Assessment of Three Methods for Near Real-Time Estimation of Leaf Area Index From AVHRR Data

Near real-time (NRT) estimation of leaf area index (LAI) is essential for monitoring rapid surface process changes within operational systems. This paper assesses the performances of three methods for the NRT estimation of LAI: 1) Whittaker (Whit); 2) Gaussian process model (GPM); and 3) the climatological temporal smoothing and gap filling (CTSGF). The methods were evaluated using Advanced Very High Resolution Radiometer time series over a selection of BELMANIP2 sites representative of seasonal patterns of global biome vegetated areas and under varying level of noise and missing observations (gaps). A simulation experiment was designed to evaluate the predictive capabilities of the three methods with an emphasis on the global and local structure of missing observations in the time series. The results show that the three methods achieve similar performances (RMSE < 0.4) when the fraction of missing data over the whole time series is lower than 65% or the length of gaps is smaller than 10 days. Conversely, for fraction of gaps higher than 65% or periods of gaps longer than 10 days, CTSGF is found to provide more accurate (RMSE < 0.4 up to 60 days with missing data) NRT estimates of LAI than Whit and GPM. CTSGF uses the baseline seasonal cycle derived from the interannual median values of LAI to fill gaps in the time series and improve the NRT projections. Our findings support the operational use of the CTSGF algorithm for NRT estimation of biophysical products at the global scale.