Weather Typing

The following material is based on the book by Mujumdar & Nagesh Kumar, Floods in a Changing Climate: Hydrologic Modeling.

Weather Typing Methods

The weather typing methods uses the types of weather/classification to define the links between large scale circulation and surface weather. The GCM variables are classified into circulation patterns which are linked to surface weather. Weather typing methods are objective, has sound physical basis and can be used for multisite downscaling. Weather typing has been used for simulating precipitation for a long time. However their drawbacks include:

  • GCM may not simulate the correct frequencies of weather types
  • Methods are unable to capture the effects of other relevant physical processes
  • The method assumes that observed relationships between the circulation types and the local climate will be unchanged in the future.

Weather typing consists of two main steps:

  1. Choice of appropriate circulation classification scheme
  2. Modeling of the circulation-surface climate relationship
  1. Choice of appropriate circulation classification scheme
    Classification of circulation patterns is performed using subjective and objective methods. Subjective methods are methods in which the knowledge and experience of meteorologists are used. The results obtained from the subjective methods cannot be reproduced and these meetings are only applicable to specific geographical methods. Lambs weather types is an example for subjective method. Objective methods on the other hand, are algorithms that operates on a dataset for classifying the same. Examples for objective methods include k-means clustering, fuzzy classification, principal component clustering, screening discriminant analysis, and neural network methods.
  2. Modeling of the circulation – surface climate relationship
    Modeling the relationship between the circulation types and surface variables are done using Markov chain models/hidden Markov models, regression analysis, fuzzy rule based approaches, canonical correlation analysis, or sampling from observed current analog data.

Hay et al., 1991, developed a method that used six weather types to simulate precipitation. Monthly statistics of the frequency and distribution of precipitation for each weather type were computed from a 30 year old period. Precipitation values are then calculated by applying these statistics to a sequence based on Markov’s model.

Bardossy & Plate, 1992, simulated daily rainfall using a space-time model, depending on the atmospheric circulation patterns. The daily rainfall was represented using a power transformed normal random variable. The model was applied in the Ruhr catchment in Germany and simulated the frequency and intensity of rainfall very well.

 Bardossy et al., 1995 classified CPs into different states using a fuzzy rule based technique. A set of rules were defined, for each CP according to which they are classified on the basis of compliance of the requirements.

Conway & Jones, 1998 generated daily rainfall series/scenarios using three versions of weather types. Two versions classified circulation over the British Isles using objective classification, using Lamb’s subjective classification. The third version used user-defined categories of vorticity. The rainfall events were classified based on whether the previous day was wet or dry and was able to successfully compute the monthly means, persistence, and inter-annual variability of daily rainfall time series.

The following example was taken from Mujumdar & Nagesh Kumar, 2010, which discusses the application of weather typing methods for downscaling daily rainfall. The study was carried out for the Guadalentin Basin in southeast Spain by Goodess & Palutikof, 1998.

Example WT 1:

The large-scale patterns of a predictor variable, gridded sea level pressure are to be downscaled to local values of a surface climate variable, which is daily rainfall at six stations. The Figure WT 1 shows the grid used in downscaling over the basin (shown hatched) in Spain. The model used in the study is the UK Meteorological Office high-resolution GCM (UKTR). Daily output is available from the years 66–75 (10 years) of the 75-year long control and perturbed simulations. In the perturbed simulation, carbon dioxide forcing is increased by 1% per annum and doubles in the year 70. The Lamb weather types, initially developed for the British Isles, are used here for classifying CPs in the Mediterranean climate regime. SLP data for years 1956–89 are interpolated to a 32-point 2.5◦ latitude by 3.75◦ longitude grid over the area 36.25◦ N–46.25◦ N and 16.88◦ W–9.38◦ E, which is the grid spacing for the GCM.

Figure WT 1: Study area for the the automated circulation typing scheme over the Guadalentin Basin (hatched) (Source: Goodess and Palutikof, 1998)

Step 1: Circulation pattern classification

Large scale patterns are classified using automated Lamb weather type classification, which was developed for the British Isles. The automated version is used to classify as it is easy to define and has a strong physical basis. Lamb’s automated weather type uses direction (with 45° resolution) and type of flow to classify the surface flow and is applicable anywhere in the Northern Hemisphere mid latitudes. The CPs are divided into eight types from the SLP data:

  1. Cyclonic (C)
  2. Hybrid-cyclonic (HYC)
  3. Unclassified/light flow cyclonic (UC)
  4. Anti-cyclonic/Hybrid Anti-cyclonic(A/HYA)
  5. Unclassified/Light flow Anti-cyclonic (UA)
  6. Westerly/northwesterly/southwesterly/northerly directional types (W/NW/SW/N)
  7.  Easterly/northeasterly directional types (E/NE)
  8. Southerly/southeasterly directional types(S/SE)

Figure WT 2: Observed and simulated monthly frequencies of the eight circulation types for years 1956–89. Solid line: observed mean frequency. Shaded area: maximum and minimum frequency range observed over any 10-year period. Dashed line: mean frequency calculated from UKTR control-run model output (Source: Goodess and Palutikof, 1998).

The observed SLP data for the period 1956-89 gives the mean monthly cycles of eight circulation type and their mean frequencies, as shown in Figure WT 2. Over a period of one year, the UC and UA types were observed to occur most frequently. The UC type of the two has a very strong seasonal cycle with a winter minimum and summer maximum. The UA type does not have a strong seasonal cycle, but occurs mostly in autumn. The A/HYA has a strong seasonal cycle with a significant winter maximum whereas C and HYC types have an unpronounced seasonal cycle and a late spring/summer maximum. The W/NW/SW/N groups have a strong seasonal cycle, with a maximum in the late autumn/winter. The E/NE and S/SE groups are less frequent, the former with no seasonal cycle and the latter occurs less frequently from May to September.

