Search Results

You are looking at 1 - 7 of 7 items for :

  • Author or Editor: Huug van den Dool x
  • Weather and Forecasting x
  • Refine by Access: All Content x
Clear All Modify Search
Jianfu Pan
and
Huug van den Dool

Abstract

A probability forecast has advantages over a deterministic forecast as the former offers information about the probabilities of various possible future states of the atmosphere. As physics-based numerical models find their success in modern weather forecasting, an important task is to convert a model forecast, usually deterministic, into a probability forecast. This study explores methods to do such a conversion for NCEP’s operational 500-mb-height forecast and the discussion is extended to ensemble forecasting. Compared with traditional model-based statistical forecast methods such as Model Output Statistics, in which a probability forecast is made from statistical relationships derived from single model-predicted fields and observations, probability forecasts discussed in this study are focused on probability information directly provided by multiple runs of a dynamical model—eleven 0000 UTC runs at T62 resolution.

To convert a single model forecast into a strawman probability forecast (single forecast probability or SFP), a contingency table is derived from historical forecast–verification data. Given a forecast for one of three classes (below, normal, and above the climatological mean), the SFP probabilities are simply the conditional (or relative) frequencies at which each of three categories are observed over a period of time. These probabilities have good reliability (perfect for dependent data) as long as the model is not changed and maintains the same performance level as before. SFP, however, does not discriminate individual cases and cannot make use of information particular to individual cases. For ensemble forecasts, ensemble probabilities (EP) are calculated as the percentages of the number of members in each category based on the given ensemble samples. This probability specification method fully uses probability information provided by the ensemble. Because of the limited ensemble size, model deficiencies, and because the samples may be unrepresentative, EP probabilities are not reliable and appear to be too confident, particularly at forecast leads beyond day 6. The authors have attempted to combine EP with SFP to improve the EP probability (referred to as modified forecast probability). Results show that a simple combination (plain average) can considerably improve upon both the EP and SFP.

Full access
Qin Zhang
and
Huug van den Dool

Abstract

Retrospective forecasts of the new NCEP Climate Forecast System (CFS) have been analyzed out to 45 days from 1999 to 2009 with four members (0000, 0600, 1200, and 1800 UTC) each day. The new version of CFS [CFS, version 2 (CFSv2)] shows significant improvement over the older CFS [CFS, version 1 (CFSv1)] in predicting the Madden–Julian oscillation (MJO), with skill reaching 2–3 weeks in comparison with the CFSv1’s skill of nearly 1 week. Diagnostics of experiments related to the MJO forecast show that the systematic error correction, possible only because of the enormous hindcast dataset and the ensemble aspects of the prediction system (4 times a day), do contribute to improved forecasts. But the main reason is the improvement in the model and initial conditions between 1995 and 2010.

Full access
Yun Fan
and
Huug van den Dool

Abstract

A simple bias correction method was used to correct daily operational ensemble week-1 and week-2 precipitation and 2-m surface air temperature forecasts from the NCEP Global Forecast System (GFS). The study shows some unexpected and striking features of the forecast errors or biases of both precipitation and 2-m surface air temperature from the GFS. They are dominated by relatively large-scale spatial patterns and low-frequency variations that resemble the annual cycle. A large portion of these forecast errors is removable, but the effectiveness is time and space dependent. The bias-corrected week-1 and week-2 ensemble precipitation and 2-m surface air temperature forecasts indicate some improvements over their raw counterparts. However, the overall levels of week-1 and week-2 forecast skill in terms of spatial anomaly correlation and root-mean-square error are still only modest. The dynamical soil moisture forecasts (i.e., land surface hydrological model forced with bias-corrected precipitation and 2-m surface air temperature integrated forward for up to 2 weeks) have very high skill, but hardly beat persistence over the United States. The inability to outperform persistence mainly relates to the skill of the current GFS week-1 and week-2 precipitation forecasts not being above a threshold (i.e., anomaly correlation > 0.5 is required).

