About

                                        
@inproceedings{erickson2021b,
    author       = {Adam Erickson and Sujay V. Kumar and Derek L. Hudson and Snorre Stamnes and Eetu Puttonen and Samuli Junttila and S\"{o}ren Pirk and Bernhard H\"{o}fle and Lukas Chrostowski and Jan Eitel and Kim Calders},
    year         = 2021,
    title        = {{Hypersurface Observation Network (Hyperon)} --- What it is and why we need it},
    booktitle    = {AGU Fall Meeting 2021},
    series       = {AGU 2021},
    month        = 12,
    day          = 13,
    organization = {American Geophysical Union},
    address      = {New Orleans, Louisiana, USA},
    eid          = {B15G-1499},
    url          = {https://agu.confex.com/agu/fm21/meetingapp.cgi/Paper/976063}
}
                                    

In the terrestrial domain, large biogeochemical and energetic uncertainties surround the soil-plant-atmosphere continuum of forests, leading to wide disagreement in the projected land carbon sink and global carbon balance. This is largely due to an absence of global observation networks providing coincident information on the structure, composition, and function of forests and adjacent planetary boundary layer (PBL) over space and time. Such observations are critical to learning new models of biospheric processes and improving our understanding and predictions of land-atmosphere exchange. Simultaneously, a new generation of hyper-temporal, -spatial, - spectral, and -angular (hypersensing) surface reference networks are needed to learn inversions of air- and space-borne measurements that resolve radiative-transfer uncertainties related to these land-atmosphere exchanges.

Existing terrestrial networks such as FluxNet, SpecNet, PhenoCam, AeroNet, and ForestGeo remain limited to single measurement points and domains. For example, FluxNet typically records CO2 and H2O exchanges at a single point in space using domain-specific instruments, making the measurements ungeneralizable, unrepresentative, and thus unsuited to up-scaling to Earth observation records or diagnosing Earth system models. Furthermore, adding new measurements to these systems often requires costly new instruments. Absent are generalizable surface-atmosphere observation systems able to retrieve a growing variety of observables over space and time.

Toward addressing these needs in unified Earth observation and system modeling (EOSM), we propose the Hypersurface Observation Network (Hyperon) — formerly 5DNet — a modular, intelligent, robotic, and coincident surface-atmosphere observation system. Initially focusing on forests, Hyperon is intended to cover a variety of surface domains. While previously impractical, powerful new low-cost instruments and embedded computers designed for edge inference provide an exciting opportunity to realize this goal. We provide an early conceptual overview of Hyperon, reaching out to the community to develop standards for instrument and site configuration options and a decentralized governance model for ensuring free and open science.

                                        
@inproceedings{erickson2021a,
    author       = {Adam Erickson and Rico Fischer and Sujay Kumar and Annikki M\"{a}kel\"{a} and Nikolay Strigul},
    year         = 2021,
    title        = {{Union Symposium 4}: Towards evolvable physics-based plants and landscape processes in terrestrial biosphere models},
    booktitle    = {EGU General Assembly},
    series       = {EGU 2021},
    organization = {European Geophysical Union},
    address      = {virtual},
    eid          = {US4},
    url          = {https://meetingorganizer.copernicus.org/EGU21/session/39990},
    eprint       = {https://meetingorganizer.copernicus.org/EGU21/sessionAssets/39990/summary.pdf}
}
                                    

The terrestrial biosphere exerts disproportionate influence on Earth's climate, making improvements in its representation key to reducing climate uncertainty. After 50 years of development, land surface models contain detailed processes of energy fluxes, photosynthesis, hydrology, C-N-P cycles, and land-use within coarse non-interacting grid cells. Remaining discrepancies in fidelity to observed carbon and water cycles appear primarily related to deficiencies in the representation of forests and human activity. These include the omission of spatial processes of disturbance, migration, adaptation, and management. Also missing is the generative process of life, evolution, which gives rise to life history strategies, trophic-metabolic networks, leaf economics, local adaptation (i.e., optimality, acclimation), and plant behaviour. Despite improvements in representing vegetation demography by utilizing emergent properties of allometric scaling, canopy geometric realism remains low. This may bias carbon and water cycles per radiative transfer and coupled processes of photosynthesis, regeneration, evapotranspiration, heterotrophic respiration, and disturbance.

