Exploratory Data Analysis of Accuracy of US Weather Forecastes

Abstract. Since 2005, one-hour temperature forecasts for the Calabria region (southern Italy), modelled by the Regional Atmospheric Modeling System (RAMS), have been issued by CRATI/ISAC-CNR (Consortium for Research and Application of Innovative Technologies/Institute for Atmospheric and Climate Sciences of the National Research Council) and are available online at http://meteo.crati.it/previsioni.html (every six hours). Beginning in June 2008, the horizontal resolution was enhanced to 2.5 km. In the present paper, forecast skill and accuracy are evaluated out to four days for the 2008 summer season (from 6 June to 30 September, 112 runs). For this purpose, gridded high horizontal resolution forecasts of minimum, mean, and maximum temperatures are evaluated against gridded analyses at the same horizontal resolution (2.5 km). Gridded analysis is based on Optimal Interpolation (OI) and uses the RAMS first-day temperature forecast as the background field. Observations from 87 thermometers are used in the analysis system. The analysis error is introduced to quantify the effect of using the RAMS first-day forecast as the background field in the OI analyses and to define the forecast error unambiguously, while spatial interpolation (SI) analysis is considered to quantify the statistics' sensitivity to the verifying analysis and to show the quality of the OI analyses for different background fields. Two case studies, the first one with a low (less than the 10th percentile) root mean square error (RMSE) in the OI analysis, the second with the largest RMSE of the whole period in the OI analysis, are discussed to show the forecast performance under two different conditions. Cumulative statistics are used to quantify forecast errors out to four days. Results show that maximum temperature has the largest RMSE, while minimum and mean temperature errors are similar. For the period considered, the OI analysis RMSEs for minimum, mean, and maximum temperatures vary from 1.8, 1.6, and 2.0 °C, respectively, for the first-day forecast, to 2.0, 1.9, and 2.6 °C, respectively, for the fourth-day forecast. Cumulative statistics are computed using both SI and OI analysis as reference. Although SI statistics likely overestimate the forecast error because they ignore the observational error, the study shows that the difference between OI and SI statistics is less than the analysis error. The forecast skill is compared with that of the persistence forecast. The Anomaly Correlation Coefficient (ACC) shows that the model forecast is useful for all days and parameters considered here, and it is able to capture day-to-day weather variability. The model forecast issued for the fourth day is still better than the first-day forecast of a 24-h persistence forecast, at least for mean and maximum temperature. The impact of using the RAMS first-day forecast as the background field in the OI analysis is quantified by comparing statistics computed with OI and SI analyses. Minimum temperature is more sensitive to the change in the analysis dataset as a consequence of its larger representative error.

