共查询到20条相似文献，搜索用时 22 毫秒
1.
人工神经网络预报模型的过拟合研究 总被引：35，自引：0，他引：35
针对神经网络方法在预报建模中存在的“过拟合”(overfitting)现象和提高泛化性能 (generalizationcapability)问题 ,提出了采用主成分分析构造神经网络低维学习矩阵的预报建模方法。研究结果表明 ,这种新的神经网络预报建模方法 ,通过浓缩预报信息 ,降维去噪 ,使得神经网络的预报建模不需要进行适宜隐节点数的最优网络结构试验 ,没有“过拟合”现象 ,并且与传统的神经网络预报建模方法及逐步回归预报模型相比泛化能力有显著提高 相似文献
2.
云对中国区域卫星观测臭氧总量精度影响的检验分析 总被引：5，自引：0，他引：5
根据卫星和地基观测, 比较了我国香河、 昆明、 瓦里关和龙凤山四个站点臭氧总量自1979年以来的变化。卫星与地基观测的臭氧总量长期趋势比较一致, 表明臭氧总量均有下降趋势, 但是卫星与地基各自观测的结果仍存在着显著的差别。为研究卫星与地基臭氧总量的差别, 以地基观测臭氧总量为参考, 检验云对历史TOMS (Total Ozone Mapping Spectrometer) 和GOME (Global Ozone Monitoring Experiment) 臭氧总量精度的影响。结果显示: 云 (云量或云顶高度) 增加了卫星臭氧总量误差, 降低数据精度。随着地面云量的增加, TOMS、 GOME臭氧总量相对误差在上述四个地点呈现明显的上升趋势 (瓦里关最为明显), 但最大变化幅度没有超过2.0%。TOMS臭氧总量相对误差随地面云量变化呈现区域性特点, 香河与龙凤山 (代表着中纬度高臭氧总量区域)、 昆明与瓦里关 (代表中、 低纬度高原低臭氧总量区域) 分别为两个变化特点接近的区域。GOME臭氧总量相对误差与云之间关系的区域特征不明显。利用卫星遥测FRESCO (Fast Retrieval Scheme for Clouds from the Oxygen A\|band) 云信息检验GOME卫星臭氧总量精度的表明, 只有当云量大于5成后GOME臭氧总量才显示出相对误差增加的现象, 但无明显趋势; 随着FRESCO云顶高度的增加, GOME臭氧相对误差在香河、 瓦里关均呈现明显的上升趋势并有3%左右幅度的变化。TOMS臭氧总量相对误差随着地面有效反射率的增加而增大, 且误差幅度超过2%; TOMS\|N7臭氧总量比TOMS\|EP约高2.0%~3.0%。分析还表明, 云内和云以下臭氧柱浓度在反演的卫星臭氧总量中的贡献很可能被高估了。 相似文献
3.
JIN Long KUANG Xueyuan HUANG Haihong QIN Zhinian WANG Yehong 《Acta Meteorologica Sinica》2005,19(2):216-225
Because of overfitting and the improvement of generalization capability (GC) available in the construction of forecasting models using artificial neural network (ANN), a new method is proposed for model establishment by means of making a low-dimension ANN learning matrix through principal component analysis (PCA). The results show that the PCA is able to construct an ANN model without the need of finding an optimal structure with the appropriate number of hidden-layer nodes, thus avoids overfitting by condensing forecasting information, reducing dimension and removing noise, and GC is greatly raised compared to the traditional ANN and stepwise regression techniques for model establishment. 相似文献
4.
Temporal Variations in the Vertical Distribution of Stratospheric Ozone over Obninsk from Lidar Data
The results of lidar measurements of ozone profiles over Obninsk in the altitude range of 12–35 km in 2012–2016 are presented. Temporal variations in total ozone in the above altitude range and seasonal variations in the vertical distribution of ozone are considered. Basic attention is paid to the analysis of ozone profile variations on the daily and weekly scales. The backtrajectory analysis demonstrated that in most cases the formation of layers with low or high ozone values is explained by the direction of meridional advection. Cross-correlation coefficients for the variations in ozone and temperature relative to the current monthly mean variations are calculated. Rather high values of correlation coefficients (~0.4–0.6) are obtained for summer in the low stratosphere (100 and 160 hPa) and for winter in the upper troposphere (50 and 20 hPa). In general, variations in ozone profiles are consistent with available climatologic data. 相似文献
5.
