Complex non-linear relationships exist between the permafrost thermal state, active layer thickness, and terrestrial carbon cycle dynamics. In Arctic and boreal Alaska, large uncertainties characterizing the spatiotemporal rate, magnitude, and extent of permafrost degradation and the permafrost carbon feedback, with an increasing recognition of the importance of abrupt thaw mechanisms. The challenges of monitoring sub-surface phenomena, such as soil temperature and soil moisture profiles, with remote sensing technology further complicates the issue. There is an urgent need to understand how and to what extent permafrost degradation destabilizes the carbon balance in Alaska and characterizes the feedbacks involved. In this research, we employ our artificial intelligence (AI)-driven model GeoCryoAI to quantify permafrost thaw dynamics and greenhouse gas (GHG) emissions in Alaska. The GeoCryoAI model uses a hybridized multimodal deep learning architecture of stacked convolutionally layered memory-encoded bidirectional recurrent neural networks to simultaneously ingest and analyze in situ measurements, remote sensing observations, and process-based modeling outputs with disparate spatiotemporal sampling and data densities. Evaluation of naïve persistence, teacher forcing, and time-delayed GeoCryoAI simulations yielded promising results with the following error metrics (RMSE) for active layer thickness (ALT), carbon dioxide (CO2), and methane (CH4) respectively: 1.997cm, 1.327cm, 1.007cm [1969-2022]; 1.906µmolCO2m-2s-1, 0.697µmolCO2m-2s-1, 0.213µmolCO2m-2s-1 [2011-2022]; 0.884nmolCH4m-2s-1, 0.715nmolCH4m-2s-1, 0.694nmolCH4m-2s-1 [2003-2021]. Our approach overcomes traditional model inefficiencies and resolves spatiotemporal disparities. GeoCryoAI captures abrupt and persistent changes while providing a novel methodology for assimilating contemporaneous information at various scales. We describe GeoCryoAI, the methodology, our results, and plans for future innovations and applications.

It is well-established that positive feedbacks between permafrost degradation and the release of soil carbon into the atmosphere impact land-atmosphere feedback mechanisms, disrupt the global carbon cycle, and accelerate climate change. Permafrost dynamics are relevant to the global community because the distribution of this frozen ground substrate characterizes nearly 23 million square kilometers of the northern latitudes. The widespread distribution of thawing permafrost is causing a cascade of geophysical and biochemical disturbances with global impact. Current earth system models do not account for permafrost carbon feedback mechanisms; we are exploring, simulating, and quantifying this limitation with field-scale surveys and numerical modeling, image processing, and machine learning at scale across the tundra and taiga ecosystems (TTE). This research seeks to identify, interpret, and explain the causal links and feedback sensitivities attributed to permafrost degradation and terrestrial carbon cycling asymmetry with in situ observations, remote sensing imagery, modeling and reanalysis products, and a hybridized multimodal deep learning ensemble of recurrent, convolutionally-layered, memory-based networks (GeoCryoAI). Preliminary metrics obtained from mirroring freeze-thaw dynamics and soil carbon flux across four subdomain in Alaska yield a root mean square error of 6.3637 and 4.7973, respectively. More specifically, this data-driven modeling ensemble is composed of a convolutional neural network-filtered (CNN) long short-term memory-encoded (LSTM) recurrent neural network that integrates teacher learning from in situ observations while embedding satellite-based measurements and time series datasets into a network of activation functions and processing layers. These outputs are then trained within a variational autoencoder framework (VAE) that encodes and imputes proper decoding protocol necessary for generative adversarial training, benchmarking, and reconstructing synthetic time series data for gap-filling and feature learning. Ongoing work demonstrates the fidelity of monitoring active layer thickness (ALT) variability as a sensitive, silent-but-pronounced harbinger of change; a unique signal for characterizing and forecasting permafrost degradation, soil carbon flux, and other biogeochemical drivers facilitating land cover change and earth system feedbacks. These multimodal approaches to knowledge discovery will not only improve sensitivity analyses and disentangle the spatial processes and causal links behind drivers of change, but also reconcile disparate estimations and below-ground uncertainty across the Arctic system.

