ConvLSTM Based Temperature Forecast Modification Model for North China

The correction of model forecast is an important step in evaluating weather forecast results.In recent years,post-processing models based on deep learning have become prominent.In this paper,a deep learning model named EDConvLSTM based on encoder-decoder structure and ConvLSTM is developed,which appears to be able to effectively correct numerical weather forecasts.Compared with traditional post-processing methods and convolutional neural networks,ED-ConvLSTM has strong collaborative extraction ability to effectively extract the temporal and spatial features of numerical weather forecasts and fit the complex nonlinear relationship between forecast field and observation field.In this paper,the post-processing method of ED-ConvLSTM for 2 m temperature prediction is tested using The International Grand Global Ensemble dataset and ERA5-Land data from the European Centre for Medium-Range Weather Forecasts(ECMWF).Root mean square error and temperature prediction accuracy are used as evaluation indexes to compare ED-ConvLSTM with the method of model output statistics,convolutional neural network postprocessing methods,and the original prediction by the ECMWF.The results show that the correction effect of EDConvLSTM is better than that of the other two postprocessing methods in terms of the two indexes,especially in the long forecast time.

An ANN-Based Short-Term Temperature Forecast Model for Mass Concrete Cooling Control

Concrete temperature control during dam construction(e.g.,concrete placement and curing)is important for cracking prevention.In this study,a short-term temperature forecast model for mass concrete cooling control is developed using artificial neural networks(ANN).The development workflow for the forecast model consists of data integration,data preprocessing,model construction,and model application.More than 80000 monitoring samples are collected by the developed intelligent cooling control system in the Baihetan Arch Dam,which is the largest hydropower project in the world under construction.Machine learning algorithms,including ANN,support vector machines,long short-term memory networks,and decision tree structures,are compared in temperature prediction,and the ANN is determined to be the best for the forecast model.Furthermore,an ANN framework with two hidden layers is determined to forecast concrete temperature at intervals of one day.The root mean square error of the forecast precision is 0.15∘C on average.The application on concrete blocks verifies that the developed ANN-based forecast model can be used for intelligent cooling control during mass concrete construction.

Regional Temperature Forecast for the Next Day in Hong Kong

For a century or so, the Hong Kong Observatory （HKO） has been providing temperature forecast for the whole of Hong Kong with the HKO Headquarters as the reference location. In recent decades, due to spreading of population from the main urban center to satellite towns, there is an increasing demand for regional temperature forecasts. To support such provision, the HKO has developed a regression model to provide objective guidance to forecasters in formulating forecasts of maximum and minimum temperatures for the next day at various locations in Hong Kong. In this paper, the regression model is presented, together with the assessment of its performance. Based on the verification of one year of forecasts, it is found that the root mean square errors （RMSEs） of maximum （minimum） temperature forecasts are from about 1.3 to 2.1 （1.1 to 1.4） degrees, respectively. The regression model is shown to have generally out-performed the operational regional spectral model then operated by HKO. Regional temperature forecast methods of other meteorological or research centers are also surveyed. Equipped with the regression model, the HKO has launched an online regional temperature forecast service for the next day in Hong Kong since March 2008.

Bias correction of sea surface temperature retrospective forecasts in the South China Sea

Offline bias correction of numerical marine forecast products is an effective post-processing means to improve forecast accuracy. Two offline bias correction methods for sea surface temperature(SST) forecasts have been developed in this study: a backpropagation neural network(BPNN) algorithm, and a hybrid algorithm of empirical orthogonal function(EOF) analysis and BPNN(named EOF-BPNN). The performances of these two methods are validated using bias correction experiments implemented in the South China Sea(SCS), in which the target dataset is a six-year(2003–2008) daily mean time series of SST retrospective forecasts for one-day in advance, obtained from a regional ocean forecast and analysis system called the China Ocean Reanalysis(CORA),and the reference time series is the gridded satellite-based SST. The bias-correction results show that the two methods have similar good skills;however, the EOF-BPNN method is more than five times faster than the BPNN method. Before applying the bias correction, the basin-wide climatological error of the daily mean CORA SST retrospective forecasts in the SCS is up to-3°C;now, it is minimized substantially, falling within the error range(±0.5°C) of the satellite SST data.

