The study, led by Yuanhao Du and Xiuli Geng of the University of Shanghai for Science and Technology, together with Sheng Cheng and Hongliu Zhang, addresses two weaknesses the authors identify in the existing literature. First, most air quality index (AQI) prediction models do not adequately account for the noisy character of AQI data, which blends pollutant concentrations, meteorological influences, and measurement artifacts into a single fluctuating number. Second, previous models have largely focused on a single region, leaving the cross-regional prediction problem—arguably the more useful one for public health planning—largely unsolved. The new framework is an explicit attempt to handle both challenges within one pipeline.

The first stage of the pipeline is data preparation, and it begins with a technique called complementary ensemble empirical mode decomposition with adaptive noise, or CEEMDAN. Rather than filtering the raw AQI signal with fixed mathematical filters, CEEMDAN adaptively decomposes it into a set of simpler components at different scales, adding controlled white noise in complementary pairs so that the residual noise in the decomposition largely cancels itself out. The result, according to the researchers, is a significant reduction in the noise effects that plague raw monitoring data. Once the data has been denoised in this way, the team applies phase space reconstruction, a method drawn from nonlinear dynamics that rebuilds the hidden geometric structure of the time series by embedding past values into a higher-dimensional space. This step is intended to expose the dynamic characteristics of the AQI series—the delayed dependencies and oscillations that a predictive model must capture in order to look even one day into the future.

With the data conditioned, the framework turns to its predictive engine: a deep echo state network modified by what the authors call a dynamic activation mechanism, abbreviated DCM-DESN. Echo state networks belong to the reservoir computing family, in which a large, sparsely connected recurrent layer—the reservoir—transforms input signals into a rich set of internal states, while only the readout weights connecting the reservoir to the output are trained. This makes them fast to train and well suited to time-series problems, and stacking reservoirs into a deep architecture extends their memory and representational capacity. The catch is that deep reservoirs typically demand a large number of neurons to perform well, which inflates computational cost. The dynamic activation mechanism adjusts the nonlinear connection mode of the deep echo state network, and by doing so it reduces the model’s dependence on the number of reservoir neurons—allowing the network to reach strong performance with a leaner internal structure.

Training the readout weights of such a network is itself an optimization problem, and this is where the second custom ingredient enters. The researchers propose an improved version of atom search optimization, a nature-inspired algorithm that models the movement of atoms under interaction forces to search a solution space. Their improved atomic search optimization, or IASO, is used to tune the network weights of the DCM-DESN, replacing gradient-based tuning of the reservoir-to-output connections with a global search that is less likely to be trapped in poor local solutions. The combination—CEEMDAN and phase space reconstruction on the front end, a dynamically activated deep reservoir in the middle, and IASO on the training side—forms what the authors describe as an intelligent AQI monitoring framework built on multi-source data.

The transfer learning component is what elevates the framework from a single-city forecaster to a cross-regional one. Transfer learning, a strategy now central to modern machine learning, allows a model trained on one task or domain to be adapted to another with far less data than training from scratch would require. In this study, the researchers constructed multiple cross-city transfer-learning tasks using daily data from Xi’an, Wuhan, Changsha, and Shenzhen covering the period from January 1, 2021 to March 24, 2024. For each task, the IASO-optimized DCM-DESN is first pretrained on data from a source city. The pretrained reservoir-based model is then transferred to the target city and frozen—its internal dynamics held fixed—while only the output weights are fine-tuned using a limited target-domain adaptation set drawn from the destination city.

The design choice to freeze the reservoir and retrain only the readout is both computationally elegant and practically significant. Because reservoir computing trains only its output layer in the first place, adaptation to a new city reduces to a lightweight re-fitting of a small set of weights, using just a limited sample of local data. In operational terms, that means a city with a short monitoring history could borrow the learned atmospheric dynamics of a data-rich neighbor and still produce accurate one-day-ahead AQI forecasts. The inputs to the model are historical pollutant concentrations and related environmental variables available up to the current day, which are used to directly predict the AQI one day ahead—a horizon short enough to be actionable for public health advisories yet long enough to require genuine predictive skill.

To evaluate the framework, the team measured performance across the various transfer tasks using three standard regression metrics: mean absolute error (MAE), root mean square error (RMSE), and the coefficient of determination (R²). Together these metrics capture both the typical size of forecast errors and the proportion of variance in the observed AQI that the model explains. The authors report that the proposed model demonstrated excellent performance across the different tasks, validating the architecture’s ability to generalize across cities with distinct pollution profiles. The four cities themselves span a meaningful range of conditions: Xi’an in the northwest, Wuhan and Changsha in central China, and Shenzhen in the subtropical south, each with its own mix of emissions sources and climatic drivers.

