This study was conducted to evaluate and improve model performance for estimating abundance and occurrence of small yellow croaker (Larimichthys polyactis) based on the data collected from fishery independent bottom trawl surveys data in Haizhou Bay in the spring of 2011 and 2013-2016. According to different methods of processing spatial data, five species distribution models were formulated by combining the delta method with structured additive regression framework, they were compared in terms of their performance on fitness, predictive capacity and independdence of residuals for two commonly used response variables, namely, occurrence and abundance. Result showed that models with spatial covariates had significantly better fitting effect and lower residual spatial correlation, and positve model and delta model showed more improvement than binomial model. The AIC of geoadditive models was the lowest in both binomial model and positive model (respectively lower than GAM without the spatial term by 7.60 and 144.90). Varying coef?cient models had the highest R² (0.68) and fitting AUC (0.94). Geoadditive model had the highest AUC (0.793±0.100) and the lowest RMSE (21.65±4.83), indicating that geoadditive model's estimation n the occurrence probability and resource density of L. polyactis were most accurate. Therefore, we predicted the spatial distribution of L. polyactis in Spring using delta-geoadditive model based on FVCOM simulation data. Prediction result reflected the distribution and variations of L. polyactis was mainly distributes in the southern and western coastal areas (34.0°N-34.5°N，121.0°E-121.5°E), decreasing with the water depth increase and varying significantly over the years. This study aimed to provided a scientific basis for the development and protecttion of L. polyactis fishery resources in Haizhou Bay.