The Arctic is warming four times faster than the global average, resulting in widespread ground thaw and state changes. Due to the rapid rate and large scale of ecosystem shifts, identifying and understanding Arctic boreal zone changes and feedbacks requires frequent observations across multiple scales. The last decade has witnessed significant increases in the number and coverage of in-situ, airborne, and satellite observations. However, additional resolution, coverage, and sustained, long-term time series data records are urgently required to characterize and understand the considerable heterogeneity of Northern permafrost environments. Here, we review the physical and technical gaps that limit the ability to detect rapid state changes and tipping points in permafrost ecosystems. Understanding and accurately forecasting changes to the Arctic is an essential component of managing climate change in this rapidly transforming system.

It is well-established that positive feedbacks between permafrost degradation and the release of soil carbon into the atmosphere impacts land-atmosphere interactions, disrupts the global carbon cycle, and accelerates climate change. The widespread distribution of thawing permafrost is causing a cascade of geophysical and biochemical disturbances with global impact. Currently, few earth system models account for permafrost carbon feedback mechanisms. This research identifies, interprets, and explains the feedback sensitivities attributed to permafrost degradation and terrestrial carbon cycling imbalance with in-situ and flux tower measurements, remote sensing observations, process-based modeling simulations, and deep learning architecture. We defined and formulated high-resolution polymodal datasets with multitemporal extents and hyperspatiospectral fidelity (i.e., 12.4 million parameters with 13.1 million in situ data points, 2.84 billion ground-controlled remotely sensed data points, and 36.58 million model-based simulation outputs to computationally reflect the state space of the earth system), simulated the non-linear feedback mechanisms attributed to permafrost degradation and carbon cycle perturbation across Alaska with a process-constrained deep learning architecture composed of cascading stacks of convolutionally layered memory-encoded recurrent neural networks (i.e., GeoCryoAI), and interpreted historical and future emulations of freeze-thaw dynamics and the permafrost carbon feedback with a suite of evaluation and performance metrics (e.g., cross-entropic loss, root-mean-square deviation, accuracy). This framework introduces ecological memory components and effectively learns subtle spatiotemporal covariate complexities in high-latitude ecosystems by emulating permafrost degradation and carbon flux dynamics across Alaska with high precision and minimal loss (RMSE: 1.007cm, 0.694nmolCH4m-2s-1, 0.213µmolCO2m-2s-1). This methodology and findings offer significant insight about the permafrost carbon feedback by informing scientists and the public on how climate change is accelerating, strategies to ameliorate the impact of permafrost degradation on the global carbon cycle, and to what extent these connections matter in space and time.

It is well-established that positive feedbacks between permafrost degradation and the release of soil carbon into the atmosphere impacts land-atmosphere interactions, disrupts the global carbon cycle, and accelerates climate change. The widespread distribution of thawing permafrost is causing a cascade of geophysical and biochemical disturbances with global impact. Currently, few earth system models account for permafrost carbon feedback mechanisms. This research identifies, interprets, and explains the feedback sensitivities attributed to permafrost degradation and terrestrial carbon cycling imbalance with in situ and flux tower measurements, remote sensing observations, process-based modeling simulations, and deep learning architecture. We defined and formulated high-resolution polymodal datasets with multitemporal extents and hyperspatiospectral fidelity (i.e., 12.4 million parameters with 13.1 million in situ data points, 2.84 billion ground-controlled remotely sensed data points, and 36.58 million model-based simulation outputs to computationally reflect the state space of the earth system), simulated the non-linear feedback mechanisms attributed to permafrost degradation and carbon cycle perturbation across Alaska with a process-constrained deep learning architecture composed of cascading stacks of convolutionally layered memory-encoded recurrent neural networks (i.e., GeoCryoAI), and interpreted historical and future emulations of freeze-thaw dynamics and the permafrost carbon feedback with a suite of evaluation and performance metrics (e.g., cross-entropic loss, root-mean-square deviation, accuracy). This framework introduces ecological memory components and effectively learns subtle spatiotemporal covariate complexities in high-latitude ecosystems by emulating permafrost degradation and carbon flux dynamics across Alaska with high precision and minimal loss (RMSE: 1.007cm, 0.694nmolCH4m-2s-1, 0.213µmolCO2m-2s-1). This methodology and findings offer significant insight about the permafrost carbon feedback by informing scientists and the public on how climate change is accelerating, strategies to ameliorate the impact of permafrost degradation on the global carbon cycle, and to what extent these connections matter in space and time.

