Corrections from B4T technical emails

The B4T technical emails of 2026-04-29/30 (relaying Philip Thornton, pers. comm.) describe the actual data used and correct the published documentation, which lags:

• Drought (D-01) is WASP, not SPI — Weighted Anomaly of Standardized Precipitation (IRI), 2.5°, 1980–2000 (event = ≤50% of long-term median for 3+ months), high-risk from Dilley 2005 top-two-deciles. The doc's "SPI, 1974–2004, Ericksen" conflicts. Coarse 2.5° strengthens the modernization case.
• Future layers are CMIP5, not CMIP3 — the 2050s threshold-flips (R-01, R-02, T-02, T-03) used 17 CMIP5 GCMs, RCP8.5 (Jones & Thornton). Provenance uncertain (notebook flags "possibly pre-CMIP5") → conflicted; pre-CMIP6/SSP either way.
• Aqueduct may be older than cited — B4T flags the CRI "seems [to use] CMIP5" Aqueduct while the doc cites 4.0 (CMIP6); verify per row.

Potential weaknesses (data)

• Pre-CMIP6 / pre-SSP future layers (RCP8.5 / CMIP5); Aqueduct CMIP5–CMIP6 version to confirm.
• T-03 appears to duplicate R-01 — identical LGP-flip wording under a different hazard.
• “SPI” (D-01) is mislabelled — the actual index is WASP (a distinct IRI index).

D-01 — “Standardized Precipitation Index (SPI)” · drought-event frequency

Verdict: Adopt SPEI — extend CDH pipeline to global 0.05°  a re-run of the existing pipeline, not a new indicator; FAO ASI complementary; construct needs B4T sign-off.

The Hub already produces a suitable drought product — the CHIRPS-CHIRTS-ERA5 SPEI dataset — so the recommendation is to build on that existing asset rather than create a new layer.

Full worked detail below — this row is resolved end-to-end to set the pattern.

Present — what the CRI uses now. Supplies the drought layer — the average number of drought events per year at each location, where an “event” is three straight months getting less than half the normal rainfall. It describes the recent past, not the future.

Present data — the sources conflict. Per the B4T technical emails (relaying Philip Thornton) (the actual data used), the drought layer is the Weighted Anomaly of Standardized Precipitation (WASP) from IRI, on a coarse 2.5° grid (~275 km) over 1980–2000, with high-risk areas from the top two deciles of Dilley (2005). The CRI document instead calls it “SPI” over 1974–2004. Either way:

• It isn’t really “SPI.” Standard SPI (McKee et al. 1993, WMO-endorsed) is a statistical index; WASP is a different IRI index, and the CRI layer is a count of dry spells. Publishing it as “SPI” misdescribes it.
• It’s very coarse. A 2.5° grid is roughly 275 km per cell — far coarser than the CRI’s 0.05° working grid, so today’s layer is heavily stretched to fit.
• The provenance is unresolved — WASP / 2.5° / 1980–2000 (emails) versus SPI / 1974–2004 (document). A real question for B4T.

Recommended — CDH’s existing SPEI product (internal Digital-Atlas observational track, CHIRPS v3 + CHIRTS-ERA5). Already produced and gridded — no new derivation. Spec per the CDH team (July 2026):

Table 4. The existing CDH SPEI product (CHIRPS-CHIRTS-ERA5 observational track) and its fit to the D-01 drought layer. Internal spec, CDH team, July 2026.

Why this is the strong choice. It runs at the CRI’s native 0.05° on a proven, maintained CHIRPS-based pipeline — extending it from Africa to global is a re-run, not a new indicator, so it fits within capacity. It is also a genuine upgrade on the coarse 2.5° WASP layer: finer, more current, and temperature-aware. But it is not a like-for-like swap, and two decisions sit with B4T.

Caveats B4T must weigh. (1) It is SPEI, not SPI/WASP. SPEI subtracts evaporative demand (Hargreaves PET), so “drought” becomes temperature-sensitive — arguably better under warming, but it partly overlaps the CRI’s separate high-temperature hazard (possible double-counting — a Phase-2 question). (2) Coverage — extend globally. Currently produced for Africa, but the same pipeline re-runs at global extent (0.05°, bounded by CHIRPS to 60°N–60°S — covers essentially all B4T crops). A pipeline re-run within capacity, not a new indicator. (3) No pure SPI in the Hub yet — a precipitation-only index would be a new addition rather than an existing product.

