How to do a Difference Query in SPARQL?

RDF defines that data is not duplicated in an RDF set, so your triplestore should ignore any duplicate values if you attempt to insert them into the graph when they are already present. In that case you can do a simple INSERT/WHERE:

INSERT {
graph <http://sparql.tw.rpi.edu/source/dataset/graphB> {
   ?measurement
   pol:hasSite ?Site ;
   time:inXSDDateTime ?TimeStamp ;
   pol:hasUSGSPCode ?PCode ;
   pol:hasValue ?Value ;
   repr:hasUnit ?Unit .
 }
} WHERE {
   graph <http://sparql.tw.rpi.edu/source/dataset/graphA> {
    ?measurement
    pol:hasSite ?Site ;
    time:inXSDDateTime ?TimeStamp ;
    pol:hasUSGSPCode ?PCode ;
    pol:hasValue ?Value ;
    repr:hasUnit ?Unit .
 }

}

Or, even more simply, if you want add all of the triples from GraphA to GraphB that did not exist in GraphB to begin with then do a SPARQL ADD operation

ADD <http://sparql.tw.rpi.edu/source/dataset/graphA> TO <http://sparql.tw.rpi.edu/source/dataset/graphB>

If you are not actually attempting to modify GraphA but just calculate the difference between the two via a query, then consider the MINUS operator:

SELECT  *
WHERE {
  graph <http://sparql.tw.rpi.edu/source/dataset/graphA> {
    ?measurement
    pol:hasSite ?Site ;
    time:inXSDDateTime ?TimeStamp ;
    pol:hasUSGSPCode ?PCode ;
    pol:hasValue ?Value ;
    repr:hasUnit ?Unit .
  }
MINUS {  graph <http://sparql.tw.rpi.edu/source/dataset/graphB> {
    ?measurement
    pol:hasSite ?Site ;
    time:inXSDDateTime ?TimeStamp ;
    pol:hasUSGSPCode ?PCode ;
    pol:hasValue ?Value ;
    repr:hasUnit ?Unit .
  }
}
}