The study also connects to a broader body of work on decomposition-based air quality forecasting. Previous research has combined CEEMDAN with recurrent architectures such as independent recurrent neural networks and long short-term memory networks, and with attention-augmented convolutional models, to squeeze better forecasts out of noisy AQI series. Spatiotemporal graph convolutional networks have been applied to exploit correlations among monitoring stations, and transfer learning has been explored for predicting different pollutants and for assessing events such as COVID-19 lockdown effects on air quality. What distinguishes the new framework is the integration of these strands—adaptive decomposition, phase space reconstruction, deep reservoir computing with a dynamic activation mechanism, metaheuristic weight optimization, and cross-city transfer—into a single end-to-end pipeline aimed squarely at the one-day-ahead AQI problem.

The implications reach beyond the four cities studied. Accurate real-time understanding of AQI evolution is described by the authors as essential for atmospheric pollution prevention and urban public health management, and a framework that can be adapted to a new city with limited local data could lower the barrier for municipalities that lack long monitoring records. The work was supported by the National Natural Science Foundation of China under grant number 72271164, and the authors state they have no competing financial or non-financial interests to disclose. As cities worldwide grapple with the health burden of air pollution—exposure that the World Health Organization continues to flag as a major environmental health risk—tools that make accurate forecasting portable from one urban environment to another may prove among the most consequential applications of machine learning in atmospheric science. The study’s data will be made available on request, and the full article is published in Air Quality, Atmosphere & Health, volume 19, article number 218.

Subject of Research: Cross-city air quality index prediction using transfer learning and an optimized deep echo state network

Article Title: An intelligent AQI prediction method using transfer learning and an IASO-optimized DCM-DESN under multi-source data: A case study of four Chinese cities

Article References: Du, Y., Geng, X., Cheng, S., & Zhang, H. (2026). An intelligent AQI prediction method using transfer learning and an IASO-optimized DCM-DESN under multi-source data: A case study of four Chinese cities. Air Quality, Atmosphere & Health, 19(10), Article 218. https://doi.org/10.1007/s11869-026-02109-y

Image Credits: AI Generated

DOI: 10.1007/s11869-026-02109-y

Keywords: air quality index, transfer learning, deep echo state network, CEEMDAN, phase space reconstruction, atom search optimization, reservoir computing, cross-city prediction, machine learning, air pollution forecasting, China, public health

The research, led by Mingkun Zhu of the Department of Automation at Taiyuan University of Technology, together with Xiaoxia Han, Jinde Wu, Haonan Zhu, and Wenxin Chai, targets a problem that most data-driven forecasting models handle poorly. Conventional deep learning models encode historical patterns implicitly within their trained weights, which means that once training is complete, the model has no explicit memory of specific pollution episodes it can consult. If a city experiences a recurring type of ozone event, driven for example by a particular combination of stagnant air, high temperatures, and regional transport of precursors, a standard model must re-learn that pattern from scratch each time it appears in slightly different form. OzoneKBNet instead makes those recurring episodes searchable.

The framework works by forecasting station-level ozone concentrations over the next 48 hours based on the preceding 96 hours of observations. Its inputs include ozone itself, nitrogen dioxide, fine particulate matter with a diameter of 2.5 micrometers or less, temperature, and relative humidity, a set of variables chosen to capture both the chemical precursors of ozone formation and the meteorological conditions that govern it. When a forecast is needed, the system searches a knowledge base of historical situations for multi-scale analogs drawn from across the monitoring network, retrieving similar episodes not just from the station in question but from neighboring stations as well. This cross-station retrieval is crucial in urban agglomerations, where pollution transported from upwind cities can dominate local ozone behavior.

Retrieval alone is not enough, and the researchers built in a sophisticated filtering stage. Candidate analogs are re-ranked according to two criteria: representation similarity, which measures how closely the retrieved historical situation matches the current one in the model’s learned feature space, and local-trend consistency, which checks whether the historical episode’s near-term trajectory actually resembles what is happening now. Only analogs that pass both tests contribute to the forecast. Their known future trajectories are then used as bounded, reliability-aware corrections to a lightweight long short-term memory network, or LSTM, that provides the baseline prediction. The word bounded matters: rather than allowing retrieved evidence to override the neural forecast wholesale, the system applies corrections within limits calibrated to how trustworthy the retrieved analogs appear to be.

