Exploring ordination as a method for normalizing disparate datasets: implications for digital change detection