1. What is the 'target' for maximum allowable bias in sea surface temperature data products for climate analysis?  What are the major corrections made to sea surface temperature data analyzed in Ra03, and has that requirement been met?
2. What are the major (large-scale spatial, decadal-centennial timescale) features evident in the Ra03 SST data product? 
   
3. What is, and what is the effect of the "reduced space optimal interpolation" procedure?  Are you convinced it is a useful addition to straightforward averaging?
   
4. What is the working hypothesis in Schubert et al. (2004; hereafter Sch04) for the mechanism behind Great Plains droughts?  What is their strategy for testing this hypothesis?
5. What are the major common features in the modeling and observations?  What are the major differences?  What do you conclude about the ability of the model to realistically simulate drought over the coterminous US?  For our ability to predict or hindcast the extent and amplitude of any one drought?
   
6. Is there evidence for a regular or periodic cycle of drought activity in the Great Plains, and a decadal-timescale mechanism forcing variation in precipitation?  Why/why not?

1. What are the strengths and weaknesses of each variety of paleoproxy data?  Make yourself a table.
2. Which sources of paleoclimate proxy data do you expect to be most useful for the study of decadal climate variability?  Why?
3. What are the pros and cons of the climate field reconstruction approach described in Ch. 9,11?
4. If you could have more observations from anywhere on the planet for studying decadal climate variability, from where would you get them, and from which proxies?
   

1. Do you agree with the conclusions reached by the NRC panel concerning the value of multiproxy climate field reconstructions spanning the past 1000 years?
2. What is the value of multiple proxies?  Would we have a better reconstruction if we used only one or another type of proxy observation?
3. Are there reasons why the proxies themselves would tend to amplify or damp out decadal scale variations?  How about the process of producing climate field reconstructions?
4. Is there decadal variability in the MBH99 reconstructions?  How do pre-20th century and 20th century timescales and amplitudes compare?

[AZC 1/30/2008] There are a myriad of ways to reconstruct variability over times beyond the instrumental record. Our discussion focused on the reconstruction carried out by Mann et al. (1998). Global climate is complex so how can we reduce this complexity? By looking at patterns. How do we examine patterns? PCA and EOF analysis are a good way to do this. What is PCA? Patterns in space, pattern strength and behavior in the system. Mann et al. (1998) used the approach of breaking the climate field into EOFs and separately broke the proxy records into EOFs then attempted to predict the climate field temperature EOFs from the proxy EOFs. We discussed that signal strength could be locally dependent because of the regional nature of the proxy records but that this problem is reduced by using a global set of records. When examining the individual spatial patterns of the EOFs it was brought up that we have to be careful about assigning physical processes to these patterns since the patterns are essentially statistical constructs. We discussed synoptic patterns moving through the data and how they may not be detected. The global nature again may help or hinder in the detection of these processes depending on the calculations used. This issue could be resolved by using the frequency domain to look at the records.

We then discussed what assumptions must be made when reconstructing climate in this manner. Linear relationship must exist between the proxy and data patterns. We need few degrees of freedom and the patterns can’t move i.e. the NAM must always be represented by a ring around the North Pole otherwise the reconstruction won’t work.

Doing this analysis we choose records that show the patterns strongly. How much do we lose? Error bars get larger with time because we lose records. This could introduce a spatial bias as well if all the longer records come from one region or if weighting lends more strength to those records.

Can we look at decadal variability this way? Well we can look at the EOFs and the PCA. We could model decadal variability both by just having a model and letting it run unforced or by forcing a model and seeing what develops. Could we use a bare planet model? Yes, but how bare can you go before you lose the complexities needed to produce the necessary variability. Sometimes the complexities are necessary.

Can we rely on proxies? Well we only have 100 years of instrumental data so we must, but how does autocorrelation or dating errors affect proxy data? Autocorrelation can smooth the data bringing out more in the decadal range relative to the annual. Age model uncertainty can in some cases enhance the decadal signal relative to the annual signal as well.