Manish Kumar Goyal 《Theoretical and Applied Climatology》2014,118(1-2):25-34
Rainfall is a principal element of the hydrological cycle and its variability is important from both the scientific as well as practical point of view. Wavelet regression (WR) technique is proposed and developed to analyze and predict the rainfall forecast in this study. The WR model is improved combining two methods, discrete wavelet transform and linear regression model. This study uses rainfall data from 21 stations in Assam, India over 102 years from 1901 to 2002. The calibration and validation performance of the models is evaluated with appropriate statistical methods. The root mean square errors (RMSE), N-S index, and correlation coefficient (R) statistics were used for evaluating the accuracy of the WR models. The accuracy of the WR models was then compared with those of the artificial neural networks (ANN) models. The results of monthly rainfall series modeling indicate that the performances of wavelet regression models are found to be more accurate than the ANN models. 相似文献
6.
Predictor selection is a critical factor affecting the statistical downscaling of daily precipitation. This study provides a general comparison between uncertainties in downscaled results from three commonly used predictor selection methods (correlation analysis, partial correlation analysis, and stepwise regression analysis). Uncertainty is analyzed by comparing statistical indices, including the mean, variance, and the distribution of monthly mean daily precipitation, wet spell length, and the number of wet days. The downscaled results are produced by the artificial neural network (ANN) statistical downscaling model and 50 years (1961–2010) of observed daily precipitation together with reanalysis predictors. Although results show little difference between downscaling methods, stepwise regression analysis is generally the best method for selecting predictors for the ANN statistical downscaling model of daily precipitation, followed by partial correlation analysis and then correlation analysis. 相似文献
7.
Urbanisation has burdened cities with many problems associated with growth and the physical environment. Some of the urban locations in India are becoming increasingly vulnerable to natural hazards related to precipitation and flooding. Thus it becomes increasingly important to study the characteristics of these events and their physical explanation. This work studies rainfall trends in Delhi and Mumbai, the two biggest Metropolitan cities of Republic of India, during the period from 1951 to 2004. Precipitation data was studied on basis of months, seasons and years, and the total period divided in the two different time periods of 1951–1980 and 1981–2004 for detailed analysis. Long-term trends in rainfall were determined by Man-Kendall rank statistics and linear regression. Further this study seeks for an explanation for precipitation trends during monsoon period by different global climate phenomena. Principal component analysis and Singular value decomposition were used to find relation between southwest monsoon precipitation and global climatic phenomena using climatic indices. Most of the rainfall at both the stations was found out to be taking place in Southwest monsoon season. The analysis revealed great degree of variability in precipitation at both stations. There is insignificant decrease in long term southwest monsoon rainfall over Delhi and slight significant decreasing trends for long term southwest monsoon rainfall in Mumbai. Decrease in average maximum rainfall in a day was also indicated by statistical analysis for both stations. Southwest monsoon precipitation in Delhi was found directly related to Scandinavian Pattern and East Atlantic/West Russia and inversely related to Pacific Decadal Oscillation, whereas precipitation in Mumbai was found inversely related to Indian ocean dipole, El Ni?o- Southern Oscillation and East Atlantic Pattern. 相似文献
8.
漠河地区臭氧的观测和计算 总被引：2，自引：0，他引：2
1997年3月上旬,在黑龙江漠河地区对地面和整层臭氧、太阳辐射等进行了短期观测,以初步了解该地区臭氧和辐射的变化规律以及它们之间的相互关系.研究发现,漠河地区近地面臭氧日变化明显,其峰值出现在每日10:00(北京时间)左右,并早于紫外辐射(UV)峰值出现时间.整层大气臭氧总量的日变化特征不明显.基于UV能量守恒,建立了臭氧与其影响因子-光化学、散射、UV等因子之间较好的定量关系和经验模式,并将其用于计算地面、整层大气臭氧小时值和日平均值.结果表明,计算值与观测值吻合的都比较好,它们相对偏差的平均值分别为:地面臭氧小时值(11.9%)和日平均值(9.0%);整层大气臭氧小时值和日平均值-7.4%、1.8%.因此,地面和整层臭氧的经验算法是合理和可行的.利用散射辐射/直接辐射(D/S)和散射辐射/总辐射(D/Q)可以描述大气中的物质如气溶胶、云等的散射作用.采用D/Q表示散射作用可以提高地面臭氧和整层大气臭氧计算的准确度,特别是对云量较大的情况.
