Abstract: With the increasing frequency and scale of extreme weather events in recent years, it is particularly important to utilize advanced technologies to enhance the accuracy and efficiency of mapping and geographic information systems, so as to improve the resilience of power systems in outdoor open-air conventional transmission and substation substations. In this paper, a multi-source spatial data fusion method based on self-attentive transform network is proposed for enhancing the monitoring and prediction capability of changes in complex geographic environments and natural resources. The method represents continuous weather radar echo data as a spatio-temporal sequence, captures long term time dependencies using a self-attentive mechanism encoder, and integrates multi-scale convolution to extract short term time dependencies. In addition, a graph attention network is introduced into the transformation network to deeply analyze the relationship between different spatial variables. To verify the effectiveness of the model, this paper conducts experiments based on precipitation data and remote sensing data from 2019 to 2022 in a region, and compares it with traditional convolutional neural networks, benchmark U-Net and long-short term memory networks. The experimental results show that the method is superior to the traditional model in terms of measurement accuracy and data fusion ability, especially in the fields of GIS construction, engineering survey, mine survey, cadastral mapping and ocean survey and mapping, etc., which has a wide application potential.

Key words: radar echo extrapolation, precipitation forecast, long-short term memory, attention