Water Mapping and Change Detection Using Time Series Derived from the Continuous Monitoring of Land Disturbance Algorithm
Huong Pham a , Samuel Cheng a , Tao Hu b , and Chengbin Deng c · 2025
Read Water Mapping and Change Detection Using Time Series Derived from the Continuous Monitoring of Land Disturbance Algorithm by Huong Pham a , Samuel Cheng a , Tao Hu b , and Chengbin Deng c (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-03170.txt.
Introduction
Huong Pham and co-authors tested in 2025 whether a land-change algorithm built for forests and cities can also map water and its drift over time. The algorithm watches satellite pixels continuously and flags disturbances as they happen. Water managers need exactly that vigilance for lakes, rivers, and reservoirs under climate pressure.
The team found the method estimates how often each pixel holds water during calm periods and outlines water bodies reliably. After a disturbance, it can judge whether water frequency at a pixel is rising, falling, or holding steady. The authors say those trend calls support timely interventions and better decisions. They frame water as one more land signal the continuous monitor can read. The approach runs on existing data streams.
Catalog record
| CREATOR | Huong Pham a , Samuel Cheng a , Tao Hu b , and Chengbin Deng c |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-03170
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Water Mapping and Change Detection Using Time Series Derived from the Continuous Monitoring of Land Disturbance Algorithm Huong Pham a , Samuel Cheng a , Tao Hu b , and Chengbin Deng c Abstract Given the growing environmental challenges, accurate monitoring and prediction of changes in water bodies are essential for sustainable management and conservation. The Continuous Monitoring of Land Disturbance (COLD) algorithm provides a valuable tool for real-time analysis of land changes, such as deforestation, urban expansion, agricultural activities, and natural disasters. This capability enables timely interventions and more informed decision-making. This paper assesses the algorithm’s effectiveness to estimate water bodies and track pixel-level water trends over time. Our findings indicate that COLD-derived data can reliably estimate estimate water frequency during stable periods and delineate water bodies. Furthermore, it enables the evaluation of trends in water areas after disturbances, allowing for the determination of whether water frequency increases, decreases, or remains constant. keywords COLD algorithm; water change; water frequency † † affiliation: a School of Electrical and Computer Engineering, University of Oklahoma, OK, USA; b Department of Geography, Oklahoma State University, OK, USA; c Center for Spatial Analysis, Department of Geography, University of Oklahoma, OK, USA 1 Introduction Detecting surface water in aquatic ecosystems is essential for environmental monitoring, water resource management, and ecosystem conservation, particularly in communities where water resources are closely linked to recreational and economic activities. The decline in water levels in many lakes has raised significant concerns, highlighting the need for accurate detection methods to assess water availability, understand hydrological cycles, and manage droughts and floods. Remote sensing offers a valuable tool for measuring the spatial variability of surface water in these ecosystems. Advances in remote sensing technologies and satellite imagery have significantly improved our ability to monitor water bodies, providing critical data for hydrological analysis and identifying the drivers of water level fluctuations ( Duan et al. 2021 ) . The Continuous Monitoring of Land Disturbance (COLD) algorithm, developed using Landsat time series data, offers an effective approach for detecting various types of land disturbances as new imagery becomes available ( Zhu et al. 2020 ) . Additionally, it suports the generation of historical maps of these disturbances. Through experimentation with different data inputs and time series analysis techniques, the algorithm has been refined to enhance detection performance. Key improvements include the use of surface reflectance instead of Top-of-Atmosphere (TOA) reflectance, the integration of multiple spectral bands, and the efficient removal of outliers based on model predictions. Importantly, the algorithm can distinguish between actual land disturbances and changes resulting from vegetation regrowth, improving the accuracy of land disturbance mapping. This capability lays a robust foundation for investigating the algorithm’s potential to estimate water frequency, defined as the proportion of observations classified as water relative to the total number of stable observations, where no disturbances occur. This