This paper examines an archive containing over 40 years of 8-day atmospheric forecasts over the contiguous United States from the NCEP reanalysis project to assess the possibilities for using medium-range numerical weather prediction model output for predictions of streamflow. This analysis shows the biases in the NCEP forecasts to be quite extreme. In many regions, systematic precipitation biases exceed 100% of the mean, with temperature biases exceeding 3°C. In some locations, biases are even higher. The accuracy of NCEP precipitation and 2-m maximum temperature forecasts is computed by interpolating the NCEP model output for each forecast day to the location of each station in the NWS cooperative network and computing the correlation with station observations. Results show that the accuracy of the NCEP forecasts is rather low in many areas of the country. Most apparent is the generally low skill in precipitation forecasts (particularly in July) and low skill in temperature forecasts in the western United States, the eastern seaboard, and the southern tier of states. These results outline a clear need for additional processing of the NCEP Medium-Range Forecast Model (MRF) output before it is used for hydrologic predictions. Techniques of model output statistics (MOS) are used in this paper to downscale the NCEP forecasts to station locations. Forecasted atmospheric variables (e.g., total column precipitable water, 2-m air temperature) are used as predictors in a forward screening multiple linear regression model to improve forecasts of precipitation and temperature for stations in the National Weather Service cooperative network. This procedure effectively removes all systematic biases in the raw NCEP precipitation and temperature forecasts. MOS guidance also results in substantial improvements in the accuracy of maximum and minimum temperature forecasts throughout the country. For precipitation, forecast improvements were less impressive. MOS guidance increases the accuracy of precipitation forecasts over the northeastern United States, but overall, the accuracy of MOS-based precipitation forecasts is slightly lower than the raw NCEP forecasts. Four basins in the United States were chosen as case studies to evaluate the value of MRF output for predictions of streamflow. Streamflow forecasts using MRF output were generated for one rainfall-dominated basin (Alapaha River at Statenville, Georgia) and three snowmelt-dominated basins (Animas River at Durango, Colorado; East Fork of the Carson River near Gardnerville, Nevada; and Cle Elum River near Roslyn, Washington). Hydrologic model output forced with measured-station data were used as "truth" to focus attention on the hydrologic effects of errors in the MRF forecasts. Eight-day streamflow forecasts produced using the MOS-corrected MRF output as input (MOS) were compared with those produced using the climatic Ensemble Streamflow Prediction (ESP) technique. MOS-based streamflow forecasts showed increased skill in the snowmelt-dominated river basins, where daily variations in streamflow are strongly forced by temperature. In contrast, the skill of MOS forecasts in the rainfall-dominated basin (the Alapaha River) were equivalent to the skill of the ESP forecasts. Further improvements in streamflow forecasts require more accurate local-scale forecasts of precipitation and temperature, more accurate specification of basin initial conditions, and more accurate model simulations of streamflow.

How does inclusion of weather forecasting impact in-season crop model predictions?

We have investigated the performance of four bias correction methods for improving both maximum and minimum temperature forecasts produced by the NWP model. The objective of bias correction is to minimize the systematic error of the next forecast using bias from past errors. The need for bias corrections arises from the many sources of systematic errors in NWP modeling systems. NWP models have shortcomings in the physical parameterization of weather events and have the inability to handle sub-grid phenomena successfully. The statistical algorithms used for minimizing the bias of the next forecast are Running-Mean (RM) bias correction, Best Easy Systematic (BES) estimator, simple Linear Regression (LR) and the Nearest Neighborhood Weighted (NNW) mean, as they are suitable for small samples. Bias correction is done for four global NWP model maximum and minimum temperature forecasts. The magnitude of the bias at a grid point depends upon geographical location and season. Validation of the bias correction methodology is carried out using daily observed and bias corrected model maximum and minimum temperature forecast over India during July-September 2011. The bias corrected NWP model forecast generally outperforms direct model output (DMO). The spatial distribution of Mean Absolute Error (MAE) and Root mean squared Error (RMSE) for bias corrected forecast over India indicate that both the RM and NNW methods produces the best skill among other bias correction methods. The inter-comparison reveals that statistical bias correction methods improves the DMO forecast in terms of accuracy in forecast and has the potential for operational applications.

We evaluate the short-term weather forecast performance of three flavors of artificial neural networks (NNs): feed forward back propagation, radial basis function, and generalized regression. To prepare the application of the NNs to an operational setting, we tune NN hyperparameters using over two years of historical data. Five objective guidance products serve as predictors to the NNs: North American Mesoscale and Global Forecast System model output statistics (MOS) forecasts, the High-Resolution Rapid Refresh (HRRR) model, National Weather Service forecasts, and the National Blend of Models product. We independently test NN performance using 96 real-time forecasts of temperature, wind, and precipitation across 11 U.S. cities made during the WxChallenge, a weather forecasting competition. We demonstrate that all NNs significantly improve short-range weather forecasts relative to the traditional objective guidance aids used to train the networks. For example, 1-day maximum and minimum temperature forecast error is 20%–30% lower than MOS. However, NN improvement over multiple linear regression for short-term forecasts is not significant. We suggest this may be attributed to the small number of training samples, the operational nature of the experiment, and the short forecast lead times. Regardless, our results are consistent with previous work suggesting that applying NNs to model forecasts can have a positive impact on operational forecast skill and will become valuable tools when integrated into the forecast enterprise. Significance Statement We used approximately two years of historical weather data and objective forecasts for a number of cities to tune a series of artificial neural networks (NNs) to forecast 1-day values of maximum and minimum temperature, maximum sustained wind speed, and quantitative precipitation. We compare forecast error against common objective guidance and multiple linear regression. We found that the NNs exhibit about 25% lower error than common objective guidance for temperature forecasting and 50% lower error for wind speed. Our results suggest that NNs will be a valuable contributor to improving weather forecast skill when adopted into the existing forecast enterprise.

