Progress 09/01/23 to 08/31/24
Outputs
Target Audience:Climate scientists and researchers, data scientists and researchers, national drought information system, natural resource managers, and graduate students Changes/Problems:Research Objective 3: Change is requested for Stakeholder Engagement - As part of the funded project, we developed the Next Generation - Interactive Soil Moisture Forecasting System (NG-ISMFS). PI presented the system during this year's PD meeting in Manhattan, Kansas. Currently, PI is working with Auburn's IT department to resolve some of the performance issues with respect to the website (https://ng-ismfs.auburn.edu/). Once the website performance is fixed, we will engage with the stakeholders by presenting the website at regional and national conferences to get feedback. We plan to continue stakeholder engagement as a post-project activity. Additionally, we will seek additional funding resources including a follow-up proposal to the DSFAS program this year. What opportunities for training and professional development has the project provided?Two Ph. D students graduated - Dr. Yanan Duan (December 2023), and Dr. Sathish Akula (August 2024). One post-doctoral scientist (Dr. G.Manogaran) was supported for developing the NG-ISMFS How have the results been disseminated to communities of interest?Through peer-reviewed journal publications, and conference presenations. What do you plan to do during the next reporting period to accomplish the goals?Complete the testing and public release of the NG-ISMFS system.
Impacts
What was accomplished under these goals?
Summary of major accomplishments We successfully developed the Next Generation Interactive Soil Moisture Forecasting System (NG-ISMFS). A manuscript detailing the core components and performance results of the NG-ISMFS is currently in preparation (Manogaran et al., under prep.). Additionally,we performed a comprehensive analysis of sub-seasonal soil moisture predictability using a novel set of climate model-based sensitivity experiments. This study comprised of a set of eight sensitivity experiments designed to isolate and examine the individual contributions of land, ocean, and atmospheric components, as well as their interactions, to forecast skill. Our findings indicate that land surface initialization accounts for approximately 90% of the predictive skill in soil moisture forecasting (Richter et al., 2024; Duan et al., in review). Finally, we contributed to a comprehensive study examining the impact of intensifying drought patterns due to climate change, which are driving persistent shifts in ecosystem structures, reducing resilience, and potentially leading to irreversible ecological transformations across diverse ecosystems, including forests and grasslands (Moss et al., 2024). Research Objective 1 Major activity completed. Sub-seasonal soil moisture forecasting:A recent analysis of sub-seasonal forecast skill using the advanced CESM2-SubX forecasting system has been published (Richter et al. 2024). Additionally, Duan et al. conducted a detailed sub-seasonal soil moisture forecast analysis with CESM2-SubX forecasts and related sensitivity experiments, which is currently under review atnpj Climate and Atmospheric Science. Climate change, Land use change and drought:Singh, Kumar et al. (2024) conducted a study examining the feedback effects of land use change on regional climate under global warming scenarios using CMIP6-LUIMP climate models. Additionally, they isolated the influence of internal climate variability using CESM2 large ensemble climate data and a sensitivity experiment excluding land use change. In a complementary study, Moss et al. (2024) investigated the role of drought as a key driver of ecological transformation in the context of climate change. Data collected/generated CESM2 Large Ensemble Data and Seasonal to Multiyear Soil Moisture Forecast Data:We developed soil moisture indices for 17 NEON ecoregions using data from the CESM2 Large Ensemble (CESM2-LE). Additionally, we generated large-scale climate indices, including ENSO, PDO, and AMO, from CESM2-LE data to support a machine learning-based soil moisture forecasting study. This data was integral to Dr. Sathish Akula's Ph.D. dissertation, and two manuscripts based on this work are currently in preparation. Summary statistics and discussion of results Richter et al., 2024, and Duan et al. (in review):On a sub-seasonal timescale, land surface initialization accounts for approximately 90% of the forecast skill in predicting root zone soil moisture. In contrast, atmospheric initialization plays a more significant role in enhancing the accuracy of temperature forecasts. Singh, Kumar et al. (2024), and Moss et al. (2024): Global warming leads to widespread soil moisture depletion across multiple regions. In midlatitude areas, this results in a shift from historically wet conditions to a water-limited transitional regime. This transition reduces evapotranspiration, diminishing the cooling effects driven by land use. Moss et al. (2024) further demonstrate that drought conditions, often intensified by other stressors such as land use changes and biotic interactions, drive long-term shifts in ecosystem structure, function, and species composition. By combining ecological theory with empirical data, their study provides critical insights into the mechanisms of drought-induced transformations, including impacts on plant population dynamics, ecosystem recovery pathways, and potential management strategies to mitigate these effects. Key outcomes Four peer reviewed manuscripts- three published and one are in review. Two Ph. D students graduated -Dr. Yanan Duan (December 2023), and Dr. Sathish Akula (August 2024). Research Objective 2 Major activity completed. Next Generation Soil Moisture Forecastign System (NG-ISMFS):We successfully developed the Next Generation Interactive Soil Moisture Forecasting System (NG-ISMFS), an advanced tool that integrates high-resolution ERA-5 and CESM2 (H2OSOI) datasets with LSTM model predictions to deliver comprehensive and accurate soil moisture forecasts. The model architecture includes two LSTM layers optimized to learn from historical data, with dropout mechanisms applied to prevent overfitting. A dense layer using LeakyReLU activation, coupled with the Adam optimizer, further enhances model performance. A key innovation within NG-ISMFS is the incorporation of Software Defined Storage (SDS) and Network Attached Storage (NAS) technologies. SDS enables dynamic and flexible management of large datasets from diverse sources, ensuring scalability and efficient resource utilization. NAS provides centralized, accessible storage, facilitating data retrieval and collaborative access for multiple users. NG-ISMFS marks a significant advancement in environmental data analytics, providing a robust solution for soil moisture forecasting