NSF Award #2429932

Farm Pond Watch

A remote sensing framework for monitoring subseasonal farm pond dynamics and strengthening water resource resilience across the Southern Great Plains.

View Research Plan NSF Award Page

Institution University of Oklahoma

Principal Investigator Chengbin Deng

Award Period 2025-2026

Project Overview

Monitoring small water bodies that shape regional water resilience.

Research Need

Small farm ponds are widely used for livestock, irrigation, fish production, recreation, stormwater retention, and habitat support, yet their cumulative hydrologic impact is often overlooked.

Project Goal

The project develops an automated monitoring framework that uses high-resolution satellite time series to detect farm pond changes at regional scale.

Expected Value

Timely water-change information can support drought and flood alerts, land and water management, rural economies, and environmental conservation.

Research Approach

From Earth observation data to actionable water intelligence.

Data Foundation

Satellite Time Series

Analyze very high-resolution satellite imagery with frequent revisit cycles to track pond extent and water-cover changes over time.

Detection Method

Adaptive Change Mapping

Use a data-driven, automated, and adaptive workflow to identify changes in farm ponds, intermittent streams, and ephemeral wetlands.

Validation

NASA Collaboration

Pair the PI's remote sensing classification expertise with NASA Marshall Space Flight Center collaborators' accuracy assessment and validation expertise.

Decision Support

Resilience Applications

Translate monitoring outputs into timely information that helps explain environmental disturbance, human activity, drought, and flooding impacts.

Paper Figures

COLD water mapping figures and model results.

Methodology to predict water mapping and water change from COLD data — **Figure 1. Methodology** Methodology to predict the water mapping and water change from COLD data.

Water maps generated by classification and regression models compared with Landsat imagery — **Figure 2. Water map visualization** Water maps generated by the classification and regression models, compared with the Landsat image as reference.

Table 2. Cross-validation water-body classification metrics across regression and classification models
Metric	Regression Models			Classification Models
Metric	GBoosting	RForest	LRegression	XGBoost	RForest	SVC
Overall Acc.	0.91 +/- 0.03	0.90 +/- 0.04	0.89 +/- 0.07	0.92 +/- 0.06	0.89 +/- 0.06	0.92 +/- 0.06
WF <= 0.25	0.91 +/- 0.04	0.89 +/- 0.06	0.85 +/- 0.12	0.96 +/- 0.04	0.88 +/- 0.10	0.94 +/- 0.06
WF > 0.25	0.81 +/- 0.07	0.85 +/- 0.07	0.92 +/- 0.07	0.74 +/- 0.17	0.81 +/- 0.13	0.81 +/- 0.07
F1-score	0.72 +/- 0.09	0.72 +/- 0.11	0.70 +/- 0.19	0.77 +/- 0.15	0.70 +/- 0.21	0.79 +/- 0.11

COLD figures were extracted from the arXiv preprint PDF.

Benchmark Dataset

SWD-Net: a benchmark dataset for small water body detection.

Small Water Body Detection from very high-resolution imagery

SWD-Net provides an expert-annotated benchmark for delineating small and seasonal water bodies from aerial and satellite imagery. It supports the Farm Pond Watch goal by improving the detection of fine-scale, irregular, and fragmented ponds that are often missed by coarser global water products.

Baseline experiments compare U-Net, Mamba U-Net, SegFormer U-Net, ViT U-Net, and Swin U-Net, with Swin U-Net reporting the strongest test performance.

View DOI

0 Image-mask pairs

1024 Pixel patch size

0 Baseline architectures

0 Best F1 score class

NAIP Google Earth LoveDA U-Net Mamba U-Net SegFormer U-Net ViT U-Net Swin U-Net

SWD-Net Figures and Baseline Results

Representative SWD-Net samples from the paper — **Fig. 1. Representative SWD-Net samples** Top: 2023 NAIP summer images. Bottom: 2024 Google Earth winter images.

SWD-Net segmentation result comparison from the paper — **Fig. 2. Comparison of segmentation results** (a) Original image. (b) Ground truth mask. (c) U-Net. (d) Mamba-U-Net. (e) SegFormer-U-Net. (f) ViT-U-Net. (g) Swin-U-Net.

TABLE II. Baseline Performance on the SWD-Net Dataset
Model (%)	IoU	F1	Precision	Recall
U-Net	93.01	96.27	95.06	97.55
Mamba-U-Net	87.38	92.88	90.12	96.10
SegFormer-U-Net	92.36	95.89	94.06	97.91
ViT-U-Net	91.05	95.13	93.20	97.27
Swin-U-Net	93.53	96.56	95.02	98.23

Broader Impacts

Why this work matters beyond a single watershed.

Water Management

Regional pond dynamics can improve understanding of water balance, streamflow regimes, and resilience under irregular precipitation and frequent drought.

Research Capacity

The fellowship supports an Associate Professor and postdoctoral researcher while strengthening collaboration between the University of Oklahoma and NASA MSFC.

Public Awareness

Updated information on farm ponds can help agencies, organizations, landowners, and communities make more informed decisions about sustainable water use.

Acknowledgment

Supported by NSF and EPSCoR.

This material is based upon work supported by the U.S. National Science Foundation under Award No. 2429932 through the EPSCoR Research Fellows program.

Any opinions, findings, and conclusions or recommendations expressed in this project report are those of the authors and do not necessarily reflect the views of the U.S. National Science Foundation.

U.S. National Science Foundation

NSF EPSCoR

Research Output

Related publication

Continuous monitoring of waterbody change detection in Oklahoma using Landsat time series

Pham, Huong; Cheng, Samuel; Hu, Tao; Deng, Chengbin. Remote Sensing Letters, 2025.

View DOI

SWD-Net: A Benchmark Dataset for Small Water Body Detection from Very High-Resolution Imagery

Liu, Di; Deng, Chengbin; Olofsson, Pontus; Yan, Songkun; Wang, Jiao; Hu, Isabelle; Hong, Yang. IEEE Geoscience and Remote Sensing Letters, 2026.

View DOI