Machine Learning‐Based Modeling of Vegetation Leaf Area Index and Gross Primary Productivity Across North America and Comparison With a Process‐Based Model

Zhang, Zhicheng; Xin, Qinchuan; Li, Wanjing

doi:10.1029/2021ms002802

Cited by 19 publications

(7 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Similar strong correlations between LAI and GPP have also been found by other researchers (e.g. Qu et al 2018 ; Zhang et al 2021 ; Chen et al 2023 ).…”

Section: Discussionsupporting

confidence: 89%

Assessment of spongy moth infestation impacts on forest productivity and carbon loss using the Sentinel-2 satellite remote sensing and eddy covariance flux data

Hussain,

Gonsamo,

Wang

et al. 2024

Ecol Process

View full text Add to dashboard Cite

Background Deciduous forests in eastern North America experienced a widespread and intense spongy moth (Lymantria dispar) infestation in 2021. This study quantified the impact of this spongy moth infestation on carbon (C) cycle in forests across the Great Lakes region in Canada, utilizing high-resolution (10 × 10 m2) Sentinel-2 satellite remote sensing images and eddy covariance (EC) flux data. Study results showed a significant reduction in leaf area index (LAI) and gross primary productivity (GPP) values in deciduous and mixed forests in the region in 2021. Results Remote sensing derived, growing season mean LAI values of deciduous (mixed) forests were 3.66 (3.18), 2.74 (2.64), and 3.53 (2.94) m2 m−2 in 2020, 2021 and 2022, respectively, indicating about 24 (14)% reduction in LAI, as compared to pre- and post-infestation years. Similarly, growing season GPP values in deciduous (mixed) forests were 1338 (1208), 868 (932), and 1367 (1175) g C m−2, respectively in 2020, 2021 and 2022, showing about 35 (22)% reduction in GPP in 2021 as compared to pre- and post-infestation years. This infestation induced reduction in GPP of deciduous and mixed forests, when upscaled to whole study area (178,000 km2), resulted in 21.1 (21.4) Mt of C loss as compared to 2020 (2022), respectively. It shows the large scale of C losses caused by this infestation in Canadian Great Lakes region. Conclusions The methods developed in this study offer valuable tools to assess and quantify natural disturbance impacts on the regional C balance of forest ecosystems by integrating field observations, high-resolution remote sensing data and models. Study results will also help in developing sustainable forest management practices to achieve net-zero C emission goals through nature-based climate change solutions.

show abstract

“…Similar strong correlations between LAI and GPP have also been found by other researchers (e.g. Qu et al 2018 ; Zhang et al 2021 ; Chen et al 2023 ).…”

Section: Discussionsupporting

confidence: 89%

Assessment of spongy moth infestation impacts on forest productivity and carbon loss using the Sentinel-2 satellite remote sensing and eddy covariance flux data

Hussain,

Gonsamo,

Wang

et al. 2024

Ecol Process

View full text Add to dashboard Cite

show abstract

“…For instance, a machine-learning model might discern relationships between certain features and NEP without any ecological justification. Conversely, vegetation ecological process models are constructed based on a profound understanding of plant growth, photosynthesis, and respiration processes [62,63]. Thus, their predictions typically have clear ecological and biophysical interpretations, which grants these models an advantage in explaining and understanding ecosystem processes.…”

Section: Integrating Machine Learning and Ecological Process Models F...mentioning

confidence: 99%

Synergistic Application of Multiple Machine Learning Algorithms and Hyperparameter Optimization Strategies for Net Ecosystem Productivity Prediction in Southeast Asia

