Modelling surface turbulent heat fluxes over heterogeneous landscapes

The topic of this thesis is “Modelling Surface Turbulent Heat Fluxes over Heterogeneous Landscapes”. Monitoring of surface turbulent heat fluxes is crucial in our understanding of changes in the hydrological, energy, and carbon cycles and their associated impacts on freshwater availability, agriculture, energy production, ecosystem health, and air pollution, among other applications. Besides, quantifying surface turbulent heat fluxes is critical in minimizing and mitigating social, economic, and environmental risks associated with weather and climate-induced extreme events such as floods, droughts, and heatwaves. Modelling approaches are more applicable for quantifying spatially distributed surface turbulent heat fluxes over regional and global scales than measurements owing to the sparse distribution of observation stations in most parts of the world. Surface turbulent heat fluxes are highly sensitive to variations in surface characteristics. Often, the modelling schemes adopted rarely capture the spatialtemporal dynamics in surface characteristics introducing biases in the modelled fluxes. Uncertainties in most modeling schemes emanate from unrepresentative input data, more so in heterogeneous landscapes, and deficiencies in the model parameterization of the physical processes. These two issues are the main focus of this research.
Land surface temperature (LST) is a critical variable in the modelling of surface turbulent heat fluxes, among other applications. Mapping of LST requires thermal infra-red (TIR) data obtained through thermal remote sensors. However, thermal remote sensors are often plagued by spatial-temporal resolution trade-off. Thus, downscaling approaches are often applied to improve the spatial-temporal resolution of LST derived from TIR data. In Chapter 2, an improved Random Forest (RF) regression approach for downscaling coarse resolution TIR data in complex terrain is presented. The proposed downscaling approach aims at minimizing spatial averaging biases inherent in the conventional RF regression-based downscaling approaches applied in the downscaling of coarse TIR data. Downscaling results indicate that the proposed approach is better suited for downscaling coarse TIR data, especially in complex landscapes.
Meteorological variables are also key inputs in soil-vegetation-atmosphere transfer (SVAT) modelling schemes. Although not as highly variable in space as LST, meteorological variables tend to be highly influenced by topography among other surface characteristics. Spatially distributed meteorological variables are often obtained from General Circulation Models (GCMs) which are too coarse for local to regional applications more so in complex terrain. Consequently, downscaling approaches are also required in order to improve the resolution of GCM derived meteorological variables for local to regional applications. Unlike LST, downscaling of meteorological variables is done through dynamical modelling approaches using numerical weather prediction (NWP) models such as the Weather Research Forecasting (WRF) Model. In Chapter 3, the ability of WRF model to retrieve high-resolution surface meteorological variables from coarse resolution fifth generation of the European ReAnalysis (ERA5) dataset is evaluated. WRF-derived meteorological variables show a significantly better spatial variability consistent with the complex topography of the Kenyan highlands. It is also observed that the choice of planetary boundary layer (PBL) parameterization scheme mainly influences the retrieval of wind speed and rainfall more than air temperature and relative humidity.
In addition to issues related to the representativeness of the input data, SVAT modelling schemes are plagued by problems related to the parameterization of heat transfer processes. Chapters 4 and 5 address issues related to the parameterization of surface roughness characteristics in SVAT modelling schemes. Surface roughness characteristics are often difficult to represent in models partly due to the complexity of the processes involved and the lack of adequate data. Thus, surface heat exchange in most SVAT schemes is often represented through simple parameterizations. In Chapter 4, a more physically based parameterization of the parameter kB-1 in the Surface Energy Balance System (SEBS) model is presented. In the revised parameterization a variable canopy heat transfer coefficient (Ct) is adopted as opposed to a constant Ct as used in the original SEBS model. The innovation in the revised parameterization is its ability to account for the effect of turbulence on Ct. The revised approach significantly reduces the underestimation of sensible heat flux in tall forest canopies compared to the parameterization adopted in the original SEBS model.
The better performance with the revised parameterization in tall forest canopies indicates a better ability of the revised parameterization in accounting for turbulence-induced heat transfer.
Often, reliable information on surface roughness characteristics such as canopy height and structure is difficult to obtain, especially from remote sensing in highly fragmented landscapes. Most SVAT modelling schemes are quite sensitive to variations in canopy height and structure. Thus, the lack of reliable information on surface roughness characteristics will lead to significant biases in the modelled surface turbulent heat fluxes despite the ability of the SVAT scheme adopted. In Chapter 5, a simple surface temperature-forced Bowen Ratio Energy Balance (SBREB) method is proposed. The innovation in this approach is that it models surface turbulent heat fluxes without explicitly characterizing surface roughness characteristics. Surface roughness characteristics in SBREB are inherently accounted for through LST. SBREB performs fairly well compared to SEBS based on the revised parameterization in Chapter 4. Thus, SBREB is a suitable and reliable alternative for monitoring surface turbulent heat fluxes, especially in heterogeneous landscapes.
The general conclusion from this research is that reliable modelling of surface turbulent fluxes is largely dependent on how well a modeling scheme can describe the physical processes involved given the available input data.