Estimating the fraction of absorbed photosynthetically active radiation from multiple satellite data

Abstract

The fraction of absorbed photosynthetically active radiation (FAPAR) is a critical input parameter in many climate and ecological models. The accuracy of satellite FAPAR products directly influences estimates of ecosystem productivity and carbon stocks. The targeted accuracy of FAPAR products is 10%, or 0.05, for many applications. This study evaluates satellite FAPAR products, presents a new FAPAR estimation model and develops data fusion schemes to improve the FAPAR accuracy. Five global FAPAR products, namely MODIS, MISR, MERIS, SeaWiFS, and GEOV1 were intercompared over different land covers and directly validated with ground measurements at VAlidation of Land European Remote sensing Instruments (VALERI) and AmeriFlux sites. Intercomparison results show that MODIS, MISR, and GEOV1 agree well with each other and so do MERIS and SeaWiFS, but the difference between these two groups can be as large as 0.1. The differences between the products are consistent throughout the year over most of the land cover types, except over the forests, because of the different assumptions in the retrieval algorithms and the differences between green and total FAPAR products over forests. Direct validation results show that the five FAPAR products have an uncertainty of 0.14 when validating with total FAPAR measurements, and 0.09 when validating with green FAPAR measurements. Overall, current FAPAR products are close to, but have not fulfilled, the accuracy requirement, and further improvements are still needed. A new FAPAR estimation model was developed based on the radiative transfer for horizontally homogeneous continuous canopy to improve the FAPAR accuracy. A spatially explicit parameterization of leaf canopy and soil background reflectance was derived from a thirteen years of MODIS albedo database. The new algorithm requires the input of leaf area index (LAI), which was estimated by a hybrid geometric optic-radiative transfer model suitable for both continuous and discrete vegetation canopies in this study. The FAPAR estimates by the new model was intercompared with reference satellite FAPAR products and validated with field measurements at the VALERI and AmeriFlux experimental sites. The validation results showed that the FAPAR estimates by the new method had slightly better performance than the MODIS and the MISR FAPAR products when using corresponding satellite LAI product values as input. The FAPAR estimates can be further improved with the LAI estimates from the presented model as input. The improvements are apparent at grasslands and forests with an 8% reduction of uncertainty. The new model can successfully identify the growing seasons and produce smooth time series curves of estimated FAPAR over years. The root mean square error (RMSE) was reduced from 0.16 to 0.11 for MODIS and from 0.18 to 0.1 for MISR overall. Application of the presented model at a regional scale generated consistent FAPAR maps at 30 m, 500 m, and 1100 m spatial resolutions from the Landsat, MODIS, and MISR data. As an alternative method to improve FAPAR accuracy, in addition to developing FAPAR estimation models, two data fusion schemes were applied to integrate multiple satellite FAPAR products at two scales: optimal interpolation at the site scale and multiple resolution tree at the regional scale. These two fusion schemes removed the bias and resulted in a 20% increase in the R2 and a 3% reduction in the RMSE as compared with the average of the individual FAPAR products. The regional scale fusion filled in the missing values and provided spatially consistent FAPAR distributions at different resolutions. The original contribution of this study is that multiple FAPAR products have been assessed with a comprehensive set of measurements from two field experiments at the global scale. This study improved the accuracy of FAPAR using a new model and local pixel based soil background and leaf canopy albedos. High FAPAR accuracy was achieved through integration at both the temporal and spatial domains. The improved accuracy of FAPAR values from this study by 5% would help to decrease an equal amount of uncertainty in the estimation of gross and net primary production and carbon fluxes.