Increasing Lead Time Research Articles

Development of a city-level surface ozone forecasting system using deep learning techniques and air quality model: Application in eastern China

Utilizing regional air quality models to accurately forecast surface ozone (O3) concentrations, particularly high concentrations, is essential for protecting public health. However, forecasts of air quality model often deviate from site observations due to the limitation of grid resolution and uncertainties from emission sources, meteorological conditions, and chemical reaction mechanisms. Especially, the underestimation is significant under condition of high O3 concentrations. Moreover, such deviations tend to accumulate as forecast lead time increases, compounding the challenges associated with reliable air quality forecast. In this study, we employed AlexNet architecture, a classical convolutional neural network, combined with multiple variables related to meteorology, chemistry, emission and geography to establish a non-linear relationship between grid-scale input variables and site-scale hourly O3 forecast biases in Eastern China, aiming to realize accurate city-level ozone forecast based on a regional air quality prediction model (i.e., Nested Air Quality Prediction Model System, NAQPMS). By assigning weights to high-bias samples and high-concentration samples within the loss function, the proposed Weighted AlexNet model (W_AlexNet) effectively reduced forecast biases and enhanced its capability to predict O3 pollution levels. Compared to NAQPMS, W_AlexNet model demonstrated a 25.71% improvement in RMSE and a 7.17% increase in IOA averagely for hourly O3 (O3-1h) forecasts across four different lead times (24-h, 48-h, 72-h, and 96-h). Notably, W_AlexNet model alleviated the tendency of NAQPMS to underestimate high concentrations and showed a superior performance in improving O3-1h pollution level forecasts, particularly for the 72-h and 96-h lead times. W_AlexNet model can effectively mitigate the bias accumulation effect over increasing lead times, thereby enhancing the reliability of longer-term forecasts. Thus, the W_AlexNet model serves as a post-processing model that can calibrate forecast biases in air quality prediction models, significantly improving the accuracy of O3 high concentration forecasts and providing more precise early warnings of O3 pollution. This underscores its utility in air quality management.

Evaluating vector winds over eastern China in 2022 predicted by the CMA-MESO model and ECMWF forecast

Vector winds play a crucial role in weather and climate, as well as the effective utilization of wind energy resources. However, limited research has been conducted on treating the wind field as a vector field in the evaluation of numerical weather prediction models. In this study, we treat vector winds as a whole by employing a vector field evaluation method, and evaluate the mesoscale model of the China Meteorological Administration (CMA-MESO) and ECMWF forecast, with reference to ERA5 reanalysis, in terms of multiple aspects of vector winds over eastern China in 2022. The results show that the ECMWF forecast is superior to CMA-MESO in predicting the spatial distribution and intensity of 10-m vector winds. Both models overestimate the wind speed in East China, and CMA-MESO overestimates the wind speed to a greater extent. The forecasting skill of the vector wind field in both models decreases with increasing lead time. The forecasting skill of CMA-MESO fluctuates more and decreases faster than that of the ECMWF forecast. There is a significant negative correlation between the model vector wind forecasting skill and terrain height. This study provides a scientific evaluation of the local application of vector wind forecasts of the CMA-MESO model and ECMWF forecast.

Understanding the initial conditions contributing to the rapid intensification of typhoons through ensemble sensitivity analysis

While steady improvements have been achieved for the track forecasts of typhoons, there has been a lack of improvement for intensity forecasts. One challenge for intensity forecasts is to capture the rapid intensification (RI), whose nonlinear characteristics impose great difficulties for numerical models. The ensemble sensitivity analysis (ESA) method is used here to analyze the initial conditions that contribute to typhoon intensity forecasts, especially with RI. Six RI processes from five typhoons (Chaba, Haima, Meranti, Sarika, and Songda) in 2016, are applied with ESA, which also gives a composite initial condition that favors subsequent RI. Results from individual cases have generally similar patterns of ESA, but with different magnitudes, when various cumulus parameterization schemes are applied. To draw the initial conditions with statistical significance, sample-mean azimuthal components of ESA are obtained. Results of the composite sensitivity show that typhoons that experience RI in 24 h favor enhanced primary circulation from low to high levels, intensified secondary circulation with increased radial inflow at lower levels and increased radial outflow at upper levels, a prominent warm core at around 300 hPa, and increased humidity at low levels. As the forecast lead time increases, the patterns of ESA are retained, while the sensitivity magnitudes decay. Given the general and quantitative composite sensitivity along with associated uncertainties for different cumulus parameterization schemes, appropriate sampling of the composite sensitivity in numerical models could be beneficial to capturing the RI and improving the forecasting of typhoon intensity.摘要针对台风强度特别是快速增强 (RI) 的预报难点, 本文对2016年5个台风的6次RI过程应用集合敏感性分析 (ESA), 以分析有利于RI的初始条件. 对于不同的个例和积云参数化方案, ESA获得的有利于台风RI的集合敏感性相似但量级存在差异. 复合敏感性揭示了经历RI的台风所需的初始条件, 其有更强的主环流, 更显著的暖心及增加的低层湿度. ESA估计的集合敏感性随预报时长增加而减弱. 通过采样复合敏感性可有望改进集合初始条件及台风强度预报.

