11 min read
Preparations for Next Moonwalk Simulations Underway (and Underwater)
Faculty Advisors:
Dr. Lisa Haber, Virginia Commonwealth University
Dr. Brandon Alveshere, Virginia Commonwealth University
Dr. Chris Gough, Virginia Commonwealth University
Graduate Mentor:
Mindy Priddy, Virginia Commonwealth University
Mindy Priddy, Graduate Mentor
Angelina De La Torre
Using NDVI as a Proxy for GPP to Predict Carbon Dioxide Fluxes
Angelina De La Torre, California State University Channel Islands
Climate change, driven primarily by greenhouse gases, poses a threat to the future of our planet. Among these gases is carbon dioxide (CO₂), which has a much longer atmospheric residence time compared to other greenhouse gases. One potential factor in reducing atmospheric CO₂ enrichment is plant productivity. Gross Primary Productivity (GPP) estimates the amount of CO₂ fixed during photosynthesis. The Normalized Difference Vegetation Index (NDVI) provides insight into the health of an ecosystem by measuring the density and greenness of vegetation. Therefore, it can be inferred that there is a relationship between NDVI and GPP, as greener plants are likely more productive. In this study, we used NDVI as a proxy for GPP and analyzed the effect NDVI had on CO₂ fluxes during California’s wet season between January and March 2023 in a restored tidal freshwater wetland. GPP and CO₂ flux data were obtained from the Dutch Slough AmeriFlux tower in Oakley, California. Landsat data were used to calculate the average NDVI. The influence of NDVI on GPP was assessed using linear regression. A second linear regression was then performed using NDVI and CO₂ flux, of which GPP is one component. We anticipate that wetlands with greater vegetation density will have lower CO₂ emissions.
Because Landsat data scans in 16-day intervals, daily variation in NDVI could not be observed. This translates to a frequency discrepancy between the Landsat and AmeriFlux data, as AmeriFlux towers measure in half-hour intervals. Additionally, the wet season represented was limited by data availability, as the data before 2023 were unavailable. Despite data limitations in this study, the outlined process could be repeated in various wetland and climate classifications for further analysis of a larger sample size. This study could assist in developing strategies to increase CO₂ sequestration in an attempt to slow the effects of climate change.
Samarth Jayadev
Using Machine Learning to Assess Relationships between NDVI and Net Carbon Exchange During the COVID-19 Pandemic
Samarth Jayadev, California State University, Monterey Bay
Understanding the movement of carbon between Earth’s land surface and atmosphere is essential for ecosystem monitoring, creating climate change mitigation strategies, and assessing the carbon budget on national to global scales. Measures of greenness serve as indicators of processes such as photosynthesis that control carbon exchange and are vital in modeling of carbon fluxes. NASA’s Orbiting Carbon Observatory (OCO-2) provides high quality measurements of column-averaged CO₂ concentrations that can be used to derive net carbon exchange (NCE), a measure of CO₂ flux between terrestrial ecosystems and the atmosphere.
From OCO-2, NCE data collected at the land nadir, land glint satellite position combined with in situ sampling can provide accurate measurements on a 1°x1° scale suitable for carbon flux characterization across the contiguous United States (CONUS). Normalized difference vegetation index (NDVI), which ranges from -1 to +1, measures the greenness of vegetation, serving as an indicator of plant density and health. This can help to understand ecosystem to carbon-cycle interactions and be leveraged for determining patterns with NCE. We examined the relationship between NDVI and NCE across CONUS during 2020 using Gradient Boosting Decision Trees (GBDT) which specialize in classifying and predicting non-linear relationships. This algorithm takes multiple weak learners (decision trees) and combines their predictions in an iterative ensemble method to improve prediction accuracy. Feature and permutation importance tests found that January and August (trough and peak NDVI, respectively) were the highest weighted predictor variables related to NCE. The dataset was split in a 90% training 10% test ratio across latitude/longitude grid cells to assess and verify model performance. Using the mean squared error loss function and hyperparameters with optimal estimators, tree depth, sample split, and learning rate the algorithm was able to converge the test predictions to match the deviance of the training data. The gradient boosting model can be applied to different months and years of NDVI/NCE to further explore these relationships or a multitude of research questions. Further studies should consider integrating land use and land cover change variables such as bare land and urbanization to improve predictions of NCE.
