Back to search

University of North Texas

Detecting Floodwaters and Estimating Their Water Depth in Harris County with Sentinel-1 Imagery

Abstract

dc:description

Urban flooding has emerged as a frequent disaster across rapidly growing metropolitan areas of the United States, where several factors such as, impervious surfaces, engineered drainage networks, and extreme rainfall events altogether result into sever and prolonged flooding. Mapping flood extent and depth accurately in narrow channel systems is still a persistent issue due to limitation of satellite sensors with low spatial resolution, backscatter response of heterogeneous urban surfaces, and lack of in-situ observations. This work evaluates the possibility of multi-temporal SAR imagery [Interferometric Wide (IW) Swath Mode with 10-m spatial resolution] and LiDAR-derived DEM to map floodwater and estimate its water depth in Harris County. The study examines accuracy of unsupervised and supervised classification to generate water extent and then derive water depths for supporting urban flood assessment. Five Sentinel-1 scenes spanning pre-flood, peak-flood, and recovery phases of Hurricane Harvey, along with one pre-event flood scene from the April 2016 Tax Day Flood, were preprocessed in SNAP and analyzed through two complementary classification pipelines. Otsu thresholding was applied with a dual-polarization VH/VV decision rule to generate binary water masks for each scene, while three supervised classifiers were trained on 3,000 labeled samples points using a seven-feature stack derived from Sentinel-1 backscatter and terrain variables after multicollinearity analysis reduced twelve initial features by removing five correlated predictors. Water depth was estimated by a boundary-based water surface elevation (WSE) interpolation in which morphological erosion extracted the flood boundary, DEM elevations were assigned as WSE seeds, and a distance-transform diffusion propagated WSE inward before subtracting the DEM to obtain depth. Water extent and water depth were validated against Google Earth Imagery and six USGS gage stations, respectively. At peak flood on August 30, 2017, the Otsu classification obtained a Kappa of 0.76 and an overall accuracy of 88.0%, while accuracy dropped sharply on pre-flood and recovery dates (Kappa = 0.08 – 0.24) where the bimodal backscatter assumption no longer held. Unsupervised classifier Otsu technique was outperformed by supervised classifiers (XGBoost, RF, and SVM) achieving overall accuracies of 92.4%, 92.2%, and 89.4% and Kappa of 0.797, 0.793, and 0.725 respectively. Feature importance analysis demonstrated that classification decision in supervised classification was combinedly dominated by VV backscatter, VH backscatter, and DEM elevation, with more than 80% of the total importance. Depth validation was performed for August 30, which yielded r = 0.953. R2 = 0.908, and RMSE = 0.288 m for XGBoost, with a near-zero bias of -0.038 m, followed by RF (R2 = 0.869, RMSE = 0.470 m) and SVM (R2 = 0.786, RMSE = 0.549 m). The estimated Otsu depths for the same date achieved R2 = 0.798 and RMSE of 0.72, decreasing rapidly on dry dates. In both validation analyses, depth accuracy is directly associated with accuracy water mask, concluding that supervised XGBoost method provides the most reliable results for mapping of flood and estimation of depth utilizing Sentinel-1 imagery. The findings contribute empirical evidence that Sentinel-1 imagery, despite 10-m spatial resolution, can deliver water depth validation accuracy in urbanized watersheds when integrated with high-resolution terrain data and modern ensemble classifiers.

Degree

thesis:*
Grantor dc:publisher
University of North Texas
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Bista, Hirendra 1999-
Contributors dc:contributor
  • Jung, Jihoon
  • Pan, Feifei
  • Dong, Pinliang

Subjects

dc:subject × 3

Rights

dc:rights
Statement dc:rights
  • Public
  • Bista, Hirendra 1999-
  • Copyright
  • Copyright is held by the author, unless otherwise noted. All rights Reserved.
Language dc:language
English

Identifiers

dc:identifier.*
Identifier
local-cont-no: submission_5056
ark: ark:/67531/metadc2652545

Chain of custody

source
Harvested from
University of North Texas
Base URL
digital.library.unt.edu/oai/
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Bista, Hirendra 1999-. Detecting Floodwaters and Estimating Their Water Depth in Harris County with Sentinel-1 Imagery. University of North Texas, 2026. https://digital.library.unt.edu/ark:/67531/metadc2652545/