Combining Multiple Machine Learning Methods Based on CARS Algorithm to Implement Runoff Simulation

Runoff forecasting is crucial for water resource management and flood safety and remains a central research topic in hydrology. Recent advancements in machine learning provide novel approaches for predicting runoff. This study employs the Competitive Adaptive Reweighted Sampling (CARS) algorithm to integrate various machine learning models into a data-driven rainfall–runoff simulation model. We compare the forecasting performance of different machine learning models to improve rainfall–runoff prediction accuracy. This study uses data from the Maduwang hydrological station in the Bahe river basin, which contain 12 measured flood events from 2000 to 2010. Historical runoff and areal mean rainfall serve as model inputs, while flood data at different lead times are used as model outputs. Among the 12 flood events, 9 are used as the training set, 2 as the validation set, and 1 as the testing set. The results indicate that the CARS-based machine learning model effectively forecasts floods in the Bahe River basin. Under the prediction period of 1 to 6 h, the model achieves high forecasting accuracy, with the average NSE ranging from 0.7509 to 0.9671 and the average R2 ranging from 0.8397 to 0.9413, though the accuracy declines to some extent as the lead time increases. The model accurately predicts peak flow and performs well in forecasting high flow and recession flows, though peak flows are somewhat underestimated for longer lead times. Compared to other machine learning models, the SVR model has the highest average RMSE of 0.942 for a 1–6 h prediction period. It exhibits the smallest deviation among low-, medium-, and high-flow curves, with the lowest NRMSE values across training, validation, and test sets, demonstrating better simulation performance and generalization capability. Therefore, the machine learning model based on CARS feature selection can serve as an effective method for flood forecasting. The related findings provide a new forecasting method and scientific decision-making basis for basin flood safety.

Evaluation of WRF-Chem air quality forecasts during the AEROMMA and STAQS 2023 field campaigns

ABSTRACT A real-time air quality forecasting system was developed using the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) to provide support for flight planning activities during the NOAA Atmospheric Emissions and Reactions Observed from Megacities to Marine Areas (AEROMMA) and NASA Synergistic TEMPO Air Quality Science (STAQS) 2023 field campaigns. The forecasting system operated on two separate domains centered on Chicago, IL, and New York City, NY, and provided 72-hour predictions of atmospheric composition, aerosols, and clouds. This study evaluates the Chicago-centered forecasting system’s 1-, 2-, and 3-day ozone (O3) forecast skill for Chiwaukee Prairie, WI, a rural area downwind of Chicago that often experiences high levels of O3 pollution. Comparisons to vertical O3 profiles collected by a Tropospheric Ozone Lidar Network (TOLNet) instrument revealed that forecast skill decreases as forecast lead time increases. When compared to surface measurements, the forecasting system tended to underestimate O3 concentrations on high O3 days and overestimate on low O3 days at Chiwaukee Prairie regardless of forecast lead time. Using July 25, 2023, as a case study, analyses show that the forecasts underestimated peak O3 levels at Chiwaukee Prairie during this regionwide bad air quality day. Wind speed and direction data indicates that this underestimation can partially be attributed to lake breeze simulation errors. Surface fine particulate matter (PM2.5) measurements, Geostationary Operational Environmental Satellite-16 (GOES-16) aerosol optical depth (AOD) data, and back trajectories from the NOAA Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model show that transported Canadian wildfire smoke impacted the Lake Michigan region on this day. Errors in the forecasted chemical composition and transport of the smoke plumes also contributed to underpredictions of O3 levels at Chiwaukee Prairie on July 25, 2023. The results of this work help identify improvements that can be made for future iterations of the WRF-Chem forecasting system. Implications: Air quality forecasting is an important tool that can be used to inform the public about upcoming high pollution days so that individuals may plan accordingly to limit their exposure to health-damaging air pollutants. Forecasting also helps scientists make decisions about where to make observations during air quality field campaigns. A variety of observational datasets were used to evaluate the accuracy of an air quality forecasting system that was developed for NOAA and NASA field campaigns that occurred in the summer of 2023. These evaluations inform areas of improvement for future development of this air quality forecasting system.

