To address the problem that existing spatiotemporal data fusion methods struggle to generate high-quality results under conditions of rapid land cover change

image registration errors

and limited cloudless observation data in cloud-covered areas

a cross-scale remote sensing image spatiotemporal fusion network based on multi-source sensors is proposed.This method utilizes multi-level and multi-scale features to fuse images from multiple sensors with different resolutions

thereby overcoming the influence of geometric registration errors.Furthermore

a fine-tuning strategy and multi-constraint loss function are introduced to effectively utilize cloud-contaminated images

achieving high-precision fusion and data gap filling simultaneously.Experimental results show that in the significant change fusion task on the Landsat—7 image dataset

the proposed network achieves a relative mean absolate error(rMAE) of 6.8%

which is significantly superior to mainstream methods.In conclusion

the proposed network can effectively tackle key challenges such as rapid land cover change

inaccurate image registration

and scarcity of cloudless data

generating high-quality spatiotemporal fusion results.