Huang,

Chen,

Sun

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

The spatiotemporal patterns and shifts of net ecosystem productivity (NEP) play a pivotal role in ecological conservation and addressing climate change. For example, by quantifying the NEP information within ecosystems, we can achieve the protection and restoration of natural ecological balance. Monitoring the changes in NEP enables a more profound understanding and prediction of ecosystem alterations caused by global warming, thereby providing a scientific basis for formulating policies aimed at mitigating and adapting to climate change. The accurate prediction of NEP sheds light on the ecosystem’s response to climatic variations and aids in formulating targeted carbon sequestration policies. While traditional ecological process models provide a comprehensive approach to predicting NEP, they often require extensive experimental and empirical data, increasing research costs. In contrast, machine-learning models offer a cost-effective alternative for NEP prediction; however, the delicate balance in algorithm selection and hyperparameter tuning is frequently overlooked. In our quest for the optimal prediction model, we examined a combination of four mainstream machine-learning algorithms with four hyperparameter-optimization techniques. Our analysis identified that the backpropagation neural network combined with Bayesian optimization yielded the best performance, with an R2 of 0.68 and an MSE of 1.43. Additionally, deep-learning models showcased promising potential in NEP prediction. Selecting appropriate algorithms and executing precise hyperparameter-optimization strategies are crucial for enhancing the accuracy of NEP predictions. This approach not only improves model performance but also provides us with new tools for a deeper understanding of and response to ecosystem changes induced by climate change.

show abstract

“…The SIF‐based and ML‐based products are produced using emerging methods and have been widely used in recent years. However, the SIF‐based products are limited by the short time range and the spatial discontinuity of satellite observations (Frankenberg et al., 2014; Köhler et al., 2018), and the ML methods cannot explain the mechanism of the photosynthetic process and depend on the data used for the model training (Jung et al., 2011, 2019; Zhang et al., 2021). The process‐based products are based on models with solid biophysical mechanisms, but the massive inputs and parameter requirements usually result in an insufficient spatial resolution and poor accuracy for these products, such as the CLM5 and BIOME‐BGC products (Huang et al., 2011; Lawrence et al., 2019).…”

Section: Introductionmentioning

confidence: 99%

Improving MODIS Gross Primary Productivity by Bridging Big‐Leaf and Two‐Leaf Light Use Efficiency Models

Ma,

Guan,

Chen

et al. 2024

JGR Biogeosciences

View full text Add to dashboard Cite

Gross primary productivity (GPP) is an important component of the terrestrial carbon cycle in climate change research. The global GPP product derived using Moderate Resolution Imaging Spectroradiometer (MODIS) data is perhaps the most widely used. Unfortunately, many studies have indicated evident error patterns in the MODIS GPP product. One of the main reasons for this is that the applied big‐leaf (BL) MOD17 model cannot properly handle the variable relative contribution of sunlit and shaded leaves to the total canopy‐level GPP. In this study, we developed a model for correcting the errors in the MODIS GPP product by bridging BL and two‐leaf (TL) light use efficiency (LUE) models (CTL‐MOD17). With the available MODIS GPP product, which considers environmental stress factors, the CTL‐MOD17 model only needs to reuse the two inputs of the leaf area index (LAI) and incoming radiation. The CTL‐MOD17 model was calibrated and validated at 153 global FLUXNET eddy covariance (EC) sites. The results indicate that the modeled GPP obtained with the correction model matches better with the EC GPP than the original MODIS GPP product at different time scales, with an improvement of 0.07 in R2 and a reduction in root‐mean‐square error (RMSE) of 117.08 g C m−2 year−1. The improvements are more significant in the green season when the contribution of shaded leaves is larger. In terms of the global spatial pattern, the obvious underestimation in the regions with high LAI and the overestimation in the low LAI regions of the MODIS GPP product is effectively corrected by the CTL‐MOD17 model. This paper not only bridges the BL and TL LUE models, but also provides a new and simple method to obtain accurate GPP through reusing two inputs used in producing the MODIS GPP product.

show abstract

Machine Learning‐Based Modeling of Vegetation Leaf Area Index and Gross Primary Productivity Across North America and Comparison With a Process‐Based Model

Cited by 19 publications

References 48 publications

Assessment of spongy moth infestation impacts on forest productivity and carbon loss using the Sentinel-2 satellite remote sensing and eddy covariance flux data

Assessment of spongy moth infestation impacts on forest productivity and carbon loss using the Sentinel-2 satellite remote sensing and eddy covariance flux data

Synergistic Application of Multiple Machine Learning Algorithms and Hyperparameter Optimization Strategies for Net Ecosystem Productivity Prediction in Southeast Asia

Improving MODIS Gross Primary Productivity by Bridging Big‐Leaf and Two‐Leaf Light Use Efficiency Models

Contact Info

Product

Resources

About