Examining optimized machine learning models for accurate multi-month drought forecasting: A representative case study in the USA.

TheColorado River has experienced a significant streamflow reduction in recent decades due to climate change, resulting in pronounced hydrological droughts that pose challenges to the environment and human activities. However, current models struggle to accurately capture complex drought patterns, and their accuracy decreases as the lead time increases. Thus, determining the reliability of drought forecasting for specific months ahead presents a challenging task. This study introduces a robust approach that utilizes the Beluga Whale Optimization (BWO) algorithm to train and optimize the parameters of the Regularized Extreme Learning Machine (RELM) and Random Forest (RF) models. The applied models are validated against a KNN benchmark model forforecasting drought from one- tosix-month ahead across four hydrological stationsdistributed over the Colorado River. The achieved results demonstrate that RELM-BWO outperforms RF-BWO and KNN models, achieving the lowest root-mean square error (0.2795), uncertainty (U95 = 0.1077), mean absolute error (0.2104), and highest correlation coefficient (0.9135). Also, the current study uses Global Multi-Criteria Decision Analysis (GMCDA) as an evaluation metric to assess the reliability of the forecasting. The GMCDA results indicate that RELM-BWO provides reliable forecasts up to four months ahead. Overall, the research methodology is valuable for drought assessment and forecasting, enabling advanced early warning systems and effective drought countermeasures.

Integrating human factors into the distribution model of goods and fast-moving consumer goods for effective inventory control

Multiple factors contribute to the occurrences of disruptions in workflow management within the e-commerce system that focuses on the distribution of goods and other fast-moving consumer goods (FMCG). These disruptions have particularly affected and increased lead times and caused frequent shipment delays. The causes of these disruptions, especially in the context of distribution of goods and FMCG towards inventory control, have been widely speculated upon in recent years. Given the recent interruptions, the study presented in this paper used a methodology called statistical model analysis (SMA) to study the impact of human factors (HFs) on the system performance in a warehouse distribution system. A case study was taken on a particular company called ABC and XYZ to model the impact of some specific HF such as job fatigue, and job rotation to name a few and to uncover the underlying reasons behind the ongoing disruptions within the distribution system. Specifically, the result of this study aims to provide a data-driven understanding of these issues and one of the contributions of this study is to enhance our understanding of the significance of these HFs, rather than focusing on the equipment and materials as seen in prior research. Through this data-based approach study, stakeholders in operations management, e-commerce, and the supply chain system would be well-informed in many ways to resolve the challenges faced by humans in the system towards enhancing their overall system performance.

Probing the capacity of a spatiotemporal deep learning model for short-term PM2.5 forecasts in a coastal urban area

