Uncertainty associated with scaling spectral indices of carbon fluxes at various spatial and temporal scales

• Main Author: Kia, Seyed Hossein
• Other Authors: Milton, E. J. ; Atkinson, Peter
• Published: University of Southampton 2017
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.736692

Measurements from the global network of micrometeorological tower sites (FLUXNET) provide essential information on the ecosystems productivity (i.e. a key component for the study of the carbon cycle). However, the area sampled by instruments on a flux tower is poorly defined and varies with weather conditions. Additionally, gaps in the FLUXNET record are common, either due to unsuitable measurement conditions or instrument failure. Hence, remote sensing (RS) has been proposed as a way to enhance the FLUXNET database, as it provides complete spatial coverage and frequent repeat observations. In practice, the use of RS for this task is challenging. The integration of spatially-explicit ecosystem models, RS observations and eddy covariance (EC) flux measurements with environmental variables have facilitated the quantification of carbon cycling dynamics across multiple spatial and temporal scales. In this regard, process-based models with the aim of simulation of carbon dynamics in forest ecosystems are increasingly being used besides other tools to predict the effects of environmental factors on the forest carbon pool and forest productivity. However, despite this, decision makers must be aware of the limitations of these models by uncertainty analyses to make the process-based models more robust and to optimize them for estimating productivity at landscape level. There is a need to address various sources of uncertainty associated with such quantification, including sensor limitations in terms of support size defined by spatial and temporal resolutions; spatial heterogeneity of land surface properties; pre-processing calibration; and the structure of the model proposed and its parameterization. calibration; and the structure of the model proposed and its parameterization. This research investigated the sources of uncertainty mentioned across three domains of interest: spectral, spatial and temporal. The present research deals with this need using a combination of tower-based EC flux measurements and RS data from both airborne and satellite RS systems at a range of temporal and spatial scales. It incorporates multiple remote sensing data sets (Airborne LiDAR, Airborne Imaging Spectrometry, DMC, and MODIS) to derive indices related to canopy structure (nCHM), plant cover (NDVI) and photosynthesis processes (PRI), and attempts to relate these to the data measured by instruments on flux towers in two locations: Wytham Woods, southern England, and Chequamegon Nicolet National Forest in northern Wisconsin, USA. In terms of the spectral domain, the research adds to the evidence that NDVI alone is insufficient to fully characterize the primary productivity of plant canopies, specifically across heterogeneous landscape. The study also demonstrates the magnitude and variability of extraneous parameters (e.g. optical geometry, shadow fraction, soil background and aerosol) in RS observations of the mixed forest of Wytham Woods using a 3D forest light simulation model (FLIGHT model). The results reveal that the observed vegetation indices (NDVI and PRI) form the mixed forest is highly sensitive to variation in solar and view zenith angles and soil background, while the indices are relatively robust to aerosol scattering. In the temporal domain, the research makes use of a unique time-series of ten multispectral images acquired during a single growing season by the DMC satellite sensors. The heterogeneity of canopy cover has greatest impact on the DMC data early in the season, and this highlights the importance of understanding how the flux tower footprint varies with weather conditions. As the canopy began to green-up, the precision of temporal sampling became more important. Based on explicit representation of the time-varying flux tower footprint, prediction of flux tower measurements directly from space-borne coarse spatial resolution imagery is challenging and leads to a low predictive ability. In order to use the global FLUXNET EC dataset and RS observations to estimate the ecosystem productivity at regional and global scales, this research deals with an upscaling approach that it involves flux footprint climatology modelling and RS-based light use efficiency (LUE) model fusion. In this aspect, a large correlation is found between satellite-based PRI and the EC-based LUE of a homogeneous deciduous forest. However, estimating the regional level LUE of a heterogeneous landscape from space is still an uncertain process as the required spectral index (PRI) is affected by canopy level variables as well as the geometry of illumination and view. Furthermore, the sensitivity analysis of a simple Diagnostic Carbon Flux Model (DCFM) to seven input parameters (ε_max,a, β, R'_refγ, λ and E₀) using a five-years record of the EC data from four flux towers selected across various plant functional types (PFTs) in the Upper Midwest region of northern Wisconsin, is considered to optimize the RS-based LUE model for estimating regional productivity. The results confirm that empirical constants for the estimation of the fPAR absorbed by vegetation canopies (a and β) next to the maximum light use efficiency (ε_max) has little impact on the fluctuations of net carbon exchange within each PFT whereas DCFM model was very sensitive to so, the estimation of this factor, in comparison with the other parameters, plays the key role in the accuracy of NEE's predictions. Moreover, except for homogenous canopy cover, in other PFTs, interactions among the crucial ecophysiological parameters have minor contribution to uncertainty of NEE prediction by DCFM model. In conclusion, the results demonstrated the potential combination of the satellite based approach, flux footprint modelling and data-model fusion for improving the accuracy of regional/global productivity estimations. This approach includes four steps: (1) a RS-based LUE model for estimating productivity; (2) EC flux footprint analysis for the corresponding RS images; (3) using the footprint integration of RSbased ecosystem productivity to be comparable with the tower-based EC-derived productivity values, several key parameters of the RS-based LUE model can be optimized using the DCFM; and (4) The optimized RS-based LUE model can be applied for estimating regional productivity.