SPEIbase (CSIC) as an independent cross-check. A ready global SPEI (1901–2024, 0.5°, open) — useful to validate the extended CDH product, but no longer needed for coverage now that the CDH pipeline can run globally at the finer 0.05°.

Best-in-class agriculture-specific alternative — FAO ASI (Agricultural Stress Index System). Unlike a generic climate index, FAO’s ASI is built for agriculture: it flags crop-season drought from satellite vegetation health (VHI), masked to cropland and timed to each area’s growing season. Global, 1 km, dekadal, 1984–present, operational at FAO, and open (Google Earth Engine, ArcGIS Online, FAO Agro-informatics Platform / WMTS). It directly answers the “relevant for agriculture” ask and, being crop-masked and season-aware, also points toward B4T’s crop-specific need. A cutting-edge multi-indicator cousin is the Copernicus Combined Drought Indicator (CDI) — precipitation (SPI) + soil-moisture anomaly + vegetation (fAPAR) anomaly, crop-masked, with Watch/Warning/Alert classes (global via the Global Drought Observatory; open). Trade-off to be explicit about: ASI and CDI measure realised crop/vegetation stress, a different construct from the CRI’s precipitation-deficit metric or SPEI’s climatic water balance — arguably the most decision-relevant for breeding, but a bigger change of meaning, and both are observational (present-climate), not future projections.

Recommended baseline (decision). Adopt SPEI as the drought-hazard index. Why SPEI and not ASI/CDI: the CRI's drought slot is a hazard layer, and the CRI already scores the crop's response separately as coping capacity — so the hazard should be the climatic water deficit (SPEI), not a realised-stress index like ASI/CDI, which risks double-counting the crop's response. (This is a defensible default rather than an absolute rule — it holds insofar as the coping-capacity layer is structural/socioeconomic and independent of vegetation stress; if it is, ASI/CDI could carry extra hazard information. Worth confirming against how CCC is actually built.) SPEI is also temperature-aware, which matters under warming. Deliver it by extending CDH's existing SPEI pipeline to global 0.05° (CHIRPS-CHIRTS-ERA5) — a re-run of a proven pipeline, not a new indicator, and within capacity. Coverage is bounded by CHIRPS to 60°N–60°S, which covers essentially all B4T crop areas (the practical availability caveat; a modest compute/storage cost for the global run). Keep FAO ASI (global, 1 km, crop-masked, open) as a complementary agriculture-specific / validation layer, with SPEIbase as an independent global cross-check. Get B4T sign-off on moving from the current precipitation-deficit count to SPEI, and stop calling it “SPI.”

D-02 — Failed season (cropping reliability)

Verdict: Clarify — confirm source & role with B4T  can’t finalise until the mis-cited source and production status are confirmed; a clear modern route exists if retained.

Measures: probability of a failed growing season — germination/establishment, then fewer than 50 growing days before season end. Reported as % cropping reliability; production-specific (a season “fails” when production cost exceeds harvest value).

Alternatives: FAO GAEZ growing-period / reliability (open, ~9–10 km, historical + future); CDH internal water-balance v2 (NDWS/NDWL). Overlap (Phase 2): failed-season overlaps the LGP-flip rows and drought — confirm it is a distinct, needed input (double-count risk). Keep the production cost-vs-value test as a B4T economic assumption, separate from the climate layer.

Aqueduct 4.0 (shared) — D-03, D-04, D-05 (drought) · R-04, R-05 (changed rainfall)

Verdict: Keep — on Aqueduct 4.0 (CMIP6); confirm version  light touch if already on 4.0/CMIP6; the one open action is confirming the running version.

Measures: five WRI Aqueduct water indicators — D-03 available blue water; D-04 gross water demand / net consumption (8 monthly sectoral sets, 1960–2019); D-05 baseline water stress; R-04 intra-annual variability; R-05 seasonal variability.

Version uncertain: B4T notes “seems CMIP5 is being used in the CRI, but Aqueduct v4.0 is available using CMIP6” — so the current layers may be a CMIP5-era Aqueduct. The spec below is the CMIP6 4.0 target.

F-01 — Dartmouth flood frequency (historic)

Verdict: Replace — Global Flood Database  a modern satellite-observed successor to the same Dartmouth lineage.

Measures: historic flood-frequency proxy — relative frequency of flood occurrence per pixel. Observational, no future element.

