Zhejiang presents an ideal and demanding test case. This coastal province in southeastern China is home to roughly 66.7 million people and an economy exceeding 9 trillion CNY in 2024, yet nearly 75 percent of its land is covered by hills and mountains, with only about 20 percent plains and a sliver of rivers and lakes. Rapid urbanization compounds the challenge, since weather stations are sparse, unevenly distributed, and subject to relocations, instrumentation changes, and gaps in the historical record. Ground observations offer accuracy but limited spatial coverage; satellite land surface temperature products offer detail but generally begin only in the early 2000s and are vulnerable to cloud cover, terrain shading, and atmospheric interference; reanalysis products such as ERA5 provide continuity but at coarse spatial resolutions, typically around 0.25 degrees or coarser, far too blunt to resolve the fine thermal texture of mountainous landscapes.

The team’s solution is an elegant two-part construction they call a spatial-background residual framework. First, they built a high-resolution climatological baseline from an existing hourly, 1-kilometer gridded temperature dataset covering 2008 to 2018, which had itself been developed using the INCA data-fusion framework with reanalysis fields and dense automatic weather station observations. This baseline serves purely as a spatial background, encoding how temperature varies across the terrain on each calendar day of the year. Second, daily temperature residuals—the departures of each station observation from that climatological expectation—were calculated for every meteorological station across the full 1961 to 2020 period. Because these residuals are computed directly from observed temperatures, they inherently preserve the long-term warming trend and interannual variability, while the baseline contributes the terrain-driven spatial detail. Summing the interpolated residual field with the baseline yields the finished reconstruction.

A critical methodological question was how best to interpolate those daily residuals across space. The researchers evaluated three schemes representing different levels of topographic correction and complexity: plain inverse distance weighting, or IDW, which relies only on spatial proximity; a lapse-rate-adjusted version of IDW, in which station temperatures are first corrected to grid-cell elevation using a fixed adiabatic lapse rate of 6.0 degrees Celsius per kilometer; and a multiple linear regression incorporating longitude, latitude, and elevation as predictors. Using leave-one-out cross-validation, in which each station is successively withheld and predicted from the others, the team assessed performance with mean absolute error, root-mean-square error, and the coefficient of determination across decades, seasons, and elevation zones.

The verdict was striking: the simplest method won. IDW without any lapse-rate correction consistently delivered the lowest errors and highest skill, achieving the best performance at roughly 68 percent of stations for daily minimum temperature, 70.7 percent for mean temperature, and 74.7 percent for maximum temperature. In a representative example from 1971 to 1980, IDW reconstructed minimum temperatures with a mean absolute error of just 0.73 degrees Celsius and an R-squared of 0.987, comfortably beating both rivals. The reason lies in the behavior of the lapse rate itself. Analysis of 60 years of observations revealed that near-surface temperature lapse rates in Zhejiang are strongly non-stationary: they peak in summer, with minimum-temperature lapse rates exceeding 7.0 degrees Celsius per kilometer in mountainous areas during July and August, yet collapse toward zero or even turn negative in winter lowlands, where temperature inversions prevail. Applying a fixed correction therefore risks systematic, elevation-related biases—a caution with implications well beyond Zhejiang.

The errors that do remain follow clear and intelligible patterns. Reconstruction accuracy improved steadily from the 1960s onward as station density grew, and summer months outperformed winter months because spatial temperature gradients are weaker in warm weather. Low-elevation areas below 400 meters consistently yielded smaller errors than high-elevation zones, where complex terrain and sparse instrumentation conspire against interpolation. Spatially, larger uncertainties cluster in the mountainous southwest, including parts of Lishui and western Wenzhou, while the plains around Hangzhou, Shaoxing, and Jinhua show excellent agreement, with most stations achieving R-squared values above 0.90 and many above 0.95. Across the entire 60-year span, the annual mean error for all three temperature variables stayed within plus or minus 0.1 degrees Celsius, with no systematic drift across decades—a testament to the temporal stability of the method.