across the United States. This study underscores the transformative potential of deep learning and advanced storage solutions in addressing complex environmental forecasting challenges, demonstrating their applicability in practical, real-world contexts. A full-length manuscript was published in the2023 International Conference on Machine Learning and Applications (ICMLA). Additionally, a manuscript detailing the core components and performance results of NG-ISMFS is currently in preparation (Manogaran et al., in prep.) Summary statistics and discussion of results Manogaran et al. (2023):We combined the state-of-the-art climate model's (Community Earth System Model Version 2) forecast that incorporates the effects of the large-scale climatic drivers, including sea-surface temperature and atmosphere circulation features into soil water forecast with the LSTM-based Deep Learning model. Our Deep Learning model understands the local forecast biases using the weekly hindcast data from 1999 to 2016. We used this trained LSTM model to test its performance from 2017 to 2021 and enhanced the forecast proficiency and aid in analyzing future soil moisture anomalies, i.e., departure from climatology using data fusion and spatial downscaling. Key outcomes Next Generation Interactive Soil Moisture Forecasting System (NG-ISMFS) developed. A Ph.D. dissertation titled as "Enhancing Hydrological and Climate Predictions Through Artificial Intelligence" is completed by Dr. Sahtish Akula. A full-length manuscript published in IEEE conference proceedings. Research Objective 3 Major activity completed. Interactive web applications: The Next Generation Interactive Soil Moisture Forecasting System (NG-ISMFS) has been developed. A beta version is under testing here:https://ng-ismfs.auburn.edu/ Regional stakeholder engagement:Co-PD Rangwala actively engaged regional stakeholders through his leadership role at the North Central Climate Adaptation Center. His interactions provided valuable feedback that has informed the development of the NG-ISMFS system. New collaborations:Principal Director Kumar has been selected to serve on the external advisory board for NSF and NCAR's Earth System Predictability Across Timescales initiative. This initiative fosters new collaborations and opportunities to advance the predictability of Earth system processes across various timescales. Data collected. Data from Research Objective 1 are used here. Summary statistics and discussion of results Summary statistics and results are described in research objectives 1 and 2.
Publications
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2024
Citation:
Richter, J.H., Glanville, A.A., King, T., Kumar, S., Yeager, S.G., Davis, N.A., Duan, Y., Fowler, M.D., Jaye, A., Edwards, J. and Caron, J.M., 2024. Quantifying sources of subseasonal prediction skill in CESM2. npj Climate and Atmospheric Science, 7(1), p.59.
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2024
Citation:
Singh, A., Kumar, S., Chen, L., Maruf, M., Lawrence, P. and Lo, M.H., 2024. Land Use Feedback under Global WarmingA Transition from Radiative to Hydrological Feedback Regime. Journal of Climate, 37(14), pp. 38473866.
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2024
Citation:
Moss, W.E., Crausbay, S.D., Rangwala, I., Wason, J.W., Trauernicht, C., Stevens-Rumann, C.S., Sala, A., Rottler, C.M., Pederson, G.T., Miller, B.W. and Magness, D.R., 2024. Drought as an emergent driver of ecological transformation in the twenty-first century. BioScience, 74(8), pp.524-538.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Manogaran, G.S., Lee, W., Kumar, S., Duan, Y. and Rangwala, I., 2023, December. A Framework for Developing the Next Generation Interactive Soil Moisture Forecasting System Using the Long-Short Term Memory Model. In 2023 International Conference on Machine Learning and Applications (ICMLA) (pp. 1986-1993). IEEE.
|
Progress 09/01/22 to 08/31/23
Outputs
Target Audience:climate scientist, AI/ML applications in earth and environmental science, agriculture and water managers Changes/Problems:We are all set fo complete the project in August 2024. Thank you for your support. What opportunities for training and professional development has the project provided?Ph.D. training for two graduate students at Auburn University: PD Kumar supervises two Ph.D. students in the Earth System Science program at Auburn University. Ph.D. Candidate Ms. Yanan Duan is working on improving understanding of hydroclimate variability and predictability in the Southeastern United States. Ms. Duan is expected to graduate in Fall 23. Another Ph.D. student Mr. Montasir Maruf is working on high-resolution climate modeling. Co-PD Rangwala serves as the Ph.D. committee member for both students. Post-doctoral Research Associate at CSUN: Co-PD Dr. Lee supervised a new post-doctoral research associate, Dr. Gunasekaran Manogaran, to implement objective 2. Professional Development PD Kumar is serving on the national-level US CLIVAR Predictability, Predictions, and Application Interface (PPAI) panel for the 2022-2025 term. Kumar attended the US CLIVAR Summit meeting in Summer 23. Further, Kumar contributes to the monthly Panel webinar discussion. PD Kumar served as the primary convener for AGU Fall meeting session GC054- High-resolution Regional Earth System Modeling - Hydroclimate Variability, Extremes, and Policy Implications, in Fall 2023 Graduate students Yanan Duan, and Montasir Maruf are presenting posters during AGU Fall meeting 2023. PD Kumar presented a poster in CESM Annual Meeting, Summer 23. Post-doctoral scientist Dr. Manogaran presented his research (talk) in CESM Annual Meeting, Summer 23. Rangwala is a co-author on the Water chapter in the National Climate Assessment Phase-5 and provided analysis of hydrological projections based on CMIP6. How have the results been disseminated to communities of interest? Peer-reviewed journal publications (see accomplishment described in answering question 1) National and regional conferences and meetings (see Profession Development in question 2) Discussion with the stakeholders in the region (see profession development in question 2, and accomplishment in research objective 3) What do you plan to do during the next reporting period to accomplish the goals? Testing, validation, and publication of the NG-ISMFS system: Manogaran et al. (2023) has developed basic framework for the NG-ISMFS system. Our goal is to complete the development and testing of the NG-ISMF system. At the end of the project, we aim to release a publicly accessible webtool for the NG-ISMF system. We are also developing a Bulletin of American Meteorological Society manuscript to describe the NG-ISMFS system. Stakeholder Engagement (Task 3): Co-PD Rangwala will take a lead in organizing a stakeholder interaction workshop that will coincide with the release of the NG-ISMF System.
