Diagnostics of the Performance of the NCEP Operational Global Analysis/Forecast System for Oct. 1988

Glenn H. White

National Centers for Environmental Prediction

W/NP23, World Weather Building,
Washington DC 20233, USA

Glenn.White@NOAA.GOV


DISCLAIMER


SUBSECTIONS

Purpose of the site
Abstract

TIME-MEAN ERRORS

Zonal Mean Height and Temperature Errors
Regional Temperature Errors
Vertical Cross-sections of Temperature Errors
Temperature Errors and Diabatic Heating
Regional Height Errors
Zonal Wind Errors
Meridional Wind Errors
Vector Wind Errors
Vertical Motion Errors
Horizontal Wind Divergence
A Measure of the Noise: Squared Divergence
Forecast Bias In Specific Humidity

PERFORMANCE OF THE MODEL PHYSICS

Precipitation
Surface Air Temperature
Precipitable Water
Zonal Surface Stress
Low Cloud Cover
Total Cloud Cover
Evaporation
Sensible Heat Flux
Surface Net Shortwave
Surface Net Heat Flux


Purpose

The purpose of this site is to present diagnostics of the performance of the NCEP global analysis/forecast system that highlight errors in the system so that possible solutions to these errors may be found. It is not intended to present a complete, balanced picture of model performance nor is it intended to be a polished, final product. The global analysis/forecast system, as many people have shown elsewhere, is a state-of-the-art analysis system and numerical weather prediction model that have improved immensely over the past few decades. Computer models now play the central role in operational weather forecasting and, together with improvements in observations, are primarily responsible for the improvement in weather forecasting in the last two decades. If one compares the analysed and forecast fields presented below, it can be difficult to find substantial differences. In the interest of improving the errors in the system, this site will concentrate on the differences.

This site is intended to be a working site and as such is subject to errors and to frequent changes. I would urge caution in accepting the interpretations of the fields shown below. To paraphase Abraham Lincoln, I may have my facts completely correct, but my interpretation completely wrong. Comments are always very welcome.


ABSTRACT

This page presents diagnostics of the performance of the NCEP global model during Oct. 1998. A change was made to the operational system at 12Z Oct. 5, in which the resolution was reduced from T170 42 levels to T126 28 levels, the number of iterations in the analysis was increased and the weighting of the first guess in the analysis was decreased. The reduction in resolution was necessary to introduce the improvements to the analysis. The diagnostics shown here reflect purely the new system, making use of parallel runs of the new system before Oct. 5. Since the parallel runs were used, forecasts longer than 5 days were not available for the entire month. For future months, errors for forecasts longer than 5 days will be shown.
.

TIME-MEAN ERRORS

This section compares monthly means from the analyses for Oct. 98 and from forecasts verifying in Oct. 98. Errors are defined as the difference between the analyses and the forecasts. Some uncertainty exists in the analyses produced by the global analysis/forecast system, especially for fields involving the divergent component of the flow and/or moisture.


Zonal mean height and temperature errors

Fig. 1a and Fig. 1b display the zonal mean difference between the time-averaged 1 day and 5 day forecasts of geopotential height for Oct. 98 and the time-averaged analyses of geopotential height. Both forecasts and analyses are valid at 000 GMT. Strong negative biases can be seen in the stratosphere; in the troposphere a positive bias appears in the Arctic and a negative bias develops in the tropics. Fig. 2a and Fig. 2b display the corresponding plots for temperature, the zonal mean errors in 1 day and 5 day forecasts of temperature verifying in Oct. 98. The model displays a cold bias in extratropical statosphere, and a substantial cold bias at 800 mb at 30S. A warm bias is seen near the north pole. In previous months the model exhibited more warm bias in the northern hemisphere midlatitudes. This difference may reflect the change in season more than model or analysis changes.


Regional Temperature Errors

Fig. 3a and Fig. 3b display the regional distributions of the monthly mean day 1 and day 5 temperature errors at 700 mb, respectively; Fig. 4a and Fig. 4b display the day 1 and day 5 errors at 800 mb; and Fig. 5a and Fig. 5b display the monthly mean day 1 and day 5 errors at 925 mb. Over the United States a cold bias can be seen of up to 2 deg. C in day 5 forecasts at 925 and 800 mb. A warm bias appeared over the western United States this summer, but now appears in the Arctic Ocean and over the northern borders of Russia and Canada and over Siberia. This may reflect a seasonal transition more than changes in the analysis or model. Cold biases are evident over the oceans, especially near 30S. They may have some relation to low-level stratus clouds which have proven difficult to model accurately. Warm biases appear over the ocean near Antarctic. Fig. 4c and Fig. 6 display the monthly mean analysed (top) and 5 day forecast (bottom) temperature fields at 800 and 925 mb, respectively with the zonal mean removed to emphasize the stationary waves. The thermal contrast between land and ocean is intensified in the subtropics in the forecasts, especially at 800 mb.


