Hi,

I'm experiencing physically impossible behavior in my WEAP model that appears to be a MILP solver numerical problem.
Setup:
Large Chilean watershed with streamflow gauge "Estacion1" in upper basin
14 demand nodes distributed throughout watershed
Tested with both lpsolve and XA solvers
The Problem:
WITH demand node E3 active (8.15 Mm³/year, near gauge):
NSE = 0.78, KGE = 0.89, excellent calibration
WHEN I ADD nodes E1 and E2 (located several kilometers DOWNSTREAM of gauge):
E1: 121.7 Mm³/year
E2: 39.9 Mm³/year
NSE drops to 0.26
Streamflow at upstream gauge decreases dramatically
This is physically impossible - downstream extractions cannot affect upstream measurements
Key Evidence:
E3 (8.15 Mm³/yr) works perfectly
E1 and E2 are kilometers downstream, spatially separated from gauge
Reducing E1 and E2 to 1 Mm³/yr each recovers NSE to 0.78
Other downstream nodes E4-E14 with smaller demands cause no issues
Problem is magnitude-related, not topology-related
Question:
Can very large demand values cause MILP solver numerical instability that propagates calculation errors throughout the river network, affecting physically independent upstream nodes? This appears to be a poorly scaled model issue where the solver produces results violating basic mass balance principles.
Has anyone encountered this with over-allocated water rights systems? Any solver settings or configuration suggestions?
Thank you,

Elizabet,

One possibility is that a downstream demand affects the availability of water for upstream demands, depending on how your priorities are set up. For example, if you add a demand downstream with a priority of 1, it may cause water to flow past an upstream demand with a priority of 2 if there is not enough water to fulfill both demands. The supply allocation can be affected by the demand priorities of demand nodes, flow requirements, and reservoirs.

Another possibility is a structure in the LP solver called "slack". Which, yes, a very large numerical value magnitudes larger than the rest of the model can cause small inaccuracies- but they are more like slightly less precise results- ie. a demand node only being filled 98% when it should be 100%. I would advise you to check your Streamflow results along each Node and Reach and investigate precisely how and where your inflows and outflows are acting.

Cheers and good luck!
-Doug

Elizabet,

One thing to keep in mind is that WEAP does not report the modeled streamflow flow at the location of the gauge but at the reach downstream of the first node BELOW the gauge. Therefore, if one of your demand node’s withdrawals is the first node downstream of the gauge, WEAP would be reporting the flow BELOW the withdrawal node.

Jack