Validation of the MODIS LAI product in the Murrumbidgee Catchment, Australia

Terrestrial vegetation plays an important role in the water, energy and carbon cycles at the land surface. Leaf Area Index (LAI), a measure of canopy structure, is an important biophysical parameter in ecosystem models which represent these cycles. However, field observations of LAI are rarely available at the resolution and coverage required for regionalor global-scale ecosystem modelling. Poor LAI estimates will introduce greater uncertainty into models, potentially compromising the accuracy of energy balance partitioning and related fluxes such as evapotranspiration. The MODerate resolution Imaging Spectroradiometer (MODIS) uses surface reflectance observations to provide 8-day composites of remotelysensed LAI estimates at a nominal spatial resolution of 1-km. However, even though various sources of estimation error exist, such as sub-pixel heterogeneity and approximations in retrieval algorithms, error estimates are not provided with the remotely-sensed LAI products. Previous validation studies in the USA and Australia have yielded contradictory results, with one demonstrating a positive bias in the MODIS LAI product, and the other a negative bias. In this paper, the latest MODIS LAI estimates (MYD15A2) are compared to Australian field observations across the Murrumbidgee Catchment in New South Wales. The ground observations were collected during three field campaigns: the National Airborne Field Experiment in November 2006 (NAFE’06), the Australian Airborne Cal/val Experiments for SMOS (AACES) in January/February and September 2010 and the Soil Moisture Active Passive Experiments (SMAPEx) in July and December 2011. Using these data, we conduct a characterization of the error properties of the MODIS LAI product using ground observations over varying land surface conditions and seasons. The results show that an assumption of Gaussian LAI errors for this region is reasonable. We also show that MODIS overestimates LAI for lightly vegetated regions (LAI < ~0.6) but underestimates LAI for densely vegetated regions (LAI > ~0.6). This result suggests that the contradictory results from previous validation studies may be associated with the different ranges of LAI observed in the studies. Overall, the bias in MODIS LAI is approximately -0.35 and the standard deviation of the error is 0.82. Observed LAI error properties from this study are expected to improve model predictions of the land-surface energy balance in ensemble model prediction studies underway for the Murrumbidgee Catchment.