Makai Ogoshi
Deep-learning Derived Spaceborne Canopy Structural Metrics Predict Forest Carbon Fluxes
Makai Ogoshi, University of Minnesota, Twin Cities
Terrestrial and airborne lidar data products describing canopy structure are potent predictors of forest carbon fluxes, but whether satellite data products produce similarly robust indicators of canopy structure is not known. The assessment of contemporary spaceborne lidar and other remote sensing data products as predictors of carbon fluxes is crucial to next generation instrument and data product design and large-spatial scale modeling. We investigated relationships between deciduous broadleaf forest canopy structure, derived from deep-learning models created with lidar data from GEDI and optical imagery from Sentinel-2, and forest carbon exchange. These included comparisons to in-situ continuous net ecosystem exchange (NEE), gross primary production (GPP), and net primary production (NPP). We find that the mean canopy height from the gridded spaceborne product has a strong correlation with forest NPP, similar to prior analysis with ground-based lidar (portable canopy lidar; PCL). For comparison to NPP, heights taken from the gridded spaceborne product were compared by overlapping the product with nine terrestrial forest sites from the National Ecological Observatory Network (NEON). We used standard deviation of canopy height as a measure of canopy structural complexity. Complexity derived from the gridded spaceborne product does not show the same strong correlation with NPP as found when using PCL. Mean annual GPP and NEE across five years were compared to the gridded spaceborne product at six Fluxnet2015-tower sites with continuous, gap-filled carbon flux data. When compared to in-situ flux tower data, neither mean canopy height nor structural complexity strongly correlate to annual NEE or GPP. Primarily, the finding that derived spaceborne products exhibit a strong correlation between forest canopy height and NPP will advance global-scale application of forest-carbon flux predictions. Secondarily, a variety of limitations highlight shortcomings in the current terrestrial flux data network. A small number of available study sites, both spatially and temporally, and lack of resolution in vertical complexity of canopy structure both contribute to uncertainty in assessing the relationships to NEE and GPP.
Sebastian Reed
Porewater Methane Concentrations Vary Significantly Across A Freshwater Tidal Wetland
Sebastian Reed, University of Alaska Anchorage Honors College
Methane is a potent greenhouse gas that is over 80 times more powerful than CO₂ at trapping heat and accounts for an estimated 30% of global temperature rise associated with climate change. The largest natural source of methane worldwide is wetlands. Despite the role of methane in driving climate change, the magnitude of global annual wetland methane flux remains highly uncertain. This study analyzes the effects of greenness (assessed using Normalized Difference Vegetation Index; NDVI), plant species composition, rooting depth, atmospheric methane concentration, and plant longevity on porewater methane concentration at the Kimages Rice Rivers Center tidal freshwater wetland. Samples for atmospheric and porewater concentrations were conducted in situ in June 2024. For each sampling location (n = 23) we collected whole air samples (WAS) 2m above the marsh surface and porewater samples 5cm below the marsh surface. We visually assessed species composition at each sample location, with 12 species of wetland plants present overall. We used the TRY plant database to find the rooting depth, leaf nitrogen content, and lifespan of each species. Drone multispectral data from 2023 was used to estimate NDVI values. These variables were compared to the pore water methane concentration via stepwise linear regression. Leaf N content, NDVI, plant species, and WAS sampling did not show statistically significant correlation to porewater methane concentration. Rooting depth showed a slight positive correlation with porewater methane (alpha = 0.1, p = 0.08, R^2 = 0.1). Samples with only perennial plants (as opposed to annual plants) had a higher mean value of porewater methane (p = 0.1). Analyzing porewater methane provides insight as to what wetland components affect methanogenesis and methane release, which aids in assessing which plant functional traits are most responsible for driving or mitigating climate change. Results from this study and future research in this area has the potential to more accurately assess how methane cycles through wetlands to the atmosphere.