Impacts
What was accomplished under these goals?
Summary of major accomplishments We improved our process-based understanding of soil moisture variability and predictability in a changing climate and seasonal climate forecast system (Kumar et al., 2023 Earth's Future). Additionally, we developed hybrid Physics-AI model that combines the strengths of process-based biophysical models with new-age Artificial Intelligence / Machine Learning (AI/ML) techniques to improve streamflow (Duan, Akula, et al., 2023, Artificial Intelligence for the Earth Systems) and soil moisture forecast on sub-seasonal timescales (Manogaran et al., 2023; accepted in ICMLA 2023). Research Objective 1 Major activity completed. Soil moisture variability and predictability under climate change:Kumar et al. (2023) presented a new assessment of soil moisture variability and predictability under changing climate. Analysis of sub-seasonal forecast system:A new analysis of sub-seasonal forecast skill using state-of-the-art CESM2-SubX forecasting system has been completed (Richter et al., npj Climate and Atmospheric Science, in revision. Additionally, Duan et al. (in prep.) has completed sub-seasonal soil moisture forecast analysis using CESM2-SubX forecast and related sensitivity experiments. A high-resolution point scale model developed:PI contributed to the development and testing of point scale CLM model at National Ecological Observatory Network (NEON) sites. The new research is published in Geoscientific Model Development (Lombardozzi et al. 2023) Data collected. EAR5 Land Soil Moisture data -This is a long-term reanalysis-based soil moisture data from European Center for Medium-Range Weather Forecast (ECMWF). We downloaded hourly data, aggregated to daily time scale, and regridded to CESM2 grid resolution. We use ERA5-Land soil moisture data as observation in this study. CESM2 SubX soil moisture forecast- We downloaded and pre-processed CESM2-SubX root zone soil moisture forecast for DL model training and validation. Summary statistics and discussion of results Kumar et al., 2023, Earth's Future:Global warming is projected to increase ENSO and its teleconnected precipitation variability over North America. However, the corresponding change in soil moisture variability is relatively small or even decreases because of a concurrent projected reduction in land surface memory. Richter et al., npj Climate and Atmospheric Science (in revision), and Duan et al. (in prep.): Atmospheric initial condition contributes most to the temperature and precipitation predictability (Richter et al., in revision) and land initial condition contributed most of the soil moisture predictability at sub-seasonal time scales (Duan et al., in prep.). Key outcomes Two peer-reviewed journal articles published. Research Objective 2 Major activity completed. A container-based software-defined storage (CB-SDS) developed:We embraced the Network Assisted Storage (NAS) system, a sophisticated solution known for its software-control unit's ability to automate and pool storage. This method dramatically simplifies the process of provisioning and management. Our chosen device for this task was the Asustor Lockerstor 8-bay NAS Device (AS6508T). This powerful device houses eight local disks, each boasting an impressive storage capacity of up to 144 TB. To further illuminate its specifications, the AS6508T is powered by an Intel Denverton-based Atom C3538 Quad-Core CPU and features 8GB of DDR4-2133 SO-DIMM memory, which offers a 30% speed advantage over DDR3. Additionally, it is equipped with dual Intel 10-Gigabit Ethernet ports and Realtek 2.5-Gigabit Ethernet ports and supports both Wake on LAN and Wake on WAN functionalities. Our reliance on the Asustor Lockerstor is paramount. The device's primary role is to store two vital data sets: Observation Data ERA-5 and Forecast Data CAM6 H2OSOI obtained from the National Centre for Atmospheric Research (NCAR). When it comes to model training, the stored data is easily accessible from the NAS device, allowing for a seamless flow of information. As a result of this setup, we have been able to generate accurate prediction graphs for a specified number of days. Leveraging the capabilities of this NAS system ensures that our substantial datasets remain not just organized but also easily retrievable, laying a strong foundation for precise and timely model projections. Summary statistics and discussion of results Duan, Akula et al. (2023):A densely connected neural network model consisting of 6 layers (Deep Learning, DL) is developed using biophysical characteristics, and National Water Model (NWM) forecast as inputs, and the forecast errors as outputs. A temporal and spatial split of the gauged data shows that the probability of capturing the observations in the forecast range improved significantly in the hybrid NWM-DL model (82±3 %) than in the NWM-only forecast (21±1 %). Manogaran et al. (2023):We combined the state-of-the-art climate model's (Community Earth System Model Version 2) forecast that incorporates the effects of the large-scale climatic drivers, including sea-surface temperature and atmosphere circulation features into soil water forecast with the LSTM-based Deep Learning model. Our Deep Learning model understands the local forecast biases using the weekly hindcast data from 1999 to 2016. We used this trained LSTM model to test its performance from 2017 to 2021 and enhanced the forecast proficiency and aid in analyzing future soil moisture anomalies using data fusion. Key outcomes Three peer-reviewed papers published or accepted (Kadiyala et al., 2022; Duna, Akula et al., 2023; Manogaran et al. 2023) Research Objective 3 Major activity completed. Interactive web applications: The Next Generation Interactive Soil Moisture Forecasting System (NG-ISMFS) developed (Manogaran et al. 2023). The system is under testing and further fine-tuning before we release its publicly available url. Regional stakeholder meeting:PD Kumar participated in Alabama Water Resources Conference and interacted with the stakeholders and presenters during the meeting. New collaborations:The research team engaging with researchers John Bradford and Daniel Schlaepfer at the USGS Southwest Biological Science Center to explore potential applications of the CLM model for developing downscaled soil moisture projections for the 21stcentury based on CMIP projections. Data collected. Data from Research Objective 1 are used here. Summary statistics and discussion of results Interactive web application- We have developed a state-of-the-art web application, the Next Generation Interactive Soil Moisture Forecasting System (NG-ISMFS) using Long-Short-Term Memory. NG-ISMFS stands out as a revolutionary web-based platform, granting users an in-depth perspective on Soil Moisture (SM) predictions, specially curated for different geographical regions throughout the United States. By harnessing SM data directly from the National Centre for Atmospheric Research (NCAR) and integrating it with advanced deep learning models, we ensure that the forecasting provided by the platform is of unparalleled precision. The technical foundation of NG-ISMFS is robust; it integrates the resilience of Angular's frontend framework with the flexibility of Django's backend framework, culminating in an unparalleled user experience. While the primary audience of this innovative tool comprises individuals and entities seeking precise SM forecasts within the United States, our vision extends to broadening this reach in future iterations. The NG-ISMFS stands as a testament to the power of blending data acquisition, sophisticated machine learning techniques, and user-centric design.
