Modelling terrestrial ecosystem productivity using remote sensing data

摘要

Production efficiency models (PEMs) have been developed to aid with the estimation of terrestrial ecosystems productivity where large spatial scales make direct measurement impractical. One of the key datasets used in these models is the fraction of photosynthetic active radiation absorbed by vegetation (FAPAR). FAPAR is the single variable that represents vegetation function and structure in these models and hence its accurate estimation is essential. This thesis focused on improving the estimation of FAPAR and developing a new PEM model that utilises the improved FAPAR data. Foremost, the accuracy of operational LAI/FAPAR products (i.e. MGVI, MODIS LAI/FAPAR, CYCLOPES LAI/FAPAR, GLOBCARBON LAI/FAPAR, and NN-MERIS LAI TOC algorithm) over a deciduous broadleaf forest was investigated. This analysis showed that the products varied in their prediction of in-situ FAPAR/LAI measurements mainly due to differences in their definition and derivation procedures. The performance of three PEMs (i.e. Carnegie-CASA, C-Fix and MOD17GPP) in simulating gross primary productivity (GPP) across various biomes was then analysed. It was shown that structural differences in these models influenced their accuracy. Next, the influence of two FAPAR products (MODIS and CYCLOPES) on ecosystem productivity modelling was analysed. Both products were found to result in overestimation of in-situ GPP measurements. This was attributed to the lack of correction for PAR absorbed by the non-photosynthetic components of the canopy by the two products. Only PAR absorbed by chlorophyll in the leaves (FAPAR chlorophyll) is used in photosynthesis and hence it was hypothesised that deriving and using this variable would improve GPP predictions. Therefore, various components of FAPAR (i.e. FAPAR canopy, FAPAR leaf and FAPAR chlorophyll) were estimated using data from a radiative transfer model (PROSAIL-2). The FAPAR components were then related to two sets of vegetation indices (i.e. broad-band: NDVI and EVI, and red-edge: MTCI and CIred-edge). The red-edge based indices were found to be more linearly related to FAPAR chlorophyll than the broad-band indices. These findings were also supported by data from two flux tower sites, where the FAPAR chlorophyll was estimated through inversion of net ecosystem exchange data and was found to be better related to a red-edge based index (i.e. MTCI). Based on these findings a new PEM (i.e. MTCIGPP) was developed to (i) use the MTCI as a surrogate of FAPAR chlorophyll and (ii) incorporate distinct quantum yield terms between the two key plant photosynthetic pathways (i.e. C3 and C4) rather than using species-specific light use efficiency. The GPP predictions from the MTCIGPP model had strong relationship with the in-situ GPP measurements. Furthermore, GPP from the MTCIGPP model were comparable to the MOD17GPP product and better in some biomes (e.g. croplands). The MTCIGPP model is simple and easy to implement, yet provides a reliable measure of terrestrial GPP and has the potential to estimate global terrestrial carbon flux.