Lectures - Monday and Wednesday, 11:00 AM - 12:15 PM

Lab - Tuesday, 4:10 PM -7 PM

The warm phase of the El Niño - Southern Oscillation (ENSO) phenomenon is a warming of the eastern tropical Pacific Ocean that occurs every 2 to 7 years, and lasts about 18 months. By upsetting normal patterns of atmospheric circulation, ENSO warm and cold phase events influence weather patterns in regions quite remote from the tropical Pacific Ocean. In these areas, critical and climate-sensitive industries such as agriculture could be adversely affected. One major goal of predicting and understanding ENSO is to be able to predict ENSO events before they happen. With this knowledge, we might be able to offset or mitigate their potentially damaging effects on the world's food supply. The purpose of this lab is to see why and how the warm phase of ENSO affects the climate of remotely located regions and their agricultural systems.

The ocean heats and cools the atmosphere right where they meet -- therefore,
one indicator of the effect of changes in ocean dynamics upon the atmosphere
is through sea surface temperature (SST). Since we are interested in departure
of sea surface temperature from the typical annual cycle, we look at SST
*anomalies*. By anomalies, we mean the difference between the SST measured
that month (e.g. September of 1997) and the averaged SST for all Septembers
for which we have data; in the case of the data below, it is January 1970-September
1998. The horizontal coverage is 124E to 70W longitude and 29S to 29N latitude,
with a resolution of 2°. To learn more about these data, visit the documentation page.

What is the pattern of SST anomalies in the tropical Pacific during an El Nino event? Use the viewer to look at an animation of monthly SST anomalies. Take note of the latitudinal and longitudinal range of the map view here. Go to the white box above the SST anomaly map and type in "January 1970 to Sep 1998" and click on the redraw button to the left. You will see a progression of SST anomaly maps drawn, starting from January 1970 and continuing to the present. Concentrate on the largest spatial-scale anomaly patterns that you see. Hit "Stop" on the Netscape browser to stop the animation.

Describe the pattern of SST anomalies across the tropical Pacific. Where are the anomalies the largest? What is the typical range of the anomalies in degrees? What is their amplitude? When during the year do anomalies start to develop? How long do the ENSO events tend to last? Do all the events you observe look the same, or are there differences in the timing, duration and strength of events?

Now consider the regularity of ENSO warm phase events. One way to do so
is via a time vs. longitude plot of SST anomalies. Using the viewer, make
a time vs. longitude plot for SST anomaly data right on the Equator. *Make
sure that the viewer is at 0 degrees* (the default is 29S).

Can you pick out the signatures of the main ENSO warm phase events? How much time passes between events? Is this time interval constant, or does it vary?

Finally, let's look at December 1982 - February 1983 averaged sea surface temperature (not anomaly). This is during a recent strong ENSO warm phase event.

Where is SST the warmest? How does the area of the Pacific covered by the warmest SST change during ENSO warm phase events? How does the location of the warmest SST compare to the location of the largest SST anomaly during ENSO events? Keep this comparison in mind as we look at how the atmosphere reacts to ENSO warm phase conditions.

During warm phase of ENSO, the atmosphere in the tropics is heated via increased SST, so the atmosphere must respond and distribute the extra heating. It does so via enhanced convective activity. One way we can observe changes in convection is via satellite observations of the outgoing longwave radiation (OLR). The OLR measurement tells us the temperature of the surface that the satellite sees. If the atmosphere between the satellite and the Earth's surface is clear, then the satellite essentially sees the OLR from the ocean or land surface. However, if the atmosphere is full of thick clouds, then the satellite sees the top of the clouds. (Recall the images of emitted longwave radiation you looked at in the energy balance lab earlier in the semester.) Since the cloud tops may be 5-10 km above the surface, they are much colder than the surface. Therefore, when we see patches of low OLR, we interpret this as the locations of thick thunderclouds.

Now look at the OLR anomaly for December 1982 -February 1983, in the middle of a strong ENSO warm phase event (this is the same period for which you just examined sea surface temperature data).

Where are the lowest OLR anomalies found? Do the patterns of the OLR anomalies more closely resemble those of the sea surface temperatures, or those of the sea surface temperature anomalies? Why do you think this is?

Scientists often wish to quantify the linear association between variables. For example, below you will examine the association between ENSO and crop yields. For this task we will use the correlation statistic. Given two variables, x and y, the correlation (r) is a measure of how well a line drawn through a scatter plot of y vs. x fits the data. It is computed from the following formula:

where and are the mean values of x
and y respectively, σ_{x} and σ_{y} are the standard deviations of the data, and
n is the number of data points:

r can take on any value between -1 and 1. If r is positive, then the data are positively correlated. This means that an increase in variable x corresponds to an increase in y, whose magnitude depends on the slope of the line drawn on a plot of y vs. x. If r is negative, then the data are negatively correlated: an increase in variable x corresponds to a decrease in variable y, again with magnitude depending on the slope of the line drawn through the data. If r=0, then there is no relationship between the data y and x: we can't make any prediction about how y should change if we vary x. In practice r almost never exactly equals -1, 0 or 1, but falls somewhere in between. Then we have to decide whether there really is any relationship between y and x. It would help to know something about the problem itself, such as its physics, to interpret and explain the correlation coefficient. The correlation coefficient itself says nothing about causality, however: it merely suggests whether changes in x are associated with changes in y (or the other way around).

