GFI 2.0

GFI 2.0 is an enhanced version of the Geomorphic Flood Index designed to overcome one of the main limitations of DEM-based flood mapping: the inability of purely geomorphic methods to reproduce confluence-controlled backwater effects. By introducing an iterative correction guided by the Horton–Strahler stream hierarchy, GFI 2.0 improves the physical consistency of inundation mapping while preserving the parsimony, low data requirements, and computational efficiency that made the original GFI attractive for large-area screening in data-scarce regions.

Highlights

• New iterative confluence module that propagates backwater information upstream along tributaries based on stream order.
• Validated across 12 Italian basins against the official ISPRA flood hazard maps, using four DEM products from 90 m to 10 m.
• Detailed benchmark in the Bradano River basin against 2D HEC-RAS simulations for return periods of 30, 200 and 500 years.
• AUC improves from 0.917 → 0.952 (Tr 30), 0.918 → 0.960 (Tr 200), and 0.914 → 0.959 (Tr 500); RMSE on flood depth drops from 5.25 → 3.24 m (Tr 200) and 5.66 → 3.58 m (Tr 500).
• The new backwater step contributes < 0.3 % of the total runtime, even at 10 m resolution over the ~17,400 km² North Po basin.

Motivation

Reliable flood hazard information underpins risk management, spatial planning, and emergency response, yet detailed two-dimensional hydraulic models remain data- and resource-intensive and difficult to deploy consistently over large domains. DEM-based geomorphic indices — and in particular the Geomorphic Flood Index (GFI) introduced by Manfreda et al. (2015) and Samela et al. (2017) — provide a parsimonious alternative for rapid delineation of flood-prone areas. However, the original formulation associates every cell with its nearest drainage line and therefore misrepresents inundation near river confluences, where main-stem backwater and tributary–main stem interactions can flood secondary floodplains well beyond what local drainage connectivity would suggest.

What is new in GFI 2.0

GFI 2.0 retains the original definition of the index, GFI = ln(hr/H), where h_r is the bankfull depth of the nearest river segment (scaled with the contributing area through a Leopold–Maddock relationship) and H is the elevation difference between the point of interest and that segment. On top of this baseline, the new framework introduces an iterative confluence module that, for every pixel:

1. Detects the nearest downstream confluence using the Horton–Strahler ordering of the river network.
2. Recomputes h_r, H and the corresponding water-depth estimate WD using the descriptors of the confluence rather than those of the local tributary reach.
3. Adopts the confluence-based descriptors only when they imply a deeper WD than the original ones — i.e., only where main-stem backwater is hydraulically relevant.
4. Iterates downstream confluence by confluence, until the tributary meets the main stem or the condition is no longer satisfied.

The correction is therefore targeted: in the Bradano basin only 0.97 % to 2.45 % of cells are modified (depending on return period), yet they are precisely the cells that control the realism of the inundation pattern around confluences.

Multi-basin validation across Italy

The framework was tested on 12 Italian basins covering a wide range of physiographic settings (drainage areas from 381 to 17,208 km², mean elevations from 316 to 1095 m a.s.l.), using four DEM products of different origin and resolution: SRTM (90 m), MERIT (90 m), FABDEM (30 m), and TINITALY v1.1 (10 m). The reference benchmark is the national ISPRA flood-hazard product (HPH ≈ 30–50 yr, MPH ≈ 100–200 yr, LPH > 200 yr). The main findings of this part of the analysis are:

• The added value of GFI 2.0 is largest in valley and hillslope environments, where floodplain expansion and tributary–main stem interactions matter most.
• Improvements are more limited in strongly confined mountainous basins, where the valley geometry already constrains flood spreading and GFI 1.0 is already a competitive approximation.
• The relative advantage of GFI 2.0 grows when coarser DEMs are used, because the confluence module partly compensates for the loss of local topographic detail.

The Bradano River case study

A dedicated 2D hydrodynamic model was implemented with HEC-RAS 5.0.7 downstream of the San Giuliano dam on the Bradano River (~2998 km², Basilicata, southern Italy), using a 5 m DEM and a variable-size mesh (50 m base, 20 m along breaklines). Three unsteady scenarios were run for return periods of 30, 200 and 500 years (peak discharges of 1770, 3375 and 4070 m³/s) and used as a physically based benchmark for both flood extent and maximum water depth.

