Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Research › peer-review
Surrogate Neural Network Models of Atmospheric Chemistry From the Output of Data Assimilation Algorithms. / Tarraf, Deniel; Shaabo, Saraa; Penenko, Alexey et al.
Proceedings - 2025 21st International Asian School-Seminar on Optimization Problems of Complex Systems (OPCS). Institute of Electrical and Electronics Engineers Inc., 2025. p. 1-5.Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Research › peer-review
}
TY - GEN
T1 - Surrogate Neural Network Models of Atmospheric Chemistry From the Output of Data Assimilation Algorithms
AU - Tarraf, Deniel
AU - Shaabo, Saraa
AU - Penenko, Alexey
AU - Dultseva, Galina
N1 - Conference code: 21
PY - 2025/11/5
Y1 - 2025/11/5
N2 - Data assimilation algorithms combine mathematical models of considered processes and available observation data to obtain more accurate predictions. In the course of operation, they produce output that can be used to refine the processes models under study. Two machine-learning approaches to refine production-loss-type mathematical models based on the output provided by data assimilation algorithms are considered. In the first approach, we construct surrogate production and loss operators based on the corresponding uncertainty functions that can be estimated by data assimilation algorithms to fit the measurement data. In the second, more traditional approach, we construct a surrogate corrections operator that partially reproduces the dynamics of a more complex model using the state function estimates that can be provided by a data assimilation algorithm. Surrogate models are presented by artificial neural networks with two dense layers. The approaches were tested on atmospheric chemistry models of different complexity. In both cases surrogate models were able to reproduce the behavior of their prototypes.
AB - Data assimilation algorithms combine mathematical models of considered processes and available observation data to obtain more accurate predictions. In the course of operation, they produce output that can be used to refine the processes models under study. Two machine-learning approaches to refine production-loss-type mathematical models based on the output provided by data assimilation algorithms are considered. In the first approach, we construct surrogate production and loss operators based on the corresponding uncertainty functions that can be estimated by data assimilation algorithms to fit the measurement data. In the second, more traditional approach, we construct a surrogate corrections operator that partially reproduces the dynamics of a more complex model using the state function estimates that can be provided by a data assimilation algorithm. Surrogate models are presented by artificial neural networks with two dense layers. The approaches were tested on atmospheric chemistry models of different complexity. In both cases surrogate models were able to reproduce the behavior of their prototypes.
UR - https://www.scopus.com/pages/publications/105023701365
UR - https://www.mendeley.com/catalogue/8e123581-d993-3932-b148-42209417efbe/
U2 - 10.1109/opcs67346.2025.11219378
DO - 10.1109/opcs67346.2025.11219378
M3 - Conference contribution
SN - 979-8-3315-8982-0
SP - 1
EP - 5
BT - Proceedings - 2025 21st International Asian School-Seminar on Optimization Problems of Complex Systems (OPCS)
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 21st International Asian School-Seminar on Optimization Problems of Complex Systems (OPCS)
Y2 - 7 July 2025 through 17 July 2025
ER -
ID: 72689581