Colorado (and Interior Southwest) forecasts

Outline for latest forecast webpage (updated on April 21st, 2008; next update by May 24th, 2008)

This webpage consists of six parts:

1. Status and Outlook for ENSO (El Niño/Southern Oscillation), the most important global climate variability factor on year-to-year time scales;

2. Background information on regional climate variability of the interior western U.S;

3. "Official" CPC (Climate Prediction Center) forecasts for May through September 2008;

4. Experimental forecast guidance for APR-JUN and JUL-SEP 2008 precipitation in the interior southwestern U.S.;

5. Discussion of experimental forecasts, and useful links to shorter-term forecasts;

6. Executive Summary of this webpage.

The most recent forecasts are based on data through March 2008. This website will remain online until further notice.

You are welcome to use any of the material from this website, but proper acknowledgment would be appreciated, especially when referring to figures, forecasts, and assessments unique to this website. PROPER ACKNOWLEDGMENT should state that the used material "was provided by Klaus Wolter at the NOAA-CIRES Climate Diagnostics Center, Boulder, Colorado, from his website at http://www.cdc.noaa.gov/people/klaus.wolter/SWcasts/". The forecasts presented in section 4 are new and experimental - users are cautioned that no responsibility for any losses can be assumed by CDC. For questions, contact me under (Klaus.Wolter@noaa.gov)

1. Current status of the El Niño/Southern Oscillation (ENSO) phenomenon and prospects for the next six to nine months

Recent sea surface temperature (SST) and wind conditions in the tropical Pacific from Australia to just west of South America show negative SST anomalies below -1C between 165E and 145W along the Equator (much reduced from last month), while positive SST anomalies above +0.5C have expanded east of 110W. At the same time, wind anomalies show slightly enhanced trade winds mostly west of the dateline. Compared to last month, cold SST anomalies have weakened further in the central Pacific, while coastal warming has continued near equatorial South America. This figure is taken from the ( TAO/TRITON website) that is updated daily.

Since this figure reflects conditions over the last five days, the reader may be interested in an ENSO discussion that pays more attention to the longer time scales of this phenomenon. A monthly updated ENSO discussion provides for this by the end of the first week of each month, and features a comparison figure of similar ENSO situations near the end of that web site. The Multivariate ENSO Index (MEI) has dropped substantially over the last two months, indicating the third strongest La Niña conditions since 1950 for February-March, anchored by substantial atmospheric circulation anomalies that exceed the SST anomalies shown here in the standardized sense. At this point, persistence of at least moderate La Niña conditions is still likely for the next month or two.

Latest European coupled model forecast for central Pacific tropical SST anomalies. The so-called "anomaly plume" (a.k.a. "spaghetti plot") refers to fifty different forecasts that were computed during the last month from slightly perturbed initial conditions, in order to encompass the range of possible outcomes over the next six months. As the stippled blue line indicates, Nino3.4 SST anomalies have risen dramatically from almost -2C in February to just under -1C in March. While earlier forecast runs missed the mid-winter drop in SST, this recent increase was well predicted in February. Beyond March, a slow drift towards near-neutral conditions is anticipated in this model, with two ensemble members reaching positive anomalies by September 2008. This figure is taken from the ( ECMWF seasonal forecast website) that is updated monthly around the 21st - not in time for this month's SWcasts update.

Latest comparison plot of now 15 dynamical and 8 statistical model forecasts for central Pacific tropical SST anomalies. For those models that are run in ensemble mode (like the ECMWF forecast - note the one shown here is from LAST month, summarizing the information from the previous figure), this graph shows the average outcome. The range of possible outcomes continues to favor at least weak La Niña, but there are now two models that quickly rise above the weak El Niño threshold by Jun-Aug, only to continue rising through the remainder of the year, and two further models that reach this threshold by late 2008. As before, there is a split between dynamical models that trend more towards near-normal conditions by late 2008 and statistical models that tend to perpetuate at least weak La Niña conditions through the remainder of 2008. This figure is taken from the ( IRI ENSO model website) that is updated monthly.

For an independent interpretation of the current situation, I recommend reading the latest NOAA ENSO Advisory which represents the official and most recent Climate Prediction Center opinion on this subject. The April 2008 CPC statement describes La Niña conditions as being "expected to continue for the next 3 months".

