Exploring the Potential of Long Short-Term Memory Networks for Predicting Net CO2 Exchange Across Various Ecosystems With Multi-Source Data DOI Open Access

Chengcheng Huang,

Wei He, Jinxiu Liu

et al.

Authorea (Authorea), Journal Year: 2023, Volume and Issue: unknown

Published: Nov. 14, 2023

Upscaling flux tower measurements based on machine learning (ML) algorithms is an essential approach for large-scale net ecosystem CO2 exchange (NEE) estimation, but existing ML upscaling methods face some challenges, particularly in capturing NEE interannual variations (IAVs) that may relate to lagged effects. With the capacity of characterizing temporal memory effects, Long Short-Term Memory (LSTM) networks are expected help solve this problem. Here we explored potential LSTM predicting across various ecosystems using data over 82 sites North America. The model with differentiated plant function types (PFTs) demonstrates capability explain 79.19% (R2 = 0.79) monthly within testing set, RMSE and MAE values 0.89 0.57 g C m-2 d-1 respectively (r 0.89, p < 0.001). Moreover, performed robustly cross-site variability, 67.19% can be predicted by both models without distinguished PFTs showing improved predictive ability. Most importantly, IAV highly correlated observations 0.81, 0.001), clearly outperforming random forest -0.21, 0.011). Among all nine PFTs, solar-induced chlorophyll fluorescence, downward shortwave radiation, leaf area index most important variables explaining variations, collectively accounting approximately 54.01% total. This study highlights great improving carbon multi-source remote sensing data.

Language: Английский

Satellite-Observed Increase in Aboveground Carbon over Southwest China during 2013-2021 DOI Creative Commons
Lei Fan, Guanyu Dong, Frédéric Frappart

et al.

Journal of Remote Sensing, Journal Year: 2024, Volume and Issue: 4

Published: Jan. 1, 2024

Over the past 4 decades, Southwest China has fast vegetation growth and aboveground biomass carbon (AGC) accumulation, largely attributed to active implementation of ecological projects. However, been threatened by frequent extreme drought events recently, potentially countering expected large AGC increase caused Here, we used L-band optical depth quantify dynamics over during period 2013-2021. Our results showed a net sink 0.064 [0.057, 0.077] Pg C year −1 (the range represents maximum minimum values), suggesting that acted as an study period. Note loss 0.113 [0.101, 0.136] was found 2013-2014, which could mainly be negative influence droughts on changes in China, particularly Yunnan province. For each land use type (i.e., dense forests, persistent nonforests, afforestation, forestry), largest stock 0.032 [0.028, 0.036] owing their widespread cover rate China. density per unit area), afforestation areas 0.808 [0.724, 0.985] Mg ha , reflecting positive effect increase. Moreover, karst exhibited higher increasing than nonkarst areas, ecosystems have high capacity

Language: Английский

Citations

4
The role of OCO-3 XCO2 retrievals in estimating global terrestrial net ecosystem exchanges DOI Creative Commons
Xingyu Wang, Fei Jiang, Hengmao Wang

et al.

Atmospheric chemistry and physics, Journal Year: 2025, Volume and Issue: 25(2), P. 867 - 880

Published: Jan. 22, 2025

Abstract. Satellite-based column-averaged dry-air CO2 mole fraction (XCO2) retrievals are frequently used to improve the estimates of terrestrial net ecosystem exchanges (NEEs). The Orbiting Carbon Observatory 3 (OCO-3) satellite, launched in May 2019, was designed address important questions about distribution carbon fluxes on Earth, but its role estimating global NEE remains unclear. Here, using Global Assimilation System, version 2, we investigate impact OCO-3 XCO2 estimation by assimilating alone and combination with OCO-2 retrievals. results show that when only is assimilated (Exp_OCO3), estimated land sink significantly lower than from experiment (Exp_OCO2). estimate joint assimilation (Exp_OCO3&amp;2) comparable a scale Exp_OCO2. However, there significant regional differences. Compared observed annual growth rate, Exp_OCO3 has largest bias Exp_OCO3&amp;2 shows best performance. Furthermore, validation independent observations biases larger those Exp_OCO2 at middle high latitudes. reasons for poor performance include lack beyond 52° S N, large fluctuations number data, varied observation time. Our study indicates leads an underestimation sinks latitudes afternoon required better NEE.