Full access
Li-Chuan Gwen Chen
and
Huug van den Dool

Abstract

In this study, an optimal weighting system is developed that combines multiple seasonal probabilistic forecasts in the North American Multimodel Ensemble (NMME). The system is applied to predict temperature and precipitation over the North American continent, and the analysis is conducted using the 1982–2010 hindcasts from eight NMME models, including the CFSv2, CanCM3, CanCM4, GFDL CM2.1, Forecast-Oriented Low Ocean Resolution (FLOR), GEOS5, CCSM4, and CESM models, with weights determined by minimizing the Brier score using ridge regression. Strategies to improve the performance of ridge regression are explored, such as eliminating a priori models with negative skill and increasing the effective sample size by pooling information from neighboring grids. A set of constraints is put in place to confine the weights within a reasonable range or restrict the weights from departing wildly from equal weights. So when the predictor–predictand relationship is weak, the multimodel ensemble forecast returns to an equal-weight combination. The new weighting system improves the predictive skill from the baseline, equally weighted forecasts. All models contribute to the weighted forecasts differently based upon location and forecast start and lead times. The amount of improvement varies across space and corresponds to the average model elimination percentage. The areas with higher elimination rates tend to show larger improvement in cross-validated verification scores. Some local improvements can be as large as 0.6 in temporal probability anomaly correlation (TPAC). On average, the results are about 0.02–0.05 in TPAC for temperature probabilistic forecasts and 0.03–0.05 for precipitation probabilistic forecasts over North America. The skill improvement is generally greater for precipitation probabilistic forecasts than for temperature probabilistic forecasts.

Full access
Huug M. Van Den Dool
and
Zoltan Toth

Abstract

It has been observed by many that skill of categorical forecasts, when decomposed into the contributions from each category separately, tends to be low, if not absent or negative, in the “near normal” (N) category. We have witnessed many discussions as to why it is so difficult to forecast near normal weather, without a satisfactory explanation ever having reached the literature. After presenting some fresh examples, we try to explain this remarkable fact from a number of statistical considerations and from the various definitions of skill. This involves definitions of rms error and skill that are specific for a given anomaly amplitude. There is low skill in the N-class of a 3-category forecast system because a) our forecast methods tend to have an rms error that depends little on forecast amplitude, while the width of the categories for predictands with a near Gaussian distribution is very narrow near the center, and b) it is easier, for the verifying observation, to ‘escape’ from the closed N-class (2-sided escape chance) than from the open ended outer classes. At a different level of explanation, there is lack of skill near the mean because in the definition of skill we compare the method in need of verification to random forecasts as the reference. The latter happens to perform, in the rms sense, best near the mean. Lack of skill near the mean is not restricted to categorical forecasts or to any specific lead time.

Rather than recommending a solution, we caution against the over-interpretation of the notion of skill-by-class. It appears that low skill near the mean is largely a matter of definition and may therefore not require a physical-dynamical explanation. We note that the whole problem is gone when one replaces the random reference forecast by persistence.

We finally note that low skill near the mean has had an element of applying the notion forecasting forecast skill in practice long before it was deduced that we were making a forecast of that skill. We show analytically that as long as the forecast anomaly amplitude is small relative to the forecast rms error, one has to expect the anomaly correlation to increase linearly with forecast magnitude. This has been found empirically by Tracton et al. (1989).

Full access
Huug van den Dool
,
Emily Becker
,
Li-Chuan Chen
, and
Qin Zhang

Abstract

An ordinary regression of predicted versus observed probabilities is presented as a direct and simple procedure for minimizing the Brier score (BS) and improving the attributes diagram. The main example applies to seasonal prediction of extratropical sea surface temperature by a global coupled numerical model. In connection with this calibration procedure, the probability anomaly correlation (PAC) is developed. This emphasizes the exact analogy of PAC and minimizing BS to the widely used anomaly correlation (AC) and minimizing mean squared error in physical units.

Full access
Yun Fan
,
Vladimir Krasnopolsky
,
Huug van den Dool
,
Chung-Yu Wu
, and
Jon Gottschalck

Abstract

Forecast skill from dynamical forecast models decreases quickly with projection time due to various errors. Therefore, postprocessing methods, from simple bias correction methods to more complicated multiple linear regression–based model output statistics, are used to improve raw model forecasts. Usually, these methods show clear forecast improvement over the raw model forecasts, especially for short-range weather forecasts. However, linear approaches have limitations because the relationship between predictands and predictors may be nonlinear. This is even truer for extended range forecasts, such as week-3–4 forecasts. In this study, neural network techniques are used to seek or model the relationships between a set of predictors and predictands, and eventually to improve week-3–4 precipitation and 2-m temperature forecasts made by the NOAA/NCEP Climate Forecast System. Benefitting from advances in machine learning techniques in recent years, more flexible and capable machine learning algorithms and availability of big datasets enable us not only to explore nonlinear features or relationships within a given large dataset, but also to extract more sophisticated pattern relationships and covariabilities hidden within the multidimensional predictors and predictands. Then these more sophisticated relationships and high-level statistical information are used to correct the model week-3–4 precipitation and 2-m temperature forecasts. The results show that to some extent neural network techniques can significantly improve the week-3–4 forecast accuracy and greatly increase the efficiency over the traditional multiple linear regression methods.

Restricted access