2. Regional climate background information

Average timing of wettest three-month season during the course of the year is shown for just over 1,000 historical climate records in the interior Southwest. This figure includes both cooperative observing stations (mostly in the lower elevations) and automated SNOTEL sites (in the montane and subalpine regions of the Rocky Mountains). The number shown refers to the CENTRAL month of a three-month season (JAN=1, FEB=2,...,DEC=B). Note the spatial extent of the winter/early spring wet season (red numbers) over Colorado and points west: mostly at high elevations. By comparison, the summer monsoon season brings the most moisture to almost all of New Mexico and much of southern Colorado, southeastern Utah and eastern Arizona (blue numbers).

This companion map to the average seasonal wetness peak shows the average driest season during the course of the year. Note the lack of precipitation over eastern Colorado and New Mexico during the winter season, while Arizona experiences its driest season during late spring prior to the monsoon.

This map shows regions of similar precipitation variability for the Interior Southwestern U.S. for April through June, a season that requires nine core regions to describe the bulk of co-variability across the domain. COOP stations are indicated by circles, and SNOTEL sites by triangles. The amount of color in each station symbol represents the amount of local variance that is explained by the index time series created for each core region.

ditto for July through September, a season that requires ten core regions to describe the bulk of co-variability across the domain.

3. Most recent Climate Prediction Center forecasts for May through September 2008

The most recent U.S. government (CPC) temperature forecast for May 2008 shows a tilt of the odds towards warmer temperatures covering most of Arizona and New Mexico, while most of Utah and Colorado are left blank ("EC"), translating into equal chances of below-normal, near-normal, and above-normal temperatures. This is mostly due to the current La Niña phase, as well as long-term spring warming trends in this region. The precipitation forecast for May 2008 (below) shows below-normal covering almost the same region as the above-normal tempearture forecast (Arizona and New Mexico), with an extension into Nebraska that covers easternmost Colorado as well. The remainder of Colorado and Utah are left "EC". This forecast reflects various statistical tools, but is mostly consistent with La Niña composites for the spring season. Both forecasts are taken from the monthly outlook website that is updated on the third Thursday of each month.

From the related seasonal outlook website, the next set of forecast maps shows the national CPC forecasts for July through September 2008. Given the expectation of weakening La Niña conditions through that season, these forecast maps reflect their reliance on long-term trends for temperatures, and La Niña composites for precipitation patterns. In the case of temperatures, this translates into a large tilt of the odds towards warm conditions in much of the western U.S. for this late summer season (see first map below), leaving only the easternmost plains of Colorado and New Mexico "EC", while exhibiting a strong tilt (17% or higher = 50%+ probabilities for the warmest tercile) over Utah and Arizona. In the case of precipitation, this covers the northern third of our domain (most of Utah and the northern half of Colorado) with slightly below-normal odds, while leaving the southern two thirds, including all of Arizona and New Mexico "EC".

Other links one might want to browse are the IRI forecasts, as well as our in-house model-derived CDC forecasts that are based on statistical modeling of impacts of global SST on atmospheric circulation models. While the IRI forecasts normally mimic the CPC forecasts shown above, the CDC model-derived forecasts for the next three months often do not. In this case, they favor dry conditions everywhere in the interior southwest, with slightly lower probabilities in northwestern Utah.

4. Most recent experimental forecast guidance - for April-June and July-September 2008 (last complete month of data used: March 2008)

The forecast guidance presented here is based on a screening stepwise multiple regression procedure that requires that each new predictor explains at least an additional 10% of the predictand's variance. These calculations were performed for the full record (1951/2-98/9), as well as for five subsets with a decade of data being withheld. The predictors I use include a variety of ENSO indices (including spatial differences and recent tendencies in time), regional SST averages in the Indian Ocean, Eastern Pacific and Caribbean, regional precipitation anomalies within the southwestern U.S., and a few general sea level pressure indices such as the North Atlantic Oscillation and centers of action both east and west of Australia.