Publications
- Type:
Journal Articles
Status:
Published
Year Published:
2023
Citation:
Duan, Y., Akula, S., Kumar, S., Lee, W., & Khajehei, S. (2023). A Hybrid PhysicsAI Model to Improve Hydrological Forecasts. Artificial Intelligence for the Earth Systems, 2(1), e220023.
- Type:
Journal Articles
Status:
Published
Year Published:
2023
Citation:
Kumar, S., Dewes, C. F., Newman, M., & Duan, Y. (2023). Robust changes in North America's hydroclimate variability and predictability. Earth's Future, 11(4), e2022EF003239.
- Type:
Journal Articles
Status:
Accepted
Year Published:
2023
Citation:
Lombardozzi, D. L., Wieder, W. R., Sobhani, N., Bonan, G. B., Durden, D., Lenz, D., SanClements, M., Weintraub-Leff, S., Ayres, E., Florian, C. R., Dahlin, K., Kumar, S., Swann, A. L. S., Zarakas, C., Vardeman, C., and Pascucci, V. (2023). Overcoming barriers to enable convergence research by integrating ecological and climate sciences: The NCAR-NEON system Version 1. EGUsphere, 2023, 1-37.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Guna Shekar M., Wonjun Lee, Sanjiv Kumar, Imtiaz Rangwala, Yanan Duan, 2023: A Framework for Developing the Next Generation Interactive Soil Moisture Forecasting System Using the Long-Short Term Memory Model, accepted for publications in 22nd International Conference on Machine Learning and Applications (ICMLA 2023) proceedings, and presentation. December 15-17, 2023, Jacksonville, Florida.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
S. P. Kadiyala, X. Li, W. Lee, A. Catlin, Securing Microservices Against Password Guess Attacks Using Hardware Performance Counters , Proceedings of the 35th IEEE International System-On-Chip Conference (SOCC-22), Belfast, Northern Ireland, Sep. 2022
|
Progress 09/01/21 to 08/31/22
Outputs
Target Audience:climate scientist, AI/ML applications in earth and environmental science, agriculture and water managers Changes/Problems:We experienced some delayin implementingResearch Objective 2 during the project's first two years, partly relatedto COVID. To make speedy progress during the remaining time - Co-PD Lee has hired a new post-doctoral research associate at CSUN. The post-doctoral associate started the new position in September 2022. So, we hope to make up for some delays in our progress. Additionally, we plan to apply for a one-year no-cost extension for the project. What opportunities for training and professional development has the project provided?Training Opportunity Ph.D. training for two graduate students at Auburn University: PD Kumar supervises two Ph.D. students in the Earth System Science program at Auburn University. Ph.D. Candidate Ms. Yanan Duan is working on improving understanding of hydroclimate variability and predictability in the Southeastern United States. Ms. Duan is expected to graduate in Fall 22. Another Ph.D. student Mr. Montasir Maruf is working on high-resolution climate modeling. Co-PD Rangwala serves as the Ph.D. committee member for both students. Data science training for Deep Learning model development: PD Kumar and Co-PD Lee jointly advise a data scientist and a Ph.D. candidate Mr. Sathish Akula at Auburn University. Mr. Akula is developing a deep learning model for forecast improvement using National Water Model data. Post-doctoral Research Associate at CSUN: Co-PD Dr. Lee has hired a new post-doctoral research associate, Dr. Gunasekaran Manogaran, to implement objective 2. Dr.Manogaran has started his work in Sept. 2022. Professional Development PD Kumar has been selected to serve on the national-level US CLIVAR Predictability, Predictions, and Application Interface (PPAI) panel for the 2022-2025 term. Kumar attended the US CLIVAR Summit meeting in Spring 22. Further, Kumar contributes to the monthly Panel webinar discussion. PD Kumar attended Southwest Drought Early Warning System (SE DEWS) Partners Dialogue on August 9-10, organized by the National Integrated Drought Information System (NIDIS) in Atlanta, GA. There were 70+ participants, providing an opportunity to learn about stakeholder needs and ongoing regional drought and water-related research. Two noteworthy observations related to the current project are: (a) prediction of flash drought is challenging in the drought monitor, and (b) there was interest in developing a high-resolution soil moisture dataset. PD Kumar served as the primary convener for AGU Fall meeting session GC23: Drivers and Mechanisms of Terrestrial Water Cycle Change at Global and Regional scales. Graduate student Yanan Duan presented at the AGU Fall meeting, 2021, and CESM Annual Workshop, 2022. Graduate student Montasir Maruf presented at the CESM Annual Workshop meeting, 2022. How have the results been disseminated to communities of interest? Peer-reviewed journal publications (see accomplishment described in answering question 1) National and regional conferences and meetings (see Profession Development in question 2) Discussion with the stakeholders in the region (see profession development in question 2, and accomplishment in research objective 3) What do you plan to do during the next reporting period to accomplish the goals?1. Downscaling of sub-seasonal climate forecast and running high-resolution CLM run Research Task 1.2: We will downscale the CESM2-SubX data using the statistical downscaling method and run the high-resolution CLM for the hindcast period from 2000 to 2021. We will investigate the improvements due to downscaling by comparing the high-resolution forecast run with the default SubX run. 2, Next, we will implement the AI/ML model to improve soil moisture forecast at the local scale (Research Task 2.2). Research Task 2.2: Our specific goal is to test the feasibility of real-time soil moisture at the local scale. We will use the National Ecological Observatory Networks (NEON) Talladega National Forest site as the testbed. PD Kumar has developed an active collaboration with regional neon manager Mr. David Mitchell, who agreed to provide real-time access to their soil moisture data. We will build an AI/ML model that uses low-resolution climate/soil moisture forecast, biophysical characteristics, and observed soil moisture data as input to improve the forecast skill. Using standard data assimilation techniques, we plan to assimilate observations in the initial condition file, then run the forecast using the updated initial condition file and downscale sub-seasonal climate forecast data. We will also investigate if the Deep Learning model can bring improvement in the forecast. We have already developed a point-scale CLM simulation for NEON sites. Point scale simulations are very computationally efficient. Hence we can thoroughly test different ideas and AI/ML applications to improve soil moisture forecast skills at the regional scale. 3. Stakeholder Engagement (Task 3) We will work with NIDIS SE DEWS and Auburn University Water Resources Center to engage stakeholders in our work. PD will attend regional meetings and develop collaborations with regional partners. Co-PD Rangwala is actively working with the stakeholders in the Northern Great Plains, especially with the Grasscast system. We will seek their input as we make progress in our work.
Impacts
What was accomplished under these goals?
Summary of major accomplishments We improved our understanding of soil moisture variability and predictability in a changing climate and seasonal climate forecast system (Singh et al., Nature Communications, in revision; Kumar et al., AGU Advances, submitted; Duan et al., in prep.) Additionally, we developed a new AI technology (Kadiyala et al., IEEE proceedings, 2022; Lee et al., in prep.) and a hybrid Physics-AI model that combines the strengths of process-based biophysical models with new-age Artificial Intelligence / Machine Learning (AI/ML) techniques to improve forecast reliability (Duan, Akula, et al., Artificial Intelligence for the Earth Systems, in revision). The two parallel developments - (1) theoretical understanding and (2) AI/ML technique will contribute to designing the Next Generation - Interactive Soil Moisture Forecasting System. Research Objective 1 Major activity completed Analysis of sub-seasonal soil moisture forecast: Duan et al. (in prep.) has completed a new sub-seasonal soil moisture forecast analysis using CESM2-SubX forecast and related sensitivity experiments. Soil moisture variability and predictability under climate change: Kumar et al. (2022, AGU Advances, submitted) have completed the analysis of two large ensemble climate data (CESM1-LE and GFDL-CM3-LE) to assess soil moisture variability and predictability under changing climate. High-resolution soil moisture data: Maruf et al. (in prep.) completed developing and validating a high-resolution (12.5km) CLM configuration for the United States. Data collected Sub-seasonal soil moisture forecasts - CESM2 SubX and its sensitivity to land, ocean, and atmosphere initializations. Climate change data - Two large ensemble data (CESM1-LE and GFDL-CM3-LE) to understand soil moisture variability and predictability under global warming scenarios. Observations/reanalysis of soil moisture data: High-resolution daily soil moisture data from four different sources: (1) ERA5-Land, (2) Smerge-Noah-CCI, (3) MERRA2, and (4) GLEAM3. Water cycle data: Evapotranspiration from ERA5, MERRA2, LERI, and NEON; Sensible heat flux data from ERA5, MERRA2; soil moisture data from ERA5-Land; Runoff data from Livneh et al. 2015. Summary statistics and discussion of results Kumar et al., 2022, AGU Advances, submitted: Global warming is projected to increase ENSO and its teleconnected precipitation variability over North America. However, the corresponding change in soil moisture variability is relatively small or even decreases because of a concurrent projected reduction in land surface memory. Duan et al. in prep: The soil moisture forecast skills in the control experiment (CESM2-SubX) that incorporated all three components' initializations: Ocean, Land, and Atmosphere, are compared with three sensitivity experiments: ocean and land initializations, land only initializations, ocean and atmospheric initializations. The land-only initialization contributes 80.3 ± 6.8 of the total skill in the control experiment. The soil moisture forecast skill is most significant in the Great Plain and central southern US. Maruf et al., in prep: The CLM high-resolution shows an improvement in simulating interannual variability of evapotranspiration, sensible heat fluxes, runoff, and soil moisture compared to the low-resolution CLM configuration. The model error has decreased by more than 20%. In addition, the soil moisture memory has increased in the high-resolution experiment. Key outcomes Four peer-reviewed journal articles (Two submitted or in revision, and two are in prep.) Research Objective 2 Major activity completed A container-based software-defined storage (CB-SDS) developed: To support highly scalable and mobile container-based software, a microservice benchmark application (e.g., DeathStarBench) is implemented using docker swarm and kubernetes orchestrators. In this system, we were able to run container-based caching and storage microservices such as Memcached, Redis, and MongoDB. We also implemented a cloud cluster using docker Swarm and Google kubernetes to manage container servers. In our experiment environment, the container-based microservice system run in the Swarm cluster comprises 31 microservices. A hybrid Physics-AI model is developed to improve the hydrological forecast: A densely