The runoff model - and the split
Generated by build_runoff_model.py. Two linear reservoirs in parallel off one split fraction, calibrated against FLOW (which closes a water balance) on the first 70% of each…
Generated by build_runoff_model.py. Two linear reservoirs in parallel off one split fraction, calibrated against FLOW (which closes a water balance) on the first 70% of each record and scored on the last 30%.
- catchments fitted: 9
- validation NSE > 0.5: 8
- median validation NSE: 0.70
The split
Median quick fraction f = 0.57 (IQR 0.56-0.63). So about 43% of effective rainfall goes to recharge and the rest runs off quickly.
Water balance, median across catchments: rain 1034, AET 476, recharge 202 mm/yr.
What this unlocks.
build_wt_coupled.pyconstrainsrecharge / Syto about 29.5 m/yr but not either term. With recharge now estimated INDEPENDENTLY from flow, Sy follows by division - and it is derived from a different dataset (river flow) than the head record that produced the ratio, so the two are genuinely independent.
| catchment | km2 | f | NSE cal | NSE val | bias % | BFI | recharge mm/yr |
|---|---|---|---|---|---|---|---|
| Milverton | 28 | 0.56 | 0.82 | 0.88 | +2.0 | 0.43 | 264 |
| Fenny Castle | 60 | 0.47 | 0.82 | 0.82 | +3.0 | 0.53 | 343 |
| Bishops Hull | 244 | 0.56 | 0.68 | 0.79 | +10.0 | 0.43 | 201 |
| Lovington | 147 | 0.62 | 0.75 | 0.70 | -3.7 | 0.37 | 168 |
| Pen Mill | 215 | 0.65 | 0.74 | 0.70 | +14.4 | 0.35 | 170 |
| Ashford Mill | 89 | 0.57 | 0.71 | 0.68 | +6.0 | 0.42 | 241 |
| Halsewater | 53 | 0.26 | 0.68 | 0.66 | -14.3 | 0.74 | 469 |
| Chiselborough | 75 | 0.66 | 0.71 | 0.66 | -7.7 | 0.34 | 202 |
| Somerton | 23 | 0.63 | 0.19 | 0.18 | -59.4 | 0.37 | 189 |
Known holes in THIS model
- daily timestep. Peaks are often sub-daily; the 15-minute stage now on disk is what refines them, and it must not be done by tuning
fto compensate for a timestep error. - no routing. Flow is generated at the outlet, not routed to it, so travel time is absorbed into
kq. Fine for catchments this size, wrong if they are nested or combined. - PET is a calibrated sinusoid, not Penman - there is no radiation or temperature series in this project.
- no tidal boundary. Lowland outlets discharge through gates; nothing here knows that.