# Spatially-informed interpolation for reconstructing lake area time series using semantic neighborhood correlation

## TL;DR

This paper introduces a new method called SNCI to reconstruct missing lake area data by using spatial correlations between lakes, improving accuracy compared to existing methods.

## Contribution

The novel contribution is the development of Semantic Neighborhood Correlation-based Interpolation (SNCI) for lake area reconstruction using spatial correlations.

## Key findings

- SNCI achieves lower interpolation errors compared to polynomial fitting, Random Forest, and Long Short-Term Memory.
- SNCI reduces mean absolute error by 50.1% and root mean square error by 28.3% for East Lake compared to the best baseline.
- SNCI demonstrates superior accuracy and correlation, especially under data-sparse conditions.

## Abstract

Long-term, high-resolution records of lake surface area are essential for characterizing the spatiotemporal dynamics of inland water bodies. Although Synthetic Aperture Radar has substantially improved water extent detection under adverse conditions, optical remote sensing imagery remains the dominant data source owing to its higher spatial resolution. Nevertheless, optical data are frequently compromised by persistent cloud cover and sensor limitations, leading to substantial observational gaps. To effectively address this challenge, this study introduces a novel spatially-informed interpolation method termed Semantic Neighborhood Correlation-based Interpolation (SNCI), which leverages spatial correlations among hydrologically interconnected lakes to reconstruct missing lake area observations. By explicitly modeling the inherent hydrological and climatic coherence among neighboring lakes, SNCI provides robust, accurate, and scalable interpolations even in the presence of extensive temporal data losses. The method was evaluated on monthly lake area data from 54 lakes in the Wuhan region between 2000 and 2020, using the Global Surface Water dataset, and validated against high-resolution Dynamic World observations. Several representative lakes were analyzed in detail to assess SNCI’s robustness across diverse seasonal and interannual conditions. Compared with polynomial fitting, Random Forest, and Long Short-Term Memory, SNCI consistently achieves lower interpolation errors. In the case of East Lake, SNCI reduces mean absolute error by 50.1% and root mean square error by 28.3% relative to the best-performing baseline. Across all lakes, SNCI demonstrates superior accuracy and correlation, particularly under data-sparse conditions. These results underscore SNCI’s potential to enhance lake area reconstruction accuracy and support broader applications in hydrological modeling, environmental monitoring, and climate impact assessment.

The online version contains supplementary material available at 10.1038/s41598-025-09410-3.