Zelalem A. Mekonnen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Zelalem A. Mekonnen has a total of 182 co-authors. Here are the top co-authors by number of shared publications:
- William J. Riley
Earth and Environment Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, California, 94720, USA
30 shared publications - Robert F. Grant
Department of Renewable Resources University of Alberta Edmonton Alberta Canada
17 shared publications - Bhavna Arora
Earth and Environmental Sciences Area, Lawrence Berkeley National Laboratory, Berkeley, CA, USA
8 shared publications Qing Zhu
Earth and Environmental Sciences Area, Lawrence Berkeley National Laboratory, Berkeley, CA, USA
8 shared publications- Margaret S. Torn
Climate and Ecosystem Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, USA
7 shared publications - Nicholas J. Bouskill
Department of Biological and Ecological Engineering Oregon State University Corvallis Oregon USA
7 shared publications
Latest Publications
- Open Access
- 18 September 2026
- Published by Springer Nature in Research Square
Disturbances Increase Net Carbon Emissions in the Arctic-Boreal Zone - An Expert Assessment
Carbon emissions from disturbances in the Arctic-Boreal Zone (ABZ) are not adequately included in global carbon budgets due to insufficient measurements and model representation. Due to this data gap, we conducted an expert assessment to examine the potential trajectories of multiple disturbance types in the ABZ and their impacts on net carbon fluxes over the next one to three centuries. Across five disturbance categories – anthropogenic (e.g., oil and gas infrastructure), biotic, fire, abrupt permafrost thaw, and weather – experts agreed that the extent, frequency, range, and severity of disturbances will increase for at least the next century. Taken together, our assessment suggests that ABZ disturbances could result in 154–417 TgC yr − 1 (median expert estimates) greater net carbon emissions by 2100 relative to 2000–2020 for three climate change mitigation scenarios – global mean temperature increase held below (1) 1.5°C by 2100, (2) 2°C by 2100, and (3) Representative Concentration Pathway (RCP) 8.5 (~ 2.6–4.8°C by 2100 with rising temperatures beyond that). Critically, the experts agreed these impacts will persist long after global net zero anthropogenic emissions are reached, causing continued carbon emissions that are currently unaccounted for. Key research priorities for reducing uncertainty in these estimates were 1) improved spatio-temporal monitoring of ABZ disturbances to quantify their extent and distribution, 2) increased measurements of ecosystem carbon fluxes during and following disturbance events, and 3) incorporation of disturbance processes into models.
- 18 September 2026
- Published by Springer Nature in Research Square
Open AccessDisturbances Increase Net Carbon Emissions in the Arctic-Boreal Zone - An Expert Assessment
Carbon emissions from disturbances in the Arctic-Boreal Zone (ABZ) are not adequately included in global carbon budgets due to insufficient measurements and model representation. Due to this data gap, we conducted an expert assessment to examine the potential trajectories of multiple disturbance types in the ABZ and their impacts on net carbon fluxes over the next one to three centuries. Across five disturbance categories – anthropogenic (e.g., oil and gas infrastructure), biotic, fire, abrupt permafrost thaw, and weather – experts agreed that the extent, frequency, range, and severity of disturbances will increase for at least the next century. Taken together, our assessment suggests that ABZ disturbances could result in 154–417 TgC yr − 1 (median expert estimates) greater net carbon emissions by 2100 relative to 2000–2020 for three climate change mitigation scenarios – global mean temperature increase held below (1) 1.5°C by 2100, (2) 2°C by 2100, and (3) Representative Concentration Pathway (RCP) 8.5 (~ 2.6–4.8°C by 2100 with rising temperatures beyond that). Critically, the experts agreed these impacts will persist long after global net zero anthropogenic emissions are reached, causing continued carbon emissions that are currently unaccounted for. Key research priorities for reducing uncertainty in these estimates were 1) improved spatio-temporal monitoring of ABZ disturbances to quantify their extent and distribution, 2) increased measurements of ecosystem carbon fluxes during and following disturbance events, and 3) incorporation of disturbance processes into models.
- Open Access
- 15 May 2026
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 12 (20) , eaee4995
Wind stilling shapes grassland water use efficiency by enhancing soil moisture retention
Covering more than 40% of Earth’s vegetated surface, grasslands critically regulate terrestrial carbon and water cycles. Their ecosystem water use efficiency (WUE eco ), the ratio of carbon uptake to water loss, governs drought resilience in these water-limited ecosystems. While global terrestrial wind speed declined substantially from the 1960s to 2000s followed by a recovery in the subsequent decade, the extent and mechanisms of its influence on grassland WUE eco remain poorly understood. Using site observations, satellite data, Earth system models, and wind manipulation experiments, we found a consistent negative sensitivity of grassland WUE eco to wind speed. Mechanistically, declining winds reduce evaporative water loss, improve soil moisture, and promote stomatal opening, thereby enhancing carbon uptake. Wind speed changes accounted for 7.7 to 25.7% of WUE eco increases under historical and future climates, making wind the second most important climatic driver after atmospheric carbon dioxide. Since Earth system models underestimate observed wind speed declines, future WUE eco increases and drought resilience may exceed current projections for global grassland.
- 15 May 2026
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 12 (20) , eaee4995
Open AccessWind stilling shapes grassland water use efficiency by enhancing soil moisture retention
Covering more than 40% of Earth’s vegetated surface, grasslands critically regulate terrestrial carbon and water cycles. Their ecosystem water use efficiency (WUE eco ), the ratio of carbon uptake to water loss, governs drought resilience in these water-limited ecosystems. While global terrestrial wind speed declined substantially from the 1960s to 2000s followed by a recovery in the subsequent decade, the extent and mechanisms of its influence on grassland WUE eco remain poorly understood. Using site observations, satellite data, Earth system models, and wind manipulation experiments, we found a consistent negative sensitivity of grassland WUE eco to wind speed. Mechanistically, declining winds reduce evaporative water loss, improve soil moisture, and promote stomatal opening, thereby enhancing carbon uptake. Wind speed changes accounted for 7.7 to 25.7% of WUE eco increases under historical and future climates, making wind the second most important climatic driver after atmospheric carbon dioxide. Since Earth system models underestimate observed wind speed declines, future WUE eco increases and drought resilience may exceed current projections for global grassland.
- Open Access
- 9 May 2026
- Published by American Geophysical Union (AGU) in Journal of Geophysical Research: Biogeosciences
- Vol. 131 (5)
Informing Robust Functional Relationship Benchmarks: An Evaluation of the Temperature Sensitivity of Ecosystem Respiration Across the Arctic‐Boreal Region
During land model development, simulated carbon dynamics are often benchmarked against observational data sets to evaluate model performance. Functional relationship benchmarks are the relationship between a driving variable (e.g., temperature) and a response variable (e.g., ecosystem respiration) and are a promising tool for assessing model performance by evaluating modeled sensitivities to changing environmental conditions. However, observed functional relationships can be influenced by choices made during data collection and throughout the benchmarking process, impacting the inferred skill of land models. To avoid misrepresenting a model's true performance, it is necessary to systematically evaluate best practices when constructing functional relationship benchmarks. We developed a set of guidelines for constructing functional relationship benchmarks, considering the choice of data set, number of daily observations, temporal extent, and temporal resolution across Alaska and Canada over a 20‐year period from 2001 to 2020. The temperature sensitivity of ecosystem respiration from observations, evaluated through an apparent Q 10 , is highly variable both spatially and as a result of the data processing approach applied in the benchmark formation. When benchmarking 13 models from the Warming Permafrost Model Intercomparison Project (WrPMIP), the range in inferred model skill is substantially impacted by the choices applied in constructing functional relationship benchmarks. The inferred performance of a given model is most sensitive to the number of daily observations and temporal extent, followed by choice of benchmark data set and temporal averaging. Results from this analysis can guide the development of consistent and robust functional relationships for future model evaluation studies.