We believe that physics-based botanical models, forest landscape models, and terrestrial biosphere models may soon merge into new multi-scale models. While low-dimensional representations of forests are often used to improve computational efficiency and cope with a dearth of 4-D forest observatories, deep learning may be combined with new autonomous scanning systems - proximal and/or remote - such as our proposed global tower-based '5DNet' to infer evolvable 4-D physics-based models. This includes learning multi-generation tree models with 4-D traits from image and/or laser scanning time-series. To date, 4-D ontogeny has been inferred from individual scans of mature trees, multi-plant phenological events have been tracked in real-time, and the self-similar and -organizing nature of plants has been used to efficiently compress tree models down to their generating parameters. Achieving leaf-to-global scaling may require co-processor acceleration and fusing deep learning with 3-D radiative transfer modeling to infer global surface properties. An additional focus on evolution and human activity comes as 21st century land surface models mature into general simulations of life on Earth.

This Union Symposium presents exciting work toward achieving this moonshot in Earth observation and systems modeling.

                                        
@article{cawsenicholson2021,
    author    = {Kerry Cawse-Nicholson and Philip A. Townsend and David Schimel and Ali M. Assiri and Pamela L. Blake and Maria Fabrizia Buongiorno and Petya Campbell and Nimrod Carmon and Kimberly A. Casey and Rosa Elvira Correa-Pabón and Kyla M. Dahlin and Hamid Dashti and Philip E. Dennison and Heidi Dierssen and Adam Erickson and Joshua B. Fisher and Robert Frouin and Charles K. Gatebe and Hamed Gholizadeh and Michelle Gierach and Nancy F. Glenn and James A. Goodman and Daniel M. Griffith and Liane Guild and Christopher R. Hakkenberg and Eric J. Hochberg and Thomas R.H. Holmes and Chuanmin Hu and Glynn Hulley and Karl F. Huemmrich and Raphael M. Kudela and Raymond F. Kokaly and Christine M. Lee and Roberta Martin and Charles E. Miller and Wesley J. Moses and Frank E. Muller-Karger and Joseph D. Ortiz and Daniel B. Otis and Nima Pahlevan and Thomas H. Painter and Ryan Pavlick and Ben Poulter and Yi Qi and Vincent J. Realmuto and Dar Roberts and Michael E. Schaepman and Fabian D. Schneider and Florian M. Schwandner and Shawn P. Serbin and Alexey N. Shiklomanov and E. Natasha Stavros and David R. Thompson and Juan L. Torres-Perez and Kevin R. Turpie and Maria Tzortziou and Susan Ustin and Qian Yu and Yusri Yusup and Qingyuan Zhang},
    year      = 2021,
    title     = {{NASA's Surface Biology and Geology Designated Observable}: A perspective on surface imaging algorithms},
    journal   = {Remote Sensing of Environment},
    volume    = 257,
    pages     = {112349},
    publisher = {Elsevier},
    address   = {Amsterdam, North Holland, Netherlands},
    issn      = {0034-4257},
    doi       = {https://doi.org/10.1016/j.rse.2021.112349},
    url       = {https://www.sciencedirect.com/science/article/pii/S0034425721000675}
}
                                    

The 2017–2027 National Academies' Decadal Survey, Thriving on Our Changing Planet, recommended Surface Biology and Geology (SBG) as a “Designated Targeted Observable” (DO). The SBG DO is based on the need for capabilities to acquire global, high spatial resolution, visible to shortwave infrared (VSWIR; 380–2500 nm; ~30 m pixel resolution) hyperspectral (imaging spectroscopy) and multispectral midwave and thermal infrared (MWIR: 3–5 μm; TIR: 8–12 μm; ~60 m pixel resolution) measurements with sub-monthly temporal revisits over terrestrial, freshwater, and coastal marine habitats. To address the various mission design needs, an SBG Algorithms Working Group of multidisciplinary researchers has been formed to review and evaluate the algorithms applicable to the SBG DO across a wide range of Earth science disciplines, including terrestrial and aquatic ecology, atmospheric science, geology, and hydrology. Here, we summarize current state-of-the-practice VSWIR and TIR algorithms that use airborne or orbital spectral imaging observations to address the SBG DO priorities identified by the Decadal Survey: (i) terrestrial vegetation physiology, functional traits, and health; (ii) inland and coastal aquatic ecosystems physiology, functional traits, and health; (iii) snow and ice accumulation, melting, and albedo; (iv) active surface composition (eruptions, landslides, evolving landscapes, hazard risks); (v) effects of changing land use on surface energy, water, momentum, and carbon fluxes; and (vi) managing agriculture, natural habitats, water use/quality, and urban development. We review existing algorithms in the following categories: snow/ice, aquatic environments, geology, and terrestrial vegetation, and summarize the community-state-of-practice in each category. This effort synthesizes the findings of more than 130 scientists.