A particularly careful aspect of the study is its handling of temporal data leakage, a common pitfall in machine learning applications to time series. If a model is trained and tested on data drawn from overlapping periods, it can inadvertently peek at information from the future, inflating its apparent accuracy. The team partitioned their data strictly by year: observations from 2023 were used for normalization, for pretraining the retrieval encoder, and for constructing the knowledge base; data from 2024 were reserved for supervised training; and a fixed set of samples from 2025 served as the test set. This design means the model was evaluated on genuinely unseen future conditions, a far more demanding test than random splitting of the data.

The results span seven Chinese urban agglomerations, offering a broad regional assessment of the framework’s capabilities. Across all stations, OzoneKBNet achieved macro-average mean absolute error of 20.06 micrograms per cubic meter and root mean square error of 24.45 micrograms per cubic meter over the 48-hour forecast horizon. It posted the lowest regional mean absolute error in six of the seven regions evaluated, suggesting that the retrieval-based approach generalizes well across geographically and climatologically diverse areas rather than being tuned to a single city’s quirks. Compared with a station-specific LSTM baseline, the framework reduced mean absolute error by 1.6 percent and root mean square error by 1.2 percent.

Those percentage improvements may sound modest, but their significance lies in what they demonstrate about the architecture of forecasting systems. The baseline LSTM is already a strong performer, trained on the same data, and squeezing additional accuracy out of it through any independent mechanism is difficult. The gains from OzoneKBNet come from an entirely complementary source: explicit regional historical evidence that the parametric model cannot access on its own. In effect, the study shows that a model’s learned parameters and a searchable library of past episodes contain partially non-overlapping information, and combining the two yields forecasts more accurate than either alone. This principle echoes a broader trend in artificial intelligence, where retrieval-augmented generation has transformed language models by letting them consult external documents rather than relying purely on memorized knowledge.

The intellectual lineage of the analog approach stretches back decades in atmospheric science. Long-range weather forecasting based on analogs, matching current conditions to historical situations and borrowing their outcomes, was explored as early as the late 1980s, and modified k-nearest-neighbor methods have been used for real-time weather prediction. What is new is the marriage of this old idea with modern deep learning: an encoder trained to map multivariate pollution and meteorology sequences into a representation space where similarity search becomes meaningful, combined with a re-ranking scheme and a residual correction mechanism that treats retrieved evidence with appropriate caution. The framework also connects to recent work on retrieval-augmented time series forecasting in the machine learning community, where researchers have begun applying similar ideas to general forecasting foundation models.

Why does station-level forecasting matter so much for ozone? Regional averages can hide enormous local variation. Ozone concentrations at a single monitoring station depend on the interplay of local emissions, photochemistry, boundary-layer dynamics, and transport from surrounding areas, and public health warnings are ultimately issued for specific places where people live and breathe. A framework that can deliver accurate 48-hour forecasts at the station level, across an entire urban agglomeration, gives authorities the granularity needed to warn residents before pollution episodes peak, to time traffic restrictions or industrial curbs more effectively, and to study how ozone responds to control measures. The authors suggest their approach may support short-term ozone-pollution episode warning, and the 48-hour horizon is well matched to the timescale on which such interventions operate.

The team has also made an unusually strong commitment to openness. The source code of OzoneKBNet and the complete fixed 2025 test set are available in a public repository on GitHub, along with the model implementation, running scripts, data-format instructions, and fixed test samples that allow others to inspect the experimental pipeline and evaluation protocol. The raw air quality monitoring data and ERA5 reanalysis data analyzed in the study come from publicly available providers, although redistribution restrictions prevent the processed training datasets from being shared directly. This transparency matters in a field where evaluation practices vary widely and where, as the strict 2023-2024-2025 split shows, the details of how models are tested can determine whether reported accuracy reflects real forecasting skill or subtle leakage. For a problem as consequential as urban ozone, reproducible methods and honest evaluation are not luxuries but necessities, and this work offers both alongside a genuinely new architectural idea.

Subject of Research: Retrieval-augmented deep learning for station-level ground-level ozone forecasting in urban agglomerations

Article Title: A retrieval-augmented historical knowledge-base framework with residual correction for station-level ozone forecasting across urban agglomerations

Article References: Zhu, M., Han, X., Wu, J., Zhu, H., & Chai, W. (2026). A retrieval-augmented historical knowledge-base framework with residual correction for station-level ozone forecasting across urban agglomerations. Air Quality, Atmosphere & Health, 19(10), Article 215. https://doi.org/10.1007/s11869-026-02105-2

Image Credits: AI Generated

DOI: 10.1007/s11869-026-02105-2

Keywords: ozone forecasting, air quality, machine learning, retrieval-augmented forecasting, LSTM, urban agglomerations, knowledge base, time series, China, pollution warning, deep learning, meteorology