By generating high-resolution remote sensing images, Spatio-temporal fusion technology provides an efficient and low-cost solution for continuously monitoring dynamic changes in the Earth. With the continuous development of spatiotemporal fusion technology, various methods have been proposed in recent years, yet accurately simulating and predicting spatial details and textures in complex land areas remains challenging. This paper introduces ACLSDF, a spatio-temporal data fusion method suitable for complex land. ACLSDF computes temporal and spatial changes through linear regression and guided filter, respectively, and assigns different spatio-temporal weights to pixels undergoing changes in land cover type using a change index. It then combines these changes with the change index to calculate the final predicted values. In order to evaluate the performance of the ACLSDF method in heterogeneous areas, this paper conducts a comparative analysis experiment between the ACLSDF fusion results and ground-measured spectral data. Results show that ACLSDF achieves a correlation coefficient above 0.93, a root mean square error below 0.07, and structural similarity indices over 0.98 in complex surface areas. The root mean square error compared to in-situ measured spectral data is less than 0.04, second only to the original GF-1 WFV images, confirming the feasibility and efficacy of the proposed method.