A machine learning model using the snapshot ensemble approach for soil respiration prediction in an experimental Oak Forest
Soil respiration plays a key role in regulating ecosystem CO2 emissions and atmospheric concentrations. However, because it is highly sensitive to the environment and has large temporal and spatial variability, its contribution to the global carbon cycle remains highly uncertain. Recent advances in...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-03-01
|
| Series: | Ecological Informatics |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1574954124005338 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832595400914108416 |
|---|---|
| author | S.N. Ferdous J.P. Ahire R. Bergman L. Xin E. Blanc-Betes Z. Zhang J. Wang |
| author_facet | S.N. Ferdous J.P. Ahire R. Bergman L. Xin E. Blanc-Betes Z. Zhang J. Wang |
| author_sort | S.N. Ferdous |
| collection | DOAJ |
| description | Soil respiration plays a key role in regulating ecosystem CO2 emissions and atmospheric concentrations. However, because it is highly sensitive to the environment and has large temporal and spatial variability, its contribution to the global carbon cycle remains highly uncertain. Recent advances in ecosystem process models have uniquely constrained soil respiration rates, largely based on results from tracing ecosystem CO2 losses using radioisotope and stable isotope methods. However, the computational requirements of process-based models at large scales are costly, limiting their application and prediction capability. To mitigate this requirement for high computational power for the assimilation of high-throughput data to identify patterns and trends in soil respiration and guide in deploying effective strategies to reduce Greenhouse gas (GHG) emission, the application of machine-learning (ML) models can help reduce the requirement of intensive computational capacity and improve the accuracy of predictions. Here, we developed a novel ML model that integrated a hybrid Prophet-ANN and snapshot ensemble approach. First, we extracted temporal features using the Prophet Forecasting model. We then used the Artificial Neural Network (ANN) regression model to correct the forecasting model errors and perform the final prediction using the snapshot ensemble approach. This model can serve as a surrogate for the ForCent model applied to the Oak Ridge National Laboratory deciduous forest. Our proposed model accurately predicted soil CO2 flux for the four studied sites: East Low, West Low, East High, and West High. We show the potential of our proposed ML model to significantly improve soil CO2 flux predictions, which in turn can help us develop more effective strategies for managing soil carbon. |
| format | Article |
| id | doaj-art-656a0f3b7ebc4847b1eed86b3dde18e2 |
| institution | Kabale University |
| issn | 1574-9541 |
| language | English |
| publishDate | 2025-03-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Ecological Informatics |
| spelling | doaj-art-656a0f3b7ebc4847b1eed86b3dde18e22025-01-19T06:24:44ZengElsevierEcological Informatics1574-95412025-03-0185102991A machine learning model using the snapshot ensemble approach for soil respiration prediction in an experimental Oak ForestS.N. Ferdous0J.P. Ahire1R. Bergman2L. Xin3E. Blanc-Betes4Z. Zhang5J. Wang6Lane Department of Computer Science and Electrical Engineering, West Virginia University, Morgantown, WV, 26505, USA; Corresponding author.Biological Systems Engineering Department, University of Wisconsin, Madison, WI, 53706, USA; Forest Products Laboratory, USDA Forest Service, Madison, WI 53726, USAForest Products Laboratory, USDA Forest Service, Madison, WI 53726, USADepartment of Computer Science, University at Albany, New York, NY 12222, USAInstitute for Sustainability, Energy, and Environment, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA; Center for Advanced Bioenergy and Bioproducts Innovation, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USABiological Systems Engineering, University of Wisconsin, Madison, WI 53706, USADepartment of Forest Biomaterials, North Carolina State University, Raleigh, NC 27695, USASoil respiration plays a key role in regulating ecosystem CO2 emissions and atmospheric concentrations. However, because it is highly sensitive to the environment and has large temporal and spatial variability, its contribution to the global carbon cycle remains highly uncertain. Recent advances in ecosystem process models have uniquely constrained soil respiration rates, largely based on results from tracing ecosystem CO2 losses using radioisotope and stable isotope methods. However, the computational requirements of process-based models at large scales are costly, limiting their application and prediction capability. To mitigate this requirement for high computational power for the assimilation of high-throughput data to identify patterns and trends in soil respiration and guide in deploying effective strategies to reduce Greenhouse gas (GHG) emission, the application of machine-learning (ML) models can help reduce the requirement of intensive computational capacity and improve the accuracy of predictions. Here, we developed a novel ML model that integrated a hybrid Prophet-ANN and snapshot ensemble approach. First, we extracted temporal features using the Prophet Forecasting model. We then used the Artificial Neural Network (ANN) regression model to correct the forecasting model errors and perform the final prediction using the snapshot ensemble approach. This model can serve as a surrogate for the ForCent model applied to the Oak Ridge National Laboratory deciduous forest. Our proposed model accurately predicted soil CO2 flux for the four studied sites: East Low, West Low, East High, and West High. We show the potential of our proposed ML model to significantly improve soil CO2 flux predictions, which in turn can help us develop more effective strategies for managing soil carbon.http://www.sciencedirect.com/science/article/pii/S1574954124005338Biogeochemical modelCO2Forest ecosystemML modelSoil carbon |
| spellingShingle | S.N. Ferdous J.P. Ahire R. Bergman L. Xin E. Blanc-Betes Z. Zhang J. Wang A machine learning model using the snapshot ensemble approach for soil respiration prediction in an experimental Oak Forest Ecological Informatics Biogeochemical model CO2 Forest ecosystem ML model Soil carbon |
| title | A machine learning model using the snapshot ensemble approach for soil respiration prediction in an experimental Oak Forest |
| title_full | A machine learning model using the snapshot ensemble approach for soil respiration prediction in an experimental Oak Forest |
| title_fullStr | A machine learning model using the snapshot ensemble approach for soil respiration prediction in an experimental Oak Forest |
| title_full_unstemmed | A machine learning model using the snapshot ensemble approach for soil respiration prediction in an experimental Oak Forest |
| title_short | A machine learning model using the snapshot ensemble approach for soil respiration prediction in an experimental Oak Forest |
| title_sort | machine learning model using the snapshot ensemble approach for soil respiration prediction in an experimental oak forest |
| topic | Biogeochemical model CO2 Forest ecosystem ML model Soil carbon |
| url | http://www.sciencedirect.com/science/article/pii/S1574954124005338 |
| work_keys_str_mv | AT snferdous amachinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest AT jpahire amachinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest AT rbergman amachinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest AT lxin amachinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest AT eblancbetes amachinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest AT zzhang amachinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest AT jwang amachinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest AT snferdous machinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest AT jpahire machinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest AT rbergman machinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest AT lxin machinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest AT eblancbetes machinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest AT zzhang machinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest AT jwang machinelearningmodelusingthesnapshotensembleapproachforsoilrespirationpredictioninanexperimentaloakforest |