resolution of MODIS LAI product (1000m pixel size), through the regression analysis applied to MODIS EVI (1000m) and LAI data and the use of the estimated regression equations in a downscaling model using images from the EVI product of Landsat (30m) and several land-cover maps. Regression analysis was applied in 5 selected study sites around the world, which differ with respect to the climate conditions. Several scenarios were tested in order to find the important parameters affecting the LAI-EVI relationship (vegetation type, seasonality) and the data were found to be described best by a linear fit. During the downscaling process the estimated LAI-EVI equations were used to calculate LAI values and subsequently create LAI maps at the Landsat spatial resolution level (30m). The results of the study showed that vegetation type has the highest influence on the EVI-LAI relationship, as well as that the sensitivity of EVI to LAI is lower in periods of high biomass production. The created LAI maps showed visual similarity of high level in patterns of LAI value distribution, when compared to the corresponding lower resolution LAI maps (MODIS). The comparison of field data with the model estimated values of LAI showed high correlation especially during the dry period. The lowest correlation was observed during the rainy season when the availability of cloud free pixels in the LAI images was low. For most of the cases examined, the model gave statistically significant results (at 0.05 and 0.001 level) with the r coefficient values ranging from negative (-0.1783, one case) and relatively low (2 cases, 0.25 and 0.32) values, to moderate (3 cases, 0.4-0.7) and high (5 cases, 0.7-0.935). Limited samples per vegetation type on specific dates, the diversity of vegetation species within the same vegetation type, as well as saturated EVI values were evaluated to be the most possible factors affecting the regression analysis results. The model estimated LAI values were found to correlate better with the field measurements in study areas of the Southern Hemisphere.