In　environmental　monitoring　applications,　selecting　appropriate　locations　to　sense　is　important　relating　to　data　quality　and　Sensing　cost.　This　paper　addresses　the　challenge　by　collecting　data　from　a　subset　of　locations,　then　leveraging　the　spatial　and　cross-domain　correlations　to　deduce　data　of　other　locations,　thus　can　obtain　acceptable　data　quality　with　lower　sensing　cost.　Referring　to　active　learning,　the　proposed　framework　is　constructed　by　two　types　modules　(i.e.,　estimators　and　selectors)　and　a　cyclic　process　of　estimating　and　selecting.　Estimators　based　on　kriging　interpolation　and　regression　tree　are　implemented,　and　their　corresponding　selectors　are　designed.　We　evaluate　the　effectiveness　of　the　framework　by　taking　air　quality　sensing　as　an　example.　Results　show　that　to　reach　data　quality　of　about　25%　MAPE,　the　framework　only　needs　15%　locations,　while　random　selector　needs　25%　locations.　©　2019　IEEE.