- 9 May 2026
- Published by American Geophysical Union (AGU) in Journal of Geophysical Research: Biogeosciences
- Vol. 131 (5)
Open AccessInforming Robust Functional Relationship Benchmarks: An Evaluation of the Temperature Sensitivity of Ecosystem Respiration Across the Arctic‐Boreal Region
During land model development, simulated carbon dynamics are often benchmarked against observational data sets to evaluate model performance. Functional relationship benchmarks are the relationship between a driving variable (e.g., temperature) and a response variable (e.g., ecosystem respiration) and are a promising tool for assessing model performance by evaluating modeled sensitivities to changing environmental conditions. However, observed functional relationships can be influenced by choices made during data collection and throughout the benchmarking process, impacting the inferred skill of land models. To avoid misrepresenting a model's true performance, it is necessary to systematically evaluate best practices when constructing functional relationship benchmarks. We developed a set of guidelines for constructing functional relationship benchmarks, considering the choice of data set, number of daily observations, temporal extent, and temporal resolution across Alaska and Canada over a 20‐year period from 2001 to 2020. The temperature sensitivity of ecosystem respiration from observations, evaluated through an apparent Q 10 , is highly variable both spatially and as a result of the data processing approach applied in the benchmark formation. When benchmarking 13 models from the Warming Permafrost Model Intercomparison Project (WrPMIP), the range in inferred model skill is substantially impacted by the choices applied in constructing functional relationship benchmarks. The inferred performance of a given model is most sensitive to the number of daily observations and temporal extent, followed by choice of benchmark data set and temporal averaging. Results from this analysis can guide the development of consistent and robust functional relationships for future model evaluation studies.
- Open Access
- 1 December 2025
- Published by IOP Publishing in Environmental Research Letters
- Vol. 20 (12) , 124079
Observations and modeling reveal that heatwaves reduce photosynthesis, plant carbon reserves, and net carbon uptake
Heatwaves threaten ecosystem carbon balances, yet the mechanisms driving short-term carbon flux responses remain poorly understood. Here, integrating high-frequency eddy covariance (EC) data from 140 global flux tower sites (872 site-years) with detailed process-based modeling, we examine ecosystem responses during and immediately after heatwaves. We show that heatwaves caused a −40% (range [−29%, −128%]) reduction in net ecosystem productivity (NEP) compared to pre-heatwave values, with this reduction persisting over the following two weeks (−38% range [+3%, −154%]). We attributed NEP decreases to photosynthesis decreases more than to ecosystem respiration (RE) increases. Forest sites had greater NEP decreases during heatwaves than non-forest sites, but remained carbon sinks afterwards, indicating resilience. Our modeling analysis of extreme heatwaves at selected EC sites shows that decreased photosynthesis, increased maintenance respiration, and decreased plant non-structural carbon reserves during heatwaves drive carbon cycle changes that persist for weeks. Consistent with phenocam observations, we modeled a reduction in leaf area index caused by reduced non-structural carbon reserves, leading to early leaf senescence and longer-term impacts. Ongoing increases in heatwaves are therefore likely to reduce NEP across a range of ecosystems, exacerbating carbon cycle feedback.
- 1 December 2025
- Published by IOP Publishing in Environmental Research Letters
- Vol. 20 (12) , 124079
Open AccessObservations and modeling reveal that heatwaves reduce photosynthesis, plant carbon reserves, and net carbon uptake
Heatwaves threaten ecosystem carbon balances, yet the mechanisms driving short-term carbon flux responses remain poorly understood. Here, integrating high-frequency eddy covariance (EC) data from 140 global flux tower sites (872 site-years) with detailed process-based modeling, we examine ecosystem responses during and immediately after heatwaves. We show that heatwaves caused a −40% (range [−29%, −128%]) reduction in net ecosystem productivity (NEP) compared to pre-heatwave values, with this reduction persisting over the following two weeks (−38% range [+3%, −154%]). We attributed NEP decreases to photosynthesis decreases more than to ecosystem respiration (RE) increases. Forest sites had greater NEP decreases during heatwaves than non-forest sites, but remained carbon sinks afterwards, indicating resilience. Our modeling analysis of extreme heatwaves at selected EC sites shows that decreased photosynthesis, increased maintenance respiration, and decreased plant non-structural carbon reserves during heatwaves drive carbon cycle changes that persist for weeks. Consistent with phenocam observations, we modeled a reduction in leaf area index caused by reduced non-structural carbon reserves, leading to early leaf senescence and longer-term impacts. Ongoing increases in heatwaves are therefore likely to reduce NEP across a range of ecosystems, exacerbating carbon cycle feedback.