1. What is meant by deterministic vs. stochastic components of climate variability?
2. What is "red noise"?  What does it look like on a power spectrum?
3. In the hypothetical example, the thermal mass of the bucket must be small relative to the thermal mass of the atmosphere for the model to be valid.  Why?  What happens if this assumption is violated?
   

1. Consider how the different insolation forcing cycles affect temperature variability on diurnal, annual and Milankovitch (obliquity, eccentricity, precession) timescales as a function of latitude.
2. Consider Toby's demonstration of the effect that age model uncertainty has on the power spectrum of 3,7,and 10 year cycles.  How might this biase the power spectrum, especially for long-term climate proxies?
3. How might deterministic processes at certain timescales translate into stochastic processes at other timescales?  More specifically, how might decadal timescale climate variation could be generated from the annual cycle?

Climatic processes can be described as either deterministic or stochastic; they are essentially antonyms. Deterministic processes are governed by consistent mechanisms that are understood and should be predictable; the same inputs into a deterministic mechanism should always have the same outcome (unless the system is chaotic). A stochastic mechanism has a random component to it, so it is unpredictable on an individual basis, so is characterized statistically.  White noise is completely random data, where each value has no predictive capability for the next value, and is flat on power spectrum (i.e. it has equal power at all frequencies). Red noise has more power at lower frequencies (red light is low frequency for visible light), so the power spectrum of red noise is higher at lower frequencies.  Because atmospheric heat input into the ocean is an essentially stochastic process (ignoring ocean-atmosphere feedbacks), and because the thermal diffusivity of water is relatively low, the ocean effectively filters high frequency atmospheric heat input, causing a tendency towards low-frequency (decadal) variability (red noise). Understanding how much of a role this phenomenon plays in observed decadal variability is a key question, which is discussed by Pierce (2001). By observing the autoregressive characteristics seen in SSTs and feeding white noise atmospheric input, Pierce (2001) generates artificial time series that show the same sort of variability as the PDO (fig. 3 and 4.). Using coupled ocean-atmosphere models of increasing complexity, Pierce (2001) tested the relative importance of the white-noise response, finding enhanced variability in 20-year variability, generally in the same region as in the white noise response.  The bucket example relies only on the thermal response of water to atmospheric heat forcing. This ignores the role of ocean circulation, but also requires that the bucket be small; in the real ocean the amount of heat stored in the surface ocean is so large relative to the atmosphere that the ocean drives atmospheric heat content, rather than the other way around.

We concluded that both Mann et al. (2000) reconstructions (cold season tropical; warm season N. American average) show decadal variability, and that it is observable both in the raw data series, and even more so in the filtered series.  The series were correlated, showing the same first-order trend, and appear to covary on decadal timescales as well, although the mechanism for this correlation is unclear. First, the two datasets are derived from the same records, so in that sense the co-variation is to be expected. It has also been suggested (e.g. Schubert et al., 2004) that Pacific SSTs drive warm season drought and temperature in North America, however Schubert et al. suggest an inverse relation (i.e. cold tropical Pacific SSTs correspond to warm NAm summers, and vice versa), which is contradictory to the positive correlation observed between the two datasets.  We might have to remove the warming trend in both series to see support for Schubert.

[NM 2/4/08]   Huybers and Curry (2006) used a suite of instrumental and proxy records to look at periodic variability in the climate system, both spatially and temporally. A significant, if intuitive, finding was the increase in power of the annual cycle with latitude and other continents, and the consequent change in the shape of power spectra from high vs. low latitudes. Higher latitudes and continents experience greater climate variability; particularly during the annual cycle, but also on orbital timescales. Because the power of the annual cycle is greater, the slope of a log-log plot of the power spectrum (i.e. the power law coefficient, β) is lower at high latitudes and over land.