connected neural network model consisting of 6 layers (Deep Learning, DL) is developed using biophysical characteristics, and National Water Model (NWM) forecast as inputs, and the forecast errors as outputs. The DL model successfully learns location invariant transferrable knowledge from the domain trained in the gauged locations and applies the learned model to estimate forecast errors at the ungauged locations. Data collected National Water Model streamflow and soil moisture forecast from December 2018 to the present for Alabama and Georgia region. Remotely sensed SMAP soil moisture observations - Soil moisture data is Soil Moisture Active Passive (SMAP) L2 half-orbit enhanced 5-cm soil moisture with 9-km resolution (access provided by the Descartes Lab) USGS streamflow observations for 389 stations - downloaded from the USGS website. Summary statistics and discussion of results Lee et al. (in prep): We implemented a complete set of microservice benchmark programs and showed a microservice-specific Denial of Service attack that exploits the complexity of internal traffic between microservices. Kadiyala et al. (IEEE proceedings, 2022): Scaling applications to make them user-friendly, adaptable to various geographical locations, and serving a large customer base is of utmost necessity. We implemented a complete microservices system and detected cyber attacks on microservices using performance counter data. Duna, Akula, et al. (Artificial Intelligence for the Earth Systems, 2022, submitted): A temporal and spatial split of the gauged data shows that the probability of capturing the observations in the forecast range improved significantly in the hybrid NWM-DL model (82±3 %) than in the NWM-only forecast (21±1 %). Key outcomes Two peer-reviewed papers were published or submitted (Kadiyala et al., 2022; Duan, Akula, et al., AIES, in revision), and one was in preparation (Lee et al.) Research Objective 3 Major activity completed Interactive web applications are developed for projecting climate and drought metrics (Thota and Rangwala, 2022) Regional stakeholder meeting: PD Kumar participated in the Southwest Drought Early Warning System (SE DEWS) Partners Dialogue meeting organized by National Integrated Drought Information System (NIDIS) in Atlanta, GA. Stakeholder inputs through an online survey: PD Kumar collaborated with Auburn University Water Resources Center (AU WRC) and conducted an online survey (Qualtrics) about the stakeholder inputs for water availability projections in Alabama. Data collected High-resolution meteorological observations - gridMET (4-km resolution) Downscaled climate projections - MACAv2-METDATA (gridded, 4 km resolution) from 20 global climate models for 2 emission scenarios (RCP 4.5 and 8.5) Stakeholder survey data and regional meeting discussions - AU WRC Survey and SE DEWS meeting Summary statistics and discussion of results Interactive web application - (1) A user can visualize and download time-series data associated with observations and future projections for any point location within the Contiguous United States (CONUS), (2) A user can select across 20 GCMs and 2 RCPs - a total of 40 different climate scenarios. Survey result - emphasis on uncertainty reduction from sub-seasonal to inter-annual climate forecast/predictions Meeting discussion - Interest in developing high-resolution soil moisture data Key outcomes Interactive web applications are published at https://droughtindexportal.colorado.edu/
Publications
- Type:
Journal Articles
Status:
Submitted
Year Published:
2022
Citation:
Kumar, S., C. F. Dewes, M. Newman, and Y. Duan, Robust Changes in North Americas Hydroclimate Variability and Predictability, AGU Advances, Submitted
- Type:
Journal Articles
Status:
Under Review
Year Published:
2022
Citation:
Duan, Y., S. Akula, S. Kumar, W. Lee, and S. Khajehei, A Hybrid Physics-AI model to improve hydrological forecast, Artificial Intelligence for Earth Systems
- Type:
Journal Articles
Status:
Submitted
Year Published:
2022
Citation:
Singh, A., S. Kumar, M. Maruf, L. Chen, P. Lawrence, and M.-H. Lo, Soil moisture modulates land-use change impact under global warming, Nature Communications
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2021
Citation:
Duan, Y., & Kumar, S. (2021, December). Assessment of internal climate variability in CESM2-LE and its comparison with multi-model CMIP6 large ensemble. In AGU Fall Meeting 2021. AGU.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2021
Citation:
Kumar, S., Dewes, C., & Newman, M. Changing drought and pluvial risks in North America linked to mean-state changes in ENSO and soil moisture. In AGU Fall Meeting 2021. AGU.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2021
Citation:
Singh, A., Kumar, S., & Chen, L. (2021, December). Increased uncertainty in land-use change impacts on near-surface air temperature due to global warming in CMIP6-LUMIP experiments. In AGU Fall Meeting 2021. AGU.
- Type:
Journal Articles
Status:
Submitted
Year Published:
2022
Citation:
Moss, W., S. Crausbay, I. Rangwala and others (2022), Transformational Ecological Drought: an emergent driver of ecosystem change in the 21st century, Global Change Biology, Submitted,
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
S. P. Kadiyala, X. Li, W. Lee, A. Catlin, Securing Microservices Against Password Guess Attacks Using Hardware Performance Counters, Proceedings of the 35th IEEE International System-On-Chip Conference (SOCC-22), Belfast, Northern Ireland, Sep. 2022, NIFA Award acknowledged.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Thota P. and I. Rangwala (2022, September), Interactive Web Applications for Projecting Climate and Drought Metrics into the 21st Century. MtnClim 2022 Conference, Gothic, Colorado.