- Open Access
- 5 November 2025
- Published by Copernicus GmbH in Geoscientific Model Development
- Vol. 18 (21) , 8157-8173
Development of a model framework for terrestrial carbon flux prediction: the Regional Carbon and Climate Analytics Tool (RCCAT) applied to non-tidal wetlands
Wetlands play a pivotal role in carbon sequestration but emit methane (CH4), creating uncertainty in their net climate impact. Although process-based models offer mechanistic insights into wetland dynamics, they require extensive site-specific parameterisation (e.g. soil carbon profiles, pore-water chemistry, vegetation-specific model parameters), as well as high-resolution hydrological and meteorological inputs that are often difficult to obtain outside of well-instrumented research sites, which makes regional upscaling challenging. In contrast, data-driven models provide a scalable alternative by leveraging available datasets to identify patterns and relationships, making them more adaptable for large-scale applications. However, their performance can vary significantly depending on the quality and representativeness of the data, as well as the model design, which raises questions about their reliability and generalizability in complex wetland systems. To address these issues, we present a data-driven framework for upscaling wetland CO2 and CH4 emissions, across a range of machine learning models that vary in complexity, validated against an extensive observational dataset from the Sacramento-San Joaquin Delta. We show that artificial intelligence (AI) approaches, including Random Forests, gradient boosting methods (XGBoost, LightGBM), Support Vector Machines (SVM) and Recurrent Neural Networks (GRU, LSTM), outperform linear regression models, with RNNs standing out, achieving an R2 of 0.73 for daily CO2 flux predictions compared to 0.64 for linear regression, and an R2 of 0.53 for CH4 flux predictions compared to 0.47 for linear regression. Interestingly, linear regression performed better than random forest for methane flux, which highlights the necessity for comparison. The, interannual variability is less well captured, with annual mean absolute error of 176 gC m−2 yr−1 for CO2 fluxes and 9 gC-CH4 m−2 yr−1 for CH4 fluxes. By integrating vertically-resolved atmospheric, subsurface, and spectral reflectance information from readily available sources, the model identifies key drivers of wetland CO2 and CH4 emissions and enables regional upscaling. These findings demonstrate the potential of AI methods for upscaling, providing practical tools for wetland management and restoration planning to support climate mitigation efforts.
- 5 November 2025
- Published by Copernicus GmbH in Geoscientific Model Development
- Vol. 18 (21) , 8157-8173
Open AccessDevelopment of a model framework for terrestrial carbon flux prediction: the Regional Carbon and Climate Analytics Tool (RCCAT) applied to non-tidal wetlands
Wetlands play a pivotal role in carbon sequestration but emit methane (CH4), creating uncertainty in their net climate impact. Although process-based models offer mechanistic insights into wetland dynamics, they require extensive site-specific parameterisation (e.g. soil carbon profiles, pore-water chemistry, vegetation-specific model parameters), as well as high-resolution hydrological and meteorological inputs that are often difficult to obtain outside of well-instrumented research sites, which makes regional upscaling challenging. In contrast, data-driven models provide a scalable alternative by leveraging available datasets to identify patterns and relationships, making them more adaptable for large-scale applications. However, their performance can vary significantly depending on the quality and representativeness of the data, as well as the model design, which raises questions about their reliability and generalizability in complex wetland systems. To address these issues, we present a data-driven framework for upscaling wetland CO2 and CH4 emissions, across a range of machine learning models that vary in complexity, validated against an extensive observational dataset from the Sacramento-San Joaquin Delta. We show that artificial intelligence (AI) approaches, including Random Forests, gradient boosting methods (XGBoost, LightGBM), Support Vector Machines (SVM) and Recurrent Neural Networks (GRU, LSTM), outperform linear regression models, with RNNs standing out, achieving an R2 of 0.73 for daily CO2 flux predictions compared to 0.64 for linear regression, and an R2 of 0.53 for CH4 flux predictions compared to 0.47 for linear regression. Interestingly, linear regression performed better than random forest for methane flux, which highlights the necessity for comparison. The, interannual variability is less well captured, with annual mean absolute error of 176 gC m−2 yr−1 for CO2 fluxes and 9 gC-CH4 m−2 yr−1 for CH4 fluxes. By integrating vertically-resolved atmospheric, subsurface, and spectral reflectance information from readily available sources, the model identifies key drivers of wetland CO2 and CH4 emissions and enables regional upscaling. These findings demonstrate the potential of AI methods for upscaling, providing practical tools for wetland management and restoration planning to support climate mitigation efforts.
