Precipitation–soil-moisture dominance and hydrograph signatures of floods: large-sample evidence across diverse catchments

Abstract. Precipitation and antecedent soil moisture jointly shape floods, but how their relative influence varies among individual floods, and how that variation is expressed in flood behavior, remains poorly resolved. We developed an event-based framework to quantify their relative predictive contributions to annual-maximum floods across diverse non-snow catchments. A catchment-shared Light Gradient Boosting Machine (LightGBM) model predicted daily streamflow percentile from 3-day accumulated precipitation (t − 2 to t) and soil moisture in four layers at t − 3. Shapley additive explanations (SHAP) decomposed annual-maximum predictions into model-attributed contributions, combined into a continuous event-level dominance index (D) ranging from −1 (soil-moisture dominance) to +1 (precipitation dominance). Applied to GRDC-Caravan, the model reached a median catchment-wise Nash–Sutcliffe efficiency (NSE) of 0.594 in percentile space. After quality control, we analyzed 32,930 annual-maximum events from 1,209 catchments worldwide. Precipitation was the strongest single predictor in 56.5 % of events, yet the four soil-moisture layers’ combined net contribution exceeded it in 71.4 % (66.6–87.8 % across tested predictor configurations). High streamflow percentiles arose from either moderate precipitation over very wet soils or very high precipitation over drier soils; the highest percentiles combined high precipitation with wet antecedent conditions. Increasing precipitation dominance was moderately associated with faster pre-peak rises (ρ = 0.466) and weakly with higher peaks (ρ = 0.191), faster recessions (ρ = 0.182), and precipitation closer to the peak (ρ = 0.156). Within catchments, the rising-limb association persisted (ρ = 0.417), while the peak-magnitude association was much weaker (ρ = 0.092). Precipitation-dominated floods began from drier soils and showed rapid shallow-layer wetting, while deeper layers stayed drier throughout the event window. Catchment-mean dominance shifted toward precipitation in steeper (partial ρ = 0.286) and smaller (partial ρ = −0.185) catchments, with weaker differentiation among climate zones. Yet 94.2 % of the 1,154 catchments with at least 10 events contained floods on both sides of the index, and within-catchment differences accounted for 73.1 % of the variation in D. Flood behavior therefore reflects an event-specific balance between forcing and antecedent storage rather than a fixed catchment property.

Authors

Institutions

Publication Details

Published
2026-11-09
DOI
https://doi.org/10.5194/egusphere-2026-6579
Primary Topic
Hydrology and Watershed Management Studies
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Precipitation–soil-moisture dominance and hydrograph signatures of floods: large-sample evidence across diverse catchments

Efrat Morin, Yifat Kimchi
Hydrology and Watershed Management Studies
preprint

Precipitation–soil-moisture dominance and hydrograph signatures of floods: large-sample evidence across diverse catchments

Efrat Morin, Yifat Kimchi
preprint en

Abstract

Abstract. Precipitation and antecedent soil moisture jointly shape floods, but how their relative influence varies among individual floods, and how that variation is expressed in flood behavior, remains poorly resolved. We developed an event-based framework to quantify their relative predictive contributions to annual-maximum floods across diverse non-snow catchments. A catchment-shared Light Gradient Boosting Machine (LightGBM) model predicted daily streamflow percentile from 3-day accumulated precipitation (t − 2 to t) and soil moisture in four layers at t − 3. Shapley additive explanations (SHAP) decomposed annual-maximum predictions into model-attributed contributions, combined into a continuous event-level dominance index (D) ranging from −1 (soil-moisture dominance) to +1 (precipitation dominance). Applied to GRDC-Caravan, the model reached a median catchment-wise Nash–Sutcliffe efficiency (NSE) of 0.594 in percentile space. After quality control, we analyzed 32,930 annual-maximum events from 1,209 catchments worldwide. Precipitation was the strongest single predictor in 56.5 % of events, yet the four soil-moisture layers’ combined net contribution exceeded it in 71.4 % (66.6–87.8 % across tested predictor configurations). High streamflow percentiles arose from either moderate precipitation over very wet soils or very high precipitation over drier soils; the highest percentiles combined high precipitation with wet antecedent conditions. Increasing precipitation dominance was moderately associated with faster pre-peak rises (ρ = 0.466) and weakly with higher peaks (ρ = 0.191), faster recessions (ρ = 0.182), and precipitation closer to the peak (ρ = 0.156). Within catchments, the rising-limb association persisted (ρ = 0.417), while the peak-magnitude association was much weaker (ρ = 0.092). Precipitation-dominated floods began from drier soils and showed rapid shallow-layer wetting, while deeper layers stayed drier throughout the event window. Catchment-mean dominance shifted toward precipitation in steeper (partial ρ = 0.286) and smaller (partial ρ = −0.185) catchments, with weaker differentiation among climate zones. Yet 94.2 % of the 1,154 catchments with at least 10 events contained floods on both sides of the index, and within-catchment differences accounted for 73.1 % of the variation in D. Flood behavior therefore reflects an event-specific balance between forcing and antecedent storage rather than a fixed catchment property.

Hebrew University of Jerusalem (IL)
Hydrology and Watershed Management Studies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon