Droughts and flooding rains at Manilla NSW were related in a way that is remarkable and unexpected.

Part III. Predicting drought from heavy rain

[Back to Part II: Scatter-plots]

The graph above is derived from the first graph in this series (copied here) by using the blue regression trend-line from the scatter plot of selected data (also copied here). (For data details, see Note 1, below.)

The equation of the trend line, y = 0.030x is used AS IF to use the daily rainfall excesses to predict the drought frequency five years later. The graph shows the “error” of this “prediction”. (In Note 2, below, I concede that this data set could not support such prediction.)
As expected from the previous graphs, the “prediction” is accurate at most data points to 1975. It is correct to the nearest percentage whole number at nine of the eighteen points. From 1940 to 1955, droughts are uniformly more frequent than predicted. After 1975, the error curve swings wildly up and down.

Could droughts have been predicted from heavy rainfalls?

By about 1915, it is conceivable that this relationship could have been discovered, either by analysis of such data, or by modelling of the climate system. Then, the data for the next 20 years, up to 1935, would seem to confirm it. Data from 1940 to 1955 would cause doubts, but data from 1960 to 1975 would restore confidence. Then the utter failure of the model in the following four decades would have led to its abandonment, at least for the time being.

Climate shifts of 1975

This local rainfall data set shows a sudden, dramatic change in 1975. A remarkable relation between heavy rainfall events and year-long droughts, which had persisted for almost a century, broke down then, to be followed by no relation whatever for at least four decades.

In or near the year 1975 several features of world climate also suffered a “climate shift”, changing from one mode to another . At that date, the cumulative movement in the Southern Oscillation Index (SOI) changed decisively from La Nina dominance to El Nino dominance. At the same time, the Inter-decadal Pacific Oscillation (IPO) changed sharply from negative to positive mode. Global temperature (HadCRUT series), which had been falling for three decades, stopped falling, and rose again for the next three decades.
(Both the IPO and the HadCRUT global warming curve suffered reversals also in 1879, 1909, 1943 and (perhaps) in 2005.)

Since 1975 was the year that the most recent “global warming” trend began, one might think that increasing human emissions of carbon dioxide caused some or all of the climate shifts in that year. Unfortunately for this case, the rate of growth of human emissions did not increase at that date, but abruptly decreased (in 1973), from 2.14 log-cycles per century to 0.77 log-cycles per century.

---

Note 1. Data

A. “Droughts”
(See this earlier post.)

The graphical log of 12-month droughts shows the percentage of months in each decade that have a 12-month rainfall total below the 5th percentile. They have a median value of 5% but range above and below it. Values are plotted against the mid-year of each decade, with decades overlapped by five years. They have been smoothed (1:2:1)/4.

B. “Flooding Rains”
(See this earlier post.)

From the 130-year record of daily rainfalls at Manilla, NSW up to December 2014, I selected all 125 “very wet days” that have readings exceeding 50 mm. I listed them by date.
Since it is only the excess rainfall that runs off, leading to flooding, I have subtracted 50 mm from each rainfall amount. Then I have summed all such excesses for each half-decade. I summed these in pairs to give a decade sum (in mm) centered on the years 1885, 1890, 1895, etc.

 Note 2. Prediction not possible

Although heavy daily rainfalls lead 12-month droughts by five years, this is not a clear gap that could allow prediction using this data. Each variable is a decade mean. Thus, the latest data required for the rainfall estimate is not available until the middle of the decade for which drought is to be predicted. The overlap is made worse by the use of a (1:2:1)/4 smoothing. Shorter-period data might work, but it may be too unstable.