For this go-around, I include calculations for the following targets (or predictands):

A. April-June 2008 precipitation for seasonal core regions 1 - 9.

B. July-September 2008 precipitation for seasonal core regions 1 - 10.

FORECAST TABLE:

First column is predictand (name of regional rainfall anomaly index);

2nd column is predicted standardized anomaly for the named region based on full training period (1951-98);

3rd column is predicted standardized anomaly based on the median of the cross-validated forecasts (i.e., the middle of five predicted values;

4th column is the range of predicted values from all six forecasts;

5th column is Heidke Skill Score (HSS) for full training period (based on 3x3 contingency table using terciles - a score of 0 means that the forecasts are no better than what would be expected by change (one hit (correctly predicted tercile) out of three tries), a score of +100 would mean "perfect" forecasts (three hits out of three forecasts), and -50 would be the worst possible forecast skill (no hits at all); for a comprehensive discussion of forecast verification issues, I recommend this link , while a "light-weight" definition of this score can be found here.

6th column is HSS for all five cross-validated decades combined (independent verification);

7th column is the calibrated shift in the probability distribution towards positive or negative anomalies (upper or lower tercile);

8th column shows the change in the 7th column from last month, if applicable;

9th and last column shows a preliminary HSS for the last eight years (1999/2000-2006/2007); the number of forecasts with a tilt of at least 3% is indicated in brackets.

Note that the 7th column mimics the CPC forecast scheme in which the middle tercile category is left at 33.3% (unless otherwise indicated), and the upper and lower terciles add up to 66.7%, with the gain of the upper tercile meaning the loss of the lower tercile, and vice versa. If the data were normally distributed, and the skill in forecasting completely unbiased, a median predicted anomaly of +0.26 would translate into a shift in the terciles of +10%, with the lowest tercile being less likely than "normal" at 23.3%, while the highest tercile would be more likely at 43.3%.

See also the discussion section following this table. 

PRECIPITATION FORECAST GUIDANCE FOR APR-JUN 2008 (A) AND JUL-SEP 2008 (B)


TIME-REGION              PRED. ANOMALY   PRED.   HSS     HSS     CROSS-VALIDATED   CHANGE FROM             HSS
PREDICTAND               FULL   MEDIAN   RANGE   FULL    IND.     TERCILE SHIFT  PREVIOUS FORECAST      2000-2007


A1. AMJ-1=NW UT/SW W       -.2     -.4    1.5      +34     +9         -8%               -3%               +50*(6)


A2. AMJ-2=NW CO/Ctl UT     -.2     0.2    2.0#     +22     +9	      -8%          	-8%                +6 (8)


A3. AMJ-3=SW UT            0.1     0.4    1.1      +38    +16         +9%               +7%                -7#(7)


A4. AMJ-4=most of AZ       0.4     0.4    1.8      +28     +3	      +1%               -3%               +25 (4)


A5. AMJ-5=SE AZ/S NM       -.2     -.7    1.5      +34    +22        -11%*              -1%               +36*(7)


A6. AMJ-6=NC NM+           0.1     0.4    2.1#     +41     +6         +8%               +2%               -29#(7)


A7. AMJ-7=SW CO+           -.1     -.7    1.1      +44    +22         -9%               -7%               +25 (4)


A8. AMJ-8=E CO Plains      1.1     0.9    2.4#     +44     +3        +13%*              +2%               +44*(8)


A9. AMJ-9=NE CO/SE WY+     -.7     -.3    1.3      +19     +6         -2%              +/-0%              +10 (5)
==================================================================================================================


B1. JAS-1=NW UT            -.7     -.1    2.3#     +34    +19        +/-0%               /                 -7#(7)


B2. JAS-2=SE UT/NW CO     -1.4     -.4    2.1#     +44     +9         -3%                /                -29#(7)


B3. JAS-3=most of AZ       -.6     -.4    1.6      +31     +6         -6%                /                -25#(6)


B4. JAS-4=SW NM            1.1     1.1    2.2#     +22    +19        +11%*               /                +14 (7)


B5. JAS-5=NE NM/Sangres    -.8     -.5    1.1      +31    +6         -10%                /                +14 (7)


B6. JAS-6=SE NM/NW TX     -1.2     -.6    1.4      +44   +22         -20%*               /                -13#(8)


B7. JAS-7=NC NM/SC CO      0.1     0.0    2.2#     +38    +3          +1%                /                  0#(0)


B8. JAS-8=SW/C CO (Mtn)    0.5     0.3    1.5      +34    +6          +4%                /                -13#(4)


