An array database is effective for managing a massive amount of sensor data, and the window aggregate is a popular operator. We propose an efficient window aggregate method over multi-dimensional array data based on incremental computation. We improve five types of aggregates by exploiting different data structures: list for summation and average, heap for maximum and minimum, and balanced binary search tree for percentile. We design and fully implement the proposed method in SciDB using the plugin mechanism. In addition, we evaluate the performance through experiments using the synthetic and JRA-55 meteorological datasets. The results of our experiments on SciDB are consistent with our analytic findings. The proposed method achieves a 17.9x, 12.5x, and 10.2x performance improvement for minimum, summation, and percentile operators, respectively, compared with SciDB built-in operators. These results align with our time-complexity analysis results.