Global high-resolution parameter maps


All hydro¬≠log¬≠i¬≠cal mod¬≠els are sim¬≠pli¬≠fi¬≠ca¬≠tions of a com¬≠plex real¬≠i¬≠ty and there¬≠fore need to be cal¬≠i¬≠brat¬≠ed to obtain sat¬≠is¬≠fac¬≠to¬≠ry stream¬≠flow sim¬≠u¬≠la¬≠tions. Here we present high-res¬≠o¬≠lu¬≠tion (0.05¬į) opti¬≠mized para¬≠me¬≠ter maps for the con¬≠cep¬≠tu¬≠al hydro¬≠log¬≠i¬≠cal mod¬≠el HBV cov¬≠er¬≠ing the entire land sur¬≠face includ¬≠ing ungauged regions.
The maps were pro¬≠duced using a nov¬≠el para¬≠me¬≠ter region¬≠al¬≠iza¬≠tion approach that involves the opti¬≠miza¬≠tion of trans¬≠fer equa¬≠tions link¬≠ing mod¬≠el para¬≠me¬≠ters to cli¬≠mate and land¬≠scape char¬≠ac¬≠ter¬≠is¬≠tics. The opti¬≠miza¬≠tion was per¬≠formed in a ful¬≠ly spa¬≠tial¬≠ly dis¬≠trib¬≠uted fash¬≠ion at high res¬≠o¬≠lu¬≠tion, instead of at lumped catch¬≠ment scale, using an unprece¬≠dent¬≠ed data¬≠base of dai¬≠ly observed stream¬≠flow from 4229 small catch¬≠ments worldwide. 
The region¬≠al¬≠ized para¬≠me¬≠ters result¬≠ed in a medi¬≠an Kling-Gup¬≠ta Effi¬≠cien¬≠cy (KGE) improve¬≠ment of +0.29 (rel¬≠a¬≠tive to uncal¬≠i¬≠brat¬≠ed para¬≠me¬≠ters) in 4229 ful¬≠ly inde¬≠pen¬≠dent val¬≠i¬≠da¬≠tion catch¬≠ments. Improve¬≠ments were obtained for 88 % of the val¬≠i¬≠da¬≠tion catch¬≠ments. Sub¬≠stan¬≠tial improve¬≠ments were obtained even for val¬≠i¬≠da¬≠tion catch¬≠ments locat¬≠ed far from the catch¬≠ments used for opti¬≠miza¬≠tion, under¬≠scor¬≠ing the val¬≠ue of the derived para¬≠me¬≠ters for poor¬≠ly gauged regions. For more infor¬≠ma¬≠tion, see the fol¬≠low¬≠ing open-access publication: 


The lat¬≠est ver¬≠sion (0.9) of the para¬≠me¬≠ter maps, includ¬≠ing the HBV mod¬≠el code used to derive the maps, can be down¬≠loaded here. If the dataset forms a key com¬≠po¬≠nent of your research, we kind¬≠ly ask that you give us the oppor¬≠tu¬≠ni¬≠ty to com¬≠ment on your results pri¬≠or to pub¬≠li¬≠ca¬≠tion. The data and code are released under the CC BY-NC 4.0 license and thus may not be used for com¬≠mer¬≠cial pur¬≠pos¬≠es. By using the data in any pub¬≠li¬≠ca¬≠tion you agree to cite the cor¬≠re¬≠spond¬≠ing paper. 


The fol­low­ing orga­ni­za­tions are thanked for pro­vid­ing stream­flow and/or catch­ment bound­ary data: the Unit­ed States Geo­log­i­cal Sur­vey (USGS), the Glob­al Runoff Data Cen­tre (GRDC), the Brazil­ian Agen­cia Nacional de Aguas, EURO-FRIEND-Water, the Euro­pean Com­mis­sion Joint Research Cen­tre (JRC), the Water Sur­vey of Cana­da (WSC), the Aus­tralian Bureau of Mete­o­rol­o­gy (BoM), and the Chilean Cen­ter for Cli­mate and Resilience Research (CR2, CONICYT/FONDAP/15110009).

Frequently asked questions

Runoff pro­jec­tions from HBV with region­al­ized para­me­ters are prob­a­bly in most cas­es more reli­able than those from HBV with unop­ti­mized para­me­ters. The fol­low­ing caveats should, how­ev­er, be kept in mind. First, since the region­al­ized para­me­ters were opti­mized to pro­duce bet­ter runoff sim­u­la­tions, they do not nec­es­sar­i­ly rep­re­sent hydro­log­i­cal process­es more real­is­ti­cal­ly. Sec­ond­ly, the valid­i­ty of the region­al­ized para­me­ters, which were opti­mized using his­toric cli­mate con­di­tions, may be less under sig­nif­i­cant­ly dif­fer­ent future cli­mate con­di­tions. See Vaze et al. (2010) and Fowler et al. (2020) for fur­ther discussion.

The ten sets of glob¬≠al para¬≠me¬≠ter maps rep¬≠re¬≠sent dif¬≠fer¬≠ent cross-val¬≠i¬≠da¬≠tion folds, mean¬≠ing they were derived from slight¬≠ly dif¬≠fer¬≠ent (but over¬≠lap¬≠ping) sub¬≠sets of catch¬≠ments. All ten sets of para¬≠me¬≠ter maps are equal¬≠ly uncer¬≠tain and there¬≠fore should pro¬≠duce equal¬≠ly ‚Äúgood‚ÄĚ sim¬≠u¬≠la¬≠tions. Ide¬≠al¬≠ly, all ten sets of para¬≠me¬≠ter maps would be used in an ensem¬≠ble mod¬≠el¬≠ing frame¬≠work to obtain an ensem¬≠ble of out¬≠puts, the spread of which pro¬≠vides an indi¬≠ca¬≠tion of the para¬≠me¬≠ter uncer¬≠tain¬≠ty. See McIn¬≠tyre et al. (2005) and Yang et al. (2011) for more dis¬≠cus¬≠sion on ensem¬≠ble modeling.

Inter¬≠ac¬≠tions between mod¬≠el para¬≠me¬≠ters and sim¬≠u¬≠la¬≠tion results are gen¬≠er¬≠al¬≠ly high¬≠ly com¬≠plex and non¬≠lin¬≠ear. The out¬≠puts of a low-res¬≠o¬≠lu¬≠tion mod¬≠el with resam¬≠pled para¬≠me¬≠ter maps will there¬≠fore like¬≠ly sig¬≠nif¬≠i¬≠cant¬≠ly dif¬≠fer from the resam¬≠pled out¬≠puts of a high-res¬≠o¬≠lu¬≠tion mod¬≠el with 0.05¬į para¬≠me¬≠ter maps. Thus, we do not rec¬≠om¬≠mend using the para¬≠me¬≠ter maps at oth¬≠er spa¬≠tial res¬≠o¬≠lu¬≠tions. See Beven (1995), Kling and Gup¬≠ta (2009), and Fre¬≠und et al. (2020) for fur¬≠ther discussion.