DELIGHT (Deep-learning-informed Glacio-Hydrological Threat) Framework

Glaci­er retreat around the world due to cli­mate change ex­poses a third of the world’s popu­lation to a range of hy­dro­logi­cal haz­ards: chief­ly changes in water quan­tity and quali­ty, and also the risk of dev­astat­ing glaci­er lake out­burst floods (GLOFs). The DE­LIGHT Framework will for the first time model the entire hy­dro­logi­cal path­way from pre­cipita­tion, through glaci­ers, to rivers, using ma­chine learn­ing to accel­erate the pro­cess, there­by gen­er­ating more accu­rate fore­casts of future risks.

The Inter­na­tional Junior Re­search Group, fund­ed by the Elite Net­work of Ba­varia, is con­duct­ing re­search into de­velop­ing a fully cou­pled suite of mod­els that trace the entire hy­dro­logi­cal path­way of glaci­erised catchments, from pre­cipita­tion, through glaci­ers, to rivers. A third of the world’s popu­lation live in catchments with glaci­erised headwa­ters and will con­se­quent­ly be ex­posed to changes in water quan­tity and quali­ty as glaci­ers re­treat, as well as the possi­bility of GLOFs as water fills un­stably dammed previ­ously ice-covered de­pres­sions.

Accu­rately pre­dicting these glaci­o-hy­dro­logi­cal haz­ards re­quires mod­elling of the entire hy­dro­logi­cal pro­cess chain, some­thing not before possi­ble as it would take too long to run the simu­la­tions. With the ad­vent of ma­chine learn­ing and the result­ing possi­bility to great­ly accel­erate the mod­elling of physi­cal sys­tems, this fun­damental barrier is now being lifted, mak­ing more accu­rate pre­dic­tions feasi­ble for the first time.