We'll investigate this question by computing the correlation between NINO3 and an index of anomalous atmospheric sea level pressure difference between the eastern and western tropical Pacific, called the Southern Oscillation Index. NINO3 is the average SST anomaly over the region 150W to 90W, 5N to 5S. (From the map views of SST anomaly, you can see that this region has the strongest SST anomalies associated with ENSO warm phase events.) The SOI is the difference between the atmospheric sea level pressure anomaly at Tahiti (in the southeastern tropical Pacific), and the atmospheric sea level pressure anomaly at Darwin, Australia (in the western Pacific). Pressure is usually low in the western tropical Pacific, and high in the eastern tropical Pacific. (From lecture, can you explain why, given the pattern of sea surface temperatures across the tropical Pacific?) The SOI tells us when this atmospheric pressure pattern departs from normal conditions, or in other words, when atmospheric sea level pressure is simultaneously lower in the eastern tropical Pacific, and higher in the western tropical Pacific, or vice versa. The NINO3 data (click for documentation) were computed by Alexey Kaplan at LDEO from the Global Ocean Surface Temperature Atlas SST anomaly dataset. The SOI data are from the Australian Bureau of Meteorology. We will use the data for the period 1961 through 1990.

Save the data (Table 1) in a place you can easily access (e.g., the desktop). There are three columns of data. The first is the calendar year. The second column is the NINO3 averaged SST anomaly for Sep, Oct, Nov of the corresponding years, and the third column gives like averages of the SOI (Jun-Nov). These intervals of months correspond to the seasons when maximum excursions of SST and SOI occur during most El Niño and La Niña episodes.

First make a scatter plot of NINO3 vs. SOI. Do you see a negative or positive
correlation? Weak or strong? Try to estimate the correlation on a scale from
0 to 1 (0 to -1 for negative correlation). Add a linear regression line to
the scatter plot and include a correlation coefficient (r^{2}) and
equation for the line.

Compute the correlation between SOI and NINO3 using Excel's **=correl()** function.

Is the correlation positive or negative? Weak or strong? Your eye was probably pretty accurate at judging the strength of the correlation. What is the significance of this correlation in terms of sea surface temperature and sea level pressure in the eastern Equatorial Pacific?

If we don't know anything about the physics (or even if we do), we still would like to test the reality of the relationship between NINO3 and SOI. Is it really different from zero? Typically a statistical test is phrased as follows. We obtained a value r for the correlation coefficient between the variables x and y.

Suppose instead of the series y, we had a series z of randomly chosen numbers. In this case, what value of correlation coefficient would you expect -- what is the most likely value for the correlation between x and z? We want to know the likelihood that we obtained a correlation as high as r. If the probability that we obtained a correlation as high as r by correlating x and z is low, then we are pretty sure the relation between x and y is significant -- that is, that the relationship we observed is not likely to be due to chance. If the probability is high, however, then we can't be confident that there is a real relationship between x and y.

We took a time series of annual averages of NINO3 SST anomaly for 1961-1990 (30 years) as x, and generated 10,000 random time series of 30 numbers each -- these are our z's. Here is a histogram showing the number of observations of the magnitude of r between 0-0.1, 0.1-0.2, 0.2-0.3, etc. For instance, reading from the histogram, there were 4039 occurrences of |r| between 0 and 0.1. (The bars around r mean "absolute value of".)

What is the median of r (i.e. the value of r where half the data has higher r, and half the data has lower r)? What percentage of the values fall between 0.4 and 0.5? What percentage of values of r is less than 0.1? 0.2? 0.3?...1.0?

Now recall the correlation between SOI and NINO3 that we calculated earlier. What percentage of the correlations between NINO3 and the random time series is less than the correlation between SOI and NINO3? What percentage is greater than the correlation between SOI and NINO3? This percentage is the chance that a random time series will be correlated as well with NINO3 as SOI is correlated with NINO3. How confident are you that the correlation between SOI and NINO3 is real?

The atmosphere has to get rid of the extra heating supplied by the eastern tropical Pacific Ocean during an ENSO warm phase event. It does this in certain preferred patterns, which affect temperatures and precipitation in many places around the world. As a result, weather-sensitive human activities, such as agriculture, can be affected. In this part of the lab, we'll examine how closely ENSO parameters are associated with remotely-located agricultural yields. We'll do so by examining the correlation between SOI, and selected wheat yields from agricultural model time series in Australia. View a map of Australia obtained from the National Mapping Division of Geoscience Australia to learn more about the various regions you'll be examining in today's lab.