                                        
@inproceedings{erickson2020poster2,
    author       = {Adam Erickson and Sujay Kumar},
    year         = 2020,
    title        = {Synthetic spectranomics: deep learning of surface {3-D} geometry, chemistry, and hyperspectra to inform next-generation land models},
    booktitle    = {AGU Fall Meeting 2020},
    series       = {AGU 2021},
    month        = 12,
    day          = 9,
    organization = {American Geophysical Union},
    address      = {virtual},
    eid          = {B031-0004},
    url          = {https://agu.confex.com/agu/fm20/meetingapp.cgi/Paper/710593}
}
                                    

Machine learning, and in particular deep learning, has transformed our approach to Earth observation and systems modeling, or EOSM. Trained on near-to-remote sensing observations, physical models, or hybrids of both, machine learning may improve existing model formulations while creating new classes of models. Potential applications include data geolocalization and assimilation, model optimization, model emulation/surrogates, upsampling/downscaling, land-surface reconstruction for physics-based 4-D models, learning new governing equations for latent or difficult-to-simulate dynamics, simulating observing systems with realistic noise (hybrid network + GAN), inverse modeling, causal inference, and model code generation, among others. Full utilization of protected global inventory datasets is possible with privacy preserving networks, while biases and adversarial robustness must be addressed. One important application is observation synthesis, allowing land models to benefit from the increased spatial, temporal, and spectral/polarimetric resolution proposed for future observing systems. This provides the continuous historical record needed to constrain, calibrate/validate, and develop models. Despite the many datasets released for remote sensing computer vision challenges to date, large labeled datasets remain scarce for a number of tasks relevant to Earth systems and planetary science generally. To address this limitation, we leverage complementary coincident observations from air- and space-borne platforms to generate labeled data relevant to modeling tasks. We focus on the information-theoretic task of maximizing mutual information between predictors X and targets Y, or feature extraction. First, we will utilize airborne spectranomics observations from NASA G-LiHT, NEON AOP, and/or ASU GAO to learn canopy geometry, chemistry, and hyperspectral reflectances from pseudo-multispectral and/or RGB imagery. Second, we aim to utilize NASA EO-1 Hyperion and ALI observations to learn the mapping from multi- to hyper-spectral. The trained model may then be applied to generate synthetic hyperspectral imagery using Landsat-8 OLI observations, for which ALI was a demonstrator. Subsequent work may extend this approach to commercial satellite constellations.

                                        
@inproceedings{Erickson2020_02,
    author = {Adam Erickson and Benjamin Poulter and David Thompson and Gregory Okin and Shawn Serbin and Weile Wang and David Schimel},
    title = {A software framework for optimizing the design of spaceborne hyperspectral imager architectures},
    booktitle = {EGU General Assembly Conference Abstracts},
    organization = {European Geophysical Union},
    year = 2020,
    month = may,
    eid = {19665},
    url = {https://meetingorganizer.copernicus.org/EGU2020/EGU2020-19665.html},
}
                                    