|
Progress 09/01/20 to 08/31/21
Outputs
Target Audience:1) Graduate students: Earth System Science (Hydrology and Water Resources), and Computer Science 2) Scientist: Land-atmosphere interaction and Climate Scientist, Computer Science Scientist 3) Stakeholder community - Drought Monitoring and Forecasting Changes/Problems: Co-PD Lee has changed his institution. This may delay the delivery of the research goal 2. We will develop a hybrid approach for climate model application. Climate model simulations will be performed at the NCAR supercomputer, and real-time soil moisture model calibration and forecasting will be performed at the big-data analytics platform. This change is needed because of the large input data requirements for the climate model simulations. Our earlier attempt to run the climate model on the local machine has failed. What opportunities for training and professional development has the project provided?Training Opportunity Ph.D. training for two graduate students at Auburn University: PD Kumar supervises two Ph.D. students in the Earth System Science program at Auburn University. Ph.D. Student Yanan Duan is working on downscaling of soil moisture forecast. Another Ph.D. student Montasir Maruf is working on high-resolution climate modeling for an improved soil moisture forecast. Data science training for Deep Leaning model development: PD Kumar and Co-PD Lee jointly advise a data scientist Mr. Sathish Akula at Auburn University. Mr. Akula is developing a deep learning model for forecast improvement using National Water Model data. Undergraduate training: Co-PD Lee trained one undergraduate student, Ethan, from the School of Computer Science at Yeshiva University during the summer. He learned about various concepts and tools required to implement deep learning platforms such as Docker container, microservice, orchestration, pipeline, Amazon web service, Google Cloud platform. High school student training: Janet Choi, a high school student at Richard Montgomery High School in Maryland, joined as a summer intern to co-PD Lee's lab and learned about deep learning, especially recurrent neural network and containerized microservices that will be implemented in our project. Participating in the project helped her decide her major as Computer Science at the University of Maryland, College Park, this year. Summer Internship opportunity for climate App development: Co-PD Rangwala provided summer graduate student internship training to Mr. Prasad Thoda in developing the climate app. ?Professional Development PD Kumar participated in the national level meeting and conferences: American Geophysical Union (AGU) Fall meeting 2020, American Meteorological Society (AMS) Annual meeting 2021, and CESM Annual Workshop 2021. These meetings provided the opportunity to share results with the scientific community and learn about the latest science in soil moisture forecasting. PD Kumar served as the lead convener for the AGU 2021 Fall meeting Global Environmental Change session titles: "Drivers and Mechanism of Terrestrial Water Cycle changes at global and regional scales." Co-PD Lee attended the Knowledge-guided machine learning (KGML) workshop on the zoom on August 9-11, 2021. How have the results been disseminated to communities of interest? Peer-reviewed publication: We published three pre-reviewed publications. Results from the study were presented at AGU, AMS, and CESM Annual meetings. Our publication on soil moisture forecasting (Esit et al., 2021) received local and regional media attention, e.g., YellowHammer News, Auburn University news, and University of Colorado news What do you plan to do during the next reporting period to accomplish the goals?Plan for Goal 1 Develop the soil moisture downscaling methodology for soil moisture forecasting: We will test the methodology in the following sequence: Level 1A forecast -> CESM2-LE subX (sub-seasonal forecast) and SMYLE (seasonal to multi-year large ensemble) (spatial resolution: 1 degree) Level 1B forecast -> Downscale SubX and SMYLE climate data (statistical downscaling) -> run CLM high resolution (0.125 degree) Level 1C forecast -> Analog method for climate model downscaling-> run CLM high resolution (H2SOI) (0.125 degree) Level 2 A -> apply machine learning to 1A for downscaling to SCAN sites Level 2 B -> apply machine learning to 1B for downscaling to SCAN sites Level 2C -> apply machine learning to 1C for downscaling to SCAN sites Develop the high-resolution assessment of the CLM soil moisture data: We will assess the CLM high-resolution simulations (12.5 degree) using network soil moisture observations, NEON, and LERI ET data. The above two will contribute towards the completion of Research Activity 1.2 and 1.3. We will also test different merging methodologies for soil moisture data (Research Activity 1.1). Plan for Goal 2 While for the first year, we tried various approaches to see proof of concepts, during the next reporting period, we will concentrate to actually implement proven concepts and tools to show the execution of the proposal. For the SDS, we will use containerized CESM model to run input data separated in different clusters, including local drive, network-attached storage, and cloud-based storage doing the same for output. Next, the containerized microservices orchestrated by Kubernates will be integrated with Deep Learning models that train soil moisture data from different sources, including user input, climate model forecast data, and remotely sensed soil moisture observations. This year, we will implement an interactive user interface to make the Deep Learning platform more intelligent. Starting from applying recurrent neural network (RNN), we will try a convolution neural network (CNN) to see whether it enhances the accuracy. As the last step, we will try combining two different algorithms (RNN + CNN). Plan for Goal 3 We plan to organize our 2nd stakeholder interaction meeting in Auburn, AL, during summer 2022. We also plan to incorporate high-resolution CLM soil moisture data Grassland Productivity and Climate App.
Impacts
What was accomplished under these goals?