All the eight circulation types have a characteristic underlying pressure pattern, physically distinct, and verified using the SLP composite maps. This underlying pressure produces the probable type and direction of flow over the southeast Spain and the Guadalentin Basin.

Step 2: Project changes in CP frequencies

The occurrence of all the circulation types, both seasonal and monthly were calculated from the SLP data of the perturbed run. The perturbed minus control run (Table WT 1) gives the mean seasonal changes, and significant changes if any, are identified Mann–Whitney/Wilcoxon rank sum test. Largest changes were observed in summer and a notable increase in the frequency of the high rainfall types, C and HYC, due to the lower mean SLP over the Iberian Peninsula. The E/NE types and S/SE types occurs frequently in summer, but not much changes has been spotted. 

Table WT 1: Mean seasonal changes (perturbed-control run) in the frequency (number of days) of the eight circulation-type groups (Source: Goodess and Palutikof, 1998)

TypeWinterSpringSummerAutumn
C−0.5−0.7+4.4−0.8
HYC+0.4−1.5+3.1−0.4
UC−0.8+0.3+3.6+2.4
A/HYA0.0−0.8−4.3−2.7
UA+3.5+2.3−7.3−2.7
W/NW/SW/N−4.5+3.1−1.0+3.5
E/NE−0.2−1.4+1.1−0.4
S/SE+2.1−1.1+0.4+1.2

Step 3: Weather generator for rainfall occurrence

As the occurrence of rainfall depends upon the type of the circulation, it is represented using a conditional weather generator (CWG). The transition from one circulation type to another is modeled as Markov’s chain process.

Two parameters, (i) the probability of next day, being one of the eight circulation types represented as a transition matrix, Pct1-8 and (ii) the mean probability of rain, PPRECct, for each type, is used to determine the occurrence of rainfall. A random number is selected each day determining the circulation type following the next day. A second random number is then selected to find out if the day will be wet or dry.

The underlying assumption of CWG is that the changes in the frequency of the circulation type will affect the changes in the occurrence and intensity of rainfall.

Step 4: Projection of rainy day changes

A hundred 30 year long simulations were performed in the Guadalentin Basin stations. The transition matrix and the random number generator determine the order of circulation types in each of the 30 year simulations. Thus each simulation has hundred different sequences and outputs from the simulation sets are used for:

  1. The first set gives the CWG parameters, Pct1-8 andPPRECct from the observed data.
  2. The second set gives the transition probabilities Pct1-8 which are calculated from the control-run output of UKTR GCM and PPRECct which is calculated from the observations. The frequency of the circulation types are also simulated by the GCM.
  3. The transition probability, Pct1-8 of the third set is calculated from the perturbed run GCM output and the probability of precipitation, PPRECct is calculated from the observations.
  4. The difference between the perturbed run and the control simulation (Table WT 2) are calculated and they provide the rainy day climate change scenarios. Based on these scenarios, it has been estimated the average non-rainy days in the Guadalentin Basin could increase by 10–18% in summer in a future warmer world, and decrease by 5–9% in spring. A very small increase (2–4%) is indicated in winter and little change in autumn (0–2%).

Table WT 2: Mean change (Pert. – Cont.) from the 10 simulations of the mean and SD of the number of rain days simulated by the CWG for six stations in the Guadalentin Basin (Modified from Goodess and Palutikof, 1998)

Station ID Winter Spring Summer Autumn
Mean
Station 1+0.3−0.7+0.3  0.0
Station 2+0.3−0.7+0.7+0.3
Station 3+0.3−1.0+0.6+0.1
Station 4+0.2−0.5+0.6+0.1
Station 5+0.3−0.8+0.6+0.1
Station 6+0.3−0.7–0.4  0.0
SD
Station 1+0.1−0.20.0  0.0
Station 2+0.1−0.2+0.1+0.2
Station 3  0.0−0.2+0.2  0.0
Station 4  0.0−0.1+0.2  0.0
Station 5  0.0−0.1+0.2+0.1
Station 6+0.2−0.2+0.1  0.0

References

Bardossy, A., & Plate, E. J. (1992). Space-Time Model for Daily Rainfall Using Atmospheric Circulation Patterns. Water Resources Research, 28(5), 1247–1259.

Bardossy, A., Duckstein, L., & Bogardi, I. (1995). Fuzzy rule‐based classification of atmospheric circulation patterns. International Journal of Climatology, 15(10), 1087–1097. https://doi.org/10.1002/joc.3370151003

Conway, D., & Jones, P. D. (1998). The use of weather types and air flow indices for GCM downscaling. Journal of Hydrology, 212–213(1–4), 348–361. https://doi.org/10.1016/S0022-1694(98)00216-9

Goodess, C. M., & Palutikof, J. P. (1998). Development of daily rainfall scenarios for southeast Spain using a circulation-type approach to downscaling. International Journal of Climatology, 18(10), 1051–1083. https://doi.org/10.1002/(SICI)1097-0088(199808)18:10<1051::AID-JOC304>3.0.CO;2-1

Hay, L. E., McCabe, G. J., Wolock, D. M., & Ayers, M. A. (1991). Simulation of precipitation by weather type analysis. Water Resources Research, 27(4), 493–501. https://doi.org/10.1029/90WR02650

Mujumdar, P. P., & Nagesh Kumar, D. (2010). Floods in a changing climate: Hydrologic modeling. Floods in a Changing Climate: Hydrologic Modeling. Cambridge University Press. https://doi.org/10.1017/CBO9781139088428

Scroll Up