Perhaps the most profound implication of Huybers and Curry’s (2006) result, is the possibility that what they call “continuum temperature variability” (i.e. variability on timescales greater than annual but less than orbital) is a response to deterministic annual and Milankovitch-scale deterministic forcing. This would mean that internal climate variability (e.g., ocean circulation and “memory”, ice sheet dynamics, etc.) serves as a mechanism to transfer deterministic annual and orbital variability to other frequencies. Both their high- and low-latitude compilations indicate a “kink” in the power law at frequencies of ~100 yr, perhaps suggesting that climate variability at timescales <100 yr (> 100 yr) is the continuum response to annual (orbital) deterministic forcing.  It also may suggest that although ultimately it's all solar, the sub-centennial variations are driven by internal mechanisms distinct from those at lower frequencies.  It is in this way that chaotic processes in the climate system may translate deterministic forcing at one frequency to stochastic at others.

We also continued our discussion of how white noise can be manifest as lower-frequency variability mathematically, specifically discussing a couple functions (i.e., random walk and coherence resonance) that receive white noise input but have red noise output. I’ve attached some matlab code that lets us play with these functions.

1. ENSO is a big part of anomalies in the global circulation described here.  How do the authors remove ENSO-related variations to explore patterns and timescales of lower-frequency variations?
2. What are the primary similarities and differences in ENSO and ENSO-like interdecadal climate variations?
3. What observed phenomena do GB99 link to ENSO and decadal ENSO-like variations?  What is the evidence these are based in the tropical Pacific?
   
4. Does the evidence require mechanisms for decadal variability distinct from those producing ENSO?

1. What are the mechanisms by which atmosphere-ocean models produce decadal-scale variability in the Pacific?
2. What is the fundamental difference between the mechanisms L98 reviews, and that of N03?  How is the N03 null hypothesis different from
   
3. What implications does the N03 expanded null hypothsis have for the ability of proxy data to record North Pacific decadal variability?
   

[NM 2/22/08]  We discussed two mechanisms for the Pacific Decadal Oscillation, following the discussion of Latif (1998) and Newman et al. (2003). Latif (1998) reviewed the model simulations of a primarily extratropical mechanism, driven by the subtropical ocean gyre. The mechanism is fairly simple: beginning with an anomalously strong subtropical gyre, warm tropical waters are transported north along the western boundary current and it’s extension. This warms SSTs in the NW pacific, resulting in a stronger gyre, and a positive feedback loop, allowing for the development of the Pacific-North American pattern. The atmosphere responds with a negative feedback, a wind stress curl anomaly, which dampens the poleward transport of warm equatorial water, weakening the gyre. This mechanism may explain the persistent decadal oscillations observed in the PDO, and Latif (1998) shows that it is simulated by models with ranging complexity, providing support for the hypothesis.

The analysis performed by Newman et al. (2003) takes the discussion in the other direction, suggesting that the decadal variability observed in the PDO is a function of the influence of interannual tropical variability (ENSO) on the SSTs in the North Pacific and the reemergence of the previous year’s SST anomalies in the North Pacific. They test this hypothesis by developing a simple model, where they attempt to model the PDO as a function of this year’s ENSO index, last year’s PDO index, and white noise. The resulting signal is remarkably similar to the observed PDO over the past 100 yr, and the power spectrum of multiple iterations of this analysis suggests that the decadal variability in the PDO can be explained by this simple hypothesis. There are two consequences of these results, first, this is a new null-hypothesis for the PDO, and it is unclear whether or not other mechanisms are needed to explain decadal variability in SSTs in the North Pacific. Second, it raises the question about decadal variability in proxy series that have been attributed to the PDO, because climate integrators which are sensitive to ENSO, and that are autocorrelated may show PDO-like variability, but may not be forced by decadal scale variability in SSTs in the North Pacific.

1. What are the pros and cons of linear inverse modeling Pacific decadal variability?
   
2. What are the mechanistic and diagnostic features of the ocean-atmosphere system which are captured by the LIM?  Which aren't?
3. How can a system exhibit decadal variability and yet have skillful predictability for perhaps only 1 year lead times?
   

