Water availability has long been a crucial determining factor of ecosystem dynamics, specifically vegetation photosynthesis (GPP) and ecosystem respiration activities (RECO) of above and below-ground components. Nevertheless, during the last 20 years, water availability has not been included in the FLUXNET GPP-RECO nighttime partitioning method, resulting in its inability to effectively estimate GPP and RECO from water-limited sites such as US-Ton (R-squared for half-hourly RECO estimation is 0.210) and US-AR1 (0.267). The nighttime partitioning method only uses temperature as the predictor for RECO, therefore, there is an urgent need to examine whether adding other climatic factors besides temperature could improve the model performance. In this study, Evaporative Fraction (EF) and Soil Water Content (SWC) were tested as the second predictor for RECO under various forms of possible equations (reciprocal, quadratic, cubic functions) that mimic the RECO-water availability relationship. We used non-linear regression with gradient descent algorithm to estimate parameters for each running window, then depending on whether the parameter is time-dependent or not, we decided to linear-interpolate it or take the parameter value with the lowest RSE. The results show that over 53 studied sites, EF better represents the relationship between RECO and water availability than SWC. In addition, not all flux towers have SWC measurement while in all towers, EF can be calculated as the ratio between latent heat and total land surface energy. Compared to the FLUXNET nighttime method, the model with the reciprocal function form of EF yields a higher R-squared and lower RMSE in grassland, woody savanna, and shrubland areas, as well as in growing seasons of most sites. This indicates that EF could capture the variability of RECO caused by water limitation and improve the respiration model performance. Nevertheless, the reciprocal function of EF still has not been able to capture most abrupt changes in RECO values due to rain patterns; hence, there is a lot of room for improvement of the current partitioning algorithm.