Perhaps the most compelling validation came from real disasters. The team tested ZJ-DAT against two extreme events from 2007, using more than a thousand automatic weather stations as independent ground truth while deliberately excluding any stations that had contributed to the reconstruction. During the cold wave of 4 to 9 March 2007, ZJ-DAT tracked the south-to-north advance of the cold air, accurately reproducing the observed cold centers around Lishui, with R-squared values of 0.65 to 0.79 and root-mean-square errors of 1.01 to 1.67 degrees Celsius. By comparison, the CDAT national dataset managed only moderate agreement, while ERA5 performed poorly, with near-zero or negative correlations and errors approaching 3 degrees Celsius. The heat wave of 30 June to 10 July 2007 told the same story: ZJ-DAT best resolved the core hot zones above 37 degrees Celsius over Jinhua, Shaoxing, and Ningbo and the inland-coastal thermal contrast, while CDAT smoothed away local extremes and ERA5 drifted with warm biases and excessive homogenization. Case studies of cold and heat events in January and July 2020 at four environmentally distinct stations—an island, a mountain site, and two plain stations—further confirmed the reconstruction’s fidelity, with discrepancies generally under 2 degrees Celsius.

Beyond validation, the dataset enabled a first-of-its-kind hazard assessment for the province. Using Gumbel distribution analysis of return periods, the researchers mapped the intensity of extreme cold and heat expected at 5-, 20-, and 50-year recurrence intervals. The results expose stark geographic contrasts in climate risk. Extreme low-temperature hazards concentrate in the northwestern inland regions, where 50-year minimum temperatures plunge below minus 15 degrees Celsius, while the southeastern coast stays comparatively mild. Extreme heat hazards show the opposite pattern, dominated by low-altitude basins in central and northern Zhejiang, where 50-year maximum temperatures climb above 43 to 44 degrees Celsius—figures that carry sobering weight given projections of accelerating heatwave duration under global warming. These maps, grounded in kilometer-scale temperature data rather than coarse reanalysis, offer planners a far sharper picture of where adaptation investments are most needed.

The authors are candid about limitations. Anchoring the reconstruction to a climatology drawn from 2008 to 2018 means the reference field does not reflect the climate state of earlier decades, though because it functions only as a spatial scaffold while temporal signals come from station residuals, warming trends and variability remain intact. The team suggests that future refinements could employ temporally adaptive reference fields. The broader significance, however, is clear: ZJ-DAT demonstrates that a simple, computationally efficient interpolation of station anomalies, layered onto a modern high-resolution climatology, can outperform more elaborate schemes in complex terrain—provided the scheme is chosen with local lapse-rate physics in mind. The framework, and the publicly available dataset released through Zenodo, is designed to be transferable to other topographically complex, observation-limited regions, offering a practical foundation for climate monitoring, extreme-event risk assessment, and adaptation planning as the planet continues to warm.

Subject of Research: High-resolution daily near-surface air temperature reconstruction for Zhejiang Province, China, from 1961 to 2020 using statistical residual interpolation

Article Title: A high-resolution daily temperature reconstruction for Zhejiang Province during 1961–2020 using statistical residual interpolation

Article References: Li, Y., Guo, H., Dong, M., Wu, J., Deng, F., Chen, Y., & Chen, F. (2026). A high-resolution daily temperature reconstruction for Zhejiang Province during 1961–2020 using statistical residual interpolation. Theoretical and Applied Climatology, 157(10), Article 632. https://doi.org/10.1007/s00704-026-06521-3

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06521-3

Keywords: Zhejiang Province, air temperature reconstruction, inverse distance weighting, spatial interpolation, temperature lapse rate, extreme temperature events, heat wave, cold wave, climate monitoring, hazard assessment, ERA5, high-resolution gridded dataset

