Integration of Sentinel-1 and Sentinel-2 for temporal identification of aquacultural ponds

Accurate classification of ponds under aquaculture practices (AP) and not under aquaculture practices (NAP) is essential for effective monitoring and sustainable management practices. In this study, we have integrated optical and SAR data alongside key spectral indices, such as the Normalized Difference Turbidity Index (NDTI) and the Normalized Difference Chlorophyll Index (NDCI), to distinguish between AP and NAP ponds across different seasons. The analysis reveals distinct spectral and surface characteristics between the two classes for both optical and SAR modalities. Especially for SAR data, NAP ponds exhibit higher anisotropy ( A ) and lower entropy ( H ), while AP ponds show lower A and higher H , reflecting complex management practices. Optical indices further highlight differences, with higher water clarity and nutrient enrichment in AP ponds. The Random Forest classifier obtained a maximum overall accuracy of 94 % by combining optical and SAR data, significantly outperforming other classifiers. The advantages of combining optical and SAR data, as supported by the t-SNE plots showing enhanced separability between the AP and NAP ponds. Seasonal variability also plays a critical role, with AP pond areas expanding during the monsoon season and contracting in the summer due to maintenance and evaporation. A notable shift in aquaculture practices was observed, with AP ponds covering 4617.47 ha in January and increasing to 4686.73 ha in September, highlighting the influence of seasonal factors. Spatial analysis revealed a dynamic shift in the usage of ponds, with aquaculture practices peaking during favorable monsoon conditions and declining during summer. The integration of multisource data significantly improves classification accuracy and captures the nuanced variability of aquaculture practices. Although environmental conditions, seasonal changes, and pond management practices influence classification performance, the proposed methodology offers a robust, scalable approach to monitoring aquaculture systems.