Nohemi Rodarte
Understanding the vertical profile of CO₂ concentration: How carbon dioxide levels change with altitude
Nohemi Rodarte, Adams State University
Carbon dioxide (CO₂) is one of the main greenhouse gasses that contribute to global warming.While the relationship between CO₂ concentrations and land cover types, such as forests and urban areas, is well documented, there is limited knowledge of how CO₂ concentrations vary with altitude at fine spatial scales. Guided by our hypothesis that CO₂ levels vary with altitude and increase with elevation, we used airborne data collected from the B200 aircraft, which flew at different altitudes (400 to 1200 feet) above the urban area of Hopewell, Virginia, between 9:40 AM and 10:40 AM. We analyzed the CO₂ concentrations recorded by the flight to obtain the median and range for each 100 feet of altitude. Our results reveal that carbon dioxide concentrations varied significantly across the range of altitudes investigated. Within the area studied, CO₂ concentrations were found to range between 410 and 470 ppm. The distribution of these concentrations along the altitude gradient shows a bimodal pattern, with notable peaks at altitudes of 700 to 800 feet and 1100 to 1200 feet. Although CO₂ levels were present at all measured altitudes, there was a noticeable drop in the mean concentration at 800 feet,which then stabilized until reaching 1,000 feet before rising again. This pattern indicates that the concentrations of this greenhouse gas are not uniformly distributed with altitude, but rather vary significantly, showing higher concentrations at certain elevations and lower concentrations at others. The CO₂ distribution fluctuates with altitude, showing higher or lower levels at specific heights rather than a smooth gradient, indicating that altitude impacts CO₂ concentrations. While we did not identify the drivers of this change, future studies could evaluate how factors such as surface emissions, atmospheric mixing, and local conditions may contribute to vertical CO₂ profiles, since the altitudes we considered in this research are within the troposphere.
Camille Shaw
Linking NDVI with CO₂ and CH₄ Fluxes: Insights into Vegetation and Urban Source-Sink Dynamics in the Great Dismal Swamp
Camille Shaw, Brigham Young University
In recent years, carbon dioxide, methane, and other greenhouse gases have gained attention because of their contribution to the rise in Earth’s global mean temperature. Methane and carbon dioxide have various sources and sinks, but an expanding array of sources have created a need to assess ongoing change in carbon balance. This study aims to quantify the relationship between Normalized Difference Vegetation Index, or NDVI, and methane and carbon dioxide fluxes. We measured carbon dioxide and methane concentrations within the boundary layer using the PICARRO instrument, focusing on the Great Dismal Swamp, a forested wetland, and surrounding areas in the Eastern Mid-Atlantic Region. Data collection occurred at various times of day and along different flight paths in 2016, 2017, and 2024, with each year representing data from a single season, either spring or fall, for temporal analysis. We calculated methane and carbon dioxide fluxes along the flight paths using airborne eddy covariance, a method for capturing accurate flux measurements while accounting for the mixing of gases in the boundary layer caused by heat. Additionally, we calculated NDVI for this area using NASA’s Landsat 8 and 9 satellite imagery. Analysis of the afternoon flight data revealed a negative linear correlation between NDVI and carbon dioxide flux. Urban areas, characterized by low NDVI, exhibit a positive carbon dioxide flux as a consequence of emissions from vehicles, while forested areas, with high NDVI, show a negative carbon dioxide flux because of photosynthesis. In contrast, methane flux shows minimal correlation with NDVI. The lack of correlation arises because forested wetlands, with high NDVI, emit substantial amounts of methane, while urban areas, despite having low NDVI, still produce significant methane emissions from landfills and industrial activities. Future research could further investigate how seasonal and diurnal variations influence the correlations between NDVI and greenhouse gases by collecting comprehensive data across all seasons within a given year and at various times of the day.