In the Arctic, the spatial distribution of boreal forest cover and soil profile transition characterizing the taiga-tundra ecological transition zone (TTE) is experiencing an alarming transformation. The SIBBORK-TTE model provides a unique opportunity to predict the spatiotemporal distribution patterns of vegetation heterogeneity, forest structure change, arctic-boreal forest interactions, and ecosystem transitions with high resolution scaling across broad domains. Within the TTE, evolving climatological and biogeochemical dynamics facilitate moisture signaling and nutrient cycle disruption, i.e. permafrost thaw and nutrient decomposition, thereby catalyzing land cover change and ecosystem instability. To demonstrate these trends, in situ ground measurements for active layer depth were collected to cross-validate below-ground-enhanced modeled simulations from 1980-2017. Shifting trends in permafrost variability (i.e. active layer depth) and seasonality were derived from model results and compared statistically to the in situ data. The SIBBORK-TTE model was then run to project future below-ground conditions utilizing CMIP6 scenarios. Upon visualization and curve-integrated analysis of the simulated freeze-thaw dynamics, the calculated performance metric associated with annual active layer depth rate of change yielded 76.19%. Future climatic conditions indicate an increase in active layer depth and shifting seasonality across the TTE. With this novel approach, spatiotemporal variation of active layer depth provides an opportunity for identifying climate and topographic drivers and forecasting permafrost variability and earth system feedback mechanisms.

In Alaska, pervasive irregularities of permafrost coverage and associated boreal forest heterogeneity within the North American Taiga-Tundra Ecological Transition Zone (TTE) are becoming more apparent as the climate warms. These anomalies correspond to extensive shifts in active layer thickness (ALT), carbon cycle disruption, and ecosystem response patterns. The feedback complexities associated with these climate-induced disturbances are evaluated with the integration of remote sensing, modeling, field observations, data assimilation and harmonization techniques, and artificial intelligence technology. In this study, to improve our understanding of shifting belowground dynamics and how they associate with aboveground vegetation patterns, we used the SIBBORK-TTE model to derive permafrost degradation and ecosystem transiency at high-resolution in this study. The intercomparison of model version output was first examined; then, multiple verification and validation methodologies revealed distinct historical and future implications resulting from ALT variability within four regions of the Alaska TTE domain (North Slope, Yukon Delta, Seward Peninsula, Interior). To quantify historical thaw variability and identify seasonality patterns across these regions of interest, in situ ALT point measurements were collected from two campaigns (CALM, SMALT) to cross-validate ALT-derived SAR data (AirMOSS, UAVSAR) and below-ground SIBBORK-TTE simulations between 1990-2020. Future conditions were then projected with a warming climate function and CMIP6 data from CNRM-CERFACS SSP126/585 scenarios. Initial results for derived and measured annual maximum ALT yield a mean-error performance metric of 0.2294. Paradoxically, future climate conditions advance the ubiquity of permafrost thaw and seasonality widening across the TTE. With this investigative approach, spatiotemporal variability in ALT provides a unique signal to enhance model precision and lower uncertainty through fine-tuning driver forcing and modular parameterization, forecast permafrost distribution, and identify the climatic and topographic mechanisms of earth system feedbacks and land cover change.