- Open Access
- 21 October 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-12
Significant sensitivity of global vegetation productivity to terrestrial surface wind speed changes
Decadal changes in terrestrial surface wind speed since the 1960s have been extensively documented, while the mechanism and the extent of their impacts on terrestrial ecosystem productivity remain unclear. Here, we systematically explore the impacts of wind speed changes on terrestrial gross primary production (GPP) using satellite-derived data, reanalysis datasets, flux tower observations, CMIP6 models, and exploratory simulation experiments. Our results show a negative sensitivity of terrestrial GPP to wind speed change, ranging from −156.67 to −65.82 g C m−2 yr−1 (m s−1)−1, across different data sources from 1983 to 2100. This is mainly attributed to the impacts of wind speed decline on stomatal conductance by reducing atmospheric dryness and soil drying. We find that during 1983 – 2010, wind speed decline is the most important factor, after rising atmospheric CO2 concentrations, in its contribution to the increasing trend in GPP (6.0% –7.8%). With further declines in wind speed, this contribution to GPP is projected to rank between second and third during 2031 – 2100. Among seven plant functional types, grasslands contribute most to the wind-induced changes in the GPP trend under current and future climates. These findings highlight the substantial effects of wind speed on centennial-scale global carbon dynamics.
- 21 October 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-12
Open AccessSignificant sensitivity of global vegetation productivity to terrestrial surface wind speed changes
Decadal changes in terrestrial surface wind speed since the 1960s have been extensively documented, while the mechanism and the extent of their impacts on terrestrial ecosystem productivity remain unclear. Here, we systematically explore the impacts of wind speed changes on terrestrial gross primary production (GPP) using satellite-derived data, reanalysis datasets, flux tower observations, CMIP6 models, and exploratory simulation experiments. Our results show a negative sensitivity of terrestrial GPP to wind speed change, ranging from −156.67 to −65.82 g C m−2 yr−1 (m s−1)−1, across different data sources from 1983 to 2100. This is mainly attributed to the impacts of wind speed decline on stomatal conductance by reducing atmospheric dryness and soil drying. We find that during 1983 – 2010, wind speed decline is the most important factor, after rising atmospheric CO2 concentrations, in its contribution to the increasing trend in GPP (6.0% –7.8%). With further declines in wind speed, this contribution to GPP is projected to rank between second and third during 2031 – 2100. Among seven plant functional types, grasslands contribute most to the wind-induced changes in the GPP trend under current and future climates. These findings highlight the substantial effects of wind speed on centennial-scale global carbon dynamics.
- Open Access
- 17 October 2025
- Published by IOP Publishing in Environmental Research Letters
- Vol. 20 (11) , 114039
Hydrology controls thermokarst and alters carbon cycling and methane emissions in peatlands near the southern limit of permafrost
Permafrost peatlands store vast amounts of frozen carbon across northern landscapes. When ground ice melts, surface subsidence produces thermokarst landforms that expand wetlands at the edges of permafrost plateaus. Thermokarst represents an accelerating climate feedback, but uncertainties remain about how ground ice, hydrology, and vegetation interact to shape landscape change and carbon fluxes. We extended the process-based model ecosys to simulate thermokarst dynamics in laterally coupled 2D transects at a well-characterized boreal peatland site in Canada’s Northwest Territories. After benchmarking against site observations, we varied ground ice content and hydrologic boundary conditions across ranges typical near the southern permafrost limit. Simulations revealed distinct degradation regimes governed by the elevation difference between the frost table and the external water table. Rates of lateral retreat, the thaw-driven encroachment of wetlands into adjacent plateaus, ranged from 0 to >2 m yr−1 under identical weather forcing, consistent with observations and highlighting the strong role of WT and ground ice. Simulated vegetation dynamics indicate that black spruce mortality cannot be explained by anoxia alone, pointing to additional stressors such as root damage, pathogens, or physical destabilization. Despite large hydrologic shifts, net ecosystem CO2 exchange remained a slight sink after collapse, while methane (CH4) emissions rose by one to two orders of magnitude. As a result, lateral retreat substantially increases the greenhouse warming potential of permafrost peatlands (1.7 million km2 in area), with simulated emissions of 0.1–10 Mt CO2-eq decade−1 depending on hydrology and retreat rates. These results underscore the need to account for both ground ice and hydrologic dynamics when assessing thermokarst-driven climate feedbacks.
