NOAA Privacy Policy | NWS Disclaimer
Link to National Oceanic and Atmospheric Agency Main WW3 logo Link to National Weather Service

NEW vs. NOAA's Operational WAM Model

 Point of contact: NCEP.EMC.waves@NOAA.gov
Last update: September 01, 2006
This page presents results of a parallel comparison between the NOAA Experimental Wave Model (NEW) and NCEP's operational WAM cycle 4, using buoy and altimeter observations of the wave height. More information on these data can be found on the NWW3 validation page. A comprehensive report is in preparation and will be published shortly.
 
The main differences between the two wave forecast systems are:
  • The wave models used: WAVEWATCH-III version 1.15 versus WAM cycle 4
  • The model resolutions: NEW uses a discrete spectrum with 25 frequencies (0.04 - 0.4Hz) and 24 directions on a 1ox1.25o latitude-longitude grid, whereas the operational WAM model uses 12 directions and a 2.5ox2.5o spatial grid
  • The operational WAM model includes assimilation of the buoy and altimeter data starting Feb. 9, 1998. Therefore, these data cannot be considered as independent validation data for the hindcasts after this time
Both wave models utilize the same wind fields, although there are some small but systematic differences in the processing of these wind fields. For historical reasons, the operational system will be simply denoted as WAM. This does not imply that all differences between the forecasts systems are attributed to the underlying wave models only.
 
The parallel comparison considers the period of Jan. 12 1998 through June 1998. Below, first a model comparison based on conventional buoy data is presented. Secondly, the models are compared using ERS2 altimeter data. More information on these data can be found on the NWW3 validation page. This page closes with a discussion of the results.
 

Buoy Data

The two wave forecasts systems are compared to the buoy data in three ways. First, a table with time series plots is available. This table contains all model results and data available for the validation period. Note that some data drop out occurred for both the buoy data and the model results. The latter is strictly due to glitches in the archiving system. Model continuity was maintained for both forecast systems throughout the validation period.
 
Also available are plots of mean statistics of both models against the buoy data. Considered are the model bias, rms error and scatter index (rms error normalized with the averaged observed wave height) for each month. These parameters have been calculated for the entire buoy data set, and for sub-sets corresponding to separate geographical regions. Groups of buoys making up separate regions are identified in the table with time series plots.
 
Finally, the model errors are assessed as a function of the wave height using an error corrected bin-averaging (BA) technique as described in Tolman (1998a,d) (see references). As discussed in the validation page, such an analysis only makes sense for hindcast results (i.e., not for forecasts). Furthermore, this analysis is not relevant after the data assimilation scheme started in WAM. Therefore, only hindcast results for the period of 01/12/98 through 02/09/98 are considered in this analysis. The results of the BA analysis, together with measured and calculated wave height distributions are presented in two pages, one considering the buoys in/near Japan, Hawaii and Gulf of Mexico, the other considering buoys in the NE Pacific and Atlantic Ocean.
 
Top Of Page | Main NWW3 Page | NWW3 Validation Page
 

Altimeter data

The wave forecast systems are compared to the ERS2 altimeter data in two ways. First , the global distribution of biases, rms errors and scatter indices are estimated as described in the NWW3 validation page. To obtain good global coverage and statistically stable results, periods of two to three months are required. The table below accesses results for the hindcasts (or analysis in the case of WAM) and different forecast time ranges for the period of January 12 through march 1998.
altimeter bias rms S.I
hindcast NEW / WAM NEW / WAM / dif NEW / WAM / dif
0-12h forecast NEW / WAM NEW / WAM / dif NEW / WAM / dif
18-30h forecast NEW / WAM NEW / WAM / dif NEW / WAM / dif
42-54h forecast NEW / WAM NEW / WAM / dif NEW / WAM / dif
60-72h forecast NEW / WAM NEW / WAM / dif NEW / WAM / dif

 
The second table contains similar results for the period of April through June 1998.
altimeter bias rms S.I.
hindcast NEW / WAM NEW / WAM / dif NEW / WAM / dif
0-12h forecast NEW / WAM NEW / WAM / dif NEW / WAM / dif
18-30h forecast NEW / WAM NEW / WAM / dif NEW / WAM / dif
42-54h forecast NEW / WAM NEW / WAM / dif NEW / WAM / dif
60-72h forecast NEW / WAM NEW / WAM / dif NEW / WAM / dif
 
Secondly, the model errors as a function of the wave height are assessed using a BA analysis and wave height distributions as above for the buoy data (hindcast for 01/12/98 through 02/09/98 only). In this analysis. areas with large model errors due to unresolved island chains are masked out as in Tolman (1998d). The results are presented on two pages, one for the northern hemisphere data and one for the southern hemisphere data.
 
 Return to top of page.

Discussion

One of the best ways to assess model behavior usually is to look at time series. For the wave forecast systems, the most interesting time series are those of contiguous hindcasts, as they present a continuous and meaningful time series. For the two forecast systems considered here, such a comparison is useful only before Feb 9, as after this date the WAM hindcasts incorporates the validation data (i.e., has become an analysis rather than hindcast).
 
The hindcasts time series at the buoy locations for NEW and WAM for the month of January show significantly different model behavior in several areas. Most striking are the differences around Hawaii , where the wave field in the last weeks of this month is dominated by swells generated roughly 1500 km north of Hawaii. In such conditions where swell is generated relatively nearby, WAM is known to significantly underestimate maximum wave heights and overestimate minima, i.e., significantly underestimate the variability of the wave height. Around Hawaii, NEW captures the wave height variability much better, as is particularly clear for buoy 51001. The other buoys around Hawaii (51002 and 51003 ) are partially sheltered by Hawaii. Because neither NEW nor WAM resolves Hawaii, this sheltering is not modelled, and the swells are therefore expected to be overestimated by both models, as is confirmed by the corresponding time series. For these two locations, NEW also captures the variability of the wave heights much better than WAM.
 
Similar model differences are found for the buoys near Japan (21004 and 22001 ). Note that in particular for these locations systematic wave height overestimations might be expected due to the unresolved Ryukyu Islands.
 
Similar differences between NEW and WAM are also found in the Gulf of Mexico (buoys 42001, 42002 and 42003 ). The differences between both models are more surprising here, as the wave fields in the Gulf of Mexico are not generally identified with swells. Possibly, some of the difference in model behavior here can be attributed to the fact the NEW grid resolves this area much better. It should also be noted that both models are completely missing some wave height events, for instance at January 14 and 15 for buoys 42001 and 42002 and at January 23 for buoy 42003 . This is probably due to the fact that the wave field in the Gulf of Mexico is often generated by relatively small scale atmospheric features that are poorly or not at all resolved by the wind fields used here.
 
For the NE Pacific and NE Atlantic buoys, which are mostly located in or near the predominant storm tracks, differences between the models are less obvious, and in many cases of the same order as the differences between models and observations. Even here, however, the NEW model is generally more responsive, showing a more realistic variance of the wave height. Also extreme events are generally better captured by NEW (e.g., January 27-28 at 46003, January 16 at 46004, January 19 at 46006 and January 16-18 at 62029 and 62081).
 
For the NW Atlantic buoys, the differences between the models are generally small. This might be expected, because wave conditions in this area are dominated by local wave generation, for which both models behave very similarly. For the southern buoys (e.g., 41002), where swell is virtually nonexistent, the models behave virtually identical. For the northern most buoys (e.g., 44138), where swells become more important, differences between the models resemble those as discussed above.
 
The systematic low biases for high wave heights of WAM and high biases for low wave heights are confirmed by the results of the error analysis as a function of the wave height (first and second panel of Fig.), and by the corresponding wave height distributions that are systematically too narrow. Such errors are generally reduced in NEW, but not completely removed. This improvement is not limited to the areas covered by buoy observations. Similar results are obtained against global altimeter data for both the northern and southern hemisphere.
 
After February 9 the WAM hindcast in effect becomes an analysis as the buoy data has been assimilated. In the comparison with buoy data this becomes clear as (a) the WAM model hindcast / analysis more closely follows the data, and (b) incorporates significantly more variability at the time resolution of the data (1 hour). Most of the assimilated data, however, is rapidly lost from the model in the forecast. In many of the time series this is obvious, as the forecast wave height from WAM are internally consistent, but differ significantly from the analysis, whereas all of the results of NEW are consistent from hind- to forecast. Whereas this can be observed to some degree in most time series after February 9, it is particularly clear in the June results near Hawaii (51001 , 51002, 51003 and 51004)
 
The effects of data assimilation and the rapid demise of its impact is also clear in the global model validation using altimeter data. It is not surprising that the WAM analysis for April through June that includes this data shows systematically smaller scatter indices than the NEW hindcast that does not incorporate this data. In fact, it is amazing that in spite of this, NEW actually shows systematically smaller scatter indices in the southern tropical Pacific Ocean. After an average of only 6 hours of forecasting, the WAM model has lost most of the positive impact of the data assimilation, and virtually everywhere shows systematically larger scatter indices than NEW. Note that the larger scatter indices for NEW in the NE Pacific in January through March are most likely due to wind errors in February as discussed on the validation page .
 
A side effect of the more responsive model behavior of NEW is that wind errors are expected to translate into larger wave errors than with WAM. This is expected to result in a more rapid error growth with forecast time in NEW . This can indeed be observed in the mean statistics of both models against buoy data which generally suggest that errors of NEW are smaller for the 24 hour forecast, but are larger for the 72 hour forecast. The more rapid error growth is also obvious in the validation with altimeter data. The 60-72 hour forecast for the last three months therefore shows systematically smaller scatter indices for WAM in the souther hemisphere storm tracks, where wave height errors are mostly governed by local wind speed errors. In the tropics, however, where the dominant swell fields are virtually independent of the local wind fields (and errors), there is virtually no error growth in either model, and NEW shows the smallest scatter indices.
Return to top of page.