月降水是气候系统中最重要的变量之一，提前数月的月降水预测对水资源管理、农业生产、旅游业、防灾减灾工程建设及其它对气候敏感的社会活动具有重要影响。目前，相对于温度预测研究，月降水预测的研究依然有待提高。关于月降水预测的研究普遍认为，月降水的可预测性非常有限，月降水是否可预测仍然是一个非常具有争议的科学问题。

本论文利用全球站点月观测资料（月平均温度、月降水）作为预报因子，在提前1-24个月的预报时效上，验证了全球站点月降水的可预测性，并进一步对典型柯本气候区月降水的可预测性进行了分析。此外，本论文采用两因子预报残差最小逐步回归预测模型，对预报量站点的月降水进行尝试预测，并将预测模型推广到全球范围。主要结论如下：

（1）   基于全球站点月观测资料作为预报因子的全球预报量站点月降水可预测性分析，揭示了全球站点的月降水具有可预测性这一特征。对于GHCN-M数据库中全球范围内的8000多个有效组对的预报量站点中的每一个站点，使用GHCN-M数据库中所有的月降水站点和月平均温度站点作为预报因子，在提前1-24个月份上分析了预报量站点每个月份月降水的可预测性。结果表明对于任意一个预报量站点的任意月份，在提前1-24个月份的预报时效上都存在数十—数百个预报因子站点，其月降水与预报量站点月降水之间存在显著的强相关（R² > 0.35，P < 0.01）。这说明，对于世界范围内的预报量站点，其月降水是可以预测的。同时，当使用月降水作为预报因子时，预报因子与预报量之间多呈正相关；当使用月平均温度作为预报因子时，预报因子与预报量之间正、负相关各占约50%。

（2）   采用柯本气候分类方案，基于CRU的高分辨格点气象资料对世界范围内的气候区进行了划分，并据此探讨了典型柯本气候区月降水的可预测性。从柯本气候分区来看，在全球五个主要气候带中（热带多雨气候带（A）、干带（B）、暖温带（C）、冷温带（D）、极地带（E）），B带的可预测性最好（R²约为0.4011），其次是A带（R²约为0.3614），C带（R²约为0.3546）、D带（R²约为0.3536）相差不大，最小的是E带（R²约为0.3417）。从全年所有月份来看，12月份的月降水可预测性最好（R²约为0.3758），冬季（按北半球季节划分）月降水的可预测性优于夏季。同时，对区域柯本气候区月降水可预测性分析结果表明，很可能存在某个预报因子区域，该区域内的月降水与相应的柯本气候区月降水之间存在显著的联系。

Monthly precipitation is one of the most important variables of climate. Water resources management, agricultural activity, tourism, disaster prevention and reduce projection, and other social activity sensitive to the climate heavily relies on the monthly precipitation prediction with the lead time from months to years. Until now, the study on precipitation prediction is still rare relative to temperature forecasting studies. Studies on monthly precipitation prediction generally believe that the predictability of monthly precipitation is little, and whether monthly rainfall is predictable remains a very controversial scientific issue.

Based on the global site monthly observation data (monthly mean temperature, monthly total precipitation) as the predictor to analysis the predictability of monthly total precipitation of the global site with the lead time from 1 to 24 months. Furthermore, the predictability of monthly total precipitation in the typical K?ppen climate types was analyzed. Besides, this study employs a two-factor stepwise optimal linear regression predicting model to try to predict the monthly total precipitation of the forecasted sites, using two predictors and to extend the predicting model to a global scale. The main conclusions as follows:

(1)      Predictability analysis of predictand sites based on monthly precipitation observations of the global site reveals that monthly precipitation of predictand sites are predictable. For each month's total precipitation of each site of the more than 8,000 total predictand sites worldwide in the GHCN-M v2 database, all monthly precipitation sites and monthly mean temperature sites in the GHCN-M database were used as predictors, which predictability was analyzed in advance from 1 to 24 months. The results show that there is a strong correlation between the predictor site and predictand site (R² > 0.35), dozens of predictor sites have the potential for becoming the predictor for each predictand site, by month, at each lead time from 1 to 24 months. Not surprisingly, our results show that the global monthly precipitation is predictable, for an arbitrary month of the year at an arbitrary lead time, when we examine the evidence for predictors around the world. In addition, the correlation is a trend to positive when using monthly precipitation as the predictor and the positive and negative correlation each account for approximately 50% when using monthly average temperature as the predictor.

(2)     The predictability of monthly precipitation in the typical K?ppen climate types is discussed, CRU-based high-resolution meteorological data is used to classify the climate types around the world, using the K?ppen climate classification scheme. In the five major climatic zones of the world (Tropical rainy climates (A), Dry climates (B), Temperate climates (C), Cold continental climates (D), Polar climates (E)), the predictability of B is best (R2 is about 0.4011), followed by A (R² is about 0.3614), C (R² is about 0.3546) and D (R² is about 0.3536), the smallest is E (R² is about 0.3417). From the perspective of the whole year, the precipitation in December is the best predictable (R² is about 0.3758), and the predictability of precipitation in winter (northern hemisphere) is slightly high than that in summer (northern hemisphere). In addition, the predictive analysis of the monthly precipitation in the regional K?ppen climate zone shows that there possibly exist a predictor hot region, and there is a significant relationship between the monthly precipitation in the predictor hot region and the monthly precipitation in the regional K?ppen climate zone.

(3)     This paper tries to predict the monthly precipitation of the global forecasting site using a two-factor stepwise optimal linear regression forecasting model, and a comprehensive forecast equation is established for each month of each predictand site. The results show that there is a good agreement between the prediction and the observation of the predictand site on the global scale (the average R² is about 0.6). On the global predictand site average, the mean absolute error (MAE) between the prediction and observation is about 26 mm, and the relative error (MAPE) is about 55%. On a regional scale, the forecasting model has a high skill(R² above 0.6) in the Indian region, the eastern and western parts of Australia, the southern part of Africa and the Sahel region, the northeastern part of South America, and the central part of North America and other areas are R² from 0.5 to 0.6. In the different months of the same region, the MAE and MAPE of the prediction model have slightly different. The larger MAE in India is mainly in June, July, August, and September, and the MAPE also increases, which may be mainly due to the monsoon system. The complexity of the internal monsoon system and monsoon systems has led to greater difficulty in predicting monthly precipitation during the rainy season in India. In Africa, with seasonal change with larger monthly rainfall, the relative error MAPE also has a seasonal change. The MAPE is smaller when the monthly precipitation in Africa is larger, and vice versa. In addition, in the 5-fold cross-validation, the forecasting model on each verification data set can maintain the same predictability on all data sets, and there is no significant difference in the performance of the forecasting model in different verification data sets, indicating that the forecasting model has good robustness.