Secondary option: Google Flood Hub “Inundation History” — per-pixel frequency of inundation (1999–2020, 128 m, wet ≥5 / ≥1 / ≥0.5% = High/Med/Low), Google Cloud Storage, CC-BY 4.0. Its README confirms it is a repackaging of the GLAD Global Surface Water Dynamics dataset (Pickens et al. 2020, UMD; Landsat 5/7/8, 30 m native → 128 m, smoothed) — i.e. a surface-water-occurrence product (peer of JRC Global Surface Water), not a flood-event catalogue, coverage band excludes far north/south. Useful context; complements GFD, doesn't replace it. (Distinct from Flood Hub's operational Sentinel-1/2 nowcast model.)

Flood hazard (shared) — F-02 river · F-03 coastal (Aqueduct Floods)

Verdict: Modernize CMIP5 → CMIP6  current Aqueduct Floods is RCP/CMIP5-era; move to a CMIP6/SSP flood product.

Measures: WRI Aqueduct Floods (Ward et al. 2020) — a 1-in-10-year flood hazard, no protection assumed. Two layers: F-02 river, F-03 coastal.

F-02 — river flood hazard

F-03 — coastal flood hazard

Cross-checks: Kirezci 2020 is CMIP5/RCP and not openly packaged; GIRI is riverine-only. Harmonise the return-period/horizon with the rest of the CRI.

Changed-rainfall LGP flips (shared) — R-01 (>120 d) · R-02 (>90 d)

Verdict: Modernize CMIP5 → CMIP6 (GAEZ v5)  legacy Thornton LGP layers → ready CMIP6 agro-climatic LGP.

Measures: length of growing period (LGP). R-01 flags LGP flipping >120→<120 days/yr by the 2050s; R-02 the >90→<90 threshold (growing day = mean air temp >6 °C and actual/potential ET > 0.35; R-02 per Nachtergaele et al. 2002).

Why GAEZ v5: it computes LGP directly from an agro-climatic water balance — the exact construct — as a ready CMIP6 product. (Thermal LGPt Ta>5/10/20/30 °C are separate layers.)

Note — likely duplicate + overlap. T-03 (under High temperature) is word-for-word the R-01 LGP-flip — resolve before implementing. LGP adequacy also overlaps failed-season (D-02) and drought — a Phase-2 consolidation question.

R-03 — Annual rainfall coefficient of variation (historic)

Verdict: Re-derive on CHIRPS v3  modern open precipitation, same stack as D-01.

Measures: flags areas where the coefficient of variation of annual rainfall exceeds the global 75th percentile (28%). Historic; current variability used as a proxy for the future (little info on how variability itself changes).

T-01 — Extreme humid heat days (GEHE)

Verdict: Keep — modern & open  Tuholske GEHE is a current, open WBGT humid-heat product.

Measures: annual count of days with max Wet Bulb Globe Temperature (WBGTmax) above ISO thresholds (>28 / 30 / 32 °C) — physiologically-relevant humid heat; manually masked.

T-02 — Growing-season Tmax flip over 30 °C

Verdict: Modernize CMIP5 → CMIP6  legacy Thornton heat-threshold layer → CMIP6 from an internal track.

Measures: areas where growing-season mean daily Tmax flips from <30 °C to >30 °C, masked where LGP > 40 days (30 °C = critical threshold; Boote 1998, Prasad 2008).

Salinity, soil water & irrigation (off-the-shelf) — S-01/02, W-01–04, I-01

Verdict: Keep — off-the-shelf  all seven are modern products from authoritative stewards; open access confirmed for all seven; one licence still to confirm on the GloSIS platform (S-01 GSASmap); no build needed. The work is cataloguing + consolidation, not replacement.

For these seven the recommendation is to keep the current dataset and catalogue it as-is — so Present and Recommended are the same dataset — except W-01 / W-02, where we recommend updating to the current SoilGrids 2.0. No CDH build. ("Present" = what the CRI uses now.)

Consolidation flags (Phase 2). Two overlaps to resolve: two salinity products (S-01 vs S-02) and four soil-water products (W-01 retention, W-02 AWC, W-03 rooting-zone, W-04 PASW) all describe soil water-holding capacity — likely more layers than the CRI needs. Confirm which are load-bearing rather than modernising all of them. B4T also notes the crop coping-capacity (CCC) soil-water term uses a soil water-balance model (waterlogging / water stress) with MapSPAM crops — confirm how the W-01…04 layers relate to it (possible overlap/double-count).

Recommendation. Keep all seven off-the-shelf (catalogue; update W-01/W-02 to SoilGrids 2.0; verify W-04's dataset link). No CDH build required. In Phase 2, consolidate the salinity pair and the soil-water quartet to what the index actually uses; add the projected AEI (SSP) if a future irrigation layer is needed.