Floods_Endorheic_Salento

Data and Routines related to "The latent drainage network: mapping the hidden connectivity that controls flooding in endorheic karst" by De Michele C., Menduni G., Blöschl G. et al. 2026 submitted. All rasters are GeoTIFF on the same 5-m grid: - Coordinate system: WGS 84 / UTM zone 33N (EPSG:32633)- Upper-left corner: 710467.5 E, 4489868.348 N- Grid: 17,590 columns × 14,571 rows, cell size 5 m- Nodata: −9999 for float rasters, 255 for byte rasters The endorheic analysis domain has 77.5 million cells (1,936 km²). The flooded-area shapefile has 39 polygons; 38 of them contain cells of the domain (1.93 million flooded cells, 48 km²). ## 01_DEM| File | Content ||---|---|| dem_original_5m.tif | Original DEM (m a.s.l.) || dem_filled_5m.tif | DEM with all depressions filled, no z-limit (Priority-Flood with a small gradient on flats; Barnes et al., 2014) || flow_direction_D8.tif | D8 flow directions of the filled DEM, ESRI coding: 1 = E, 2 = SE, 4 = S, 8 = SW, 16 = W, 32 = NW, 64 = N, 128 = NE, 0 = no downslope neighbour || flow_accumulation_cells.tif | Number of upstream 5-m cells, excluding the cell itself | ## 02_latent_network| File | Content ||---|---|| latent_network_0.25km2.tif / .shp | Latent drainage network: cells whose contributing area exceeds 0.25 km² (flow accumulation > 10,000 cells). The raster is 1 = channel, 0 = other cell. The shapefile has one polyline per link between channel heads and junctions, with attributes SEG_ID, LENGTH_M and AREA_KM2 (contributing area at the downstream end). || sensitivity/latent_network_{0.0625,0.125,0.5,1,2}km2.tif | Networks used in the threshold sensitivity analysis (Supplementary Section S3, Table S2) || ctr_network_rasterized.tif | Drainage network of the Regional Technical Map (CTR), rasterized on the 5-m grid | ## 03_terrain_attributes| File | Content ||---|---|| hand_latent_network.tif | Height above the nearest drainage (m), latent network || hand_ctr_network.tif | Height above the nearest drainage (m), CTR network || depression_depth.tif | Filled DEM minus original DEM (m) || topographic_wetness_index.tif | ln(a/tan β): a = specific contributing area (m), β = local slope of the original DEM, minimum tan β = 10⁻³ || slope_degrees.tif | Local slope of the original DEM (degrees, Horn method, rounded to 0.01°) || curve_number_AMCII.tif | SCS Curve Number for antecedent moisture condition II, used for the runoff of the design storms (converted to AMC III in the scripts) | In both HAND rasters, HAND is nodata where the D8 flow paths never reach the network. ## 04_susceptibility| File | Content ||---|---|| factor_classes/*.tif | Factor class rasters of all weight-of-evidence models, for the endorheic domain only (0 = outside the domain) || woe_weights_distance_based.csv | W⁺, W⁻ and contrast C of the distance-based factors, computed on all cells of the endorheic domain || woe_weights_hand_based.csv | W⁺, W⁻ and contrast C of the factors of the HAND-based models and of the models without a network, computed on all flooded cells and a weighted random sample of 6 million non-flooded cells || decile_breaks.csv | Decile class limits of HAND, depression depth and TWI || models.csv | Factors of the six models of Table 2 | The factor classes are: - **Distance-based models:** slope class, Curve Number class, distance classes (latent network: limits 70, 150, 250 and 400 m) and relative-elevation classes. Relative elevation is the elevation minus the focal minimum within 326 m for the latent network and within 1,641 m for the CTR network.- **HAND-based models and models without a network:** Curve Number class, plus deciles of HAND, depression depth and TWI. Class 11 = HAND undefined. The susceptibility index of a model is the sum of the contrasts C of its factors. The weights given here are in-sample. Cross-validated AUCs re-estimate them on the training blocks (scripts in 06_scripts/python/susceptibility). ## 05_flooded_areas| File | Content ||---|---|| flooded_areas.shp | Mapped flooded areas of the endorheic domain (39 polygons, regional flood hazard plan) || flooded_area_id.tif | Polygon index + 1 of each flooded cell (record order of flooded_areas.shp); 0 = not flooded || endorheic_domain.tif | 1 = endorheic analysis domain || calibration_validation_split.tif | Random 70/30 split of the flooded cells used in the previous version of the manuscript (1 = calibration, 2 = validation), replaced by spatial block cross-validation || nonflooded_sample.tif | Random sample of non-flooded cells used for the HAND-based models; weight = number of non-flooded cells / 6 × 10⁶ || active_sinkholes.shp | The 43 active sinkholes of the endorheic domain || spatial_blocks/folds_*.csv | Square blocks (5, 10 and 20 km) and their fold assignment (1–5) in every partition, with block bounds in metres | The fold files correspond to three procedures: - distance-based models: 50 partitions (cvdist_new.py);- HAND-based models: 8 partitions (cvphys.py);- curves of Fig. 5: 10 assignments of 10-km blocks (fig6data.py). ## 06_scripts| Folder | Content ||---|---|| c/ | hydro.c: Priority-Flood filling, D8 directions and accumulation. hydroflat.c: the same, with flat resolution after Barnes et al., 2014. hand2.c: HAND. edt.c: Euclidean distance to the network. pond.c: volume-based fill-and-spill routing. || python/data_preparation | Reading of the original grids and shapefiles || python/network_and_terrain | Terrain attributes, threshold sensitivity, null models for proximity, flat-resolution test || python/susceptibility | WOE models, spatial block cross-validation, success-rate curves, equal-interval comparison || python/fill_and_spill | Runoff of the design storms (SCS-CN, AMC III), fill-and-spill scenarios, null model for the flooded areas reached || python/design_rainfall | Annual maxima at Maglie: GEV and Gumbel fits, simple scaling (Avanzi et al., 2015) || python/figures | Figures 1, 2, 4 and 5 and Supplementary Figures S1–S10 | The C programs are compiled with `gcc -O2 -o hydro hydro.c -lm` and read and write raw binary rasters (float32 or uint8, 14,571 × 17,590, row-major). The Python scripts need numpy, scipy, pandas and matplotlib. Paths in the scripts refer to the original working directories.