B9. JAS-9=E CO (plains)    0.2     0.7    2.1#     +41   +22          +5%                /                 -7#(7)


B10.JAS-10=N Front Range   -.0     0.3    1.4      +31   +31*         +8%                /                 +6 (8)

I have marked with an asterisk (*) those forecast probability shifts that equal or exceed 11% (or, roughly, a doubled risk of one tercile category vs. the opposite one). The same applies to cross-validated as well as 1999-2006 verification Heidke skill scores above +30, as well as to predicted ranges of 0.5 or less. For AMJ'08, three out of nine predictions qualify for an asterisk: regions 1, 5, and 8 have shown good verification skill in the last eight years, as well as a significant forecast tilt (5, 8). For JAS'08, three out of ten predictions qualify for an asterisk as well, but none of them for good verification skill. Instead, there are two significant forecast tilts (4,6) and one good cross-validated skill score (10).

In contrast, a pound (#) symbol indicates a very large spread (at least 2 standard deviations), weak cross-validated or 1999-06 HSS's (0 or less), or a large change from last month's forecast. Such forecasts should be taken with a 'grain of salt'. For AMJ'08, four out of nine regions are handicapped in this fashion, mostly due to a large range in forecast members (2,6, and 8), as well poor verification skill (3 and 6). For JAS'08, EIGHT out of ten regions share this stigma, all of them due to poor verification skill at this lead-time (all but 5 and 10), and an additional five due to their wide range in ensemble forecast members (1,2,4,7, and 9). Sometimes, regional forecasts feature both symbols (* and #), which reduces the confidence one should place in those forecasts. In this forecast round, AMJ'08 features only one such contradiction: region 8 has good verification skill and a large tilt in the odds, but the range of ensemble members is too large. For the JAS'08 forecasts, there are two such juxtapositions: region 4 has a large tilt in the odds, but also a large range in ensemble members, and region 6 also has a large shift, but poor verification skill.

The map below shows the calibrated shifts in the tercile probabilities for AMJ 2008 (top) and JAS 2008 (bottom):

Forecasted shifts in tercile probabilities for April-June 2008. In order to be shown on this map, a forecast tilt in the odds has to reach at least 3% either towards wet, dry, or near-normal. Shifts towards the wettest (driest) tercile are indicated in green (red), with a green plus sign for shifts between +3% and +5% (none), and a red minus sign for equivalent shifts towards the negative (none). Tilts towards near-normal are indicated by the letter "N", if at least by 3% (none). Question marks denote a forecast with a greater likelihood of being either wetter OR drier than near-normal at the expense of near-normal odds (none). Positive or negative shifts of over 5% are contoured in 5% increments. If any shift reaches over 10%, it is considered significant. There are two regions that reach that threshold: eastern Colorado on the positive ("wet") side, and southern New Mexico on the negative ("dry") side.

ditto for July-September 2008. This one has a question mark in northwest Utah, indicating a more likely outcome in the tails of the distribution than the middle, and one plus sign (in southwest Colorado) and one minus sign (in eastern Utah) each. Eastern New Mexico is covered by significant negatively tilted odds (10-20%) and southwestern New Mexico by positively tilted odds in excess of 10%.

NEW: To access an archive of earlier public experimental forecasts and verifications, click on the season and lead-time of interest, once you get to the respective directory. Average Heidke Skill scores for each forecast season are listed here. I will update this archive in 2008.

5. Discussion of forecast guidance

To repeat from the beginning of this webpage: This forecast guidance is experimental, and should be used with caution. The Climate Diagnostics Center cannot assume any responsibility for losses incurred related to them.

===========================================================

Discussion of Spring (April-June 2008) forecast guidance:

===========================================================

AMJ-1 (NW UT/SW WY) shows a moderate tilt of the odds (-8%) towards dryness this spring, along with modest cross-validated skill (HSS-IND=+9), and very high verification skill: HSS'00-07=+50. In other words, a dry spring is more likely than it was in earlier forecasts (with a -5% tilt, and lower verification skill).

AMJ-2 (NW CO/Ctl UT) has a moderate tilt in the odds towards dryness (-8%), down from a neutral forecast last month, but with only modest cross-validated skill (HSS-IND=+9) and similar verification skill (+6). In addition, the spread among the forecast ensemble members is suspiciously large (2.0 standard deviations). This leaves a cautiously dry forecast in place for a region that tends to be dry with La Niña.

