Documenting improvement in leaf area index estimates from MODIS using hemispherical photos for Australian savannas

This paper compares estimates of Leaf Area Index (LAI) obtained from the MODIS (Moderate Resolution Imaging Spectroradiometer) collections 4.8 (MC4) and 5.0 (MC5) with ground-based measurements taken along a 900km north-south transect through savanna in the Northern Territory, Australia. There was excellent agreement for both the magnitude and timing in the annual variation in LAI from MC5 and biometric estimates at Howard Springs, near Darwin, whereas MC4 overestimated LAI by 1-2m²m^-2 for the first 200 days of the year. Estimates of LAI from MC5 were also compared with those obtained from the analysis of digital hemispherical photographs taken during the dry season (September 2008) based on algorithms that included random and clumped distribution of leaves. Linear regression of LAI from MC5 versus that using the clumping algorithm yielded a slope close to 1 (m=0.98). The regression based on a random distribution of leaves yielded a slope significantly different from 1 (m=1.37), with higher Mean Absolute Error (MAE) and bias compared to the clumped analysis. The intercept for either analysis was not significantly different from zero but inclusion of five additional sites that were visually bare or without green vegetation produced a statistically significant offset of +0.16m²m^-2 by MC5. Overall, our results show considerable improvement of MC5 over MC4 LAI and good agreement between MC5 and ground-based LAI estimates from hemispherical photos incorporating clumping of leaves. © 2011 Elsevier B.V.