Station spacing bounds the recovery of rainfall extremes by interpolation: an independent-network test with consequences for landslide thresholds
Abstract
Gridded rainfall products interpolated from gauge networks underpin landslide early warning, and a common view holds that their main deficiency in mountains is the omission of orographic enhancement. This study tests that view using an independent network: 464 mountain stations operated by the Korea Forest Service that do not enter the operational grid of the Korea Meteorological Administration. Over five summers (2020–2024, 92,663 wet station-days after quality control), daily rainfall interpolated from the 615 meteorological-agency stations alone was compared with measurements from the mountain stations. Three findings follow. First, summer daily rainfall showed no systematic dependence on elevation: the median elevation gradient within the mountain network ranged from −0.05 to +0.10 mm per 100 m across the five years, and interpolation from the lowland network reproduced the day-to-day gradient in both sign and magnitude. Second, interpolation attenuated extremes in proportion to their intensity, capturing 112% of 1–10 mm days but 75% of days above 200 mm; the operational configuration classified only 26% of 200-mm days as exceeding 200 mm. Third, this loss was not recoverable: terrain-weighted regression, a PRISM-type local regression and ordinary kriging performed no better than inverse-distance weighting, and quadrupling gauge density raised exceedance detection from 0.25 to 0.37. Forty-four percent of the mountain extremes had no gauge within 10 km that recorded a comparable total. Radar reflectivity confirmed that these extremes were spatially real. At 9,656 landslide sites the operational field placed 25% above 200 mm; adding the mountain network raised this to 49%. The deficiency of the gauge-interpolated field for landslide forecasting is the attenuation of extremes between gauges, which denser networks and better interpolators do not cure and which the inclusion of mountain observations partly does. Keywords: rainfall interpolation; gauge network density; extreme precipitation; orographic enhancement; landslide early warning; independent validation; Korea
Related
- Landslide Susceptibility Modelling and Infrastructure Exposure in the Zillertal, Austrian Alps
- Machine-learning modelling of regional landslide susceptibility in eThekwini, South Africa
- Landslide Susceptibility Modeling and Mapping: A Comparison of Frequency Ratio and Decision Tree Models
- Brief communication: Landsifier 2 – Python-based Software for Automated Landslide Type Classification
- InSAR observations from multiple tracks reveal rainfall modulation of post-failure reactivation of the Xiongwadashan landslide
BibTeX
@misc{min2026station,
title = {{Station spacing bounds the recovery of rainfall extremes by interpolation: an independent-network test with consequences for landslide thresholds}},
author = {Sunghyun Min},
howpublished = {EarthArXiv},
year = {2026},
month = sep,
doi = {10.31223/x5ff7k},
url = {https://eartharxiv.org/repository/view/14846/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Station spacing bounds the recovery of rainfall extremes by interpolation: an independent-network test with consequences for landslide thresholds
AU - Sunghyun Min
PY - 2026
DA - 2026/09/06
PB - EarthArXiv
DO - 10.31223/x5ff7k
UR - https://eartharxiv.org/repository/view/14846/
AB - Gridded rainfall products interpolated from gauge networks underpin landslide early warning, and a common view holds that their main deficiency in mountains is the omission of orographic enhancement. This study tests that view using an independent network: 464 mountain stations operated by the Korea Forest Service that do not enter the operational grid of the Korea Meteorological Administration. Over five summers (2020–2024, 92,663 wet station-days after quality control), daily rainfall interpolated from the 615 meteorological-agency stations alone was compared with measurements from the mountain stations. Three findings follow. First, summer daily rainfall showed no systematic dependence on elevation: the median elevation gradient within the mountain network ranged from −0.05 to +0.10 mm per 100 m across the five years, and interpolation from the lowland network reproduced the day-to-day gradient in both sign and magnitude. Second, interpolation attenuated extremes in proportion to their intensity, capturing 112% of 1–10 mm days but 75% of days above 200 mm; the operational configuration classified only 26% of 200-mm days as exceeding 200 mm. Third, this loss was not recoverable: terrain-weighted regression, a PRISM-type local regression and ordinary kriging performed no better than inverse-distance weighting, and quadrupling gauge density raised exceedance detection from 0.25 to 0.37. Forty-four percent of the mountain extremes had no gauge within 10 km that recorded a comparable total. Radar reflectivity confirmed that these extremes were spatially real. At 9,656 landslide sites the operational field placed 25% above 200 mm; adding the mountain network raised this to 49%. The deficiency of the gauge-interpolated field for landslide forecasting is the attenuation of extremes between gauges, which denser networks and better interpolators do not cure and which the inclusion of mountain observations partly does. Keywords: rainfall interpolation; gauge network density; extreme precipitation; orographic enhancement; landslide early warning; independent validation; Korea
ER -