Longitude-pressure cross-sections

Equator

> Fig. 7a and Fig. 7b display the monthly mean error in temperature at the equator as a function of pressure for 1 day and 5 day forecasts. A cold bias develops quickly in the eastern Pacific near 750 mb, while warm biases appear at 100 and 200 mb.

30S

Fig. 8a and Fig. 8b display the monthly mean temperature error at 30S in 1 day and 5 day forecasts verified against the analysis. Fig. 9 displays the analysed (top) and 5 day forecast monthly mean temperatures at 30S as a function of pressure with the zonal mean removed at each vertical level. Cold biases are most intense at 800 mb and in the eastern subtropical oceans, in regions where low-level stratus clouds appear. Fig. 10 shows the vertical profile of temperature at 30S 100E in the monthly mean analyses and 5 day forecasts for Oct. A slight inversion can be seen in the analyses between 800 and 900 mb (green); the forecasts (white) however weaken the inversion, a problem seen before in the MRF global model.

60S

Fig. 11a and Fig. 11b display the monthly mean temperature error at 60S in 1 day and 5 day forecasts verified against the analysis. Low-level warm biases appear over the ocean near Antarctica.


DIABATIC HEATING AND TEMPERATURE ERRORS

60-90N

Fig. 70a displays the bias in temperature averaged over 60-90N as a function of sigma layer for day 1 (dashed-dotted green), day 3 (dashed-dotted red), day 5( dashed-dotted white), day 7 (solid blue), day 10 (solid green), day 12 (solid red) and day 15 (solid white) forecasts verifying Oct. 16-31, 1998. The top most sigma layer (1) represents the vertical integral. The bias is against the analysed temperatures for 000 GMT. In the bottom 14 layers, a warm bias increases with forecast length, although day 1 forecasts show a slight cold bias. In the upper atmosphere a cold bias develops. The biases grow throughout the forecasts. Fig. 70b displays diabatic heating averaged over 60-90N as function of sigma layer for various forecast lengths for the month of Oct. The solid white curve gives heating from the first model timestep 4 times a day; the solid red curve displays heating from the first guess forecast (0-6hr) 4 times a day and therefore, unlike the red curve, includes the complete diurnal cycle. The solid green curve is from the 0-24 hour forecast (day1), while the solid blue curve is from the forecast from 48 to 72 hours (day 3). Solid orange denotes day 5 (from 96 to 120 hr), dashed white day 10 (from 216 to 240 forecast hours), and dashed red day 15 (from 336 to 360 forecast hours). Heating during the first timestep has a distinctly difference shape from the others, with cooling in the first layer and maximum heating in the second layer. The others have maximum heating in the first layer. Heating increases until day 3 and throughout the forecast remains larger than in the first 6 hours.

30-60N

Fig. 71a displays the bias in temperature averaged over 30-60N as a function of sigma layer for day 1 (dashed-dotted blue), day 3 (dashed-dotted green), day 5( dashed-dotted red), day 7 (dashed-dotted white), day 8(solid blue), day 10 (solid green), day 12 (solid red) and day 15 (solid white) forecasts verifying Oct. 16-31, 1998. A warm bias appears in the top layer, while a cold bias appears below the top layer in the upper atmosphere. These biases continue to grow throughout the forecast, but a weak cold bias in the lower atmosphere stops growing early in the forecast. Fig. 71b displays diabatic heating averaged over 30-60N as function of sigma layer for various forecast lengths for the month of Oct. The solid white curve gives heating from the first model timestep 4 times a day; the solid red curve displays heating from the first guess forecast (0-6hr) 4 times a day. The solid green curve is from the 0-24 hour forecast (day1), while the solid blue curve is from the forecast from 48 to 72 hours (day 3). Solid orange denotes day 5 (from 96 to 120 hr), dashed white day 10 (from 216 to 240 forecast hours), and dashed red day 15 (from 336 to 360 forecast hours). Heating during the first timestep has a distinctly difference shape again from the other curves, with a maximum in layer 2. Heating increases from day 1 to day 3. Days 10 and 15 shows more heating above the bottom 5 layers than days 3 and 5, but less heating in the bottom layers.

30S-30N

Fig. 72a displays the bias in temperature averaged over the tropics (30S-30N) as a function of sigma layer for day 1 (dashed-dotted blue), day 3 (dashed-dotted green), day 5( dashed-dotted red), day 7 (dashed-dotted white), day 8(solid blue), day 10 (solid green), day 12 (solid red) and day 15 (solid white) forecasts verifying Oct. 16-31, 1998. A warm bias appears in the top layer, while a cold bias appears below the top layer in the upper atmosphere. These biases continue to grow throughout the forecast. A tropospheric cold bias saturates about day 8. Fig. 72b displays diabatic heating averaged over 30S-30N as function of sigma layer for various forecast lengths for the month of Oct. The solid white curve gives heating from the first model timestep 4 times a day; the solid red curve displays heating from the first guess forecast (0-6hr) 4 times a day. The solid green curve is from the 0-24 hour forecast (day1), while the solid blue curve is from the forecast from 48 to 72 hours (day 3). Solid orange denotes day 5 (from 96 to 120 hr), dashed white day 10 (from 216 to 240 forecast hours), and dashed red day 15 (from 336 to 360 forecast hours). Heating during the first timestep has a distinctly difference shape from the other curves, which show a maximum heating in layer 2. Fig. 72c displays heating due to deep convection averaged over 30S-30N as function of sigma layer for various forecast lengths for the month of Oct. The solid white curve gives heating from the first model timestep 4 times a day; the solid red curve displays heating from the first guess forecast (0-6hr) 4 times a day. The solid green curve is from the 0-24 hour forecast (day1), while the solid blue curve is from the forecast from 48 to 72 hours (day 3). Solid orange denotes day 5 (from 96 to 120 hr), dashed white day 10 (from 216 to 240 forecast hours), and dashed red day 15 (from 336 to 360 forecast hours). The figure implies a downward shift in convective heating between day 1 and day 3.


REGIONAL HEIGHT ERRORS

1000 mb

Fig. 12a and Fig. 12b display the monthly mean height bias at 1000 mb in day 1 and day 5 forecasts for Oct. 98. Fig. 13 shows the monthly mean 1000 mb height in analyses (top) and 5 day forecasts for Oct. Negative biases develop quickly over the Caribbean, northern South America and southern Asia; positive biases appear in the 1 day forecasts over the northern coasts of Eurasia and North America and the coast of Antarctica. Day 5 forecasts display too low 1000 mb height in the tropics and over India, weaken the Icelandic low and shift the Aleutian low inland, and strengthen the North Pacific and North Atlantic highs.

500 mb

Fig. 14a and Fig. 14b display the monthly mean height bias at 500 mb in day 1 and day 5 forecasts for Oct. 98. Fig. 15 shows the monthly mean 500 mb height in analyses (top) and 5 day forecasts (bottom) with the sonal means removed. The 5 day forecasts weaken the time-mean ridges over Alaska, in the eastern North Atlantic and in the South Atlantic. In day 1 forecasts negative errors appear in the tropics, near the Himalayas and Rockies and over the southeast United States, while positive biases appear in the Arctic and along the coast of Antarctica.

250 mb

Fig. 16a and Fig. 16b display the monthly mean height bias at 250 mb in day 1 and day 5 forecasts for Oct. 98. Fig. 17 show the monthly mean 250 mb height in analyses (top) and 5 day forecasts (bottom) with the zonal means removed. As at 500 mb, a zonal wavenumber 2 pattern can be seen in the 5 day error, suggesting that the analyses and 5 day forecasts may differ in their depiction of the semidiurnal tide. The monthly means shown here include analyses and forecasts valid at 000 GMT only. A large error can be seen in day 1 forecasts over the eastern United States.


ZONAL WIND ERRORS

Zonal mean

Fig. 18 displays the zonal mean zonal wind in analyses (top) and 5 day forecasts (bottom) for Oct. 98. Fig. 19a and Fig. 19b show the 1 day and 5 day zonal mean errors in the zonal wind. The 1 day forecasts' bias displays tendencies to strengthen the equatorial easterlies near 150 mb and to shift the subtropical jets equatorward and upward, tendencies that increase in the day 5 forecasts. The 5 day forecasts also display stronger low-level easterlies at the equator.

Equator

Fig. 20 displays the monthly mean zonal wind at the equator in analyses (top) and 5 day forecasts (bottom) for Oct. 98. Fig. 21a and Fig. 21b show the 1 day and 5 day monthly mean errors in the zonal wind at the equator. The 1 day forecasts display a large easterly bias near 100 mb over the Indian Ocean. Low level easterlies over the eastern equatorial Pacific are 33% stronger in the 5 day forecasts than in the verifying analyses, while aloft westerlies increase. Upper level easterlies over the Indian Ocean display an easterly bias of as much as 8 m/s in the 5 day forecasts. The pattern suggests increased rising motion near Indonesia and more sinking in the east Pacific, as is seen below in Fig. 34, Fig. 35a, and Fig. 35b.

Fig. 21c shows the vertical profile of the zonal wind at 170W on the equator. Monthly mean analyses (white), 1 day (red), 3 day (green) and 5 day (blue) forecasts for Oct. 98 are shown. An easterly bias in the 5 day forecasts can be seen below 600 mb and a westerly bias above. Much of the bias develops between day 3 and day 5.

Tropical

Fig. 21d displays the growth of the tropical zonal wind bias (with respect to the analysis) with forecast length as a function of sigma layer for forecasts verifying Oct. 16-31, 1998. The vertical coordinate is sigma layer, with the topmost being the vertical integral. Dot-dashed curves represent day 1 (blue), day 3 (green), day 5 (red) and day 7 (black); solid curves represent day 8 (blue), day 10 (green), day 12 (red), and day 15 (black). The bias grows throughout the forecast and is westerly in the top 10 sigma layers and easterly below.

250 mb

Fig. 22 displays the monthly mean zonal wind at 250 mb in analyses (top) and 5 day forecasts (bottom) for Oct. 98. Fig. 23a and Fig. 23b displays the monthly mean bias in 1 and 5 day forecasts of the 250 mb zonal wind. The day 1 forecast error exhibits a substantial weakening of the Asian jet over northern China and Korea and a positive bias over much of the North Atlantic. The 5 day forecasts generally but not always display weaker time-mean jet maxima and weaken the intensity of splits in the flow, such as over the northwest United States and to the west of Spain. Thus the jet maxima over Australia, east Asia and the northeast Atlantic are weaker in the forecasts than the analyses; however, jet maxima over the North Pacific and the southern tip of South America are stronger than observed.


MERIDIONAL WIND ERRORS

Zonal mean

Fig. 24a displays the zonal mean meridional wind in analyses (top) and 5 day forecasts (bottom) for Oct. 98 while Fig. 24b displays the zonal mean meridional wind in 1 day (top) and 3 day forecasts (bottom) for Oct. 98. Fig. 25a and Fig. 25b show the 1 day and 5 day zonal mean difference from the analysis in the meridional wind. The day 1 forecasts show a substantially stronger Hadley cell than the analyses, increasing the upper branch by as much as 1 m/s while the lower branch increases by more than .4 m/s. The increase in the upper branch is lost by day 3, but the lower branch keeps its increase through day 5. In the 5 day forecasts the lower branch of the Hadley cell has a stronger southerly flow across the equator and a slight decrease in the upper level return flow across the equator.

Equator

Fig. 26 displays the monthly mean meridional wind at the equator in analyses (top) and 5 day forecasts (bottom) for Oct. 98. Fig. 27a and Fig. 27b show the 1 day and 5 day forecast monthly mean difference from the analysis in the meridional wind at the equator. The day 1 forecasts clearly show increases in the upper and lower level cross-equatorial flows. The 5 day forecasts decrease the upper level northerly flow near Indonesia and increase the low-level southerly flow in the east Pacific. A complex two-celled vertical structure can be seen in the east Pacific in both the analyses and forecasts; whether this structure is real or not is unclear.


VECTOR WIND ERRORS

150 mb

Fig. 28a and Fig. 28b display the monthly mean winds at 150 mb between 40S and 40N in the eastern and western hemispheres respectively from analyses (top) and 5 day forecasts (bottom) for Oct. 98. Fig. 29a, Fig. 29b and Fig. 29c show the errors in the 1, 3 and 5 day forecasts of 150 mb wind verified against the analysed wind. All three forecast lengths show anticyclonic biases in the vicinity of the upper tropospheric troughs over the western oceans north of the equator. The 5 day forecasts shift the upper tropospheric trough in the west Pacific to the southeast. The 5 day forecasts also act to split the anticyclones over southern Asia and Central America. The 5 day forecasts display stronger westerlies near 30 deg in each hemisphere and stronger easterlies near the equator over the Indian Ocean.

850 mb

Fig. 30a and Fig. 30b display the monthly mean winds at 850 mb between 40S and 40N in the eastern and western hemispheres respectively from analyses (top) and 5 day forecasts (bottom) for Oct. 98. Fig. 31a, Fig. 31b and Fig. 31c show the errors in the 1, 3 and 5 day forecasts of 850 mb wind verified against the analysed wind. The 5 day forecasts intensify lows over India and to the wast of Indochina, intensify the time-mean circulations near Central America, and intensify equatorial easterlies in the Pacific and the Atlantic.