Language: Английский

Citations

0
Random Forest-Based Retrieval of XCO2 Concentration from Satellite-Borne Shortwave Infrared Hyperspectral DOI Creative Commons

Wenhao Zhang,

Zhengyong Wang,

Tong Li

et al.

Atmosphere, Journal Year: 2025, Volume and Issue: 16(3), P. 238 - 238

Published: Feb. 20, 2025

As carbon dioxide (CO2) concentrations continue to rise, climate change, characterized by global warming, presents a significant challenge sustainable development. Currently, most shortwave infrared CO2 retrievals rely on fully physical retrieval algorithms, for which complex calculations are necessary. This paper proposes method predict the concentration of column-averaged (XCO2) from hyperspectral satellite data, using machine learning avoid iterative computations method. The training dataset is constructed Orbiting Carbon Observatory-2 (OCO-2) spectral XCO2 OCO-2, surface albedo and aerosol optical depth (AOD) measurements 2019. study employed variety including Random Forest, XGBoost, LightGBM, analysis. results showed that Forest outperforms other models, achieving correlation 0.933 with products, mean absolute error (MAE) 0.713 ppm, root square (RMSE) 1.147 ppm. model was then applied retrieve column 2020. 0.760 Total Column Observing Network (TCCON) measurements, higher than 0.739 product verifying effectiveness

Language: Английский

Citations

0
Towards verifying and improving estimations of China's CO2 and CH4 budgets using atmospheric inversions DOI Creative Commons
Yilong Wang, Yuzhong Zhang,

Xiangjun Tian

et al.

National Science Review, Journal Year: 2025, Volume and Issue: 12(4)

Published: March 7, 2025

This paper reviews the application of atmospheric inversions for estimating national CO₂ and CH₄ fluxes with a focus on China. After describing fundamental principles methodologies technique, we synthesize recent progress in China's budgets through inversion, compare these estimates greenhouse gas (GHG) inventory (NGHGI) reports. The inverted total CO2 CH4 emissions amount to 8.35 ± 1.39 Pg a-1 60.8 5.9 Tg a-1, respectively, last decade, which are general consistent NGHGIs. However, large uncertainties spatial temporal disaggregation hinder effectiveness method verifying GHG improving NGHGI estimates. These largely driven by differences inversion models, observational coverage methodological assumptions. We recommend networks, conducting model intercomparison exercises refining methods better support reporting future climate goals.

Language: Английский

Citations

0
Diurnal and seasonal dynamics of regional CO2 drawdown at Harvard Forest: Integrating remote sensing and modeling perspectives DOI
Yang Li,

Ethan Manninen,

Jonathan E. Franklin

et al.

The Science of The Total Environment, Journal Year: 2025, Volume and Issue: 981, P. 179580 - 179580

Published: May 8, 2025

Language: Английский

Citations

0
Top-down Constraint on Regional Fossil Fuel CO2 Emissions in China using GOSAT and OCO-2 Satellite XCO2 Retrievals: A Case of the COVID-19 Lockdown DOI
Wenyuan Chang,

Dongxu Yang,

Xiao Tang

et al.

Advances in Atmospheric Sciences, Journal Year: 2025, Volume and Issue: unknown

Published: May 20, 2025

Language: Английский

Citations

0
Satellite-detected large CO2 release in southwestern North America during the 2020–2021 drought and associated wildfires DOI Creative Commons
Hui Chen, Wei He, Jinxiu Liu

et al.

Environmental Research Letters, Journal Year: 2024, Volume and Issue: 19(5), P. 054047 - 054047

Published: April 10, 2024

Abstract Southwestern North America (SWNA) continuously experienced megadroughts and large wildfires in 2020 2021. Here, we quantified their impact on the terrestrial carbon budget using net biome production (NBP) estimates from an ensemble of atmospheric inversions assimilating in-situ CO 2 Carbon Observatory – (OCO-2) satellite XCO retrievals (OCO-2 v10 MIP Extension), two satellite-based gross primary (GPP) datasets, fire emission datasets. We found that 2021 drought associated SWNA led to a loss, mean 95.07 TgC estimated by both nadir glint (LNLG) within OCO-2 MIP, greater than 80% SWNA’s annual total sink. Moreover, loss was mainly contributed emissions while impacts uptake. In addition, indicated huge forests grasslands along with uptake reductions due shrublands. This study provides process understanding how some droughts following affect regional scale.