相似文献
9.
FY-3A气象卫星臭氧总量数据的质量控制方案及其在台风Tembin(2012)和Isaac(2012)中的应用 总被引：1，自引：0，他引：1
我国第二代极轨气象卫星“风云三号”A星(FY-3A)上搭载的紫外臭氧总量探测仪(Total Ozone Unit,TOU)每天可以提供一次覆盖全球的臭氧总量观测。为了在数值预报中应用TOU的臭氧资料,从资料同化角度发展了一套质量控制方案。首先基于臭氧总量和平均位势涡度的高相关性建立了逐日动态更新的臭氧线性回归预报模型,然后使用双权重算法对臭氧资料进行质量控制。将该质量控制方案应用于台风Tembin(2012)和Isaac(2012)个例,试验结果说明该方案可以体现出臭氧总量和平均位势涡度之间相关关系的逐日变化,识别出的离群资料百分比随时间变化较稳定,可以保留原始资料的主体信息,并且显著降低了原始资料的标准差。同时,质量控制后的臭氧数据与统计拟合量更加一致,观测减拟合的概率密度函数分布形式也更接近高斯分布,有利于后续的资料同化。 相似文献
10.
城市低层大气臭氧生成的模拟研究 总被引：7，自引：0，他引：7
利用一个城市边界层臭氧模式，以南京市为对象，研究了臭氧在城市低层大气中的形成机制和规律，得出了南京市18个交通点的日臭氧浓度最大值的分布图。结果表明，南京市白天有相当长的时间是处于臭氧超标的状态，不少交通点的日最大臭氧浓度值是标准的2－3倍。文章还模拟了PAN，OH自由基和HCHO这三种重要物种的日变化规律。同时，本文选择南京市东郊的紫金山风景区，实次探讨了植物排放有机物对臭氧生成的贡献。从模式的模拟研究可得，与人为因素相比，植物排放对臭氧生成的作用也是不可忽视的，值得作深入一步的探讨。 相似文献
11.
A Tibetan ozone low was found in the 1990s after the Antarctic ozone hole.Whether this ozone low has been recovering from the beginning of the 2000s following the global ozone recovery is an intriguing topic.With the most recent merged TOMS/SBUV(Total Ozone Mapping Spectrometer/Solar Backscatter Ultra Violet) ozone data,the Tibetan ozone low and its long-term variation during 1979-2010 are analyzed using a statistical regression model that includes the seasonal cycle,solar cycle,quasi-biennial oscillation(QBO),ENSO signal,and trends.The results show that the Tibetan ozone low maintains and may become more severe on average during 1979-2010,compared with its mean state in the periods before 2000,possibly caused by the stronger downward trend of total ozone concentration over the Tibet.Compared with the ozone variation over the non-Tibetan region along the same latitudes,the Tibetan ozone has a larger downward trend during 1979-2010,with a maximum value of-0.40±0.10 DU yr 1 in January,which suggests the strengthening of the Tibetan ozone low in contrast to the recovery of global ozone.Regression analyses show that the QBO signal plays an important role in determining the total ozone variation over the Tibet.In addition,the long-term ozone variation over the Tibetan region is largely affected by the thermal-dynamical proxies such as the lower stratospheric temperature,with its contribution reaching around 10% of the total ozone change,which is greatly different from that over the non-Tibetan region. 相似文献
12.