- 17 October 2025
- Published by IOP Publishing in Environmental Research Letters
- Vol. 20 (11) , 114039
Open AccessHydrology controls thermokarst and alters carbon cycling and methane emissions in peatlands near the southern limit of permafrost
Permafrost peatlands store vast amounts of frozen carbon across northern landscapes. When ground ice melts, surface subsidence produces thermokarst landforms that expand wetlands at the edges of permafrost plateaus. Thermokarst represents an accelerating climate feedback, but uncertainties remain about how ground ice, hydrology, and vegetation interact to shape landscape change and carbon fluxes. We extended the process-based model ecosys to simulate thermokarst dynamics in laterally coupled 2D transects at a well-characterized boreal peatland site in Canada’s Northwest Territories. After benchmarking against site observations, we varied ground ice content and hydrologic boundary conditions across ranges typical near the southern permafrost limit. Simulations revealed distinct degradation regimes governed by the elevation difference between the frost table and the external water table. Rates of lateral retreat, the thaw-driven encroachment of wetlands into adjacent plateaus, ranged from 0 to >2 m yr−1 under identical weather forcing, consistent with observations and highlighting the strong role of WT and ground ice. Simulated vegetation dynamics indicate that black spruce mortality cannot be explained by anoxia alone, pointing to additional stressors such as root damage, pathogens, or physical destabilization. Despite large hydrologic shifts, net ecosystem CO2 exchange remained a slight sink after collapse, while methane (CH4) emissions rose by one to two orders of magnitude. As a result, lateral retreat substantially increases the greenhouse warming potential of permafrost peatlands (1.7 million km2 in area), with simulated emissions of 0.1–10 Mt CO2-eq decade−1 depending on hydrology and retreat rates. These results underscore the need to account for both ground ice and hydrologic dynamics when assessing thermokarst-driven climate feedbacks.
- Open Access
- 17 June 2025
- Published by Copernicus GmbH
- Vol. 2025, 1-28
Reply on RC2
Wetlands play a pivotal role in carbon sequestration but emit methane (CH4), creating uncertainty in their net climate impact. Although process-based models offer mechanistic insights into wetland dynamics, they are computationally expensive, uncertain, and difficult to upscale. In contrast, data-driven models provide a scalable alternative by leveraging extensive datasets to identify patterns and relationships, making them more adaptable for large-scale applications. However, their performance can vary significantly depending on the quality and representativeness of the data, as well as the model design, which raises questions about their reliability and generalizability in complex wetland systems. To address these issues, we present a data-driven framework for upscaling wetland CO2 and CH4 emissions, across a range of machine learning models that vary in complexity, validated against an extensive observational dataset from the Sacramento-San Joaquin Delta. We show that artificial intelligence (AI) approaches, including Random Forests, gradient boosting methods (XGBoost, LightGBM), Support Vector Machines (SVM) and Recurrent Neural Networks (GRU, LSTM), outperform linear regression models, with RNNs standing out, achieving an R2 of 0.71 for daily CO2 flux predictions compared to 0.62 for linear regression, and an R2 of 0.60 for CH4 flux predictions compared to 0.54 for linear regression. Despite that, interannual variability is less well captured, with annual mean absolute error of 193 gC m-2 yr-1 for CO2 fluxes and 11 gC-CH4 m-2 yr-1 for CH4 fluxes. By integrating vertically-resolved atmospheric, subsurface, and spectral reflectance information from readily available sources, the model identifies key drivers of wetland CO2 and CH4 emissions and enables regional upscaling. These findings demonstrate the potential of AI methods for upscaling, providing practical tools for wetland management and restoration planning to support climate mitigation efforts.
