Machine learning models inaccurately predict current and future high-latitude C balances
The high-latitude carbon (C) cycle is a key feedback to the global climate system, yet because of system complexity and data limitations, there is currently disagreement over whether the region is a source or sink of C. Recent advances in big data analytics and computing power have popularized the u...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IOP Publishing
2023-01-01
|
| Series: | Environmental Research Letters |
| Subjects: | |
| Online Access: | https://doi.org/10.1088/1748-9326/acacb2 |
| _version_ | 1797747253331361792 |
|---|---|
| author | Ian A Shirley Zelalem A Mekonnen Robert F Grant Baptiste Dafflon William J Riley |
| author_facet | Ian A Shirley Zelalem A Mekonnen Robert F Grant Baptiste Dafflon William J Riley |
| author_sort | Ian A Shirley |
| collection | DOAJ |
| description | The high-latitude carbon (C) cycle is a key feedback to the global climate system, yet because of system complexity and data limitations, there is currently disagreement over whether the region is a source or sink of C. Recent advances in big data analytics and computing power have popularized the use of machine learning (ML) algorithms to upscale site measurements of ecosystem processes, and in some cases forecast the response of these processes to climate change. Due to data limitations, however, ML model predictions of these processes are almost never validated with independent datasets. To better understand and characterize the limitations of these methods, we develop an approach to independently evaluate ML upscaling and forecasting. We mimic data-driven upscaling and forecasting efforts by applying ML algorithms to different subsets of regional process-model simulation gridcells, and then test ML performance using the remaining gridcells. In this study, we simulate C fluxes and environmental data across Alaska using ecosys , a process-rich terrestrial ecosystem model, and then apply boosted regression tree ML algorithms to training data configurations that mirror and expand upon existing AmeriFLUX eddy-covariance data availability. We first show that a ML model trained using ecosys outputs from currently-available Alaska AmeriFLUX sites incorrectly predicts that Alaska is presently a modeled net C source. Increased spatial coverage of the training dataset improves ML predictions, halving the bias when 240 modeled sites are used instead of 15. However, even this more accurate ML model incorrectly predicts Alaska C fluxes under 21st century climate change because of changes in atmospheric CO _2 , litter inputs, and vegetation composition that have impacts on C fluxes which cannot be inferred from the training data. Our results provide key insights to future C flux upscaling efforts and expose the potential for inaccurate ML upscaling and forecasting of high-latitude C cycle dynamics. |
| first_indexed | 2024-03-12T15:48:11Z |
| format | Article |
| id | doaj.art-aadf6efca9614e839824ed7acaea4520 |
| institution | Directory Open Access Journal |
| issn | 1748-9326 |
| language | English |
| last_indexed | 2024-03-12T15:48:11Z |
| publishDate | 2023-01-01 |
| publisher | IOP Publishing |
| record_format | Article |
| series | Environmental Research Letters |
| spelling | doaj.art-aadf6efca9614e839824ed7acaea45202023-08-09T15:20:18ZengIOP PublishingEnvironmental Research Letters1748-93262023-01-0118101402610.1088/1748-9326/acacb2Machine learning models inaccurately predict current and future high-latitude C balancesIan A Shirley0https://orcid.org/0000-0002-2229-1414Zelalem A Mekonnen1https://orcid.org/0000-0002-2647-0671Robert F Grant2https://orcid.org/0000-0002-8890-6231Baptiste Dafflon3https://orcid.org/0000-0001-9871-5650William J Riley4https://orcid.org/0000-0002-4615-2304Climate and Ecosystem Sciences Division, Lawrence Berkeley National Laboratory , Berkeley, CA, United States of America; Department of Physics, University of California-Berkeley , Berkeley 94720-3114, CA, United States of AmericaClimate and Ecosystem Sciences Division, Lawrence Berkeley National Laboratory , Berkeley, CA, United States of AmericaDepartment of Renewable Resources, University of Alberta , Edmonton, CanadaClimate and Ecosystem Sciences Division, Lawrence Berkeley National Laboratory , Berkeley, CA, United States of AmericaClimate and Ecosystem Sciences Division, Lawrence Berkeley National Laboratory , Berkeley, CA, United States of AmericaThe high-latitude carbon (C) cycle is a key feedback to the global climate system, yet because of system complexity and data limitations, there is currently disagreement over whether the region is a source or sink of C. Recent advances in big data analytics and computing power have popularized the use of machine learning (ML) algorithms to upscale site measurements of ecosystem processes, and in some cases forecast the response of these processes to climate change. Due to data limitations, however, ML model predictions of these processes are almost never validated with independent datasets. To better understand and characterize the limitations of these methods, we develop an approach to independently evaluate ML upscaling and forecasting. We mimic data-driven upscaling and forecasting efforts by applying ML algorithms to different subsets of regional process-model simulation gridcells, and then test ML performance using the remaining gridcells. In this study, we simulate C fluxes and environmental data across Alaska using ecosys , a process-rich terrestrial ecosystem model, and then apply boosted regression tree ML algorithms to training data configurations that mirror and expand upon existing AmeriFLUX eddy-covariance data availability. We first show that a ML model trained using ecosys outputs from currently-available Alaska AmeriFLUX sites incorrectly predicts that Alaska is presently a modeled net C source. Increased spatial coverage of the training dataset improves ML predictions, halving the bias when 240 modeled sites are used instead of 15. However, even this more accurate ML model incorrectly predicts Alaska C fluxes under 21st century climate change because of changes in atmospheric CO _2 , litter inputs, and vegetation composition that have impacts on C fluxes which cannot be inferred from the training data. Our results provide key insights to future C flux upscaling efforts and expose the potential for inaccurate ML upscaling and forecasting of high-latitude C cycle dynamics.https://doi.org/10.1088/1748-9326/acacb2machine learningupscalingforecastingindependent evaluationcarbon cyclehigh-latitudes |
| spellingShingle | Ian A Shirley Zelalem A Mekonnen Robert F Grant Baptiste Dafflon William J Riley Machine learning models inaccurately predict current and future high-latitude C balances Environmental Research Letters machine learning upscaling forecasting independent evaluation carbon cycle high-latitudes |
| title | Machine learning models inaccurately predict current and future high-latitude C balances |
| title_full | Machine learning models inaccurately predict current and future high-latitude C balances |
| title_fullStr | Machine learning models inaccurately predict current and future high-latitude C balances |
| title_full_unstemmed | Machine learning models inaccurately predict current and future high-latitude C balances |
| title_short | Machine learning models inaccurately predict current and future high-latitude C balances |
| title_sort | machine learning models inaccurately predict current and future high latitude c balances |
| topic | machine learning upscaling forecasting independent evaluation carbon cycle high-latitudes |
| url | https://doi.org/10.1088/1748-9326/acacb2 |
| work_keys_str_mv | AT ianashirley machinelearningmodelsinaccuratelypredictcurrentandfuturehighlatitudecbalances AT zelalemamekonnen machinelearningmodelsinaccuratelypredictcurrentandfuturehighlatitudecbalances AT robertfgrant machinelearningmodelsinaccuratelypredictcurrentandfuturehighlatitudecbalances AT baptistedafflon machinelearningmodelsinaccuratelypredictcurrentandfuturehighlatitudecbalances AT williamjriley machinelearningmodelsinaccuratelypredictcurrentandfuturehighlatitudecbalances |