AbstractThe solar backscattered ultraviolet (SBUV/SBUV-2) merged ozone datasets, version 8.6, including column ozone and ozone profiles for the 1979–2012 period are examined for the 35°N–60°N zonal belt in the northern hemisphere mid-latitudes and four sub-regions: central Europe, continental Europe, North America, and East Asia. The residual long-term patterns for total ozone and ozone profiles are extracted by smoothing the time series of differences between the original and the modelled ozone time series. Modelled ozone is obtained using the standard trend model accounting for ozone variability due to changes in stratospheric halogens and various dynamical factors commonly used in previous ozone trend analyses. Since about 2005 spring and summer total ozone in the troposphere and lower stratosphere has decreased in some regions (central and continental Europe, North America, and the 35°N–60°N zonal belt) compared with modelled ozone. The negative departure from modelled ozone in 2010 is approximately 2–3% of the overall 1979–2012 monthly mean level. It seems that this decrease is a result of yet unknown dynamical processes rather than to chemical destruction because the differences have a longitudinal structure, and total ozone in the upper stratosphere still follows changes in stratospheric halogen loading. 相似文献
13.
ＳｉｇｎａｔｕｒｅｓｏｆａＵｎｉｖｅｒｓａｌＳｐｅｃｔｒｕｍｆｏｒＮｏｎｌｉｎｅａｒＶａｒｉａｂｉｌｉｔｙｉｎＤａｉｌｙＣｏｌｕｍｎａｒＴｏｔａｌＯｚｏｎｅＣｏｎｔｅｎｔＡ．Ｍ．ＳｅｌｖａｍａｎｄＭ．ＲａｄｈａｍａｎｉＳｉｇｎａｔｕｒｅｓｏｆａＵｎｉ... 相似文献
14.
This study investigates the ability of two different artificial neural network (ANN) models, generalized regression neural networks model (GRNNM) and Kohonen self-organizing feature maps neural networks model (KSOFM), and two different adaptive neural fuzzy inference system (ANFIS) models, ANFIS model with sub-clustering identification (ANFIS-SC) and ANFIS model with grid partitioning identification (ANFIS-GP), for estimating daily dew point temperature. The climatic data that consisted of 8 years of daily records of air temperature, sunshine hours, wind speed, saturation vapor pressure, relative humidity, and dew point temperature from three weather stations, Daego, Pohang, and Ulsan, in South Korea were used in the study. The estimates of ANN and ANFIS models were compared according to the three different statistics, root mean square errors, mean absolute errors, and determination coefficient. Comparison results revealed that the ANFIS-SC, ANFIS-GP, and GRNNM models showed almost the same accuracy and they performed better than the KSOFM model. Results also indicated that the sunshine hours, wind speed, and saturation vapor pressure have little effect on dew point temperature. It was found that the dew point temperature could be successfully estimated by using T mean and R H variables. 相似文献
15.
Global solar radiation (GSR) is essential for agricultural and plant growth modelling, air and water heating analyses, and solar electric power systems. However, GSR gauging stations are scarce compared with stations for monitoring common meteorological variables such as air temperature and relative humidity. In this study, one power function, three linear regression, and three non-linear models based on an artificial neural network (ANN) are developed to extend short records of daily GSR for meteorological stations where predictors (i.e., temperature and/or relative humidity) are available. The seven models are then applied to 19 meteorological stations located across the province of Quebec (Canada). On average, the root-mean-square errors (RMSEs) for ANN-based models are 0.33–0.54?MJ?m?2?d?1 smaller than those for the power function and linear regression models for the same input variables, indicating that the non-linear ANN-based models are more efficient in simulating daily GSR. Regionalization potential of the seven models is also evaluated for ungauged stations where predictors are available. The power function and the three linear regression models are tested by interpolating spatially correlated at-site coefficients using universal kriging or by applying a leave-one-out calibration procedure for spatially uncorrelated at-site coefficients. Regional ANN-based models are also developed by training the model based on the leave-one-out procedure. The RMSEs for regional ANN models are 0.08–0.46?MJ?m?2?d?1 smaller than for other models using the same input conditions. However, the regional ANN-based models are more sensitive to new station input values compared with the other models. Maps of interpolated coefficients and regional equations of the power function and the linear regression models are provided for direct application to the study area. 相似文献
16.