Most of Australia is semi-arid, and climate conditions are relatively unfavorable for many agricultural crops. Of the major grains, wheat requires the least amount of precipitation to grow, and thus is the most important grain crop in the country. Australia is one of only about five or six countries that produce a significant excess of wheat crop beyond domestic consumption requirements. These few countries supply most of the grain for international trade in wheat. Since ENSO is responsible for a significant portion of the interannual climate variability in parts of Australia, knowledge of ENSO conditions is critical in agricultural decisions in many regions of the country.

Using SOI as an indicator of ENSO conditions, make a scatter plot of annual precipitation amount in the catchment of the River Murray (southeastern Australia) vs SOI (Table 2). The Murray River catchment basin (map obtained from the Murray-Darling Basin Commission) precipitation and discharge data were obtained via personal communications with staff of the Australian Bureau of Meteorology and the Murray-Darling Basin Commission, respectively. Add a linear regression slope, with correlation coefficient and slope equation to the plot. Is the slope positive, negative, or zero? Is there higher rainfall during El Niño or La Niña episodes? Do you think the correlation is statistically significant? Athough the correlation statistics are between the NINO3 index and 10,000 random time series, you may assume that a similar distribution would be obtained for correlations between the SOI and 10,000 random time series. What is the approximate range of lowest to highest annual precipitation amount in this region? Which ENSO state would you think would be more favorable for higher than average wheat yields in this region of Australia?

To facilitate comparison of annual river discharge and annual precipitation in the basin of the River Murray, make a scatter plot of annual natural river discharge (Q) vs annual precipitation (P) (Table 2). [Note that "natural" discharge is computed from actual discharge measurements by attempting to account for all the changes in water amount due to human influences. Examples of these changes including adding back losses such as evaporation in water storage reservoirs, or diversions to agricultural irrigation districts.] Add a linear regression slope, with correlation coefficient and slope equation to the plot. Is the slope positive, negative, or zero? Do you think the correlation is statistically significant? What is the approximate range of lowest to highest annual river amount in this region? How does the ratio of highest to lowest river annual discharge amount compare to that ratio for precipitation in the basin of the River Murray? Does the linear regression line intercept zero for both axes? Why might the Q vs P plot have this feature?

Using SOI as an indicator of ENSO conditions, make a scatter plot of annual natural river discharge of the River Murray to South Australia (SA) vs SOI (Table 3). Note that the River Murray discharge to South Australia is downstream of the confluence of the Darling River, and thus includes runoff from the large northern portion of the catchment, as well as the runoff from the mountain range near the southeast coast of the country. Add a linear regression slope, with correlation coefficient and slope equation to the plot. Is the slope positive, negative, or zero? Is there higher river discharge during El Niño or La Niña episodes? Do you think the correlation is statistically significant? Do you think flooding would be more likely during El Niño or La Niña episodes? Do you think droughts would be more likely during El Niño or La Niña episodes?

Wheat Yields in many regions of Australia have large interannual variations, in large part because of variations in rainfall amount. Make a time-series plot (Yield vs year) for mean annual Wheat Yields per unit area of crop (Tons per hectare) in the state of New South Wales (NSW) in southeastern Australia (Table 4). These data were provided courtesy of Graeme Hammer of the Agricultural Production Systems Research Unit in Australia. How many years have had a Wheat Yield in NSW of less than 1.0 T/ha during the 20th century? How many years have had a Wheat Yield in NSW of less than 1.5 T/ha during the 20th century? Do the years of especially low yield tend to occur in clusters of several years at a time, or do they tend to have higher yields immediately preceding or following them?

Make a scatter plot of Wheat Yields in NSW vs SOI (Table 4). Do the years of higher yield tend to occur in El Nino or La Nina episodes? Do years of low yield tend to occur in El Nino or La Nina episodes? Add a linear regression slope, with correlation coefficient and slope equation to the plot. Is the slope positive, negative, or zero? Do you think the correlation is statistically significant?

Make a scatter plot of Wheat Yields in Queensland (Qld) in eastern Australia vs SOI (Table 4). Do years of higher yield tend to occur in El Nino or La Nina episodes? Do the years of low yield tend to occur in El Nino or La Nina episodes? Add a linear regression slope, with correlation coefficient and slope equation to the plot. Is the slope positive, negative, or zero? Do you think the correlation is statistically significant?

Make a scatter plot of Wheat Yields in Western Australia (WA) vs SOI (Table 4). Do years of higher yield tend to occur in El Nino or La Nina episodes? Do the years of low yield tend to occur in El Nino or La Nina episodes? Add a linear regression slope, with correlation coefficient and slope equation to the plot. Is the slope positive, negative, or zero? Do you think the correlation is statistically significant?

Using all the information and results obtained in this lab session, write a lab report (as per the Lab Report Format) summarizing the major findings of your investigation. In particular, if you had advanced knowledge of the likely ENSO state about a year in advance, what advice might you provide wheat growers in Australia in each of the following states: New South Wales, Queensland, Western Australia? Discuss why.

For the more adventurous students, you may further examine data provided to you by exploring possible strong correlations between different combinations of data sets and interpreting these results.

*Updated
July 9, 2007
To report problems, email webmaster.*