Author: Nastaran Chitsaz
Chitsaz, Nastaran, 2024 Advancing Evaporation and Runoff Simulation: Incorporating CO2 and Environmental Variables into Stomatal Conductance Utilizing Mixed Generalized Additive Models, Flinders University, College of Science and Engineering
Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.
Catastrophic floods and runoff events are increasingly prevalent due to the influence of anthropogenic climate change and variability. Evapotranspiration (ET), the process that governs the exchange of water and energy between the atmosphere, land surface, and groundwater, plays a crucial role in the simulation of runoff within hydrological models. However, accurately estimating ET remains a significant challenge for these models. Presently, many hydrological models rely heavily on potential evapotranspiration (PET) models, as observed ET data is often limited. PET refers to the maximum possible water loss from the soil and vegetation to the atmosphere when water is not limited. PET estimation is challenging due to the complexity of the processes involved and the various sources of uncertainty.
The main source of uncertainty in PET simulation is neglecting the impact of CO2 on plant water use, which leads to inaccurate runoff simulation. In response to the rising CO2 concentration, plants close their stomata and decrease stomatal conductance (gs), which can reduce the amount of water loss through transpiration. A decrease in plant transpiration and an increase in water use efficiency can result in greater antecedent soil moisture and, therefore, increased runoff. Hence, runoff simulations need to consider the relative role of climate change in ecosystems through the PET equation. However, the response of plants to CO2 varies significantly between different biomes and plant species around the world. In addition, the effects of CO2 on plant physiology and morphology have complex interactions with other environmental variables such as air temperature (TA), radiation (R), vapour pressure deficit (VPD), and soil water content (SWC). Therefore, the response of plants to CO2 is characterised by high uncertainty with significant knowledge gaps.
In the first and second chapters of this thesis, the mixed generalised additive model (MGAM) as a nonlinear machine learning technique investigates the plants' response to CO2 and environmental variables. MGAM analyses the direct and interactive effects of CO2 and environmental variables on gs with appropriate sets of statistical covariates between variables. Using eddy covariance flux tower datasets for different vegetation types including crop, deciduous broad-leaf forest, evergreen needle-leaf forest, and grass, shows that MGAM improved gs simulation by up to 50% increase in Nash-Sutcliffe Efficiency (NSE) compared with conventional gs simulation models. The MGAM model highlighted the interactive effects of CO2, VPD, and SWC for crops and grasses. The interactive effects of CO2, VPD, and TA were identified as important for trees and grasses. In the third chapter of this thesis, the simulated gs by MGAM was added to the Penman-Monteith PET equation to incorporate vegetation response to environmental variables as a part of the PET equation. The modified PET improved runoff simulation up to a 13% increase in NSE, especially in wet conditions when the role of PET is more significant in runoff fluctuation. The results of this study show that conventional PET models need modification by considering the vegetation response to interactive effects of environmental variables through gs simulation. This modification leads to a more accurate estimation of water balance elements especially under wet climatic conditions.
Keywords: Stomatal conductance, Evaporation and transpiration, Machine learning, Penman-Monteith, Eddy covariance flux tower, Land surface models, Climate change, Global sensitivity analysis, Mixed Generalized Additive Model
Subject: Environmental Science thesis
Thesis type: Doctor of Philosophy
Completed: 2024
School: College of Science and Engineering
Supervisor: Okke Batelaan