Wireless　underground　sensor　networks　(WUSNs)　consist　of　sensors　that　are　buried　in　and　communicate　through　soil　medium,　while　the　channel　quality　ofWUSNsis　greatly　impacted　by　the　underground　environment,　such　as　soil　moisture　and　composition.　Due　to　the　precipitation　and　harsh　weather,　the　underground　environments　change　frequently,　which　make　wireless　communication　inWUSNsmuch　complicated　than　that　in　terrestrial　over-the-air　wireless　sensor　networks.　To　achieve　reliable　and　energy-efficient　data　gathering　in　dynamicWUSNs,　this　article　proposes　an　optimal　transmission　policy,　where　path　loss　of　sensory　data　transmission,　energy　constraint,　and　network　load　balancing　are　the　factors　to　be　considered.　Specifically,　we　capture　the　effect　of　underground　environments　on　wireless　communications,　and　evaluate　path　probability,　energy　consumption,　and　load　balancing　factor　with　respect　to　reliability　and　efficiency　of　transmission　paths.　The　transmission　topology　can　be　reduced　to　a　multi-objective　and　multi-constrained　optimization　problem　and　solved　through　an　improved　maximum　flow　minimum　cost　algorithm.　By　using　reinforcement　learning,　we　derive　an　adaptive　transmission　policy　for　underground　sensors　to　efficiently　use　their　energy　and　avoid　transmitting　sensory　data　in　unreliable　paths　under　a　dynamic　environment.　Through　simulations　and　comparison　upon　publicly　available　real　data,　our　technique　achieves　more　reliable　wireless　communication　with　significant　reduction　of　packet　loss,　and　enables　more　energy-efficient　data　gathering　than　other　techniques,　especially　when　soil　moisture　varies　frequently.