Quantifying the capacity, and uncertainty, of proposed spaceborne hyperspectral imagers to retrieve atmospheric and surface state information is necessary to optimize future satellite architectures for their science value. Given the vast potential joint trade-and-environment-space, modeling key ‘globally representative’ points in this n-dimensional space is a practical solution for improving computational tractability. Given guidance from policy instruments such as the NASA Decadal Survey and the recommended Designated Target Observables, or DOs, the downselect process can be viewed as a constrained multi-objective optimization. The need to simulate imager architecture performance to achieve downselect goals has motivated the development of new mathematical models for estimating radiometric and retrieval uncertainties provided conditions analogous to real-world environments. The goals can be met with recent advances that integrate mature atmospheric inversion approaches such as Optimal Estimation (OE) that includes joint atmospheric-surface state estimation (Thompson et al. 2018) and the EnMAP end-to-end simulation tool, EeteS (Segl et al. 2012), which utilizes OE for inversions. While surface-reflectance and retrieval simulation models are normally run in isolation on local computing environments, we extend tools to enable uncertainty quantification into new representative environments and thereby increase robustness of the downselect process by providing an advanced simulation model to the broader hyperspectral imaging community in software-as-a-service (SaaS). Here, we describe and demonstrate our instrument modeling web service and corresponding hyperspectral traceability analysis (HyperTrace) library for Python. The modeling service and underlying HyperTrace OE library are deployed on the NASA DISCOVER high-performance computing (HPC) infrastructure. An intermediate HTTP server communicates between FTP and HTTP servers, providing persistent archival of model inputs and outputs for subsequent meta-analyses. To facilitate enhanced community participation, users may simply transfer a folder containing ENVI format hyperspectral imagery and a corresponding JSON metadata file to the FTP server, from which it is pulled to a NASA DISCOVER server for processing, with statistical, graphical, and ENVI-formatted results subsequently returned to the FTP server where it is available for users to download. This activity provides an expanded capability for estimating the various science values of architectures under consideration for NASA’s Surface Biology and Geology Designated Observable.

                                        
@article{Erickson2020_01,
    author = {Erickson, Adam and Strigul, Nikolay},
    title = {Implementation of the perfect plasticity approximation with biogeochemical compartments in R},
    journal = {Ecography},
    year = {2020},
    volume = {n/a},
    number = {n/a},
    pages = {},
    kerywords = {forest ecosystems, ecological modeling, biogeochemistry, software, R},
    doi = {10.1111/ecog.04756},
    url = {https://onlinelibrary.wiley.com/doi/abs/10.1111/ecog.04756},
    eprint = {https://onlinelibrary.wiley.com/doi/pdf/10.1111/ecog.04756}
}
                                    

Modeling forest ecosystems is a landmark challenge in science, due to the complexity of the processes involved and their importance in predicting future planetary conditions. While there are a number of open‐source forest biogeochemistry models, few papers exist detailing the software development approach used to develop these models. This has left many forest biogeochemistry models large, opaque and/or difficult to use, typically implemented in compiled languages for speed. Here, we present a forest biogeochemistry model from the SORTIE‐PPA class of models, PPA‐SiBGC. Our model is based on the perfect plasticity approximation with simple biogeochemistry compartments and uses empirical vegetation dynamics rather than detailed prognostic processes to drive the estimation of carbon and nitrogen fluxes. This allows our model to be used with traditional forest inventory data, making it widely applicable and simple to parameterize. We detail the conceptual design of the model as well as the software implementation in the R language for statistical computing. Our aim is to provide a useful tool for the biogeochemistry modeling community that demonstrates the importance of vegetation dynamics in biogeochemical models.

                                        
@inproceedings{Erickson2019_03,
    title = {A software framework for rapid prototyping of artificial intelligence in Earth system models},
    author = {Erickson, Adam and Strigul, Nick},
    booktitle = {AGU Fall Meeting 2019},
    year = {2019},
    organization = {American Geophysical Union},
    url = {https://agu.confex.com/agu/fm19/meetingapp.cgi/Paper/634946}
}
                                    

Machine or deep learning represents a promising new way to represent processes in Earth system models. These models are fundamentally data-driven or pattern-based, whereby models are learned from data. This fills a critical gap in our current modeling abilities given the absence of process understanding to produce mechanistic or physical models. While machine/deep learning practitioners commonly use Python frameworks such as Scikit-Learn, TensorFlow, and PyTorch, component models of Earth systems are commonly implemented in low-level FORTRAN and may move toward C++ for its numerous industry-standard libraries. This divide between user-friendly high-level languages and high-performance low-level languages is known as the two-language problem. We address this problem by proposing the generation of a Python interface to C++ code using CMake and pybind11. This makes high-performance C++ classes and methods accessible to Python deep learning frameworks, allowing rapid prototyping of machine/deep learning process models.

