Improving the representation of hydrology in a terrestrial ecosystem model, with applications at catchment to global scales
Zhou, Hao LU
(2025)
- Abstract
- Accurately representing and predicting hydrological processes in terrestrial ecosystems is crucial for understanding the impacts of global environmental change, particularly climate change. This doctoral thesis aims to advance the understanding of vegetation-hydrology interactions and provides new modelling approaches to enhance hydrological prediction abilities.
The thesis is grounded in the evaluation of hydrological processes in the LPJ-GUESS Dynamic Global Vegetation Model (DGVM). A comprehensive assessment of LPJ-GUESS's hydrological performance was conducted using a wide range of observation-based datasets for runoff, evapotranspiration, and soil moisture across multiple spatial and temporal scales. This evaluation identified... (More) - Accurately representing and predicting hydrological processes in terrestrial ecosystems is crucial for understanding the impacts of global environmental change, particularly climate change. This doctoral thesis aims to advance the understanding of vegetation-hydrology interactions and provides new modelling approaches to enhance hydrological prediction abilities.
The thesis is grounded in the evaluation of hydrological processes in the LPJ-GUESS Dynamic Global Vegetation Model (DGVM). A comprehensive assessment of LPJ-GUESS's hydrological performance was conducted using a wide range of observation-based datasets for runoff, evapotranspiration, and soil moisture across multiple spatial and temporal scales. This evaluation identified key areas for model improvement.
To address the model prediction limitations, the thesis explored the integration of LPJ-GUESS with Machine Learning (ML) models to enhance streamflow prediction. A hybrid modelling framework was created, linking the DGVM with five widely used ML algorithms. This approach leveraged the strengths of both process-based and data-driven methodologies, demonstrating superior performance over standalone models. The use of interpretable ML techniques, such as the SHapley Additive exPlanations (SHAP) algorithm, provided insights into key driving factors including climate and land use inputs and important processes for streamflow variability.
Furthermore, the thesis developed a novel framework for incorporating sub-grid scale topographic heterogeneity and lateral water flow processes into LPJ-GUESS. This framework divided the standard grid cells into distinct Topographic Types (TTs) based on elevation and aspect distributions. Temperature, radiation, and hydrological processes were then simulated for each TT, capturing fine-scale topographic effects. Evaluation against observational data showed significant improvements in predicting runoff, evapotranspiration, and other ecosystem variables compared to the original LPJ-GUESS model.
The thesis also developed an adaptive framework for landscape planning that integrates sustainability and resilience metrics across different governance levels. This framework coupled the land use model with LPJ-GUESS to simulate different future landscape pattern scenarios and evaluate the ecosystem resilience indicators under them. The approach revealed critical discrepancies in government management goals and expectations across national, provincial, and local administrative levels, and identified strategies for balancing socio-economic development and ecological preservation.
In summary, this doctoral thesis advances the state of the art in terrestrial ecosystem hydrological modelling through comprehensive model evaluation, integration of process-based and data-driven approaches, model application in scenario planning and representations of sub-grid scale processes. The resulting frameworks provide valuable tools for understanding and predicting hydrological responses to global environmental change.
(Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/8079d219-9ba2-48e4-aa0c-88f367219a6c
- author
- Zhou, Hao
LU
- supervisor
-
- Paul Miller LU
- Jing Tang LU
- Stefan Olin LU
- opponent
-
- Professor Xu, Chong-Yu, Department of Geoscience, University of Oslo
- organization
- alternative title
- Förbättrad representation av hydrologiska processer i en terrester ekosystemmodell, med tillämpningar från avrinningsområden till global skala
- publishing date
- 2025-03-07
- type
- Thesis
- publication status
- published
- subject
- keywords
- Ecosystem modelling, Vegetation dynamics, Hydrological processes, LPJ-GUESS, Machine learning, Sub-grid scale, Topographic heterogeneity, Land use
- pages
- 74 pages
- publisher
- Lund University
- defense location
- Pangea at the Department of Physical Geography and Ecosystem Science
- defense date
- 2025-03-07 09:00:00
- ISBN
- 978-91-89187-50-4
- 978-91-89187-49-8
- project
- Advances of hydrology in a global vegetation model
- language
- English
- LU publication?
- yes
- id
- 8079d219-9ba2-48e4-aa0c-88f367219a6c
- date added to LUP
- 2025-01-24 16:40:47
- date last changed
- 2025-04-04 15:27:39
@phdthesis{8079d219-9ba2-48e4-aa0c-88f367219a6c,
abstract = {{Accurately representing and predicting hydrological processes in terrestrial ecosystems is crucial for understanding the impacts of global environmental change, particularly climate change. This doctoral thesis aims to advance the understanding of vegetation-hydrology interactions and provides new modelling approaches to enhance hydrological prediction abilities. <br/>The thesis is grounded in the evaluation of hydrological processes in the LPJ-GUESS Dynamic Global Vegetation Model (DGVM). A comprehensive assessment of LPJ-GUESS's hydrological performance was conducted using a wide range of observation-based datasets for runoff, evapotranspiration, and soil moisture across multiple spatial and temporal scales. This evaluation identified key areas for model improvement.<br/>To address the model prediction limitations, the thesis explored the integration of LPJ-GUESS with Machine Learning (ML) models to enhance streamflow prediction. A hybrid modelling framework was created, linking the DGVM with five widely used ML algorithms. This approach leveraged the strengths of both process-based and data-driven methodologies, demonstrating superior performance over standalone models. The use of interpretable ML techniques, such as the SHapley Additive exPlanations (SHAP) algorithm, provided insights into key driving factors including climate and land use inputs and important processes for streamflow variability. <br/>Furthermore, the thesis developed a novel framework for incorporating sub-grid scale topographic heterogeneity and lateral water flow processes into LPJ-GUESS. This framework divided the standard grid cells into distinct Topographic Types (TTs) based on elevation and aspect distributions. Temperature, radiation, and hydrological processes were then simulated for each TT, capturing fine-scale topographic effects. Evaluation against observational data showed significant improvements in predicting runoff, evapotranspiration, and other ecosystem variables compared to the original LPJ-GUESS model.<br/>The thesis also developed an adaptive framework for landscape planning that integrates sustainability and resilience metrics across different governance levels. This framework coupled the land use model with LPJ-GUESS to simulate different future landscape pattern scenarios and evaluate the ecosystem resilience indicators under them. The approach revealed critical discrepancies in government management goals and expectations across national, provincial, and local administrative levels, and identified strategies for balancing socio-economic development and ecological preservation.<br/>In summary, this doctoral thesis advances the state of the art in terrestrial ecosystem hydrological modelling through comprehensive model evaluation, integration of process-based and data-driven approaches, model application in scenario planning and representations of sub-grid scale processes. The resulting frameworks provide valuable tools for understanding and predicting hydrological responses to global environmental change.<br/>}},
author = {{Zhou, Hao}},
isbn = {{978-91-89187-50-4}},
keywords = {{Ecosystem modelling; Vegetation dynamics; Hydrological processes; LPJ-GUESS; Machine learning; Sub-grid scale; Topographic heterogeneity; Land use}},
language = {{eng}},
month = {{03}},
publisher = {{Lund University}},
school = {{Lund University}},
title = {{Improving the representation of hydrology in a terrestrial ecosystem model, with applications at catchment to global scales}},
year = {{2025}},
}