[AT 2/28/08] This week, we discussed whether the PDO is a predictable pattern and how much North Pacific (NP) SST predictability is influenced by the tropics, or vice-versa. A linear inverse models (LIM) was used to analyze interactions with the tropics and NP SST anomalies because individual patterns are not ideal in our complex ocean-atmosphere system. LIMs, in a nutshell, set up a model from what you’re measuring. It models a system that is consistent with the outputs. They are simple models with empirically-derived modes that assume stable, linear dynamics.

Can LIM reproduce spectral peaks from the ENSO signal? To look into this, Newman used the HADlSST dataset to calibrate the model, ran it, and plotted the power spectra. The plots indicate that coupling is important at lower frequencies (decadal?). Uncoupled patterns (the green line in figure 4) peak around f= 0.2/yr. We determined that the NP was not influencing the tropics and that LIM does a decent job explaining the observed power spectra of annual mean tropical and NP SST on interannual and interdecadal timescales (refer to figures 4 and 5). Newman also compared LIM to several GCMs (figure 14). The results point out that current GCMs were not correctly simulating the observed PDO-ENSO correlation. One apparent reason that LIM is so good at this correlation is because it is forced with SST to correlate the PDO-ENSO interaction while the GCMs were not constrained by observed statistics of the system. We decided that GCMs are much more complex than LIMs but that a fairer comparison might be made by substituting observed SSTs into the models and running them again.

Newman correlated ENSO and PDO to determine which pattern influenced the other (figure 10). The line graph indicates that the NP is not influential in the ENSO regime (f=0.2-0.3/yr), indicating that the tropics are the influential pattern here; other results suggested that the tropical variations led the NP variations, again suggesting the tropics as the 'driver'. The question of whether we can use LIM for predictability was touched upon in the discussion of forecast skill. 2-yr forecast skill autocorrelations were noticeably less than 1-yr forecasts perhaps because autocorrelation loses skill over each year. This opens up the conversation to whether or not skillful prediction at 1-year lead time is useful. Even though we need predicting power greater than 1 year for decadal variability, a 1-year lead time is still an achievement. It allows adequate preparation time for necessary preparation on many levels (federal, state, etc.). Ultimately, however, “decadal SST predictability is not equal to decadal atmospheric predictability or predictions over land” because the atmosphere and land surface may impart shorter and longer timescales, respectively (Shey).  And it wasn't clear that in a weak PDO state, the operational LIM forecasts had much skill beyond a 3-6 month leadtime.

1. What is the mechanism by which the Indian Ocean dipole works?  Does it exhibit decadal timescale variability? 
   
2. Is the dipole mechanism unique and independent of ENSO?
   
3. Are the coral records sufficient to identify whether the IO Dipole is independent of Pacific ENSO and of the South Asian monsoon?

1. What is Webster's mechanism for the Indian Ocean dipole?
2. What aspects of the mechanism imply predictability?  What limits predictability?
3. Does the Luo et al. (2007) model produce skillful forecasts of IOD events? Why/why not?

[AZC 3/12/2008] Predictability and mechanisms of the Indian Ocean Dipole are difficult problems, because it's hard to tease out the Indian Ocean from the Pacific. Our discussion focused on the predictive mechanisms presented in Luo et al. (2007). We examined the 96-99 event and discussed how the phase of the IOD was positive in the summer of 97 with cold in the eastern Indian Ocean and warm in the west. Enhanced winds push water towards Africa enhancing upwelling off Sumatra and producing an ‘Ekman bump’ in the Autumn. In the Topex/Poseidon sea level height altimetry we examined you could also see the Kelvin waves propagating along the equator, consistent with the mechanism proposed by Webster.

We then examined the SINTEX-F1 model of the Indian Ocean, discussing how this model has a very high resolution and is the state of the art of such models. Unfortunately although SSTs and winds are modeled well the model does not do a very good job at modeling the thermocline depth in the eastern Indian Ocean. This is important since the amount and area of upwelling and its response to wind stress changes is very much dependent on the depth of the thermocline. Using this model they were able to predict the 1994 IOD but had problems with the 1997 IOD. We discussed how this might be caused by anomalous northwesterly winds reducing the upwelling off Sumatra.

