Functional array programming is a style of programming that enables massive parallelism through use of combinators (such as map and reduce) that apply functions to whole arrays. These can be readily parallelised when the functions these combinators are applied to are pure and, in some cases, also associative. We introduce reversible variants of well-known array combinators and show how these can be implemented in parallel using only reversible operations and without accumulating garbage. We introduce a simple reversible functional array programming language, Agni, and show some examples of use.