Replication provides redundancy so that in case of hardware failure, the service is still available
Backup periodically stores a copy of the data offline in case it needs to be restored
Replication is useful to protect against hardware failures
Backup is to prevent against the case where operator error or application bug or user error wipes out all replicas of a piece of data
Regarding Cassandra, replication with factor 3 means the data will be stored in 3 nodes to prevent single point failure. But when you write something wrong to the nodes and want to get your original data back, backups earlier is here to help.
Replication doesn't mean the data is in memory, instead disk mirroring is often used for replication, check RAID 1
In summary, replication itself provides high availability already while backup is for restoring to a point.