AMJ-3 (SW UT) has risen from a weak positive tilt (+2%) last month to a moderate positive tilt (+9%). While cross-validated skill is moderate (+16), the verification skill for the last eight springs has been poor (-7). In the context of typical La Niña spring dryness, I would caution against expecting a wet spring.

AMJ-4 (most of AZ) has a negligible positive tilt in the odds (+1%), down from +4% last month, along with very modest cross-validated skill (+3). The verification skill since 2000 has been quite good for those four forecasts that deviated by at least 3% from climatological odds (+25), but the small tilt in the odds does not justify anything but climatological odds this time around. Given the ongoing La Niña situation, anything but a dry spring would be a surprise, but then the winter was much wetter than expected as well.

AMJ-5 (SE AZ/S NM) shows the only significant negative tilt in forecast odds (-11%). This forecast is buttressed by the best cross-validated skill score (+22) of this go-around (tied with region 7), and has been verifying very well since 2000 (+36). Furthermore, La Niña tends to favor dry springs, so this is the most unqualified dry forecast of this go-around.

AMJ-6 (NC NM+) remains under a positive forecast tilt (+8%), slightly up from last month's +6%. Modest cross-validated (+6) and poor verification (-29) skill caution against too much optimism, especially considering the more typical lingering La Niña dryness in this season, and a very large spread among forecast ensemble members.

AMJ-7 (SW CO+) exhibits a fairly large drop from last month's forecast, with a negative tilt of 9% compared to last month's -2%. Decent cross-validated (+22) and verification (+25) support this pessimistic forecast that also fits typical La Niña assocations. Given the slowly dwindling, but still above-average snowpack, a dry spring would reduce the flooding danger.

AMJ-8 (E CO Plains) is the only region with a significant positive tilt, now at +13%. This is supported by very good verification skill (+44%), but only marginal cross-validated skill (+3). Furthermore, there is a very large spread in forecast ensemble members (2.4 standard deviations), leaving a nagging doubt about the validity of this forecast, especially considering the more typical dryness associated with La Niña springs.

AMJ-9 (NE CO/SE WY+), just to the north of AMJ-8, remains unchanged at a negligible tilt (-2%). This is now supported by only modest cross-validated (+6) and verification skill (+10), leaving climatological odds as the best bet, even though La Niña springs tend to be dry in this region as well.

=====================================================

Discussion of summer (July-September 2008) forecasts:

=====================================================

JAS-1=NW UT shows an enhanced tilt towards the tails of the distribution than the middle (+/-0%), along with verification skill (-7). Decent cross-validated skill (+19) cannot compensate for this lack of preference towards one "wet" or "dry" odds, so climatological odds are the safest bet for now.

JAS-2=SE UT/NW CO features a modest tilt of the odds towards the driest tercile (-3%), but cross-validated skill is only modest (HSS-I=+9), while the verification skill level has been very poor (-29). Another case for climatological odds.

JAS-3=most of AZ has a moderate tilt of the odds towards a dry monsoon season (-6%), along with modest cross-validated (+6), but poor verification (-25) forecast skill. This is the only forecast region with a known tendency for wet monsoons during La Niña conditions, and its dry winters also tend to be followed by a wet summer. Since this winter was not dry, the latter tendency does not apply. All in all, this is still a case for climatological odds, mainly due to the conflict between typical La Niña summers and a dry forecast with poor verification skill.

JAS-4=SW NM features a significant above-normal forecast (+11%) for this season. Skill levels have been moderate this far out in the cross-validated sense (+19), and in the verification sense (+14). Unfortunately, the spread in the ensemble members is disturbingly high (2.2 sigma), and La Niña summers are not known for wetness in this region. While some optimism appears warranted, it is worthwhile waiting for later forecasts, which have better skill as we get closer to summer.

JAS-5=NE NM/Sangres shows the opposite tilt in its foreast (-10%), supported by only limited cross-validated skill (HSS-I=+6), and slightly better verification skill (+14). Since La Niña summers tend to be drier than average, a dry monsoon season is indeed more likely than not.

JAS-6=SE NM/NW TX features the largest tilt in the precipitation odds for this season (-20%). This is backed up by decent cross-validated skill (+22), but contradicted by poor verification skill (-13). Cautious pessimism seems to be the safest bet, unfortunately in the wake of a dry fall and winter.

JAS-7=NC NM/SC CO shows a negligible tilt in the forecast odds (+1%), along with marginal cross-validated skill (+3) and no skill since 2000 (0). Climatological odds are the safest bet for now.

JAS-8=SW/C CO (Mtn) has a modest tilt of +4%, along with modest cross-validated skill (+6) and poor verification skill (-13). Another case where climatological odds are most appropriate for now.

JAS-9=E CO (plains) features a modest positive tilt in the forecast odds (+5%), supported by good cross-validated skill (+22). Unfortunately, verification skill levels have been poor (-7), and the spread in the ensemble members is bigger than two standard deviations (2.1). This is a case where climatological odds are a tad conservative, especially since neighboring region 10 also supports a wetter-than-average forecast (see below).

JAS-10=N Front Range is the only summer forecast region with significant cross-validated skill (+31), a fact which is somewhat supported by recent verified performance (+6). Since the forecast tilt is moderately high (+8), I believe cautious optimism is warranted.

=================================================

Links to further pertinent websites

=================================================

Given continued drought conditions in much of the western U.S., there is great concern about streamflow and reservoir levels. To help with the near-term assessment of precipitation prospects, I am enclosing a few links of interest:

1. Outgoing Longwave Radiation ( (OLR) anomalies for the last week) show up as blue for deep convection (over the tropics). Recent thunderstorm activity over the Indian Ocean and Western Pacific is indeed flagged by this color, while recent dry conditions west of the dateline (and Equator) and over mmost of the U.S. show up as yellowish-orange colors - all of these are consistent with La Niña.

2. Under the following link, click on "latest forecast", then on "precip" at the top (This link is based on work originated by Jeff Whitaker) and forecasts generated here at CDC, with a recalibrated version of the Medium Range Forecast (MRF) model that was originally developed at NCEP. A related user-friendly link to "zoomed-in" forecasts is maintained by ( Gary Bates) who takes the recalibrated MRF forecasts and puts them into an interface where the user can pick and choose the grid box for which a daily temperature or precipitation forecast out to fifteen days is desired. Both of these websites have had recent problems with staying current, but somewhat warmer and drier than average weather appears on tap for the southwestern U.S., including Colorado.

6. Executive Summary (updated on April 21st, 2008; next update by May 24th, 2008)

1. A moderate-to-strong La Niña event has been in place since August 2007, and is expected to continue for at least another three months, possibly right through our summer season, albeit at weaker strength.

2. The winter season brought above-normal moisture to Colorado's mountains, in excess of La Niña-based expectations. In addition, colder-than-average temperatures in much of the Western U.S. have kept more snow on the ground into April than has been typical for several years. However, March and early April have shown signs of a more typical La Niña-related pattern of drier conditions in the southwestern U.S., along with warmer temperatures. This has initiated snowmelt below about 9000 feet elevation. The remainder of April appears stuck in this less favorable pattern, so that even higher elevations will start losing their snowpack, although occasional incursions of colder air will keep this meltdown at a manageable pace.

3. My experimental forecast guidance for the late spring season (April-June 2008) is still optimistic for eastern Colorado, while western Colorado is now more likely to be on the dry side. Once we get through the remainder of April, the odds for a dry spring go up region-wide due to lingering La Niña effects, even in eastern Colorado. A first look at the summer season (July-September) is handicapped by poor verification skill since 2000, except for most of New Mexico (wet in southwest and dry in east) and northeastern Colorado (wet). As we get closer to the summer, the skill level of these forecasts goes up, and a firmer outlook should be possible.

4. Bottomline: After a surprisingly wet winter season, March and early April have finally shown more typical La Niña-related dryness in much of the southwestern U.S., with lingering wintry weather in north-central Colorado, as is not uncommon with La Niña. While the storm track remains active, it has started to shift to the north, benefiting Wyoming (and Montana) more than Colorado and adjacent states. Thus, the remainder of spring is likely to bring dry weather to the mountains of Colorado and Utah, easing the flooding threat from one of the highest snowpacks in recent memory.

Questions about this webpage should be addressed to:
(Klaus.Wolter@noaa.gov), (303) 497-6340.