What makes the IOD predictable? Knowing that an ENSO event is happening helps with predictability of SST over more than 3-4 month lead times. At shorter lead times, the state of the ocean (and Kelvin wave propagation at the thermocline) in the equatorial Indian Ocean is important.  When examining the predictability of the IOD presented in the paper we discussed that during an El Nino the IOD is more easily predicted. We then discussed how the authors used their model to predict the 2006 even accurately but didn’t tell anyone. This led to a discussion of scientific responsibility and how to present uncertainty. Communication is important; and communicating uncertainty may be just as important.

How predictable is the IOD? Luo et al. say very. We refocused on the ENSO influence on IOD predictability discussing what would happen if you ‘turned off’ the Pacific. Some ideas presented were that you could see if the IOD is separate from ENSO, you could see how predictable the IOD is in the absence of ENSO. It is clear from the paper that predictability skill would suffer in winter. What could we test without the Pacific? We could test whether the IOD is random or red noise generated by ENSO; in other words, can ENSO activity be one of many triggers for IOD activity?  The analogy of a man standing on a dock who will fall in if pushed by ENSO but then again he may still fall in anyway if not pushed was considered. We could also test if ENSO predictability is affected by the IOD.

To test these we suggested some improvements that could be made in the model, such as, a better match of mean thermocline depth in the eastern Indian Ocean, higher temporal and spatial resolution, and more dense observations for initial conditions. Resolution may matter over India (where topography controls so much), near Sumatra so that we can better catch the upwelling along the coast and in the eastern equatorial region during monsoon season so as to get the annual cycle right. To improve resolution in the model is probably not enough unless we can also improve resolution in the observations.

We then discussed the tendency in the IOD index to be skewed positive; i.e. positive event amplitudes greater than negative event amplitudes, which we concluded was because of the Ekman bump. We also noted that if El Nino seems to be a trigger it also seems that during La Nina the IOD remains unaffected.

1. What are the primary pressure, wind, and temperature patterns associated with the North Atlantic Oscillation (NAO)?   What regions are most sensitive to NAO variability and in what way? 
   
2. Describe a “typical” Northern Annular Mode (NAM) signature based on winds, pressure and SST, and differences between high- and low-index conditions (cold and warm events). What locations/countries are most affected? Is the NAO an isolated phenomenon, or is it best considered as part of the Northern Annular Mode?
3. Why is the abundance of G. Bulloides in marine sediment in the southern Caribbean more sensitive to SSTs elsewhere in the Atlantic than closer to the site?
4. Is there evidence for decadal variability in the North Atlantic? 
   

1. How does the model used in Dong and Sutton (2005; hereafter DS05) differ from that used in R99?
2. This article focuses on the mechanisms responsible for generating an approximately 25 year cycle in the N. Atlantic thermohaline circulation (THC).  How does the ocean with depth evolve through this period and what are the oceanic mechanisms responsible for this according to their model?
3. What are the two mechanisms by which it is suggested that the atmosphere can force the THC?  Are these forcings regular or random?

1. Rodwell et al. (1999; hereafter R99) used a global AGCM driven by input of SST data (GISST3.0).  How might the model represent the NAO if the atmosphere were allowed to provide feedback to the Ocean SSTs?  There is some related discussion in the article.
2. Fig. 2 of R99 shows how the NAO time-series projects onto the spatial variability of SSTs in the N. Atlantic.  How does the spatial patterns within the 95% condifidence level compare from the "observed" NAO SST projections to the "modeled" NAO projections?  What is meant by field significance?
3. What role does latent heat transport have in producing the atmospheric circulation of the NAO?  How does the SST anomaly pattern compare to the model's surface evaporation pattern in Fig. 3 of R99?

[SR 04/21/08] This week we learned use of Ingrid data catalog and how one can make spatial (regression, correlation) maps, timeseries etc. using this system. Two relevant websites are http://ingrid.ldgo.columbia.edu/ and http://climexp.knmi.nl/start.cgi?someone@somewhere . We focused more on the use of Ingrid, and worked out an example to create the NAO/SST regression map. To get started with Ingrid look here: http://iridl.ldeo.columbia.edu/dochelp/Tutorial/ .

