Complete　data　on　wastewater　quality　are　essential　for　managing　and　monitoring　wastewater　treatment　processes.　Most　management　and　monitoring　methods　involve　the　use　of　voluminous　training　data　for　imputation,　but　the　problem　is　that　the　sensors　used　in　wastewater　treatment　plants　(WWTPs)　collect　only　a　limited　amount　of　data.　The　lack　of　sufficient　training　data　can　diminish　the　accuracy　of　traditional　imputation　techniques.　To　address　this　problem,　this　study　developed　a　novel　approach　called　Miss-GBRT　(imputing　missing　values　with　gradient　boosting　regression　trees),　which　can　impute　missing　values　into　wastewater　quality　data　even　with　minimal　training　data.　The　proposed　approach　consists　of　a　preprocessing　stage　and　an　imputation　stage.　In　the　preprocessing　stage,　different　copies　of　masked　datasets　are　produced　from　raw　data　according　to　various　levels　of　missingness,　after　which　pre-imputation　is　conducted　to　ensure　the　integrality　of　training　data.　In　the　imputation　stage,　Miss-GBRT　is　used　to　combine　shallow　regression　trees　to　regress　the　residuals　of　time　and　impute　each　missing　value　into　a　masked　dataset　in　a　stepwise　manner.　We　carried　out　extensive　experiments　on　the　WWTP　datasets　of　the　University　of　California,　Irvine　and　Beijing　Drainage　Group　to　compare　Miss-GBRT　with　baseline　imputation　methods.　The　results　demonstrated　that　the　proposed　approach　improves　the　accuracy　with　which　missing　wastewater　quality　data　are　imputed　under　limited　training　data.　It　can　also　perform　better　than　other　methods　on　datasets　with　considerable　proportions　of　missing　values.