Advances in long lead-time soil moisture prediction will greatly augment the ongoing drought early warning efforts for agriculture and natural ecosystems. Our new study (Esit et al., 2021) examined longer-term (months to years) climate processes from the decadal prediction experiments of Community Earth System Modeling (CESM). This study provided new insights into the interplay between atmospheric (precipitation) and land-surface processes on those timescales that can boost the predictability (i.e., ability to forecast successfully) of soil moisture on both seasonal and multi-year timescales.A critical finding from the study is that the predictability for soil moisture on seasonal periods is three times greater than that for precipitation. Regionally, the study finds that this potential soil moisture predictability is higher for the central and western United States, mainly on longer lead times that could span multiple years. Objective 1: Soil moisture forecasting model Research Activity 1.1: Merging of multi-scale, multi-sensor soil moisture data Major activity completed:We have completed an analysis of the potential predictability of soil moisture in the Community Earth System Model-Decadal Prediction Large Ensemble (CESM-DPLE) experiment. Data collected:We have collected long-term soil moisture data from (a) Four in-situ networks: SCAN, USCRN, ARM, and ICN; (b) Two remote sensing-based data: Smerge-Noah-CCI root zone soil moisture, and Global Land Evaporation Amsterdam Model; and (c) Two climate model forecasts: CESM2-SubX, and CESM2-SMYLE. We have also collected an ancillary data product: the Landscape Evaporative Response Index (LERI) data. Merging of these products is underway. Summary statistics and discussion of results:Soil moisture predictability on seasonal to decadal (S2D) continuum timescales over North America is examined from the Community Earth System Modeling (CESM) experiments. We find that soil moisture has significant predictability on S2D timescales despite limited predictability in precipitation. On sub-seasonal to seasonal timescales, precipitation variability is an order of magnitude greater than soil moisture, suggesting land surface processes, including soil moisture memory, reemergence, land-atmosphere interactions, transform a less predictable precipitation signal into a more predictable soil moisture signal. Key outcomes or other accomplishments realized:Our soil moisture predictability analysis (Esit et al., 2021) is published in a high profile journal: npj Climate and Atmospheric Sciences, with an impact factor of 7.99. Research Activity 1.2:Downscaling of seasonal climate forecast None to report in year 1 Research Activity 1.3: Developing Community Land Model configurations at 25-km resolution and with the downscaled seasonal climate forecast forcing data Major activity completed:We have developed a high-resolution (12.5km) community land model (CLM) configuration for the conterminous U.S. We have completed soil moisture simulations from 1981 to 2018. Data collected:We have collected three data: (a) network soil moisture observations, (b) National Ecological Observation Network (NEON), and (c) LERI actual ET data to evaluate high-resolution CLM performance. Summary statistics and discussion of results:none Key outcomes or other accomplishments realized:none Research Activity 1.4: Developing and testing 25-km resolution soil moisture forecast None to report in year 1 Research Objective 2: Big-data infrastructure Research Activity 2.1: Developing container-based software-defined storage (CB-SDS) technology:The deep learning analytics platform for real-time and interactive soil moisture forecast applications requires a highly scalable big data infrastructure. The software-defined storage (SDS) supported by docker containers has a highly scalable feature to satisfy our objective 2. During the first year, we tried many different open source-based SDS to see if they fit our desired criteria by checking resource scalability, software extendibility, no vendor lock-in, and flexibility of source code. We have eventually chosen OpenSDS under Linux Foundation to implement our proposed platform. We were able to install the required software and run a containerized OpenSDS in which we can create/delete/list volumes in both local and cloud storage using a command line and dashboard. Research Activity 2.2: Developing Big-Data Analytics Platform: We have implemented an example of Deep Learning workflows in the Google Cloud Platform and Amazon Web Service using Google Kubernates as a part of Pipelines and container orchestration in the Deep Learning Analytics platform. The pipelining, one of the important functions in the Deep Learning platform layer, is to deploy and manage end-to-end deep learning workflows, providing rapid and reliable deep learning experimentation. Starting from a microservice-based airplane booking and airmileage application composed of API Gateway, AWS Lambda, Trigger Lambda, DynamoDB, and SNS, we implemented another knowledge-based microservice application where Deep Learning models are trained in one microservice, and the trained knowledge is deployed to other microservices. This containerized application scales out by replicating some microservices based on the need. Using Kubernates, we were able to scale out 77 microservices where each microservice runs in one container. Finally, we tried to run the CESM2.2 release with Jupyter integrated container in the local machine, but since it requires downloading huge input data, it was unsuccessful. Research Objective 3: Stakeholder Interactions Major activity completed:(1) we organized one stakeholder inputs meeting (using zoom) on 11/16/2020 and collected their feedback on soil moisture forecast data products. A total of 11 participants, including NOAA, NIDIS, USDA, and Auburn representatives, attended the meeting. (2) Grassland Productivity and Climate App has been developed. This app quantifies relationships, based on linear regression, between 'observed' grasslands productivity and different climate variables that includes precipitation, potential evapotranspiration (PET), PET minus precipitation) for a point location in the US Great Plains. Data collected:(a) We collected oral feedback from the participants and documented their feedback in a google doc that was circulated to all participants, (b) We collected the following data for the grassland productivity app: (1) Growing-season ANPP (1982-2019) from Grass-Cast; (2) gridded observations from gridMET; and (3) Future Projections from MACAv2-METDATA Summary statistics and discussion of results: New potential partners for soil moisture data were identified:(a) USDA LTAR (Long-Term Agroecosystem Research) network has 18 sites nationwide, many of which have soil moisture monitoring in place. A sharp group of researchers is always interested in collaboration with University partners. (b) Other potential partners can be USDA NRCS SCAN sites and Tribal SCAN sites. A range of potentially relevant applications was also discussed- (a) Whether to plant marginal croplands this year (e.g., to dryland corn or an annual forage crop), or leave them fallow; (b) -Whether/when to conduct a prescribed burn of rangelands, or what the risk of grassland fire might be. Technical issues: root zone definitions (0-10cm, 10-40 cm, and 40cm to 1m), the sub-seasonal forecast can have many agricultural applications: grassland fire risk, winter-wheat planting decisions, spring/summer rangeland production. For the multi-year forecast, the potential application could be reservoir inflows; snowpack runoff Key outcomes or other accomplishments realized:(1) Grassland Productivity App developed and documented:https://nccasc.shinyapps.io/Grasslands_Productivity_Climate_App/
Publications
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
Esit, M., Kumar, S., Pandey, A., Lawrence, D. M., Rangwala, I., & Yeager, S. (2021). Seasonal to multi-year soil moisture drought forecasting. npj Climate and Atmospheric Science, 4(1), 1-8.
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
Sutton, C., Kumar, S., Lee, M. K., & Davis, E. (2021). Human imprint of water withdrawals in the wet environment: A case study of declining groundwater in Georgia, USA. Journal of Hydrology: Regional Studies, 35, 100813.
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
Duan, Y., Kumar, S., & Kinter, J. L. (2021). Evaluation of Long-term Temperature Trend and Variability in CMIP6 Multimodel Ensemble. Geophysical Research Letters, e2021GL093227.
|
|