The topic of discussion was predictability in the Atlantic Sector. As pointed out by Tyson, the key to predictability in the Atlantic is dependent on our understanding and modeling of the mechanisms. The mechanism involves interaction between the thermohaline circulation and the subpolar gyre circulation. Since this topic was discussed last week, I will not talk about it again. For more on the mechanisms, read Dong and Sutton, 2005 (DS05) (refer Figure, 13 for a quick summary, or see last week's summary).

Rodwell et al., 1999 (R99) used a general circulation model of the atmosphere (GCM) to investigate the ocean’s role in forcing North Atlantic and European climate. The model was setup with horizontal grid resolution of 2.5^o x 3.75^o and 19 levels in the vertical and a timestep of 30min. The model was forced with sea surface temeratures (SSTs) and sea ice extents from the GISST3.0 data set. Six 128-year simulations were started from 1870, differing only in their initial atmospheric conditions. 

The authors show that the ensemble mean from the six GCM runs has good correspondence with the observed NAO (Figure, 1, R99). They also show regression maps (Figure 2, R99) and draw the same conclusion. This led to the discussion: why is it that a deterministic model needs to be run six times, and the ensembles mean taken, before the predicted values seem comparable to the observed? The answer as pointed out by Mike was given back in 1963 by a very famous Meteorologist, Edward Lorenz, who passed away just this week. To learn more see Deterministic Nonperiodic Flow (Lorenz, 1963); or the first chapter of Gleick, 1987. In summary, he showed that some deterministic systems have formal predictability limits under special conditions :-).

So now that we have a model and it seems to predict historical NAO timeseries, can it be used to predict future climate in the Atlantic and Europe? Well, the answer to this question as you guessed is not a simple (Yes!). As pointed out by Tyson, there are differences between the models used by R99 and DS05. One of the primary differences is that the model used by DS05 is a coupled ocean-atmosphere model where as, R99 forced the atmospheric model with observed SSTs. Though this may be the case, R99 argue that SST characteristics are communicated to the atmosphere through evaporation, precipitation and atmospheric heating processes. R99 then suggest that since there is significant predictability of mixed layer oceanic temperatures (which is what they are using to force their model) at multiannual timescales, there is predictability of NAO at season and perhaps multiannual timescales. They do however acknowledge that, to understand the interannual and longer timescale predictability of SST pattern, a fully coupled model of the ocean and the atmosphere is required.  We looked at the Met Office predictions using a statistical model based on the R99 hypothesis and found that the predictions are typically right about 2 times out of three, and that the prediction for this past winter was right, within error - - but the errors were quite large and straddled zero. 

1. How is the Southern Annular Mode expressed in time series and spatial patterns in wind, pressure and geopotential height field anomalies? 
   
2. How does the SAM differ from the NAM?  Why?
3. Is there decadal variability in the SAM?
   

1. What are the advantages and disadvantages of the model used in the study?  Is it appropriate for their question?
2. Why are the patterns (SLP, wind, etc.) associated with the southern annular mode so much more zonally homogeneous than the northern annular mode?
   
3. Describe the ocean-atmosphere feedback mechanisms that amplify positive SAM anomalies. (Figure 12 and summary).
   
4. What is/are the negative feedback(s) that could return the system to the
   mean or a negative state, and allow for decadal variability? Is there a
   preferred timescale for these mechanisms?

1. Why is description of patterns in the extratropical, interannual, wintertime Northern Hemisphere sea level pressure field in terms of just a few large-scale structures preferable to multiple indices?
   
2. Could the decadal patterns we have studied within the various ocean sectors be heuristically unified in a similar sense?  If so, how?
   

1. What are the sources of decadal-timescale memory in the climate system?
2. Is there evidence for uniquely decadal mechanisms of climate variation, relative to a reasonable null hypothesis (annual cycle + interannual variability + memory)?
3. Is there evidence for predictability on decadal timescales, relative to persistence?  If so, where, how and why?