The research, led by Sharmin Akther of Jahangirnagar University together with colleagues from the Bangladesh Meteorological Department, the University of Melbourne and other institutions, focused on Mymensingh district, located at 24.75 degrees north and 90.40 degrees east. The team assembled a remarkable 38-year record of daily minimum temperatures spanning 1985 to 2022, obtained from the Bangladesh Meteorological Department. Only 30 of the 13,787 daily records, a mere 0.22 percent of the dataset, contained missing values, and these were filled using time-based interpolation that respects the temporal relationship between adjacent observations. Statistical tests confirmed the series was stationary: the Augmented Dickey-Fuller test rejected a unit root with a statistic of minus 11.426, while the KPSS test failed to reject stationarity, giving the researchers a solid foundation for time series modeling.

The core of the study was a head-to-head comparison of forecasting approaches. On the statistical side, the team fitted an autoregressive integrated moving average model, selecting ARIMA(1,1,2) as optimal using the Akaike and Bayesian information criteria, and an exponential smoothing state space model, where the simplest ETS(A,N,N) configuration with additive errors, no trend and no seasonality achieved the lowest AIC. On the machine learning side, they trained support vector regression with a radial basis function kernel, random forest regression, and a long short-term memory neural network, the deep learning architecture specifically designed to capture long-range dependencies in sequential data through its gated memory cells. Each machine learning model was fed 30 days of lagged temperatures as input features, standardized with Z-score normalization, and tuned through five-fold time series cross-validation to prevent any leakage of future information.

The team also built six hybrid models that combined each machine learning predictor with a statistical correction of its residuals, following the classic hybridization strategy in which a neural network captures nonlinear patterns while ARIMA or ETS models any remaining linear structure. The final hybrid prediction was the sum of the machine learning forecast and the statistical model’s forecast of the residuals. This family of models, including LSTM+ARIMA, LSTM+ETS, SVR+ARIMA, SVR+ETS, RF+ARIMA and RF+ETS, was evaluated with the same rigorous out-of-sample protocol applied to the standalone models.

When the models were tested on data they had never seen, covering June 2011 through September 2022, the deep learning model emerged as the clear winner. The LSTM achieved a root mean square error of 1.395 degrees Celsius, a mean absolute error of 1.053 degrees, and a mean absolute scaled error of 0.906, meaning it beat a naive persistence forecast. Support vector regression came in a close second at 1.403 degrees RMSE, and random forest followed at 1.435 degrees. The statistical models, by contrast, performed poorly, with RMSE values around 6.7 to 6.8 degrees and MASE values above 4. The researchers attribute the LSTM’s success to three factors: daily minimum temperature exhibits stronger day-to-day persistence than mean or maximum temperature, the 38-year training record provides ample data for learning seasonal cycles, and the single-station design avoids errors from spatial heterogeneity.

Perhaps the most surprising finding concerned the hybrid models. Despite the theoretical appeal of combining statistical and machine learning methods, none of the six hybrids significantly improved on its standalone counterpart. The best hybrid, LSTM+ARIMA, achieved an RMSE of 1.397 degrees, essentially identical to the standalone LSTM’s 1.395 degrees. Diebold-Mariano tests, which formally compare predictive accuracy between competing forecasters, confirmed that only LSTM+ARIMA differed significantly from its base model, and in that case the standalone LSTM was actually better. The message is that when a deep learning model already captures the complex temporal structure of a temperature series, bolting on a statistical correction adds complexity without adding skill.