The 3-Hour-Interval Prediction of Ground-Level Temperature in South Korea Using Dynamic Linear Models

The 3-hour-interval prediction of ground-level temperature from +00 h out to +45 h in South Korea (38 stations) is performed using the DLM (dynamic linear model) in order to eliminate the systematic error of numerical model forecasts. Numerical model forecasts and observations are used as input values of the DLM. According to the comparison of the DLM forecasts to the KFM (Kalman filter model) forecasts with RMSE and bias, the DLM is useful to improve the accuracy of prediction.

IMPACT OF SUMMER WARMING ON DYNAMICS-STATISTICS-COMBINED METHOD TO PREDICT THE SUMMER TEMPERATURE IN CHINA

Based on NCEP/NCAR daily reanalysis data, climate trend rate and other methods are used to quantitatively analyze the change trend of China's summer observed temperature in 1983—2012. Moreover, a dynamics-statistics-combined seasonal forecast method with optimal multi-factor portfolio is applied to analyze the impact of this trend on summer temperature forecast. The results show that: in the three decades, the summer temperature shows a clear upward trend under the condition of global warming, especially over South China, East China, Northeast China and Xinjiang Region, and the trend rate of national average summer temperature was 0.27℃ per decade. However, it is found that the current business model forecast(Coupled Global Climate Model) of National Climate Centre is unable to forecast summer warming trends in China, so that the post-processing forecast effect of dynamics-statistics-combined method is relatively poor. In this study, observed temperatures are processed first by removing linear fitting trend, and then adding it after forecast to offset the deficiency of model forecast indirectly. After test, ACC average value in the latest decade was 0.44 through dynamics-statistics-combined independent sample return forecast. The temporal correlation(TCC) between forecast and observed temperature was significantly improved compared with direct forecast results in most regions, and effectively improved the skill of the dynamics-statistics-combined forecast method in seasonal temperature forecast.

Verification of an operational ocean circulation-surface wave coupled forecasting system for the China＇s seas

An operational ocean circulation-surface wave coupled forecasting system for the seas off China and adjacent areas（OCFS-C） is developed based on parallelized circulation and wave models. It has been in operation since November 1, 2007. In this paper we comprehensively present the simulation and verification of the system, whose distinguishing feature is that the wave-induced mixing is coupled in the circulation model. In particular, with nested technique the resolution in the China＇s seas has been updated to（1/24）° from the global model with（1/2）°resolution. Besides, daily remote sensing sea surface temperature（SST） data have been assimilated into the model to generate a hot restart field for OCFS-C. Moreover, inter-comparisons between forecasting and independent observational data are performed to evaluate the effectiveness of OCFS-C in upper-ocean quantities predictions, including SST, mixed layer depth（MLD） and subsurface temperature. Except in conventional statistical metrics, non-dimensional skill scores（SS） is also used to evaluate forecast skill. Observations from buoys and Argo profiles are used for lead time and real time validations, which give a large SS value（more than 0.90）. Besides, prediction skill for the seasonal variation of SST is confirmed. Comparisons of subsurface temperatures with Argo profiles data indicate that OCFS-C has low skill in predicting subsurface temperatures between 100 m and 150 m. Nevertheless, inter-comparisons of MLD reveal that the MLD from model is shallower than that from Argo profiles by about 12 m, i.e., OCFS-C is successful and steady in MLD predictions. Validation of 1-d, 2-d and 3-d forecasting SST shows that our operational ocean circulation-surface wave coupled forecasting model has reasonable accuracy in the upper ocean.

气溶胶:导致全球天气预报模式中气温预报偏差的关键因素

天气预报与人们的日常生活息息相关.尽管过去几十年大气动力模式和高性能计算技术快速发展,当前天气预报的准确度在部分地区依然存在较大差异.本研究系统比较了全球3年的短期(1-5天)预报和同化气象观测的再分析资料,发现气温预报偏差和大气中的气溶胶存在显著关联.在人为或自然排放密集的地区和季节(如化石燃料燃烧排放量巨大的中国和印度、生物质燃烧频发的南非和亚马逊等),气温预报往往存在更大的偏差,且随着预报时长而显著放大.虽然与空气污染相关的大气化学及气溶胶的理化过程在气候模式中已经普遍得到显式的表达和解析,但在天气预报模式中尚未引起足够的重视.本文以直接的"观测"证据揭示了空气污染对全球天气预报的影响,进一步证明了"化学天气"及理化过程相互作用研究的重要性.