Accurate forecast of fine particulate matter (PM2.5) is crucial for city air pollution control, yet remains challenging due to the complex urban atmospheric chemical and physical processes. Recently deep learning has been routinely applied for better urban PM2.5 forecasts. However, their capacity to represent the spatiotemporal urban atmospheric processes remains underexplored, especially compared with traditional approaches such as chemistry-transport models (CTMs) and shallow statistical methods other than deep learning. Here we probe such urban-scale representation capacity of a spatiotemporal deep learning (STDL) model for 24-hour short-term PM2.5 forecasts at six urban stations in Rizhao, a coastal city in China. Compared with two operational CTMs and three statistical models, the STDL model shows its superiority with improvements in all five evaluation metrics, notably in root mean square error (RMSE) for forecasts at lead times within 12 h with reductions of 49.8 % and 47.8 % respectively. This demonstrates the STDL model's capacity to represent nonlinear small-scale phenomena such as street-level emissions and urban meteorology that are in general not well represented in either CTMs or shallow statistical models. This gain of small-scale representation in forecast performance decreases at increasing lead times, leading to similar RMSEs to the statistical methods (linear shallow representations) at about 12 h and to the CTMs (mesoscale representations) at 24 h. The STDL model performs especially well in winter, when complex urban physical and chemical processes dominate the frequent severe air pollution, and in moisture conditions fostering hygroscopic growth of particles. The DL-based PM2.5 forecasts align with observed trends under various humidity and wind conditions. Such investigation into the potential and limitations of deep learning representation for urban PM2.5 forecasting could hopefully inspire further fusion of distinct representations from CTMs and deep networks to break the conventional limits of short-term PM2.5 forecasts.

Significant advancement in subseasonal-to-seasonal summer precipitation ensemble forecast skills in China mainland through an innovative hybrid CSG-UNET method

Reliable Subseasonal-to-Seasonal (S2S) forecasts of precipitation are critical for disaster prevention and mitigation. In this study, an innovative hybrid method CSG-UNET combining the UNET with the censored and shifted gamma distribution based ensemble model output statistic (CSG-EMOS), is proposed to calibrate the ensemble precipitation forecasts from ECMWF over the China mainland during boreal summer. Additional atmospheric variable forecasts and the data augmentation are also included to deal with the potential issues of low signal-to-noise ratio and relatively small sample sizes in traditional S2S precipitation forecast correction. The hybrid CSG-UNET exhibits a notable advantage over both individual UNET and CSG-EMOS in improving ensemble precipitation forecasts, simultaneously improving the forecast skills for lead times of 1–2 weeks and further extending the effective forecast timeliness to ∼4 weeks. Specifically, the climatology-based Brier Skill Scores are improved by ∼0.4 for the extreme precipitation forecasts almost throughout the whole timescale compared with the ECMWF. Feature importance analyze towards CSG-EMOS model indicates that the atmospheric factors make great contributions to the prediction skill with the increasing lead times. The CSG-UNET method is promising in subseasonal precipitation forecasts and could be applied to the routine forecast of other atmospheric and ocean phenomena in the future.

Краткосрочное и среднесрочное прогнозирование уровней воды на реках бассейна Тобола

In order to provide the necessary hydrological information for operational decisions on the use of water resources and protecting the population from dangerous floods for the rivers of the Tobol river basin, a methodology for daily forecasting of average daily water levels with a lead time of 1 to 10 days has been developed. The hydrograph extrapolation method was used, which takes into account water levels for the date of the forecast and for the 5 previous days. The methodology was developed for 64 sections located on the Tobol River and its tributaries of various orders. To estimate the parameters of the scheme for obtaining forecasts and their verification, hydrological observation data for the period from 1985 to 2022 were used. For most rivers in the basin, the methodology provides satisfactory results. In general, the accuracy of forecasts and their lead time increase with increasing catchment area. The methodology can be used within an automated system for preparing and issuing forecasts.

Circulating tumor DNA (ctDNA) kinetics in colorectal cancer (CRC) treated with curative intent in the VICTORI study with an ultrasensitive MRD assay.