VERTICAL VELOCITY ERRORS

Zonal mean

Fig. 32 displays the zonal mean vertical velocity wind in analyses (top) and 5 day forecasts (bottom) for Oct. 98. Fig. 33a and Fig. 33b show the zonal mean differences between the 1day and 5 day monthly mean vertical motion and the analysed vertical motion. Day 1 forecasts tend to increase rising motion just north of the equator and increase sinking motion in the subtropics. The biggest increase in rising motion occurs in the upper troposphere. The 5 day forecasts broaden the ascending branch of the Hadley cell near 15N and increase sinking near 30N and 40S. Here the biggest increase in rising motion occurs in the lower troposphere.

Equator

Fig. 34 displays the monthly mean vertical motion at the equator in analyses (top) and 5 day forecasts (bottom) for Oct. 98. Fig. 35a and Fig. 35b show the monthly mean differences between the analysed and the day 1 and day 5 vertical velocity at the equator. Day 1 forecasts act to increase rising motion over the west Pacific and increase sinking over the east Pacific and over the Atlantic. 5 day forecasts tend to increase sinking motion in the east Pacific and increase rising motion in the west Pacific and over the Atlantic.

500 mb

Fig. 36 displays the monthly mean vertical motion at 500 mb in analyses (top) and 5 day forecasts (bottom) for Oct. 98. Fig. 37a and Fig. 37b show the monthly mean difference between the 1 day and 5 day vertical velocity and the analyses at 500 mb. Only 000 GMT fields are included; thus, the diurnal cycle of vertical motion is not sampled. Day 1 forecasts show very localized differences from the analysed vertical motion. The 5 day forecasts display a very similar pattern to the analyses, showing a double ITCZ in the Pacific, intense rising motion over central Africa and rising motion over the Bay of Bengal and along the east coast of Asia. The forecasts intensify the split ITCZ in the central Pacific, increasing sinking at the equator and ascent near 10S and 10N. The forecasts also intensify rising motion over the Andes and decrease sinking motion over southeastern Canada.


HORIZONTAL DIVERGENCE

Zonal mean

Fig. 38a displays the monthly and zonal mean analysed and 5 day forecast horizontal divergence for Oct. 98. Divergence aloft and convergence at the surface can be seen near 10N, implying rising motion, while convergence aloft and surface convergence, implying descent, can be seen near 30N and 30S. The figure also implies rising motion near 60N and 60S, over the midlatitude storm tracks. The figure is of course consistent with the zonal mean vertical motion shown in Fig. 32 and the zonal mean meridional velocity shown in Fig. 24a. The 5 day forecasts appear quite similar to the analysed divergence field in the zonal mean. Fig. 38b displays the monthly and zonal mean 1 day forecast horizontal divergence for Oct. 98 and the difference between the 1 day and the analysed divergence. The Hadley cell clearly intensifies in the first 24 hours of the forecast, as can also be seen in Fig. 25a and Fig. 33a; this intensification however is lost by day 5 (Fig. 38a).

Longitude-pressure cross-sections

Fig. 39a, Fig. 39b, Fig. 39c and Fig. 39d display vertical cross-sections of the monthly mean divergence every 10 degrees from the equator to 30N from analyses and 5 day forecasts for Oct. 98. The forecasts maintain the large-scale pattern. At 10N, upper level divergence stretches from the Indian Ocean to Central America, but is interrupted near 130W, where strong surface convergence is overlain by divergence at 700 mb. A similar pattern can be seen over Africa where lower troposphere divergence overlays near-surface convergence. Upper level convergence dominates at 20 N and 30 N. Fig. 39e displays the vertical cross-section of the monthly mean divergence at 10N from 1 day forecasts for Oct. 98 and the difference between the 1 day forecasts and the analysis. The intensification of the ascending branch of the Hadley cell is clear.

200 mb

Fig. 40a and Fig. 40b display the monthly mean horizontal divergence at 200 mb in analyses and 5 day forecasts for Oct. 98 for the eastern and western hemispheres between 40S and 40N. Fig. 41a and Fig. 41b display the monthly mean differences between the 1day and 5 day forecasts and the analysed horizontal divergence at 200 mb for Oct. 98. Day 1 forecasts intensify the ITCZ north of the equator in the Pacific. Day 1 differences from the analysed divergence display many dimilarities to day 5 differences from the analysis. Monthly mean divergence in the 5 day forecasts resembles the analysed divergence on the larger scale. The 5 day forecasts increase divergence over central Africa and over the oceans near India. They create a double maxima near India and increase divergence in the east Pacific ITCZs.


Divergence Squared

Fig. 42 shows the square of the divergence as a function of sigma level for the the band 30S-30N. The top level in the plot represents the vertical integrals; below that the vertical axis is linear in sigma layer. This plot shows the monthly average of the divergence squared averaged over the tropics each day and provides a measure of the noisiness or amount of detail in the analysis. The solid lines represent the analysis (white) and 1 day (red), 3 day (green). 5 day (blue) and 7 day (orange) forecasts; the dotted-dashed lines represent day 8 (white), day 10 (red), day 12 (green) and day 15 (orange) forecasts. Unlike the plots in pressure, this represents the drift in the model, not systematic error. All the forecasts orginate in Oct. and are averaged over all 31 forecasts made from the 000 GMT analysis in Oct. 98.

Fig. 42 shows an increase in squared divergence between the analysis (solid white) and day 1 forecasts (solid red) and a continued increase from day 1 to day 3 (solid green). Little increase is seen between day 3 and day 7 (solid orange). One explanation is that the analysis may be too smooth. The horizontal resolution of the model is truncated after hour 168 (7 days) from triangular 126 waves to triangular 62 waves, so day 8 forecasts (dash-dot white) have much less detail than day 7. Some growth in noisiness is apparent between day 8 and day 10 (dash-dot red).


Forecast Bias in Specific Humidity

30S-30N

Fig. 73 displays the bias in specific humidity averaged over the tropics (30S-30N) as a function of sigma layer for day 1 (dashed-dotted blue), day 3 (dashed-dotted green), day 5( dashed-dotted red), day 7 (dashed-dotted white), day 8(solid blue), day 10 (solid green), day 12 (solid red) and day 15 (solid white) forecasts verifying Oct. 16-31, 1998. Moisture decreases in the tropics in the forecast. Day 8 has a smaller difference from the analysed moisture than day 7; precipitation decreases after the model is truncated from T126 to T62 after hour 168. Moisture continues to decrease after day 8.

PERFORMANCE OF MODEL PHYSICS

This section examines the performance of the global model physics by examining precipitation and near surface fields. Since most of these quantities vary significantly with the diurnal cycle, many of the fields below are accumulated over the diurnal cycle. For instance, precipitation is accumulated at every time step in the 6 hr forecast used in the analysis cycle, then averaged. In calculatng the monthly mean for October, all 124 6-hr forecasts are averaged together to produce a monthly mean for the first guess or for 0-6 hrs. (Often a few 6-hour forecasts are missing during a month, usually no more than one or two.) The monthly mean for day 1 forecasts is the average of the 12 hr and 24 hr flux files, each of which is accumulated over the 12 hours, and thus is an monthly mean of the 0-24 hr forecasts. Similarly, day 3 forecasts are the average of 60 and 72 hr forecasts and for accumulated fields are based on the forecasts between 48 to 72 hrs.

The fields shown above are instaneous fields. A monthly mean day 5 forecast of 500 mb height is based on 31 120 hr forecasts valid at 000 GMT and does not include the diurnal cycle.

Some fields below are instaneous fields and are therefore sampled at 2 points in the diurnal cycle only.

Differences between days 1, 3, and 5 and the first guess are shown to display model drift; it should not be assumed that physical fields from the first guess are always accurate. Comparisons to independent estimates are needed.


PRECIPITATION

Zonal Mean

Fig. 43 displays the zonal mean precipitation over land (top) and ocean (bottom) accumulated during 0-6 hr (black), 0-24 hr (red) 48-72 hr (green) and 96-120 hr (blue) forecasts verifying in Oct. 98. Over land precipitation increases in the extratropics with forecast length; near the equator precipitation tends to shift northward. Much of the increase occurs between 6 and 24 hours into the forecast. Over the ocean precipitation increases in the ITCZ until 72 hrs into the forecast, then decreases. In midlatitudes over the ocean precipitation increases until 72 hrs into the forecast.