Being able to forecast events has always been important for humans. Humans did forecasting by inspecting movements of material and non-material objects in ancient times. However, thanks to the technological developments and the increasing amount of data in recent years, forecasting is now done by computers, especially by machine learning methods. One of the areas where these methods are used frequently is numerical weather forecasting. In this type of forecast, short, medium and long-term weather forecasts are made using historical data. However, predictions are inherently error-prone phenomena and should be stated which error range the predictions fall. In this study, numerical weather forecasting was done by combining Genetic Programming and Inductive Conformal Prediction method. The effect of 10 and 20 days of historical data on short (1-day), medium (3-days) and long-term (5-days) weather forecasts was examined. Results suggested that Genetic Programming has a good potential to be used in this area. However, when Genetic Programming was combined with the Inductive Conformal Prediction method, it was shown that forecasts gave meaningful results only in short-term; forecasts made for medium and long-term did not produce meaningful results.

The land surface temperature (LT) is a crucial variable that governs the energy and radiation budget of the earth's atmosphere and influences land‐atmosphere interactions. The LT plays a crucial role mainly in the short‐range forecast of a numerical weather prediction (NWP) model. The primary research goal in this research work undertaken is to assess the impact of assimilation of LT data from the Indian satellite (INSAT‐3D) into the NCMRWF global NWP model (NCUM) through a simplified Extended Kalman Filter (sEKF) land data assimilation system (LDAS), particularly important as there are few screen‐level observations over the region. A dedicated stand‐alone pre‐processing system has been designed to prepare LT observations in a compatible format for the land surface assimilation system. The approach for LT data assimilation from the INSAT‐3D satellite reduces the uncertainty associated with the initial state of LT analysis while simultaneously improving the accuracy of forecasts of near surface atmospheric variables. An observing system experiment (OSE) was carried out during both the summer (May) and winter (February) months by assimilating the INSAT‐3D LT data in a coupled land‐atmosphere analysis‐forecast system. The results obtained from the OSE demonstrate that the use of INSAT‐3D LT data improves the forecast skill of both maximum and minimum temperature over India, particularly in areas characterized by higher LT variability. The improvement is pronounced in forecasts of maximum (minimum) temperature during “Boreal” summer (“Boreal” winter) season. The verification scores also indicate that the incorporation of INSAT LT data substantially improves the NCUM model's forecast performance. By assimilating LT, the mean error of maximum and minimum temperature forecasts in India was decreased, accompanied by enhanced forecast accuracy within a time frame of approximately 24 h. The scores for the verification measures, specifically the Probability of Detection (POD), demonstrate a ~15% improvement in both the forecasts for maximum and minimum temperatures. This improves the temperature prediction as well as the ability to forecast intense weather episodes like cold spells and heat waves.

In this paper the statistics of daily maximum and minimum surface air temperature at weather stations in the southeast United States are examined as a function of the El Niño–Southern Oscillation (ENSO) and Arctic Oscillation (AO) phase. A limited number of studies address how the ENSO and/or AO affect U.S. daily—as opposed to monthly or seasonal—temperature averages. The details of the effect of the ENSO or AO on the higher-order statistics for wintertime daily minimum and maximum temperatures have not been clearly documented. Quality-controlled daily observations collected from 1960 to 2009 from 272 National Weather Service Cooperative Observing Network stations throughout Florida, Georgia, Alabama, and South and North Carolina are used to calculate the first four statistical moments of minimum and maximum daily temperature distributions. It is found that, over the U.S. Southeast, winter minimum temperatures have higher variability than maximum temperatures and La Niña winters have greater variability of both minimum and maximum temperatures. With the exception of the Florida peninsula, minimum temperatures are positively skewed, while maximum temperatures are negatively skewed. Stations in peninsular Florida exhibit negative skewness for both maximum and minimum temperatures. During the relatively warmer winters associated with either a La Niña or AO+, negative skewnesses are exacerbated and positive skewnesses are reduced. To a lesser extent, the converse is true of the El Niño and AO−. The ENSO and AO are also shown to have a statistically significant effect on the change in kurtosis of daily maximum and minimum temperatures throughout the domain.

A new nonlinear objective prediction scheme has been developed for predicting 24h daily maximum and minimum temperature forecasts at 14 stations in Guangxi, China during Jan, 2015-Jun, 2018 using Recurrent Neural Network (RNN) and based on the daily average, maximum, minimum temperature and precipitation data. Taking the climatology and persistence predictors as primary factors, the conditional attribute reduction method of rough set theory is adopted. By eliminating the unrelated attributes, the predictors direct correlated with the predictand (maximum and minimum temperature) are taken as the RNN model input by means of attribute reduction. This new scheme is validated with 24h short-range forecasts spanning Jan to Jun, 2018. Using identical predictors and sample cases, predictions of the RNN model are compared with the stepwise regression method, and results show that the former is more accurate. The mean absolute errors of RNN at 14 stations in Guangxi are lower than those of the stepwise regression method. The mean forecast accuracy with absolute errors being less than 2°C (1°C) of RNN is higher than that of the stepwise regression method. Moreover, the number of forecast errors larger than 2°C and the system deviation of daily maximum (minimum) temperature prediction are significantly reduced by RNN model, indicating a potentially better operational weather prediction tool.

Statistical bias correction methods for numerical weather prediction (NWP) forecasts of maximum and minimum temperatures over India in the medium-range time scale (up to 5 days) are proposed in this study. The objective of bias correction is to minimize the systematic error of the next forecast using bias from past errors. The need for bias corrections arises from the many sources of systematic errors in NWP modeling systems. NWP models have shortcomings in the physical parameterization of weather events and have the inability to handle sub-grid phenomena successfully. The statistical algorithms used for minimizing the bias of the next forecast are running-mean (RM) bias correction, best easy systematic estimator, simple linear regression and the nearest neighborhood (NN) weighted mean, as they are suitable for small samples. Bias correction is done for four global NWP model maximum and minimum temperature forecasts. The magnitude of the bias at a grid point depends upon geographical location and season. Validation of the bias correction methodology is carried out using daily observed and bias-corrected model maximum and minimum temperature forecast over India during July–September 2011. The bias-corrected NWP model forecast generally outperforms direct model output (DMO). The spatial distribution of mean absolute error and root-mean squared error for bias-corrected forecast over India indicate that both the RM and NN methods produce the best skill among other bias correction methods. The inter-comparison reveals that statistical bias correction methods improve the DMO forecast in terms of accuracy in forecast and have the potential for operational applications.

