DG expand: wire Greg DG hook into NE variant (PC form, DGGD/PTBALT call site)#107
Open
holoros wants to merge 13 commits into
Open
DG expand: wire Greg DG hook into NE variant (PC form, DGGD/PTBALT call site)#107holoros wants to merge 13 commits into
holoros wants to merge 13 commits into
Conversation
Add DGDRIVER (slot 149) and MORTDRVR (150; 8-char FVS limit on MORTDRIVER) to the base keyword TABLE (keywds.f90). initre.f90 handlers mirror the GOMPMORT/GOMPON pattern: read the driver code into the new /GREGKW/ common (IDGDRV,IMORTDRV; defaults DG=cspi(2), mort=bgi(3), matching config_loader). GREGLOADDG consults IDGDRV. GREGKW.f90 common + GREGKWBD block data init to -1. Compiles clean (FVSne, 0 failures); keywords present in binary; gregkwbd_ linked. Per-driver coefficient-file resolution is the documented downstream step.
Extends the Greg DG native hook to CS and LS for the in-engine 3-arm A/B. Adapted the ne/dgf.f90 hook to each variant vars (LS also needed VARCOM.f90 for PTBALT). Added ../base/gregdghg.f90 to FVScs/FVSls sourceLists (was NE-only, caused undefined gregloaddg_/gregdgv_ at link). Hook stays non-default (env/keyword gated). A/B result: hook wins NE, marginal CS, LOSES LS (over-predicts).
… + wc LS bug: ls/dgf.f90 passed PTBALT(I) to GREGDGV but LS never populates PTBALT at DG time (no BADIST) so it was 0 -> competition term vanished -> over-predict. Fix: pass the local BAL. LS hook now RMSE 0.1222 (native 0.1212), bias -0.005 (native -0.020): competitive, not a loss. The LS over-shoot was a wiring bug, not coefficient scale. DGDRIVER: GREGLOADDG now resolves the coef file from the keyword code (1=deployed, 2=refit) and self-activates the hook; env-var fallback kept. Validated end to end. nc/wc: hook wired (+ gregdghg.f90 in their sourceLists). A/B: WC hook wins (0.1154 vs 0.1182); NC hook over-predicts (native already well-calibrated). Landscape: hook helps where native under-predicts, hurts where native is calibrated.
IE/SN/AK hook wired (local BAL, no LS-style zero-BAL). Confirming A/B: IE CONFIRM (hook 0.0973, bias +0.0003), AK CONFIRM (provisional, n=11), SN REFUTE (native under-predicts but hook overshoots +0.083 -> keep native). DGDRIVER policy deployed: config/dg_hook_policy.json + config_loader.dg_hook_policy/dgdriver_keyword, opt-in (tag byte-identical). HOOK confirmed: ne,cs,ls,wc,ie (+ak provisional); NATIVE: rest incl SN. BAKUZIS GATE FLAG: hook runs away in DBH long-term (no max-size asymptote); short-interval prediction only until a DG deceleration is added.
GREGDGV: logistic size-decel (onset 85% of per-species DBHMAX from FIA p99, new GDGMAX array in GREGMC). Long-term QMD plateaus 19-31 in at 200yr (was 28-50, native 13-31); site ordering/monotonicity intact; short-interval NE A/B preserved (0.2135 vs 0.2123). Backward compatible (sentinel if no DBHMAX col). Hook coef set greg_dg_coefficients_refit_decel.csv (refit + DBHMAX). Policy: Bakuzis caveat RESOLVED, AK CONFIRMED (34 plots). Hook now validated short AND long horizon.
… policy OP hook wired (correct local BAL, verified) but the confirming A/B REFUTES the hook-on call: bias improves ~25% but RMSE worsens (0.1354 vs native 0.1313), trades bias for variance. Policy: OP=NATIVE (like SN). dg_hook_policy.json updated (OP native, confirmed set ne/cs/ls/wc/ie/ak, mortality status note). Mortality size-extension kept as an artifact (needs all morts.f90 callers updated before merge).
eta = b0 + b1*(cr+.01)^b2 + b3*cch^b4 + b5*log(DBH) + b6*BGI; GB->7 cols (GOMPMC); GOMPLOAD reads 6th/7th cols backward-compatible (5-col=Greg crown-only default, 6=+size, 7=+BGI); BGI per-stand via FVS_GOMP_BGI. All 21 morts.f90 callers pass DBH(I). NE 4-arm: native 0.755/crown 0.451/size 0.434/size+BGI 0.422 (AUC 0.662). Crown-only (Greg source form) stays default; size/BGI opt-in. Build verified in worktree (ne/ca/sn exit 0, identical source).