- 17 June 2025
- Published by Copernicus GmbH
- Vol. 2025, 1-28
Open AccessReply on RC2
Wetlands play a pivotal role in carbon sequestration but emit methane (CH4), creating uncertainty in their net climate impact. Although process-based models offer mechanistic insights into wetland dynamics, they are computationally expensive, uncertain, and difficult to upscale. In contrast, data-driven models provide a scalable alternative by leveraging extensive datasets to identify patterns and relationships, making them more adaptable for large-scale applications. However, their performance can vary significantly depending on the quality and representativeness of the data, as well as the model design, which raises questions about their reliability and generalizability in complex wetland systems. To address these issues, we present a data-driven framework for upscaling wetland CO2 and CH4 emissions, across a range of machine learning models that vary in complexity, validated against an extensive observational dataset from the Sacramento-San Joaquin Delta. We show that artificial intelligence (AI) approaches, including Random Forests, gradient boosting methods (XGBoost, LightGBM), Support Vector Machines (SVM) and Recurrent Neural Networks (GRU, LSTM), outperform linear regression models, with RNNs standing out, achieving an R2 of 0.71 for daily CO2 flux predictions compared to 0.62 for linear regression, and an R2 of 0.60 for CH4 flux predictions compared to 0.54 for linear regression. Despite that, interannual variability is less well captured, with annual mean absolute error of 193 gC m-2 yr-1 for CO2 fluxes and 11 gC-CH4 m-2 yr-1 for CH4 fluxes. By integrating vertically-resolved atmospheric, subsurface, and spectral reflectance information from readily available sources, the model identifies key drivers of wetland CO2 and CH4 emissions and enables regional upscaling. These findings demonstrate the potential of AI methods for upscaling, providing practical tools for wetland management and restoration planning to support climate mitigation efforts.
- Open Access
- 17 June 2025
- Published by Copernicus GmbH
- Vol. 2025, 1-28
Reply on RC1
Wetlands play a pivotal role in carbon sequestration but emit methane (CH4), creating uncertainty in their net climate impact. Although process-based models offer mechanistic insights into wetland dynamics, they are computationally expensive, uncertain, and difficult to upscale. In contrast, data-driven models provide a scalable alternative by leveraging extensive datasets to identify patterns and relationships, making them more adaptable for large-scale applications. However, their performance can vary significantly depending on the quality and representativeness of the data, as well as the model design, which raises questions about their reliability and generalizability in complex wetland systems. To address these issues, we present a data-driven framework for upscaling wetland CO2 and CH4 emissions, across a range of machine learning models that vary in complexity, validated against an extensive observational dataset from the Sacramento-San Joaquin Delta. We show that artificial intelligence (AI) approaches, including Random Forests, gradient boosting methods (XGBoost, LightGBM), Support Vector Machines (SVM) and Recurrent Neural Networks (GRU, LSTM), outperform linear regression models, with RNNs standing out, achieving an R2 of 0.71 for daily CO2 flux predictions compared to 0.62 for linear regression, and an R2 of 0.60 for CH4 flux predictions compared to 0.54 for linear regression. Despite that, interannual variability is less well captured, with annual mean absolute error of 193 gC m-2 yr-1 for CO2 fluxes and 11 gC-CH4 m-2 yr-1 for CH4 fluxes. By integrating vertically-resolved atmospheric, subsurface, and spectral reflectance information from readily available sources, the model identifies key drivers of wetland CO2 and CH4 emissions and enables regional upscaling. These findings demonstrate the potential of AI methods for upscaling, providing practical tools for wetland management and restoration planning to support climate mitigation efforts.
- 17 June 2025
- Published by Copernicus GmbH
- Vol. 2025, 1-28
Open AccessReply on RC1
Wetlands play a pivotal role in carbon sequestration but emit methane (CH4), creating uncertainty in their net climate impact. Although process-based models offer mechanistic insights into wetland dynamics, they are computationally expensive, uncertain, and difficult to upscale. In contrast, data-driven models provide a scalable alternative by leveraging extensive datasets to identify patterns and relationships, making them more adaptable for large-scale applications. However, their performance can vary significantly depending on the quality and representativeness of the data, as well as the model design, which raises questions about their reliability and generalizability in complex wetland systems. To address these issues, we present a data-driven framework for upscaling wetland CO2 and CH4 emissions, across a range of machine learning models that vary in complexity, validated against an extensive observational dataset from the Sacramento-San Joaquin Delta. We show that artificial intelligence (AI) approaches, including Random Forests, gradient boosting methods (XGBoost, LightGBM), Support Vector Machines (SVM) and Recurrent Neural Networks (GRU, LSTM), outperform linear regression models, with RNNs standing out, achieving an R2 of 0.71 for daily CO2 flux predictions compared to 0.62 for linear regression, and an R2 of 0.60 for CH4 flux predictions compared to 0.54 for linear regression. Despite that, interannual variability is less well captured, with annual mean absolute error of 193 gC m-2 yr-1 for CO2 fluxes and 11 gC-CH4 m-2 yr-1 for CH4 fluxes. By integrating vertically-resolved atmospheric, subsurface, and spectral reflectance information from readily available sources, the model identifies key drivers of wetland CO2 and CH4 emissions and enables regional upscaling. These findings demonstrate the potential of AI methods for upscaling, providing practical tools for wetland management and restoration planning to support climate mitigation efforts.