Statistical postprocessing of NWP model outputs is applied to maximum and minimum temperature forecasts. Two approaches to its application are effected to local short-range weather forecasts of minimum and maximum temperatures: Model Output Statistics and modified Perfect Prognosis. The modified Perfect Prognosis method is restricted to the first step of PP because of the significant difference between the horizontal resolution of the available objective analyses and the NWP model outputs. The modified Perfect Prognosis method uses actual data from the objective analysis related to the forecast period instead of the NWP forecast. The results are compared with a simple statistical prognostic model, which does not utilize the NWP model outputs, and with simple reference methods. The forecast is verified using ground station measurements from stations providing SYNOP reports. The results show that the predictive accuracy of the Model Output Statistics method is not very different from that of the modified Perfect Prognosis method, and both are significantly more accurate than the direct predictions of the NWP model. The results have confirmed that statistical postprocessing is able to make localized predictions even if lowresolution data are used.

<p>It is very difficult to predict accurate temperature, especially for maximum and minimum temperature, due to the large diurnal temperature range in arid area. Based on the temperature forecast products from ECMWF, T639, DOGRAFS and GRAPES models and hourly temperature observations at 105 automatic weather stations in Xinjiang during 2013~2015, two kinds of error correction and integration schemes were designed by using the decaying averaging method, ensemble average and weighted ensemble average method, the effects of error correction and integration on predicted maximum and minimum temperature in fore seasons in different partitions Xinjiang were tested contrastively. The first scheme was integrating forecast temperature before correcting errors, while the second scheme was correcting forecast errors firstly and then giving an integration. The results are follows as: (1)The accuracy of temperature predictions from ECMWF model was the best in Xinjiang as a whole, while that from DOGRAFS model was the worst, and the improvement to minimum temperature predictions was higher than that of maximum temperature prediction. (2) With regarding to different partitions Xinjiang, the accuracies of predicted maximum and minimum temperature in northern Xinjiang, west region and plain areas were correspondingly higher than those in southern Xinjiang, east region and mountain areas, and the correction capability to temperature prediction in winter was higher than that in other seasons. (3) The integrated prediction of maximum and minimum temperature by weighted ensemble average method was better than that of ensemble average method. The second scheme was superior to the first scheme. (4) The improvement to maximum(minimum) temperature prediction in the extreme high(low) temperature event process from 13 to 30 July 2017(from 22 to 24 April 2014) in Xinjiang was significant by using the second scheme.</p>

Day (0600-1800 UTC) maximum and night (1800-0600 UTC) minimum temperature forecasts as well as prediction of the occurrence of precipitation are evaluated for different sites in Estonia: southern coast of the Gulf of Finland (Tallinn), West- Estonian archipelago (Kuressaare), and inland Estonia (Tartu). The forecasts are collected from Estonian weather service. Several traditional verification methods are used, first of all reliability (root mean square error (RMSE)) and validity (mean error (ME)). Detailed analysis is carried out by means of the contingency tables that enable the user to calculate percent correct, percent underestimated, and percent overestimated. The contingency tables enable the user to calculate conditional probabilities of the realizations of certain forecasts. The paper is user-oriented and does not analyse the forecast technique. On the other hand, attention is drawn to the subjectivity of such evaluation, as the results may depend on the forecast presentation style and/or on the choice of the features of the meteorological parameter under consideration. For the current case study (the coldest hour during night and the warmest hour during day chosen to validate the temperature forecast, the temperature validation bin size 3 degrees, precipitation forecast validated in three categories based on the 12 h precipitation sums) one may say that the RMSE of the short- term prediction of night minimum and day maximum temperature is 1.5°…3.1°. It was also noticed that Estonian weather service predicts lower night minimum temperature than it follows in reality. The skill of the temperature forecast is estimated by comparison of its RMSE with that of the persistence forecast (next night/day will be similar to the previous one). The RMSE of the 1st day/night forecast is by 1.3°…1.4° less for Tallinn and Tartu and 0.3°…0.8° for Kuressaare than that of the persistence forecast. For the precipitation forecast, percent correct is 60…70, the probability that dry weather forecast is followed by no precipitation is 70%…80%. At the end of the paper the long-term forecasts of two international weather portals www.gismeteo.ru (Russia) and www.weather.com (USA) are briefly analysed.