Regional Monthly Means

Fig. 44a and Fig. 44b display the monthly mean precipitation accumulated during the first 6 hours (first guess forecast), the first 24 hours (day 1), hours 48 to 72 (day 3) and hours 96 to 120 (day 5) of the forecasts verifying in Oct. 98. Fig. 45a and Fig. 45b display the difference in monthly mean precip for day1-first guess forecasts, day 3 minus first guess, day 5 minus first guess and day 5 minus day 1 forecasts verifying in Oct. 98. Near the equator precipitation increases over the Indian Ocean, over the western Atlantic and near 10N in the Pacific in the first 24 hrs. Precipitation also increases over the midlatitude oceanic storm tracks. In longer forecasts precipitation decreases in the west Pacific and near Indonesia. Precipitation over the Indian Ocean forms a double convergence zone in the 5 day forecasts with stronger maxima than is seen initially. The tendency for the NCEP global model to weaken precipitation in the western Pacific has been seen before. The forecasts also increase precipitation off the northeastern coast of South America while decreasing it over Central America. Over the United States precipitation increases west of the Mississippi after the first 24 hrs. Precipitation also increases north of the equator over Africa.


SURFACE AIR TEMPERATURE

Surface air or 2-meter temperature is an instantaneous field; thus the surface air temperature from the first guess is sampled 4 times a day, while surface air temperature from day 1, 3 or 5 is sampled twice a day. The difference in the sampling of the diurnal cycle makes a difference in surface air temperature that can exceed the model drift; therefore in this section day 3 and day 5 temperatures are compared to day 1 temperatures.

Zonal Mean

Fig. 46 displays the zonal mean difference in 2 meter temperature over land (top) and ocean (bottom) for day 3 (forecast hours 60 and 72) minus day 1 (forecast hours 12 and 24)in black and for day 5 (forecast hours 108 and 120) minus day 1 in red for October 1998. North of 60N air temperatures increase with forecast lengths; they decrease with forecast length in the tropics over land and outside polar latitudes over the oceans.

Regional Monthly Means

Fig. 47 displays the regional monthly mean difference in 2 meter temperature for day 3 minus day 1 and for day 5 minus day 1 for October 1998. A model warm bias can be seen over the Arctic Ocean, much of Canada, Siberia and portiosn of Antarctica; cold biases emerge over the oceans, particularly near 30S, and over midlatitude and tropical continents.


PRECIPITABLE WATER

Regional Monthly Means

Fig. 48 displays the monthly mean vertically integrated precipitable water from the first guess and from the day 5 forecasts for Oct. 98. Fig. 49a displays the monthly mean difference in precipitable water between the day 1 (top) and day 3 forecast (bottom) and the first guess forecast for Oct. 98. Fig. 49b displays (top) the monthly mean difference in precipitable water between the day 5 and the first guess forecast for Oct. 98 and (bottom) the zonal mean differences between the 1 (in black), 3 (red) and 5 day (green) forecasts of precipitable water and the first guess forecast. Precipitable water decreases in the tropics (in agreement with Fig. 73), especially around Indonesia and in the eastern Atlantic, and increases north of 30N. The change in precipitable water resembles the change in precipitation with forecast length ( Fig. 45a and Fig. 45b).


ZONAL SURFACE STRESS

Zonal Mean and Equator

Fig. 50 displays (top) the zonal mean zonal surface stress for the first guess (0-6 hr) in black, 0-24 hr (day1) in red, 48-72 hr (day 3) in green and 96-120 hr day 5) in blue forecasts for Oct. 98 and (bottom) the monthly mean zonal surface stress at the equator from the first guess (black), day 1 (red), day 3 (green) and day 5 (blue) forecasts verifying in Oct. 98. Much of the change in surface stress appears to occur between forecast days 1 and 3. Zonal surface stress appears to increase in magnitude with forecast length, particularly near 50S and in the equatorial east Pacific and Atlantic.

Regional

Fig. 51 displays the zonal mean surface stress for Oct. 98 from (top) the first guess forecasts and from (bottom) day 5 forecasts. Fig. 52a and Fig. 52b display the differences in monthly mean zonal surface stress between the day 1 and day 3 forecasts and the first guess and between the day 5 and first guess forecasts. The largest changes are found near 50S. Zonal surface stress also increases in the North Pacific.


LOW CLOUD COVER

Zonal Mean

Fig. 53 displays (top) the zonal mean low cloud cover for the first guess (black), day1 (red), day 3 (green) and day 5 (blue) forecasts for Oct. 98. Low level cloud cover increases with forecast length, particularly near 60N, but decreases near the equator.

Regional

Fig. 54 diplays the low cloud cover from the (top)first guess and from the (bottom) day 5 forecasts. Note that in the subtropics a minimum in cloud cover occurs along the western coasts of the continents, while just to the west over the eastern subtropical oceans maxima in low-level stratus clouds occur. Fig. 55a and Fig. 55b display the differences in monthly mean zonal low cloud cover between the day 1 and day 3 forecasts and the first guess and between the day 5 and first guess forecasts. Low-level clouds increase with forecast length near 60S and north of 45N everywhere; they also increase over the subtropical oceans. Decreases can be seen in the tropics.


