AQUAGAN: GENERATIVE ADVERSARIAL NETWORKS FOR SYNTHETIC HYDROCLIMATIC DATA GENERATION IN DATA-SCARCE WATERSHEDS

Sangeethavani B

AQUAGAN: GENERATIVE ADVERSARIAL NETWORKS FOR SYNTHETIC HYDROCLIMATIC DATA GENERATION IN DATA-SCARCE WATERSHEDS

Authors

Sangeethavani B

Assistant professor, Centre for rural technology, The Gandhigram Rural Institute, Gandhigram, Dindigul District 624302, India.

Keywords:

Generative Adversarial Networks, Hydroclimatic Data, Data-Scarce Watersheds, Synthetic Time Series, Wasserstein GAN, Temporal Coherence , Multi-Variable Correlation

Synopsis

In data-scarce watersheds, limited hydroclimatic data poses challenges for effective water resource management and predictive modeling. We propose AquaGAN, a novel Generative Adversarial Network (GAN) framework designed to generate high-fidelity synthetic time series for precipitation, streamflow, and temperature. AquaGAN employs a hybrid architecture integrating Long Short-Term Memory (LSTM) networks for temporal dependencies and convolutional neural networks (CNNs) for spatial patterns. A Multi-Variable Correlation Module (MVCM) ensures inter-variable consistency, while a Temporal Coherence Constraint (TCC) enforces realistic temporal transitions using a Wasserstein loss function with gradient penalty. The model is trained on normalized, segmented data with conditional inputs like climate indices to enhance contextual relevance. AquaGAN was evaluated on synthetic and real-world datasets from a data-scarce Sub-Saharan African watershed. Results demonstrate superior performance over baselines (Vanilla GAN, TimeGAN, and multivariate autoregressive models), achieving a Fréchet Distance of 0.12 (synthetic) and 0.15 (real-world), Autocorrelation Error of 0.03 and 0.04, and Cross-Correlation Score of 0.92 and 0.90, respectively. Hydrological utility tests show AquaGAN improves streamflow prediction by 15–18% and flood forecasting by 18–25% compared to baselines. Visual inspections confirm AquaGAN captures seasonal patterns and extreme events effectively. AquaGAN offers a robust solution for generating realistic hydroclimatic data, enhancing water resource management in data-limited regions, with potential for future transfer learning applications.

References

[1] Bougdira, B., Layan, B., Yadari, S. E., Abbou, M. B., Adouk, N., & Benaabidate, L. (2025). Assessing Climatic Variability in Data Scare Region of Morocco (UpperLarbaâ Basin): Drought Periods and Exceptional Precipitation Events from 1958 to 2023. European Scientific Journal ESJ, 21(18), 1. https://doi.org/10.19044/esj.2025.v21n18p1

[2] Da Silva, F. H. O., Lopes, F. B., Da Costa Bezerra, B. G. M., Viana, N. S., Da Silva Araújo, I. C., De Sousa Luna, N. R., Pontes, M. C., Cavalcante, R. R., De Alburquerque Aragão, F. T., & De Andrade, E.M. (2025). Impact of permanent preservation areas on water quality in a Semi-Arid Watershed. Environments, 12(7), 220. https://doi.org/10.3390/environments12070220

[3] Funes, M. F., Reyna, T. M., García, C. M., Lábaque, M., López, S., Strusberg, I., & Vanoni, S. (2025). Estimating Macroplastic Mass Transport from Urban Runoff in a Data-Scarce Watershed: A Case Study from Cordoba, Argentina. Sustainability, 17(13), 6177. https://doi.org/10.3390/su17136177

[4] Lami, A., Rogora, M., Austoni, M., & Brankovits, D. (2025). High frequency data from in-situ sensors as a support to long-term ecological research on aquatic ecosystems. ARPHA Conference Abstracts, 8. https://doi.org/10.3897/aca.8.e151707

[5] Mezali, F., Chetibi, M., Naima, K., Derdour, A., Benmamar, S., Almohamad, H., Hasher, F. F. B., & Abdo, H. G. (2025). Enhancing Groundwater Recharge Assessment in Mediterranean Regions: A Comparative Study Using Analytical Hierarchy Process and Fuzzy Analytical Hierarchy Process Integrated with Geographic Information Systems for the Algiers Watershed. Sustainability, 17(7), 3242. https://doi.org/10.3390/su17073242

[6] Putri, A. S., Suhartanto, E., & Andawayanti, U. (2025). Validation of NRECA parameters for Rainfall- to-Discharge modeling in the Rejoso Watershed. Jurnal Penelitian Pendidikan IPA, 11(5), 1081–1088. https://doi.org/10.29303/jppipa.v11i5.11107

[7] Qamar, M. U., Turner, C., & Stooshnoff, C. (2025). Amalgamation of drainage area ratio and nearest neighbors methods for predicting stream flows in British Columbia, Canada. Water, 17(10), 1502. https://doi.org/10.3390/w17101502

[8] Quesada-Román, A., Picado-Monge, A., Rivera-Solís, J., Hernández, M., Torres-Berhard, L., & Ruiz, M. (2025). Flood risk method for scarce-data catchments and municipalities. Revista Geológica De América Central, 72. https://doi.org/10.15517/rgac.2025.64923

[9] Weldegebriel, L., Kluger, D. M., & Lobell, D. (2025). Evaluating restoration success: long-term impact of sustainable land management practices in Ethiopia using synthetic control with matrix completion method. Environmental Research Letters. https://doi.org/10.1088/1748-9326/adea89

[10] Yoo, S., & Ahn, K. (2025). Exploring event-based calibration approaches for semi-distributed urban hydrologic models in data scarce urban watersheds under climate change. Journal of Hydrology, 133498. https://doi.org/10.1016/j.jhydrol.2025.133498