Abstract: Tropics are more fragile to climate change, especially in tropical island. It’s has not been investigated the change of agricultural heat and precipitation resources in future in tropical island like Hainan island, China. Because there are a lot of space biases between the raw CMIP5 data set and the observed values in Hainan island. Daily maximum temperature, minimum temperature and precipitation were obtained from the ground weather stations and the GCMs include FGOALS-g2, GFDL-ESM2G, HadGEM2-ES, MPI-ESM-MR and MRI-CGCM3 in Hainan island and its nearby waters. The observations and the raw GCMs outputs for the historical (1970-1999), RCP2.6, RCP4.5 and RCP8.5 (2020−2099) scenarios were processed and interpolated to a spatial resolution of 0.5°×0.5° as grid cells using the bilinear method. We used both systematic residuals revision methods (corrected value method or ratio method) and multi-mode ensemble averaging methods include the Bayesian model averaging (BMA) method and the equal weight averaging (EW) method in each grid cells to reduce the spatial uncertainty of the raw GCMs in the training and verification period. And then, we used the revised GCMs outputs and the agro-climatic index computing software to analysis the change of agricultural heat and precipitation resources under the scenarios of RCP2.6, RCP4.5 and RCP8.5 in both short-term (2020−2059) and long-term (2060−2099). These sources include annual mean temperature, mean temperature in January, ≥10℃ and ≥20℃ integrated temperature, annual precipitation, precipitation in January and precipitation in ≥20℃ integrated temperature period.The results showed that the correct coefficients of the raw GCMs outputs from both systematic residuals revision and the BMA method all have large spatial differences among the grid cells. The raw GCMs outputs underestimate the daily maximum temperature about 3.55℃, overestimate the daily minimum temperature about 1.19℃ and underestimate the daily precipitation which only 54.35% of the observations. It can effectively reduce the spatial uncertainty of the raw GCMs outputs by the above revision methods. The revised results of the BMA and the EW are similar and both are better than a single GCM for simulate historical climate variables. After comprehensive revision of the BMA in each grid cells, the correlation coefficients of maximum temperature, minimum temperature and precipitation are increased about 0.10, 0.07 and 0.06 respectively, and the root mean square error are reduced about 2.38℃, 1.01℃ and 1.01mm respectively, in the verification period. There are decreased about 3.25℃, 1.13℃ and 25.67mm compared with the average biases of a single GCM and closer to the observed value. In the future, the agricultural heat resources will generally show a gradual increase from the central mountains to the coast in spatial. The high temperature will distribute mainly range from the southern to the western coastal areas. The annual mean temperature will increase evenly in the whole island. The increasing amplitude of mean temperature in January, ≥10℃ and ≥20℃ integrated temperature has different patterns that will decrease from the eastern to the western, from the northern to the southern, and from the central mountains to the coast, respectively. It will increase significantly with the fastest climate trend rate under the RCP8.5, or increase first in short-term and then level off in long-term under the RCP4.5, or relatively flat without increase significantly under the RCP2.6. The precipitation resources are transforming into a pattern of gradually decreasing from the eastern to the western and with no significant trend in temporal. The precipitation variability will increase in the southern and the northern coastal areas, while decrease in the western and the central areas. With climate warming and the changes of precipitation pattern in future, the expansion of suitable crop cultivation areas will face huge challenges to agricultural production. It is necessary to arrange in advance to seek advantages and avoid disadvantages.

Key words: Climate change, GCM, Grid BMA, Heat and precipitation resources, Hainan island