                                        
@inproceedings{Erickson2019_02,
    author = {{Erickson}, Adam and {Strigul}, Nikolay},
    title = {Validation of a simple biogeochemistry variant of SORTIE-PPA in two temperate forests using the Erde modeling framework},
    booktitle = {EGU General Assembly Conference Abstracts},
    organization = {European Geophysical Union},
    year = 2019,
    month = apr,
    eid = {11708},
    url = {https://ui.adsabs.harvard.edu/abs/2019EGUGA..2111708E},
}
                                    

While vegetation model development has accelerated over the past three decades, lessprogress has been made in software interfaces that bind models and data. Such interfacesbecome necessary as model complexity increases, making models more cumbersome for newusers to parameterize, run, and analyze. Furthermore, model wrappers may provide newmodeling capabilities, such as Bayesian optimization. Toward this end, we have developed asimple geoscientific simulation model API and toolkit in R and Python known as theEarth-systems Research and Development Environment (Erde). While Erde is primarilyintended for vegetation models, its data structures and algorithms are applicable across arange of geoscientific modeling domains. We demonstrate an application of Erde with asimple biogeochemistry variant of SORTIE-PPA known as PPA-SiBGC. The PPA-SiBGCmodel combines the Perfect Plasticity Approximation with explicit above-and-below-groundbiogeochemical pools and simple flux models. We parameterized, ran, and validatedPPA-SiBGC at two research forests in the Eastern United States: (1) Harvard Forest,Massachusetts (HF-EMS) and (2) Jones Ecological Research Center, Georgia (JERC-RD).We assessed model fitness using these temporal metrics: net ecosystem exchange,aboveground net primary production, aboveground biomass, C, and N, belowgroundbiomass, C, and N, soil respiration, and, species total biomass and relative abundance.Without applying any optimization, we found that a simple biogeochemistry variant ofSORTIE-PPA was able to outperform an established forest landscape model (LANDIS-IINECN) across the metrics tested. While LANDIS-II NECN showed better NEE fit,PPA-SiBGC demonstrated better overall correspondence to field data for both sites(HF-EMS: R2=0.73, +0.07, RMSE=4.84, −10.02; JERC-RD: R2=0.76, +0.04,RMSE=2.69, −1.86)

                                        
@article{Erickson2019_01,
    author = {Erickson, Adam and Strigul, Nikolay},
    title = {A Forest Model Intercomparison Framework and Application at Two Temperate  Forests Along the East Coast of the United States},
    journal = {Forests},
    year = {2019},
    volume = {10},
    number = {2},
    pages = {180},
    url = {https://www.mdpi.com/1999-4907/10/2/180},
    issn = {1999-4907},
    doi = {10.3390/f10020180}
}
                                    

State-of-the-art forest models are often complex, analytically intractable, and computationally expensive, due to the explicit representation of detailed biogeochemical and ecological processes. Different models often produce distinct results while predictions from the same model vary with parameter values. In this project, we developed a rigorous quantitative approach for conducting model intercomparisons and assessing model performance. We have applied our original methodology to compare two forest biogeochemistry models, the Perfect Plasticity Approximation with Simple Biogeochemistry (PPA-SiBGC) and Landscape Disturbance and Succession with Net Ecosystem Carbon and Nitrogen (LANDIS-II NECN). We simulated past-decade conditions at flux tower sites located within Harvard Forest, MA, USA (HF-EMS) and Jones Ecological Research Center, GA, USA (JERC-RD). We mined field data available from both sites to perform model parameterization, validation, and intercomparison. We assessed model performance using the following time-series metrics: Net ecosystem exchange, aboveground net primary production, aboveground biomass, C, and N, belowground biomass, C, and N, soil respiration, and species total biomass and relative abundance. We also assessed static observations of soil organic C and N, and concluded with an assessment of general model usability, performance, and transferability. Despite substantial differences in design, both models achieved good accuracy across the range of pool metrics. While LANDIS-II NECN showed better fidelity to interannual NEE fluxes, PPA-SiBGC indicated better overall performance for both sites across the 11 temporal and two static metrics tested. To facilitate further testing of forest models at the two sites, we provide pre-processed datasets and original software written in the R language of statistical computing. In addition to model intercomparisons, our approach may be employed to test modifications to forest models and their sensitivity to different parameterizations.