In this paper, trends of minimum and maximum temperatures in Iran were studied using time series of daily minimum and maximum temperatures of 45 meteorological stations from 1976 to 2005 (as the baseline period). Mann-Kendall test, for maximum and minimum temperature, was obtained 1.85 and 3.56, respectively, which was positive and significant. The slope of the trend line for maximum and minimum temperature was obtained 0.23 and 0.39 °C decade−1, respectively. In this study, the trend of extreme temperature indicators was also evaluated. According to the obtained results, in annual terms, TX10, FDO, TN10, and IDO indices have had a negative trend at most stations, but TX90, TR20, TNx, TNx, TXn, TN90, SDI, and SU25 indices showed a positive trend. In the seasonal scale, the indices of cold days (TX10) and cold nights (TN10) showed significant negative trends in most seasons. Warm days (TX90) and warm nights (TN90) showed significant positive trends at most stations. The results of future simulations using several general circulation models in different time periods showed that the highest increase in maximum and minimum temperature related to the RCP8.5 scenario in periods of 2071 to 2099. The results also showed that northwest of Iran would have the highest temperature rise. The results also showed that the probability density function of the minimum and maximum temperatures will shift to warmer temperatures. This could be an indication of climate change in the future decades in Iran.

Purpose Some studies have reported a relationship between meteorological factors and the occurrence of acute aortic dissection (AAD). Nevertheless, the results of the studies are heterogeneous. Furthermore, whether the absolute values or fluctuation of meteorological factors influence the occurrence of AAD remains controversial. The aim of this study was to determine the meteorological factors associated with the occurrence of AAD. Methods Two hundred eighty-two consecutive patients (male, n = 178; female, n = 104; average age, 68 years) admitted to our hospital for AAD in the 10 years from September 1st 2008 were included in this study. One hundred fifty-seven patients had type A dissection. The correlation between the clinical data and the local meteorological data over the same period (provided by the National Meteorological Agency) was analyzed. We compared the following factors on days of AAD occurrence and non-occurrence: minimum and maximum temperature, minimum and maximum temperature difference between day of occurrence and previous day, difference between maximum and minimum temperature, atmospheric pressure and atmospheric pressure difference between day of occurrence and previous day (Δatmospheric pressure), and minimum and maximum temperature difference from climatological standard normal (CSN). Cutoff values were determined by ROC curve analyses and odds ratios (ORs) were calculated by a logistic regression analysis of meteorological factors with statistically significant differences. Results ignificant differences between the days of AAD occurrence and non-occurrence were observed for minimum and maximum temperature (p < 0.0001), atmospheric pressure (p < 0.0001) and Δatmospheric pressure (p = 0.0286), minimum temperature difference from CSN (p < 0.0001), and maximum temperature difference from CSN (p = 0.0010). The cutoff values were as follows: minimum temperature, 4°C; maximum temperature, 15.1°C; atmospheric pressure, 1008.9hPa; Δatmospheric pressure, 0.4hPa; minimum temperature difference from CSN, 1°C; and maximum temperature difference from CSN, -0.2°C. The univariate logistic regression model showed revealed the following significant predictors of the occurrence of AAD; minimum temperature (OR2.42, p < 0.0001), maximum temperature (OR2.23, p < 0.0001), air pressure (OR1.75, p < 0.0001), Δatmospheric pressure (OR 1.44, p = 0.0030), minimum temperature difference from CSN (OR1.80, p < 0.0001) and maximum temperature difference from CSN (OR1.58, p = 0.0003). However, only minimum temperature (OR1.60, 95% CI 1.00-2.53, p = 0.0478) and maximum temperature difference from CSN (OR1.45, 95% CI 1.11-1.89, p = 0.0062) remained significant in the multivariate analysis. Conclusion Meteorological factors, especially a minimum temperature under 4°C strongly influenced the occurrence of AAD. A maximum temperature difference from CSN of over -0.2°C was also a significant predictor of AAD.