…riant size+BGI policy MORTDRVR/MORTDRIVER previously carried a 0-5 driver-family scheme (0=none,1=elev,2=cspi,3=bgi,4=esi,5=emt) copied from DGDRIVER. That never matched what merged into the engine: GOMPSURV is an additive TIER structure (crown-only 5-coef -> +size 6-coef -> +size+BGI 7-coef), not a swappable single driver. Redefine MORTDRVR/IMORTDRV to select the tier: 0=crown-only (default, unchanged behavior), 1=size, 2=size+BGI. - src-converted/vbase/initre.f90: option 150 handler rewritten for the 0/1/2 tier scheme, bare keyword defaults to the recommended tier (size+BGI); new echo format names the tier instead of the driver family. - src-converted/common/GREGKW.f90: IMORTDRV comment corrected to describe the tier scheme; -1 unset sentinel unchanged for both codes. - src-converted/base/gompmort.f90: GOMPLOAD includes GREGKW.f90 and logs the keyword-selected tier (reproducibility-logging only, matching the DGDRIVER precedent -- actual tier selection is still via the FVS_GOMPIT_COEF file path). - config/mort_hook_policy.json: per-variant size+BGI policy from the 6-variant 4-arm A/B (native/crown-only/size/size+BGI) on held-out FIA remeasurement pairs. size+BGI wins on log-loss and AUC in all six confirmed variants (ne, cs, ls, wc, ie, ak); size-only is retained but not recommended standalone. ak flagged low-confidence (33-plot sample). - config/config_loader.py: mort_hook_policy()/mortdriver_keyword() helpers mirroring dg_hook_policy()/dgdriver_keyword(). Rebuilt FVSne (0 build failures); smoke-tested MORTDRVR in the keyword table, the new tier echo string, and gregkwbd_ link via strings/nm.
Greg Johnson revised the DG equations to use two principal components derived from DD_0, TD, PPT_sm, and DD_18, replacing the previous ELEV and EMT covariates (per Western Mensurationists feedback on ELEV use). Changes: - GREGMC.f90: replace GEMT, GELEV with GDD0, GPPT_SM, GDD18, GPC1, GPC2 - gregdghg.f90 GREGLOADDG: read DD0/TD/PPT_SM/DD18 env vars; compute PCs - gregdghg.f90 GREGDGV: drop ELEV/EMT args; use GPC1/GPC2 from module - dgf.f90: update GREGDGV CALL to drop GELEV, GEMT arguments New env vars required at runtime: FVS_GREG_DD0 (growing degree days > 0C) FVS_GREG_TD (temp diff MWMT-MCMT, unchanged) FVS_GREG_PPT_SM (summer precip mm) FVS_GREG_DD18 (growing degree days > 18C) Replaces: FVS_GREG_ELEV, FVS_GREG_EMT PC loadings: PC1 = 0.252*DD0 + 0.002*TD - 0.035*PPT_SM + 0.967*DD18 PC2 = 0.882*DD0 + 0.015*TD + 0.420*PPT_SM - 0.215*DD18 Coefficient CSV must be regenerated from dg_parms.RDS (see extract_dg_parms_csv.R). Shape is unchanged: SPCD, N, B0..B6, DBHMAX.
Greg Johnson revised the DG equations to use two principal components derived from DD_0, TD, PPT_sm, and DD_18, replacing the previous ELEV and EMT covariates (per Western Mensurationists feedback on ELEV use). Changes: - GREGMC.f90: replace GEMT, GELEV with GDD0, GPPT_SM, GDD18, GPC1, GPC2 - gregdghg.f90 GREGLOADDG: read DD0/TD/PPT_SM/DD18 env vars; compute PCs - gregdghg.f90 GREGDGV: drop ELEV/EMT args; use GPC1/GPC2 from module - dgf.f90: update GREGDGV CALL to drop GELEV, GEMT arguments New env vars required at runtime: FVS_GREG_DD0 (growing degree days > 0C) FVS_GREG_TD (temp diff MWMT-MCMT, unchanged) FVS_GREG_PPT_SM (summer precip mm) FVS_GREG_DD18 (growing degree days > 18C) Replaces: FVS_GREG_ELEV, FVS_GREG_EMT PC loadings: PC1 = 0.252*DD0 + 0.002*TD - 0.035*PPT_SM + 0.967*DD18 PC2 = 0.882*DD0 + 0.015*TD + 0.420*PPT_SM - 0.215*DD18 Coefficient CSV must be regenerated from dg_parms.RDS (see extract_dg_parms_csv.R). Shape is unchanged: SPCD, N, B0..B6, DBHMAX.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Wires the Greg Johnson DG hook (GREGDGV/GREGLOADDG) into the NE variant and updates all shared files to the PC-form equations, matching the pattern established by #103 (SN), #104 (PN), #105 (EM), #106 (WC).
Changes
src-converted/common/GREGMC.f90— replace GEMT/GELEV with GDD0/GTD/GPPT_SM/GDD18/GPC1/GPC2 in COMMON /GREGMR/src-converted/base/gregdghg.f90— GREGDGV drops ELEV/EMT args; GREGLOADDG reads DD0/TD/PPT_SM/DD18 env vars and computes PC1/PC2src-converted/common/GREGKW.f90— COMMON /GREGKW/ include needed by gregdghg.f90src-converted/ne/dgf.f90— update CALL GREGDGV to drop GELEV, GEMT (NE uses DGGD and PTBALT(I))config/greg_dg_coefficients.csv— regenerated from dg_parms.RDS in PC form (83 spp; SPCD 653 dropped, non-convergence)Runtime env vars
FVS_GREGDG=1(enable hook, unchanged),FVS_GREG_DD0,FVS_GREG_TD,FVS_GREG_PPT_SM,FVS_GREG_DD18(replace ELEV/EMT)Verification
FVSne built on Cardinal with gcc/12.3.0; gregdgv_ and gregloaddg_ confirmed linked (2 syms).
Notes
Do not merge without Greg Johnson sign-off (matches policy on #103-106). SPCD 653 excluded from CSV; Greg flagged as known.