Resumen de Monitoring water fluxes in complex landscapes: improving remote sensing-based evapotranspiration models for treegrass ecosystems
Vicente Felipe Burchard Levine
The availability of water is fundamental for the adequate functioning of ecosystems and, by consequence, to the life and biodiversity they support, including human society. Water scarcity is expected to be exacerbated due to increasingly frequent and severe drought events, as predicted by climate change scenarios. Droughts have multifaceted impacts ranging from biodiversity degradation to decreases in food security. Evapotranspiration (ET), the transfer of water from the land surface to the atmosphere, is a fundamental process for the Earth system and a key indicator for drought monitoring in the context of natural and agricultural land. It is a major component of the global water balance and has strong links to the carbon and energy cycles, making ET a nexus and integrated process for Earth¿s biogeochemical cycling. Remote sensing offers the most feasible set of methods to retrieve ET at diverse spatial and temporal scales. However, their large and well documented uncertainties in heterogeneous landscapes remain largely unresolved. In particular, tree-grass ecosystems (TGEs), savanna-like landscapes representing 15-20% of the Earth surface, have unique structural and phenological features that present large difficulties for Earth observation and modeling methods. These landscapes contain multiple vegetation strata, with complex geometrical features, that are often misrepresented by conventional model structures. Therefore, central to this PhD thesis is the investigation of improving remote sensing-based ET models in TGEs to enhance our understanding of the soil-grass-tree-atmosphere continuum in relation to water and heat fluxes. The remote sensing-based two-source energy balance (TSEB), shown to be a robust model for different landscape types, was used as a reference. As a first step, a comprehensive global sensitivity analysis of TSEB was conducted to understand which parameters and sub-models were most influential in propagating model uncertainty. This led to the proposal of a new modeling strategy (i.e., TSEB-2S) that better integrated the distinctive seasonal dynamics of TGEs by separating the simulation period, and parameterization, into different phenological phases. Subsequently, the effect of spatial scale, namely the intrinsic sub-pixel heterogeneity of tree-grass mixtures, was analyzed using a multi-scale approach with a series of airborne hyperspectral imagery. The increased uncertainties at coarser spatial scales (i.e., pixel resolution > 10 m) were related to the poor depiction of aerodynamic characteristics using pixel averaging approaches, due to non-linear relationships between surface roughness and turbulent fluxes. The diagnosed uncertainties at both temporal (i.e., complex phenology) and spatial (i.e., tree-grass mixing) domains demonstrated the need to formulate a new model structure. To overcome these issues, a three-source energy balance (3SEB) model was proposed to inherently represent the distinct vegetation layers in TGEs. 3SEB was evaluated across four TGE experimental sites in Australia, Spain (2) and USA, with variability in climatic regimes and vegetation. 3SEB proved to be robust (root-mean-square deviation (RMSD) of ET ~60 W/m-2), improving over TSEB-2S (RMSD ~70 W/m-2) and TSEB (RMSD ~85 W/m-2) in addition to providing a framework to investigate ET partitioning from the different landscape components. These findings should help to alleviate the disproportionally large uncertainty of global remote sensing-based ET products in these important and extensive ecosystems, providing new avenues to understand the role of complex vegetation dynamics, at both temporal and spatial scales, in modulating ecosystem level fluxes and water scarcity.