Language: Английский

Citations

3
Global Evaluation and Intercomparison of XCO2 Retrievals from GOSAT, OCO-2, and TANSAT with TCCON DOI Creative Commons
Junjun Fang, Baozhang Chen, Huifang Zhang

et al.

Remote Sensing, Journal Year: 2023, Volume and Issue: 15(20), P. 5073 - 5073

Published: Oct. 23, 2023

Accurate global monitoring of carbon dioxide (CO2) is essential for understanding climate change and informing policy decisions. This study compares column-averaged dry-air mole fractions CO2 (XCO2) between ACOS_L2_Lite_FP V9r Japan’s Greenhouse Gases Observing Satellite (GOSAT), OCO-2_L2_Lite_FP V10r the USA’s Orbiting Carbon Observatory-2 (OCO-2), IAPCAS V2.0 China’s Dioxide Observation (TANSAT) collectively referred to as GOT, with data from Total Column Network (TCCON). Our findings are follows: (1) Significant quantity differences exist OCO-2 other satellites, boasting a volume 100 times greater. GOT shows highest 30–45°N 20–30°S, but availability notably lower near equator. (2) XCO2 exhibits similar seasonal variations, concentrations during June, July, August (JJA) (402.72–403.74 ppm) higher December, January, February (DJF) (405.74–407.14 ppm). levels in Northern Hemisphere March, April, May (MAM) DJF, while slightly JJA September, October, November (SON). (3) The (ΔXCO2) reveal that ΔXCO2 TANSAT minor (−0.47 ± 0.28 ppm), whereas most significant difference observed GOSAT (−1.13 0.15 Minimal seen SON (with biggest TANSAT: −0.84 0.12 notable occur DJF −1.43 0.17 Regarding latitudinal distinctions pronounced SON. (4) Compared TCCON, relatively high determination coefficients (R2 > 0.8), having root mean square error (RMSE = 1.226 ppm, <1.5 indicating strong relationship ground-based retrieved values. research contributes significantly our spatial characteristics XCO2. Furthermore, it offers insights can inform analysis inversion sources sinks within assimilation systems when incorporating satellite observations.

Language: Английский

Citations

7
Prospects for the potential carbon sink effects of afforestation to enhance weathering in China DOI
Weihua Wu, Werner Nel,

J.-Q. Ji

et al.

Journal of Asian Earth Sciences, Journal Year: 2024, Volume and Issue: 276, P. 106370 - 106370

Published: Oct. 20, 2024

Language: Английский

Citations

2
Exploring the Potential of Long Short‐Term Memory Networks for Predicting Net CO2 Exchange Across Various Ecosystems With Multi‐Source Data DOI
Chengcheng Huang, Wei He, Jinxiu Liu

et al.

Journal of Geophysical Research Atmospheres, Journal Year: 2024, Volume and Issue: 129(7)

Published: April 1, 2024

Abstract Upscaling flux tower measurements based on machine learning (ML) algorithms is an essential approach for large‐scale net ecosystem CO 2 exchange (NEE) estimation, but existing ML upscaling methods face some challenges, particularly in capturing NEE interannual variations (IAVs) that may relate to lagged effects. With the capacity characterize temporal memory effects, Long Short‐Term Memory (LSTM) networks are expected help solve this problem. Here we explored potential of LSTM predicting across various ecosystems using data over 82 sites North America. The model with differentiated plant function types (PFTs) demonstrates capability explain 79.19% ( R = 0.79) monthly within testing set, RMSE and Mean Absolute Error values 0.89 0.57 g C m −2 d −1 respectively r 0.89, p < 0.001). Moreover, performed robustly cross‐site variability, 67.19% can be predicted by both models without distinguished PFTs showing improved predictive ability. Most importantly, IAV highly correlated observations 0.81, 0.001), clearly outperforming random forest −0.21, 0.011). Among all nine PFTs, solar‐induced chlorophyll fluorescence, downward shortwave radiation, leaf area index most important variables explaining variations, collectively accounting approximately 54.01% total. This study highlights great improving carbon multi‐source remote sensing data.

Language: Английский

Citations

2