Convergence analysis of an incremental approach to online inverse reinforcement learning

Interest in inverse reinforcement learning (IRL) has recently increased,that is,interest in the problem of recovering the reward function underlying a Markov decision process (MDP) given the dynamics of the system and the behavior of an expert.This paper deals with an incremental approach to online IRL.First,the convergence property of the incremental method for the IRL problem was investigated,and the bounds of both the mistake number during the learning process and regret were provided by using a detailed proof.Then an online algorithm based on incremental error correcting was derived to deal with the IRL problem.The key idea is to add an increment to the current reward estimate each time an action mismatch occurs.This leads to an estimate that approaches a target optimal value.The proposed method was tested in a driving simulation experiment and found to be able to efficiently recover an adequate reward function.

A STATISTICAL PREDICTION MODEL FOR SPRING RAINFALL OVER NORTHERN CHINA BASED ON THE INTERANNUAL INCREMENT APPROACH

The spring (March-April-May) rainfall over northern China (SPRNC) is predicted by using the interannual increment approach. DY denotes the difference between the current year and previous years. The seasonal forecast model for the DY of SPRNC is constructed based on the data that are taken from the 1965-2002 period (38 years), in which six predictors are available no later than the current month of February. This is favorable so that the seasonal forecasts can be made one month ahead. Then, SPRNC and the percentage anomaly of SPRNC are obtained by the predicted DY of SPRNC. The model performs well in the prediction of the inter-annual variation of the DY of SPRNC during 1965-2002, with a correlation coefficient between the predicted and observed DY of SPRNC of 0.87. This accounts for 76% of the total variance, with a low value for the average root mean square error (RMSE) of 20%. Both the results of the hindcast for the period of 2003-2010 (eight years) and the cross-validation test for the period of 1965-2009 (45 years) illustrate the good prediction capability of the model, with a small mean relative error of 10%, an RMSE of 17% and a high rate of coherence of 87.5% for the hindcasts of the percentage anomaly of SPRNC.

A skillful method for precipitation prediction over eastern China

Precipitation prediction is essential for disaster prevention,yet it still remains a challenging issue in weather and climate studies.This paper proposes an effective prediction method for summer precipitation over eastern China(PEC) by combining empirical orthogonal function(EOF) analysis with the interannual increment approach.Three statistical prediction models are individually developed for respective predictions of the three principal components(PCs) corresponding to the three leading EOF modes for the interannual increment of PEC(hereafter DY;EC).Each model is run for the month of March with two previous predictors derived from sea-ice concentration/soil moisture/sea surface temperature/snow depth/sea level pressure over specific regions.The predicted PCs are projected to the EOF modes derived from observations of DY;EC to produce a new DY;EC.This new DY;EC is then added to the observed PEC of the previous year to obtain the final predicted PEC.The spatial features of the predicted PEC are highly consistent with observations,with the anomaly correlation coefficient skill ranging from 0.32 to 0.64 during 2012-2020.The method is applied for real-time prediction of PEC in 2021.And the results indicate two rain belts located over northeastern China and the Yangtze-Huaihe River valley,respectively,although the chance for the occurrence of a "super" mei-yu with a similar intensity to that in 2020 would be rare in 2021.

Has the Prediction of the South China Sea Summer Monsoon Improved Since the Late 1970s?

Based on the evaluation of state-of-the-art coupled ocean-atmosphere general circulation models （CGCMs） from the ENSEMBLES （Ensemble-based Predictions of Climate Changes and Their Impacts） and DEME- TER （Development of a European Multimodel Ensemble System for Seasonal to Interannual Prediction） projects, it is found that the prediction of the South China Sea summer monsoon （SCSSM） has improved since the late 1970s. These CGCMs show better skills in prediction of the atmospheric circulation and precipitation within the SCSSM domain during 1979-2005 than that during 1960-1978. Possible reasons for this improvement are investigated. First, the relationship between the SSTs over the tropical Pacific, North Pacific and tropical Indian Ocean, and SCSSM has intensified since the late 1970s. Meanwhile, the SCSSM-related SSTs, with their larger amplitude of interannual variability, have been better predicted. Moreover, the larger amplitude of the interannual variability of the SCSSM and improved initializations for CGCMs after the late 1970s contribute to the better prediction of the SCSSM. In addition, considering that the CGCMs have certain limitations in SCSSM rainfall prediction, we applied the year-to-year increment approach to these CGCMs from the DEMETER and ENSEMBLES projects to improve the prediction of SCSSM rainfall before and after the late 1970s.