Accuracy in predicting temperature, however, is not the same as accuracy in detecting cold waves, and here the study delivers its most important caution. Treating cold wave detection as a binary classification problem with the 10 degree threshold, the LSTM showed excellent discriminative power, with a ROC-AUC of 0.975, meaning it ranks cold days above ordinary days with remarkable consistency. Yet its recall was only 0.215: the model correctly identified just 14 of the 65 actual cold wave days in the test period, missing 51 of them. Precision stood at 0.560, so when the model did flag a cold day it was right 56 percent of the time, and it raised only 11 false alarms. The overall accuracy of 0.985 is misleading because cold days make up only 1.6 percent of observations, a class imbalance that pushes models toward conservative behavior. Monthly analysis revealed the pattern in detail: the model detected 22 of 43 cold days in December, a 51 percent detection rate, but only 2 of 18 in November and 1 of 4 in January, suggesting systematic underestimation of early winter cold events.

Using the trained model, the researchers generated a daily minimum temperature forecast for 2027 with 80 and 95 percent prediction intervals constructed from the test RMSE. The projection captured the expected seasonal cycle, with summer values peaking around 24 to 25 degrees and winter minima between 22 and 23 degrees, and all forecasted temperatures remained above the cold wave threshold. But the authors are emphatic that this absence of forecasted cold waves must not be read as a prediction of no cold wave risk. A model that misses roughly 78 percent of historical cold days would likely miss actual cold waves in 2027 as well. The model was trained on data ending in 2011 and cannot account for climate regime shifts or changing winter patterns since then, and the winter prediction intervals, spanning roughly plus or minus 2 to 3 degrees, are wide enough that a downward fluctuation could still push temperatures below 10 degrees. The researchers stress that disaster management agencies, health authorities and local governments should continue normal winter preparedness measures from November through February regardless of this exploratory projection, and that operational decisions should rely on routine seasonal forecasts from national meteorological services.

The study’s implications reach beyond Bangladesh. The finding that deep learning substantially outperforms linear statistical models for daily temperature echoes results from across South Asia, including ARIMA-based temperature analysis in Karachi, Pakistan, and an STL-ARIMA-LSTM hybrid for heatwave forecasting in Rajshahi that achieved an RMSE of 1.18 degrees. The low recall for rare events is also not unique to this work; studies of heatwave and flood classification have reported similar struggles with imbalanced datasets, where high overall accuracy conceals poor detection of the very events that matter most. The authors recommend that future operational systems adjust the classification threshold to balance precision and recall, incorporate perceived temperature metrics such as the wind chill index, and integrate humidity, wind speed, cloud cover and large-scale climate indices like ENSO, all of which were absent from the current univariate framework.

The researchers outline a clear roadmap for strengthening the approach. Expanding the analysis to multiple stations across Bangladesh would test spatial transferability and reveal regional patterns in cold wave occurrence. Advanced techniques for handling class imbalance, including synthetic minority oversampling, focal loss and cost-sensitive learning, could raise recall at some cost to precision. Linking temperature forecasts to health outcome data such as cold-related mortality and hospitalization rates would allow health-relevant alert thresholds to be defined and would provide direct validation of the model’s usefulness as an early warning tool. Probabilistic methods such as quantile regression forests and Bayesian neural networks could extend forecast horizons while quantifying uncertainty more honestly. For now, the study positions the LSTM model as a powerful temperature forecasting instrument rather than a standalone cold wave alarm, a distinction that could shape how machine learning is deployed to protect vulnerable communities across the region.

Subject of Research: Machine learning forecasting of daily minimum temperatures and cold wave events in Mymensingh district, Bangladesh

Article Title: Forecasting cold wave in Bangladesh: a validated machine learning approach for early warning and vulnerability reduction

Article References: Akther, S., Hussain Khan, M. M., Chowdhury, S., Das, A., Rahman, M., & Rois, R. (2026). Forecasting cold wave in Bangladesh: a validated machine learning approach for early warning and vulnerability reduction. BMC Environmental Science, 3(1), Article 17. https://doi.org/10.1186/s44329-026-00058-6

Image Credits: AI Generated

DOI: 10.1186/s44329-026-00058-6

Keywords: cold wave, Bangladesh, LSTM, machine learning, temperature forecasting, early warning, Mymensingh, ARIMA, hybrid models, time series, climate extremes, public health