High utility co-location pattern (HUCP) mining refers to discovering a set of spatial features from a spatial dataset whose instances are not only close in space but also have high utility. The degree of utility of a pattern is computed by the sum of the utility values of all features participating in the pattern named participation utility ratio (PUR). Since HUCPs do not consider the length of patterns, this results in the utility values of long-size patterns being more likely to be larger than that of the short patterns, the mining result normally contains more long-size patterns while valuable short-size patterns may be neglected. To address this problem, this work considers the length of patterns in computing PUR, i.e., average utility ratio (AUR), and focuses on mining high average utility co-location patterns (HAUCPs). Since AUR does not hold the downward-closure property, unnecessary candidates cannot be effectively pruned in advance, so a dynamic upper bound is designed to prune candidates early. Moreover, a hierarchical instance tree is also designed to collect co-location instance sets of candidates efficiently. The proposed method algorithm is subjected to extensive experiments on both synthetic and real data sets simultaneously. Experimental results show that the proposed algorithm outperforms the existing algorithms in mining HUCPs and HAUCPs.