Dieter H. W. Peters Andrea Schneidereit Marianne Bügelmayer Christoph Zülicke Ingo Kirchner 《大气与海洋》2015,53(1):74-88
AbstractIn a sensitivity study, the influence of an observed stratospheric zonal ozone anomaly on the atmospheric circulation was investigated using the Fifth Generation European Centre Hamburg Model (ECHAM5) which is a general circulation model. The model was run from 1960 to 1999 (40 years) with a mean seasonal cycle of zonally symmetric ozone. In order to isolate the induced dynamical influence of the observed zonally asymmetric part of the three-dimensional stratospheric ozone, a second run was performed for the boreal extratropics using prescribed monthly means from the 40-year reanalysis dataset from the European Centre for Medium-range Weather Forecasts (ERA-40). The main findings are the interdecadal westward shift of the polar vortex at about 65°N and a significant increase in the number of stratospheric sudden warmings during the 1980–99 period. Under the action of zonally asymmetric ozone a decrease in the Arctic Oscillation was identified between the mid-1980s and the mid-1990s. The lag correlation between the mean Arctic Oscillation at the surface and the daily stratospheric northern annular mode increased in mid-winter. Furthermore, we examined the influence of the stratospheric zonal ozone anomaly on Rossby wave breaking in the upper troposphere and found a significant westward shift of poleward Rossby wave breaking events over western Europe in the winter. By this we show that the stratospheric zonal ozone anomaly has a strong influence on the tropospheric circulation as a result of enhanced dynamical coupling processes. 相似文献
17.
The aim of the present study is to develop an adaptive neuro-fuzzy inference system (ANFIS) to forecast the peak gust speed associated with thunderstorms during the pre-monsoon season (April?CMay) over Kolkata (22°32??N, 88°20??E), India. The pre-monsoon thunderstorms during 1997?C2008 are considered in this study to train the model. The input parameters are selected from various stability indices using statistical skill score analysis. The most useful and relevant stability indices are taken to form the input matrix of the model. The forecast through the hybrid ANFIS model is compared with non-hybrid radial basis function network (RBFN), multi layer perceptron (MLP) and multiple linear regression (MLR) models. The forecast error analyses of the models in the test cases reveal that ANFIS provides the best forecast of the peak gust speed with 3.52% error, whereas the errors with RBFN, MLP, and MLR models are 10.48, 11.57, and 12.51%, respectively. During the validation with the 2009 observations of the India Meteorological Department (IMD), the ANFIS model confirms its superiority over other comparative models. The forecast error during the validation of the ANFIS model is observed to be 3.69%, with a lead time of <12?h, whereas the errors with RBFN, MLP, and MLR are 12.25, 13.19, and 14.86%, respectively. The ANFIS model may, therefore, be used as an operational model for forecasting the peak gust speed associated with thunderstorms over Kolkata during the pre-monsoon season. 相似文献
18.
19.
20.
Given the coarse resolution of global climate models, downscaling techniques are often needed to generate finer scale projections of variables affected by local-scale processes such as precipitation. However, classical statistical downscaling experiments for future climate rely on the time-invariance assumption as one cannot know the true change in the variable of interest, nor validate the models with data not yet observed. Our experimental setup involves using the Canadian regional climate model (CRCM) outputs as pseudo-observations to estimate model performance in the context of future climate projections by replacing historical and future observations with model simulations from the CRCM, nested within the domain of the Canadian global climate model (CGCM). In particular, we evaluated statistically downscaled daily precipitation time series in terms of the Peirce skill score, mean absolute errors, and climate indices. Specifically, we used a variety of linear and nonlinear methods such as artificial neural networks (ANN), decision trees and ensembles, multiple linear regression, and k-nearest neighbors to generate present and future daily precipitation occurrences and amounts. We obtained the predictors from the CGCM 3.1 20C3M (1971–2000) and A2 (2041–2070) simulations, and precipitation outputs from the CRCM 4.2 (forced with the CGCM 3.1 boundary conditions) as predictands. Overall, ANN models and tree ensembles outscored the linear models and simple nonlinear models in terms of precipitation occurrences, without performance deteriorating in future climate. In contrast, for the precipitation amounts and related climate indices, the performance of downscaling models deteriorated in future climate. 相似文献