e15625 Background: Detection of molecular residual disease (MRD) with ctDNA is highly prognostic for recurrence in CRC. However, many cancers still recur without detectable ctDNA and increased lead time before clinical recurrence may create a window of opportunity to intervene. Here we apply an ultra-sensitive assay, NeXT Personal, to profile patients in the VICTORI study, a prospective cohort of patients with CRC managed with curative-intent treatment. Methods: To date, the first 33 patients with CRC have undergone panel creation. NeXT Personal was employed to construct personalized liquid biopsy panels for each patient with up to ~1,800 single nucleotide variants (SNV) identified via whole-genome sequencing. For each patient, plasma is collected before curative intervention (baseline), every 2 weeks for 2 months (MRD window), and every 3 months for up to 3 years (surveillance). VICTORI will enroll > 175 patients and include a health economics analysis. Results: Of 33 patients enrolled to date, 25 (76%) are stage I-III and 8 are stage IV treated with curative interventions. 18 colon cancers and 15 rectal cancers are included. Pre-operative baseline sensitivity was 90.6% (29/32; 1 patient excluded as pCR at the time of baseline blood draw). Of the MRD- patients at baseline, 2 had prior neoadjuvant treatment and 1 was stage 1. Preliminary results (median 7.2 month follow-up; 5 recurrences) show 80% (4/5) of recurrences are MRD+ at the 4-week landmark and 50% (2/4; 1 missing sample) were MRD+ at the 2-week landmark. In 75% (3/4) stage II-III recurrences, the first MRD+ detection was in the ultra-low range of < 100 PPM ( < 0.01% tumor fraction). MRD detection at week 4 and 8 trended towards poorer recurrence-free survival (RFS) with similar separation of patients (log rank p = 0.062 and p = 0.011, respectively). Conclusions: Preliminary results from our prospective study demonstrate strong rates of ctDNA detection at week 4 landmark analysis, potentially due to detection of very low levels of ctDNA with an ultra-sensitive MRD assay. The majority of the first detections for MRD were in the ultra-low ctDNA range, indicating the importance of high sensitivity MRD testing. [Table: see text]

Mitigating blade erosion damage through nowcast-driven erosion-safe mode control

The erosion-safe mode (ESM) is a novel mitigation strategy that reduces rainfall-induced erosion damage by lowering the tip-speed of the turbine during precipitation events. The ESM requires accurate information about future expected rainfall for its control. In current research, it is debated what method or source should be used to this end. This study explores the effectiveness of driving the ESM using a state-of-the-art weather-radar-based probabilistic rainfall nowcast provided by the Royal Netherlands Meteorological Institute (KNMI). The performance of the nowcast is assessed for various lead times with an impingement-based damage model for three sample sites in the Netherlands and for two distinct ESM strategies. The results show that the quality of the nowcast degrades with increasing lead times, where the 5- and 15-minute lead times exhibit sufficiently good accuracy and response time for adjusting turbine speeds. Overall, the results highlight that the probabilistic information in the nowcast can be employed to improve the efficiency and viability of the ESM.

Residual Spatiotemporal Convolutional Neural Network Based on Multisource Fusion Data for Approaching Precipitation Forecasting

Approaching precipitation forecast refers to the prediction of precipitation within a short time scale, which is usually regarded as a spatiotemporal sequence prediction problem based on radar echo maps. However, due to its reliance on single-image prediction, it lacks good capture of sudden severe convective events and physical constraints, which may lead to prediction ambiguities and issues such as false alarms and missed alarms. Therefore, this study dynamically combines meteorological elements from surface observations with upper-air reanalysis data to establish complex nonlinear relationships among meteorological variables based on multisource data. We design a Residual Spatiotemporal Convolutional Network (ResSTConvNet) specifically for this purpose. In this model, data fusion is achieved through the channel attention mechanism, which assigns weights to different channels. Feature extraction is conducted through simultaneous three-dimensional and two-dimensional convolution operations using a pure convolutional structure, allowing the learning of spatiotemporal feature information. Finally, feature fitting is accomplished through residual connections, enhancing the model’s predictive capability. Furthermore, we evaluate the performance of our model in 0–3 h forecasting. The results show that compared with baseline methods, this network exhibits significantly better performance in predicting heavy rainfall. Moreover, as the forecast lead time increases, the spatial features of the forecast results from our network are richer than those of other baseline models, leading to more accurate predictions of precipitation intensity and coverage area.

Application of a hybrid algorithm of LSTM and Transformer based on random search optimization for improving rainfall-runoff simulation

