estconvs2s.zip

The GRACE and GRACE-FO satellite missions play a critical role in helping us understand changes in water storage, including groundwater levels, which is crucial for managing water resources effectively. However, a gap between the data collected by these two missions poses challenges in making accurate predictions about water storage. To address this gap, we developed a new model called Enhanced Spatiotemporal Convolutional Sequence to Sequence Network (ESTConvS2S). This model leverages advanced deep learning techniques to fill in missing data and improve the accuracy of water storage predictions. Our study focused on Southwest China, a region known for its unique karst topography and diverse climate conditions, making it particularly sensitive to water storage changes. The ESTConvS2S model showed high accuracy in estimating water storage dynamics. We validated the model by comparing its predictions with actual groundwater measurements and observed a strong correlation, underscoring the reliability of the model. Our model not only effectively bridges the data gap between GRACE and GRACE-FO missions but also significantly enhances our ability to estimate groundwater data accurately. This improvement is vital for better water management, especially in regions facing water scarcity or excessive groundwater extraction.

GRACE与GRACE-FO卫星任务在助力我们认知水资源储量变化（涵盖地下水位变化）方面发挥着关键作用，而该认知对于高效开展水资源管理工作至关重要。然而，这两项任务所采集的数据之间存在的间隙，给精准预测水资源储量带来了显著挑战。为填补这一数据间隙，我们研发了一款名为增强型时空卷积序列到序列网络（Enhanced Spatiotemporal Convolutional Sequence to Sequence Network，ESTConvS2S）的新型模型。该模型依托先进的深度学习（Deep Learning）技术，实现缺失数据的有效填补，并提升水资源储量预测的精度。本研究聚焦于中国西南地区——该区域以独特的喀斯特地貌（karst topography）与多样的气候条件著称，对水资源储量变化尤为敏感。ESTConvS2S模型在估算水资源储量动态特征方面展现出了较高的精度。我们通过将模型预测结果与实际地下水位实测值进行对比验证该模型，发现二者呈现出极强的相关性，这充分凸显了模型的可靠性。本模型不仅有效弥合了GRACE与GRACE-FO任务间的数据间隙，还显著增强了我们精准估算地下水位数据的能力。这一改进对于优化水资源管理而言至关重要，尤其是在面临水资源短缺或地下水过度开采的地区。