- Open Access
- 5 May 2025
- Published by Copernicus GmbH
- Vol. 2025, 1-28
Comment on egusphere-2025-361
Wetlands play a pivotal role in carbon sequestration but emit methane (CH4), creating uncertainty in their net climate impact. Although process-based models offer mechanistic insights into wetland dynamics, they are computationally expensive, uncertain, and difficult to upscale. In contrast, data-driven models provide a scalable alternative by leveraging extensive datasets to identify patterns and relationships, making them more adaptable for large-scale applications. However, their performance can vary significantly depending on the quality and representativeness of the data, as well as the model design, which raises questions about their reliability and generalizability in complex wetland systems. To address these issues, we present a data-driven framework for upscaling wetland CO2 and CH4 emissions, across a range of machine learning models that vary in complexity, validated against an extensive observational dataset from the Sacramento-San Joaquin Delta. We show that artificial intelligence (AI) approaches, including Random Forests, gradient boosting methods (XGBoost, LightGBM), Support Vector Machines (SVM) and Recurrent Neural Networks (GRU, LSTM), outperform linear regression models, with RNNs standing out, achieving an R2 of 0.71 for daily CO2 flux predictions compared to 0.62 for linear regression, and an R2 of 0.60 for CH4 flux predictions compared to 0.54 for linear regression. Despite that, interannual variability is less well captured, with annual mean absolute error of 193 gC m-2 yr-1 for CO2 fluxes and 11 gC-CH4 m-2 yr-1 for CH4 fluxes. By integrating vertically-resolved atmospheric, subsurface, and spectral reflectance information from readily available sources, the model identifies key drivers of wetland CO2 and CH4 emissions and enables regional upscaling. These findings demonstrate the potential of AI methods for upscaling, providing practical tools for wetland management and restoration planning to support climate mitigation efforts.
- 5 May 2025
- Published by Copernicus GmbH
- Vol. 2025, 1-28
Open AccessComment on egusphere-2025-361
Wetlands play a pivotal role in carbon sequestration but emit methane (CH4), creating uncertainty in their net climate impact. Although process-based models offer mechanistic insights into wetland dynamics, they are computationally expensive, uncertain, and difficult to upscale. In contrast, data-driven models provide a scalable alternative by leveraging extensive datasets to identify patterns and relationships, making them more adaptable for large-scale applications. However, their performance can vary significantly depending on the quality and representativeness of the data, as well as the model design, which raises questions about their reliability and generalizability in complex wetland systems. To address these issues, we present a data-driven framework for upscaling wetland CO2 and CH4 emissions, across a range of machine learning models that vary in complexity, validated against an extensive observational dataset from the Sacramento-San Joaquin Delta. We show that artificial intelligence (AI) approaches, including Random Forests, gradient boosting methods (XGBoost, LightGBM), Support Vector Machines (SVM) and Recurrent Neural Networks (GRU, LSTM), outperform linear regression models, with RNNs standing out, achieving an R2 of 0.71 for daily CO2 flux predictions compared to 0.62 for linear regression, and an R2 of 0.60 for CH4 flux predictions compared to 0.54 for linear regression. Despite that, interannual variability is less well captured, with annual mean absolute error of 193 gC m-2 yr-1 for CO2 fluxes and 11 gC-CH4 m-2 yr-1 for CH4 fluxes. By integrating vertically-resolved atmospheric, subsurface, and spectral reflectance information from readily available sources, the model identifies key drivers of wetland CO2 and CH4 emissions and enables regional upscaling. These findings demonstrate the potential of AI methods for upscaling, providing practical tools for wetland management and restoration planning to support climate mitigation efforts.