Flood forecasting using traditional physical hydrology models requires consideration of multiple complex physical processes including the spatio-temporal distribution of rainfall, the spatial heterogeneity of watershed sub-surface characteristics, and runoff generation and routing behaviours. Data-driven models offer novel solutions to these challenges, though they are hindered by difficulties in hyperparameter selection and a decline in prediction stability as the lead time extends. This study introduces a hybrid model, the RS-LSTM-Transformer, which combines Random Search (RS), Long Short-Term Memory networks (LSTM), and the Transformer architecture. Applied to the typical Jingle watershed in the middle reaches of the Yellow River, this model utilises rainfall and runoff data from basin sites to simulate flood processes, and its outcomes are compared against those from RS-LSTM, RS-Transformer, RS-BP, and RS-MLP models. It was evaluated against RS-LSTM, RS-Transformer, RS-BP, and RS-MLP models using the Nash–Sutcliffe Efficiency Coefficient (NSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Bias percentage as metrics. At a 1-h lead time during calibration and validation, the RS-LSTM-Transformer model achieved NSE, RMSE, MAE, and Bias values of 0.970, 14.001m3/s, 5.304m3/s, 0.501% and 0.953, 14.124m3/s, 6.365m3/s, 0.523%, respectively. These results demonstrate the model's superior simulation capabilities and robustness, providing more accurate peak flow forecasts as the lead time increases. The study highlights the RS-LSTM-Transformer model's potential in flood forecasting and the advantages of integrating various data-driven approaches for innovative modelling.

Research on B2C Cross-Border Electronic Commerce Return Logistics Model Selection Based on Estimated Return Rate

This paper investigates the estimated return rate and optimal order quantity under three cross-border e-commerce return logistics modes: direct mail (from predecessor), in situ destruction (new), and insurance (new). The estimated return rate under each model was analyzed and it was found that different modes have different thresholds in delivery lead time (the time retailers need to deliver goods to customers), and within which the estimated return rate increases as the delivery lead time increases. And a size comparison of the estimated return rates for the three models was conducted. A profit model was constructed based on the estimated return rate model, the optimal order quantity was calculated, and the effects of different factors (tax, postage, and delivery lead time etc.) on it were analyzed. For the insurance model, the effect of bearing the insurance ratio between retailers and consumers on the optimal order quantity was examined. The goal of this paper was to construct a model of the estimated return rate for the two new modes and to compare the estimated return rate of the three modes, which provides a reference for retailers to choose among the diversified return logistics modes and then make the best ordering strategy according to the influence of different factors on the optimal order quantity.

Deep learning-based ensemble forecast and predictability analysis of the Kuroshio intrusion into the South China Sea

Abstract The Kuroshio intrusion (KI) into the South China Sea (SCS) significantly affects the environment, ecology, and climate change of the SCS. However, due to the nonlinearity of KI, its numerical prediction often requires large ensemble size to measure prediction uncertainty. The huge computational costs of large numbers of members and high-resolution numerical models pose significant challenges for KI prediction. We, therefore, construct a Kuroshio ensemble deep learning prediction system (KurNet) through taking different values of parameters to predict KI paths because the deep learning models have high prediction skills and low computational cost. The KurNet containing 64 ensemble members can not only output ensemble mean forecast result of the Kuroshio path, but also estimate probability density functions for the path types. The KurNet illustrates a high predictive ability for the KI with the mean classification accuracy of 71.9% and root mean square error of 0.913 on the testing set, which is superior to the single control prediction by ∼1.0–2.9%, although the control prediction model is selected as one of the ensemble members with the best predictive ability on the validation set. Furthermore, the predictability analysis of 10 KI events indicates that when the lead time is 3 days, the most important areas are in the east of Luzon Island due to the upstream Kuroshio transport. As the lead time increases, the most important area is in the Luzon Strait due to the eddy activity. Observing system simulation experiments reveal that the KI forecast skill can be enhanced by ∼12–18%, when uncertainties of the input data in these important regions are removed.

An explainable multiscale LSTM model with wavelet transform and layer-wise relevance propagation for daily streamflow forecasting

Developing an accurate and reliable daily streamflow forecasting model is important for facilitating the efficient resource planning and management of hydrological systems. In this study, an explainable multiscale long short-term memory (XM-LSTM) model is proposed for effective daily streamflow forecasting by integrating the à trous wavelet transform (ATWT) for decomposing data, the Boruta algorithm for identifying model inputs, and the layer-wise relevance propagation (LRP) for explaining the prediction results. The proposed XM-LSTM is tested by performing multi-step-ahead forecasting of daily streamflow at four stations in the middle and lower reaches of the Yangtze River basin and compared with the X-LSTM. The X-LSTM is formed by coupling the long short-term memory (LSTM) with the LRP. For comparison, the inputs of these two models are identified by the Boruta selection algorithm. The results show that all models exhibit good ability to forecast daily streamflow, however, the prediction performance decreases as the lead time increases. The XM-LSTM provides a better forecasting performance than the X-LSTM, suggesting the ability of the ATWT to improve the LSTM for daily streamflow forecasting. Moreover, the correlation scores analysis by the LRP shows that the ATWT can extract useful information that influences the daily streamflow from the raw predictors, and the water level has the most significant contribution to streamflow prediction. Accordingly, the XM-LSTM model can be viewed as a potentially useful approach for increasing the accuracy and explainability of streamflow forecasting.

Sub-seasonal soil moisture anomaly forecasting using combinations of deep learning, based on the reanalysis soil moisture records

Sub-seasonal drought forecasting is crucial for early warning in estimating agricultural production and optimizing irrigation management, as forecasting skills are relatively weak during this period. Soil moisture exhibits stronger persistence compared to other climate system quantities, which makes it especially influential in shaping land-atmosphere feedback, thus supplying a unique insight into drought forecasting. Relying on the soil moisture memory, this study investigates the combination of multiple deep-learning modules for sub-seasonal drought indices hindcast in the Huai River basin of China, using long-term ERA5-Land soil moisture records with a noise-assisted data analysis tool. The inter-compared deep-learning models include a hybrid model and a committee machine framework. The results show that the performance of the committee machine framework can be improved with the help of series decomposition and the forecasting skill is not impaired with the lead time increases. Overall, this study highlights the potential of combining deep-learning models with soil moisture memory analysis to improve sub-seasonal drought forecasting.

Robustness of hydrometeorological extremes in surrogated seasonal forecasts

AbstractWater and disaster risk management require accurate information about hydrometeorological extremes. However, estimation of rare events using extreme value analysis is hampered by short observational records, with large resulting uncertainties. Here, we present a surrogate world setup that makes use of data samples from meteorological and hydrological seasonal re‐forecasts to explore extremes for long return periods. The surrogate timeseries allow us to pool the re‐forecasts into 1000‐year‐long timeseries. We can then calculate return values of extremes and explore how they are affected by the size of sub‐samples as method for estimating the uncertainty. The approach relies on the fact that probabilistic seasonal re‐forecasts, initialized with perturbed initial conditions, have limited predictive skill with increasing lead time. At long lead times re‐forecasts will diverge into independent samples. The meteorological seasonal re‐forecasts are taken from the SEAS5 system, and hydrological re‐forecasts are generated with the E‐HYPE process‐based model for the pan‐European domain. Extreme value analysis is applied to annual maxima of precipitation and streamflow for return periods of 100 years. The analysis clearly demonstrates the large uncertainty in long return period estimates with typical available samples of only few decades. The uncertainty is somewhat reduced for 100‐year samples, but several 100 years seem to be necessary to have robust estimates. The bootstrap with replacement approach is applied to shorter timeseries, and is shown to well reproduce the uncertainty range of the longer samples. However, the main estimate of the return value can be significantly offset. Although the method is model based, with the associated uncertainties and bias compared to the real world, the surrogate approach is likely useful to explore rare and compounding extremes.

Hydrological Verification of Two Rainfall Short-Term Forecasting Methods with Floods Anticipation Perspective

Abstract Flood forecasting remains a significant challenge, particularly when dealing with basins characterized by small drainage areas (i.e., 103 km2 or lower with response time in the range 0.5–10 h) especially because of the rainfall prediction uncertainties. This study aims to investigate the performances of streamflow predictions using two short-term rainfall forecast methods. These methods utilize a combination of a nowcasting extrapolation algorithm and numerical weather predictions by employing a three-dimensional variational assimilation system, and nudging assimilation techniques, meteorological radar, and lightning data that are frequently updated, allowing new forecasts with high temporal frequency (i.e., 1–3 h). A distributed hydrological model is used to convert rainfall forecasts into streamflow prediction. The potential of assimilating radar and lightning data, or radar data alone, is also discussed. A hindcast experiment on two rainy periods in the northwest region of Italy was designed. The selected skill scores were analyzed to assess their degradation with increasing lead time, and the results were further aggregated based on basin dimensions to investigate the catchment integration effect. The findings indicate that both rainfall forecast methods yield good performance, with neither definitively outperforming the other. Furthermore, the results demonstrate that, on average, assimilating both radar and lightning data enhances the performance.