Developing  a  capability  to  es1mate  the  probability  of  drought  recovery  

Eric  Wood  Princeton  University  

MAPP  Research  in  support  of  RtC  Webinar  November  22,  2013  

The  Challenge:    Forecas1ng  drought  recovery  in  a  manner  useful  for  drought  decision  making  

Seasonal  forecast  models  have  poor  skill  in  predic4ng  drought  dura4on  and  recovery.    

Yet,  improved  skill  in  forecas4ng  drought  recovery  is  cri4cal  for  improved  drought  management.  

A  risk-­‐based  (probabilis4c)  drought  recovery  procedure  would  allow  users  to  assess  their  poten4al  decisions  in  a  stronger  economic  framework.    

July  2012  ini4alized  7/1/2012   Aug.  2012  ini4alized  7/1/2012   Sept.  2012  ini4alized  7/1/2012  

Verifica4on  of  Aug.  2012  forecast  Verifica4on  of  Jul.  2012  forecast   Verifica4on  of  Sept.  2012  forecast  

A  probabilis1c  approach  for  assessing  drought  recovery  

The  approach  is  presented  in  Pan  et  al.  (GRL,  2013),  and  builds  on  the  NLDAS  research  into  developing  hydrologic  forecasts  using  land  surface  models  and  ensemble  forecasts.  In  this  case,  ESP,  which  uses  current  hydrologic  ini4al  condi4ons  and  historical    meteorology  to  force  a  LSM  that  provides  predic4ons  of  future  hydrologic  states  and  our  drought  index.    

Princeton’s  Hydrologic  Monitoring  and  Predic4on  System  

Multiple GCM Ensemble Forecast

“Weather” Generator (Ensembles)

1/8 degree

daily time step

1/8 degree

Monthly time step

Historical Observations (Climatology)

Climate indices Teleconnection

1/8 degree GCM resolution Large scale

Drought Management

(Ensemble) Drought

Index

Real-time Weather

Initial Conditions

(Ensemble) River Discharge

Bayesian Merging Downscaling (Ta, P)

Land surface (hydrology)

models

23

ESP  Hydrologic  Predic4on  System  

“ Historical Weather” (Ensembles)

1/8 degree

daily time step

Historical Observations (Climatology)

1/8 degree, daily

Drought Management

(Ensemble) Drought

Index

Real-time Weather

Initial Conditions

(Ensemble) River Discharge

Land surface (hydrology)

models

23

Hydrologic  Ensemble  Forecas4ng  32

Spin-up Period

Forecast Period Observed Meteorology

Tim

e of

fore

cast

PDF  fit  to  the    ensemble  histogram  

Results  for  the  2012-­‐2013  upper  mid-­‐west  drought  Pan,  Ming;  Yuan,  Xing;  Wood,  Eric  F  2013A  probabilis4c  framework  for  assessing  drought  recovery  Geophys.  Res.  Le,s.  40(14):  3637-­‐3642  DOI:  10.1002/grl.50728,  July28.  

0.5  mo  lead   1.5  mo   2.5  mo  

4.5  mo  3.5  mo   5.5  mo  4.5  mo  

Grid-­‐scale  analysis  of  recovery  (from  2/2013)  

Approach  uses    1.  Resampling  of  the  

historical  precipita4on  and  temperature  record,    

2.  Land  surface  modeling  and  

3.   Copula  sta4s4cs  to  compute  the  probability  that  soil  moisture  will  exceed  a  specified  threshold.  

4.5  

5.5  

3.5  0.5  

1.5  

2.5  

CONUS  analysis  of  recovery  probability  above  a  threshold  (30th  percen4le)  given  the  median  probability  of  the  required  accumulated  precipita1on  (red  x  in  the  first  figure)  over  the  forecast  period  

Results  for  the  2012-­‐2013  upper  mid-­‐west  drought (2013/2)  Pan,  Ming;  Yuan,  Xing;  Wood,  Eric  F  2013A  probabilis4c  framework  for  assessing  drought  recovery  Geophys.  Res.  Le,s.  40(14):  3637-­‐3642  DOI:  10.1002/grl.50728,  July28.  

Results  for  the  2012-­‐2013  upper  mid-­‐west  drought (2013/2)  Pan,  Ming;  Yuan,  Xing;  Wood,  Eric  F  2013A  probabilis4c  framework  for  assessing  drought  recovery  Geophys.  Res.  Le,s.  40(14):  3637-­‐3642  DOI:  10.1002/grl.50728,  July28.  

4.5  

5.5  

3.5  0.5  

1.5  

2.5  

CONUS  analysis  of  recovery  above  a  threshold  (30th  percen4le)  given  that  the  accumulated  precipita1on  equaled  the  median  over  the  forecast  period  

Drought  recover  analysis  (0.5-­‐degree  boxes):    top,  Georgia  ini4alized  May  2007;  bodom  Texas  ini4alized  May  2011.  

3.5  

0.5  lead     1.5  lead     2.5  lead     5.5    

0.5  lead     1.5  lead     2.5  lead     5.5    

Summary  

NLDAS  research  at  Princeton  University  has  developed  seasonal  drought  monitoring  and  forecas4ng  products  that  have  been  transi4oned  to  EMC/NCEP,  CTB/CPC/NCEP  and  NCO  opera4ons;  

These  NLDAS  procedures  can  be  used  for  probabilis4c  drought  recovery  forecas4ng  using  ESP  methodology.    The  approach  has  been  demonstrated  with  historical  droughts.  

The  approach  can  be  used  to  increase  the  opera4onal  capabili4es  of  CPC  for  providing  drought  guidance,  and  could  be  used  with  CFSv2  forecasts,  in  an  ESP  re-­‐sampling  approach,  or  a  combined  merged  approach.