Compared to GFI 1.0, GFI 2.0 systematically improves the classification of flood-prone areas: the combined error R_FP + (1 − R_TP) decreases from 0.239 to 0.141 (Tr 30), from 0.283 to 0.174 (Tr 200), and from 0.288 to 0.183 (Tr 500). Flood-depth estimates also improve, especially for the larger return periods: RMSE drops from 5.25 to 3.24 m at Tr 200 and from 5.66 to 3.58 m at Tr 500, while KGE moves from −0.27 to 0.26 and from −0.24 to 0.22 respectively. The well-known positive bias of GFI-based depths is also clearly reduced (mean depth at Tr 200 decreases from 6.88 m to 5.21 m, against a simulated mean of 3.37 m).

Computationally lightweight, even at 10 m

A runtime benchmark on the North Po basin (~17,400 km², ~174 million pixels at 10 m) shows that the full GFI 2.0 workflow remains operationally lightweight: about 1 minute with 90 m DEMs, 7.6 minutes with FABDEM at 30 m, and 62.9 minutes with TINITALY at 10 m. Crucially, the new backwater propagation step itself consumes less than 0.3 % of the total runtime even in the most demanding configuration, so the added physical realism comes essentially for free. The dominant cost remains the computation of topological distances and the calibration/recalibration of the linear binary classifier.

Practical implications

GFI 2.0 is best interpreted as a targeted refinement of the original index rather than a universal replacement. It is particularly attractive for:

• Regional-scale flood-prone area screening in data-scarce regions, especially with globally available DEMs (SRTM, MERIT, FABDEM).
• Preliminary hazard assessment and prioritisation of areas in which more sophisticated 2D hydraulic modelling is warranted.
• Use as a physically grounded predictor layer in hybrid hydrologic–hydraulic workflows and in machine-learning / deep-learning pipelines for flood susceptibility mapping.

GFI 2.0 does not solve flow momentum, does not reproduce hydrograph timing or velocities, and does not explicitly represent levees, bridges or culverts; its outputs should therefore be read as first-order, geomorphology-aware envelopes of flood-prone areas and indicative inundation depths.

Resources

• Open-access paper (DOI): 10.1016/j.catena.2026.110242
• MATLAB code (Zenodo): 10.5281/zenodo.18903835
• Foundational GFI references: Manfreda et al., 2015 (Natural Hazards); Samela et al., 2017 (Advances in Water Resources); Manfreda & Samela, 2019 (J. Flood Risk Management).
• Related tools on this site: see the HydroTools page for the GFI / flood-mapping toolbox.

Citation

Manfreda, S., Saavedra Navarro, J., Albertini, C., Zhuang, R., Pacia, F. D., Chaturvedi, S., & Samela, C. (2026). Geomorphic flood index 2.0: enhanced tools for delineating flood-prone areas in data-scarce regions.Catena, 110242. https://doi.org/10.1016/j.catena.2026.110242

Funding

This study was carried out within the RETURN Extended Partnership, funded by the European Union Next-Generation EU (NRRP, Mission 4, Component 2, Investment 1.3 — D.D. 1243 02/08/2022, PE0000005); the ESA MOST Dragon VI programme (DTE-CLIMATE: Digital Twin Earth Approach for Monitoring and Modelling Climate Change in Water, Energy and Carbon Cycles in Eurasia); and the CHANCES project (An Integrated Modelling Approach for Mitigating Climate Change Effects through Enhanced Weathering in Southern Italy, CUP E53D23021850001). The work also received support from the PhD Programmes in Sustainable Development and Climate Change of the IUSS University School for Advanced Studies of Pavia, in cooperation with the University of Naples Federico II (XXXIX and XL cycles, NRRP — NextGenerationEU).

Authors

• Salvatore Manfreda — Department of Civil, Construction and Environmental Engineering (DICEA), University of Naples Federico II, Italy.
• Jorge Saavedra Navarro — DICEA, University of Naples Federico II, Italy.
• Cinzia Albertini — DICEA, University of Naples Federico II, Italy.
• Ruodan Zhuang — DICEA, University of Naples Federico II, Italy.
• Felice Daniele Pacia — DICEA, University of Naples Federico II, Italy & IUSS University School for Advanced Studies of Pavia, Italy.
• Sadashiv Chaturvedi — DICEA, University of Naples Federico II, Italy & IUSS University School for Advanced Studies of Pavia, Italy.
• Caterina Samela — National Research Council of Italy, Institute of Methodologies for Environmental Analysis (CNR-IMAA), Tito Scalo (PZ), Italy.