XPost: alt.global-warming   
      
   EXECUTIVE SUMMARY:   
   - We use the African Elephant Database reports of estimated animals   
     since ~1998 to predict the future of elephant over the next 20-100y.   
   - Adjusting for "effort" that went into gathering the reported numbers   
     we find African elephant apparently peaked around 2006 and have been   
     in slow decline since.   
   - Using an AI s/w with access to 10s of 1000s of datasets we find the   
     sea surface temperature in a region that includes W Greenland and   
     the Antarctic Peninsula predicts more than 80% of the year-to-year   
     variation in African elephant numbers.   
   - Unfortunately the temperature of the region is on the increase   
     around 1.4C/cent, indicating elephant numbers are predicted to 1/2   
     before that century is up.   
   - Because of animal characteristics -- social grouping and intelligence --   
     it's expected real numbers over the next 100y are likely to involve a   
     crash at some point with small groups of smaller animals perhaps   
     surviving in tiny pockets across the continent.   
      
      
   Despite a huge and continent wide effort to boost the survival of   
   African elephant, the species seems to be on a track to decline.   
      
   According to periodic reports from the AED numbers of animals peaked   
   in 2006 and have declined since. Numbers are estimated for each region   
   of their range across central Africa and the results combined for the   
   survey reports. Ground observations of animals, aircraft surveys,   
   counts of dung heaps and the opinion of local farmers and other   
   experts are all combined to obtain a total number of animals.   
      
   And we can use some data science techniques to massage that and see if   
   we can make a reasonably robust prediction about where the numbers are   
   headed.   
      
   The various reports give the total estimates since the 90s as:   
      
   Year    Estimated numbers   
   1995    285233   
   1998    301733   
   2002    402067   
   2006    471836  <-- max   
   2013    404247   
   2015    395593   
      
   Like field estimates of other endangered animals there are many   
   niggling little problems in gathering this kind of data.  E.g. the   
   majority of the "number" above comes from aircraft surveys of   
   herds. And it's a truism the harder you look the more you find.  So we   
   have to adjust the estimates by the "amount of effort" that went in to   
   assembling them. Various data are available to estimate "effort" --   
   e.g. amount of project funding (adjusted for inflation, of course),   
   manpower available, number of miles flown for aircraft surveys   
   (e.g. length of transect/transects) -- and we can combine all that   
   using high-fallutin methods to come up with:   
      
   Year    Adjusted elephant numbers   
   1995    361829   
   1998    367553   
   2002    465764   
   2006    521189  <-- max   
   2013    412833   
   2015    395593   
      
   And from these numbers we can interpolate missing years to obtain an   
   estimate for all the years from approx 1990 to 2020.  Which we can   
   then throw into a little program to discover what out there in the   
   world seems to be influencing elephant numbers in Africa.   
      
   And the top10 answers we find after search 10s of 1000s of data   
   series are:   
      
   "Suspect" dataset       Lag(y)  LogTrans  R2   
   world-70                2       y        0.83462826   
   sduah_globe6SoPol       3       y        0.81257976   
   arc-40                  5       y        0.78078380   
   sdcambodia              2                0.77924226   
   ant70                   4       y        0.73949191   
   ant80                   4       y        0.73390412   
   sdireland               3       y        0.72508670   
   presband20              3       y        0.72354974   
   world150                0                0.70189185   
   maxuah_lsNH             4                0.69806836   
      
   So the "best" simple predictor of African elephant numbers we can find   
   is the SST of oceans along the longitude segment 70W-60W that goes   
   through the waters off W Greenland and the Antarctic peninsula -- key   
   weather determiners for most of the globe.   
      
   The annual avg SST of this region lagged by 2 years matches 83% of the   
   elephant count data so we expect it will be able to predict at least   
   numbers a decade or 2 into the future given it is based on reports for   
   the last 20+ years.   
      
   But the results are not happy. The SST along the segment are on the   
   increase:   
      
   Model for world-70 1990-2020:   
   (Serial corr detected; estimated rho = 0.449530)   
   y = 0.0141806*x + -12.6978   
   beta in 0.0141806 +- 0.00665622  90% CI   
   alpha in -12.6978 +- 7.35401   
   P(beta>0.000000) = 0.999430   
   r2 = 0.31930471   
      
   I.e. SST in the region are presently increasing around 1.4 degC/cent --   
   similar to the rise of avg global temps.   
      
      
   And the relationship with African elephant numbers over the past 20y   
   is inversely related to temperature -- the higher the SST the lower   
   the elephant count. The rise in elephant numbers between 1998 and 2006   
   is matched by a decline in SST in the region during those years --   
   presumably due to factors like melting glacier ice entering the seas   
   off W Greenland and the Antarctic peninsula. After ~2006 overall ocean   
   warming took over and SST started to climb again, in parallel with the   
   reported decline in elephant numbers.   
      
   The best model linking "world-70" and AED reported numbers is:   
      
   (Log transform enabled).   
   (Lagging 2y).   
   y = 7.93112e+09*exp(-0.619429*x)   
   Halving Rate 1.12C   
   beta in -0.619429 +- 0.124802  90% CI   
   alpha in 22.7941 +- 1.9685 = [20.8256, 24.7626]   
   P(beta<0.000000) = 1.000000   
   calculated Spearman corr = -0.850490   
   Critical Spearman = 0.582500 2-sided at 1%; reject H0:no_link   
   r2 = 0.83462826   
      
   Binned data:   
   Year                    Av SST(degC)   #Elephants   Model-est #Elephants   
      1998                   15.9376       416658       409158   
      1999                   15.7879       441211       448913   
      2000                   15.7094       465764       471281   
      2001                   15.7086       479620       471515   
      2002                   15.7115       493476       470669*(1sd diff)   
      2003                   15.6984       507333       474503*   
      2004                   15.5224       521189       529159     <-- max animals   
                                                                       & min SST   
      2005                   15.6614       505710       485504*   
      2006                   15.6126       490230       500404   
      2007                   15.6867       474751       477955   
      2008                   15.682        459271       479348*   
      2009                   15.7718       443792       453413   
      2010                   15.974        428312       400036*   
      2011                   15.8595       412833       429439*   
      2012                   15.9257       404213       412185   
      2013                   15.9789       395593       398824   
      2014                   15.9022       395593       418229*   
      
      
      
   The model shows the data fit the (adjusted) report numbers well.   
   There are no 2s outliers and the 1s outliers fall on both sides of the   
   model curve.   
      
   The stats finds the model is significant at 90% (T-test) and 99% (rank   
   test) meaning there is almost 999 chances in 1000 the SST data   
   "explains" or predicts what is happening to African elephant numbers.   
      
      
   [continued in next message]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)