enables the delineation of water bodies and the analysis of water trends at the pixel level following disturbance events. Generating land cover classifications from remotely sensed data is a relatively straightforward process, but achieving high accuracy presents challenges. To enhance classification accuracy, many products incorporate multitemporal images into their algorithms ( Melichar et al. 2023 ; Liu et al. 2021 ; Shetty et al. 2021 ) . However, automating the collection of multitemporal images can introduce several potential issues. One major challenge is ensuring data consistency across images. Variations in atmospheric conditions during different image acquisition dates can affect spectral signatures, making it difficult for algorithms to maintain consistent classification over time. For instance, cloud-free images are essential for accurately classifying each pixel across multiple observations. This requirement is often unattainable, particularly for sensors with low temporal frequency, such as Landsat. As a result, acquiring suitable images may take several years, leading to the production of Landsat-based land cover maps at five- or ten-year intervals ( Zhu and Woodcock 2014 ) . Additionally, temporal variability in images, such as phenological changes in vegetation or land use changes due to human activities, can lead to misclassification if the algorithm does not account for these temporal dynamics ( Wang and Yao 2024 ) . This issue is particularly pronounced when images from long intervals are used or when mapping areas that undergo frequent changes. Consequently, land cover maps produced using conventional multitemporal methods may be unreliable for identifying land cover changes, as classification errors can be misinterpreted as actual changes. This problem is further exacerbated in smaller areas where subtle changes are more likely to occur. To address these challenges, using data derived from Continuous Monitoring of Land Disturbance (COLD) algorithm could be a promising approach for classifying water bodies, as it (1) accounts for the temporal dynamics of land cover changes, (2) is robust against data inconsistencies, and (3) reliably detects water change trends. The main contributions of this study are as follows: 1. The proposed classification and regression models can effectively identify the water frequency of various areas. 2. Estimate the changing in water frequency at the break time between two stable periods. 2 Related work Various methods for extracting water surfaces using remote sensing data have been developed, including both machine learning and traditional approaches ( Gharbia 2023 ) . Machine learning methods require large datasets and careful sample selection to ensure generalization across different regions. Traditional algorithms, in contrast, often rely on spectral differences between water and non-water areas, using multiple bands for effective extraction. A notable example is the water index method, which combines spectral bands and remote sensing indices to distinguish water bodies. This method has demonstrated high performance in studies and offers advantages such as simplicity, robustness, and ease of deployment across large, diverse areas, enabling quick and efficient extraction ( Wang et al. 2020 ) . Deep learning, with the invention of convolutional neural networks (CNNs), has been particularly useful for segmentation tasks in remotely sensed image analysis. These models can automatically extract both low- and high-level features from images, enabling the classification of water bodies with high accuracy, even in challenging environments ( Chang et al. 2024 ; Mullen et al. 2023 ; Nasir et al. 2023 ; Liu, Liu, and Hu 2024 ; Erdem et al. 2021 ; Gharbia 2023 ) . For example, the CNN-based model U-Net has been widely used in segmentation tasks ( Cao et al. 2024 ; Gharbia 2023 ) . An ensemble model, WaterNet, combines various U-Net architectures with the cGAN-based Pix2Pix to improve water body segmentation accuracy ( Erdem et al. 2021 ) . Similarly, Wang ( Wang et al. 2023 ) investigates the SER34AUnet model, an enhancement of the U-Net architecture specifically tailored for water extraction in cold and arid regions. While deep learning models are powerful for tasks such as water body extraction, they are heavily data-dependent and require large amounts of labeled data for training. Acquiring large, continuous, high-quality labeled datasets can be challenging and time-consuming ( Gharbia 2023 ) . Due to the difficulty of acquiring high-quality training datasets, these models are prone to overfitting, especially when the dataset is small or not representative of the broader context, leading to poor generalization on unseen data ( Liu, Liu, and Hu 2024 ) . Recent advances in visual foundation models like