TOTAL CLOUD COVER

Zonal Mean

Fig. 56 displays (top) the zonal mean total cloud cover for the first guess (black), day1 (red), day 3 (green) and day 5 (blue) forecasts for Oct. 98. The pattern is similar to that for zonal mean low level cloud cover ( Fig. 53).

Regional

Fig. 57 diplays the total cloud cover from the (top)first guess and from the (bottom) day 5 forecasts. The pattern resembles that for low level cloud cover ( Fig. 54) except for more distinct maxima in regions of deep tropical convection. Fig. 58a and Fig. 58b display the differences in monthly mean zonal total cloud cover between the day 1 and day 3 forecasts and the first guess and between the day 5 and first guess forecasts. These figures are similar to those for low-level cloud cover ( Fig. 55a and Fig. 55b), except for larger decreases near Indonesia, where precipitation decreases with forecast length.


EVAPORATION

Zonal Mean

Fig. 59 displays (top) the zonal mean evaporation for the first guess (black), day1 (red), day 3 (green) and day 5 (blue) forecasts for Oct. 98. Evaporation increases with forecast length, with much of the increase occurring between day 1 and day 3 forecasts.

Regional

Fig. 60 diplays the evaporation from the (top)first guess and from the (bottom) day 5 forecasts. Fig. 61a and Fig. 61b display the differences in monthly mean evaporation between the day 1 and day 3 forecasts and the first guess and between the day 5 and first guess forecasts. Evaporation increases over much of the subtropical and northern midlatitude ocean; small decreases occur in the central Pacific and Atlantic ITCZs and near the Philippines.


SENSIBLE HEAT FLUX

Zonal Mean

Fig. 62 displays (top) the zonal mean sensible heat flux for the first guess (black), day1 (red), day 3 (green) and day 5 (blue) forecasts for Oct. 98. At most latitudes sensible heat increases with forecast length.

Regional

Fig. 63 diplays the monthly mean sensible heat flux from the (top)first guess and from the (bottom) day 5 forecasts. Fig. 64a and Fig. 64b display the differences in monthly mean sensible heat flux between the day 1 and day 3 forecasts and the first guess and between the day 5 and first guess forecasts. Increases with forecast length occur over northern Eurasia, northern Canada, and the north Pacific and Atlantic.


SURFACE NET SHORTWAVE RADIATION

Regional

Fig. 65 diplays the monthly mean surface net shortwave radiation from the (top)first guess and from the (bottom) day 5 forecasts. Fig. 66a and Fig. 66b display the differences in monthly mean surface net shortwave radiation between the day 1 and day 3 forecasts and the first guess and between the day 5 and first guess forecasts. Surface net shortwave increases with forecast length over Indonesia and over the eastern tropical Atlantic, where total cloud cover and precipitation decreases. The patterns are similar, but opposite in sign to those for total cloud cover ( Fig. 58a and Fig. 58b) and for precipitation ( Fig. 45a and Fig. 45b).


SURFACE NET HEAT FLUX

Zonal Mean

Fig. 67 displays the zonal mean net heat flux for the first guess (black), day1 (red), day 3 (green) and day 5 (blue) forecasts for Oct. 98 for (top) land and (bottom) ocean. Positive values indicate a downward flux into the surface from the atmosphere. Zonal mean net heat fluxes over land are generally small and change little with forecast length, except over southern South America. The ocean absorbs heat near the equator and in the southern hemisphere and gives up heat to the atmosphere in the Northern Hemisphere. The surface net heat flux becomes more upward/less downward with increasing forecast length, with much of the change occurring between days 1 and 3.

Regional

Fig. 68 diplays the monthly mean surface net heat flux from the (top)first guess and from the (bottom) day 5 forecasts. Fig. 69a and Fig. 69b display the differences in monthly mean net surface heat flux between the day 1 and day 3 forecasts and the first guess and between the day 5 and first guess forecasts. The patterns of differences resemble the patterns of differences in evaporation ( Fig. 61a and Fig. 61b), suggesting that a dominant component in the changes in the surface energy balance with forecast length is the change in evaporation. (Note that evaporation is defined as positive upwards, the opposite sign convention from net heat flux.) Changes in net shortwave ( Fig. 66a and Fig. 66b) also appear to contribute significantly in certain regions, such as off eastern Brazil.