Subseasonal-to-seasonal (S2S) prediction of winter wheat yields is crucial for farmers and decision-makers to reduce yield losses and ensure food security. Recently, numerous researchers have utilized machine learning (ML) methods to predict crop yield, using observational climate variables and satellite data. Meanwhile, some studies also illustrated the potential of state-of-the-art dynamical atmospheric prediction in crop yield forecasting. However, the potential of coupling both methods has not been fully explored. Herein, we aimed to establish a skilled ML–dynamical hybrid model for crop yield forecasting (MHCF v1.0), which hybridizes ML and a global dynamical atmospheric prediction system, and applied it to northern China at the S2S time scale. In this study, we adopted three mainstream machining learning algorithms (XGBoost, RF, and SVR) and the multiple linear regression (MLR) model, and three major datasets, including satellite data from MOD13C1, observational climate data from CRU, and S2S atmospheric prediction data from IAP CAS, used to predict winter wheat yield from 2005 to 2014, at the grid level. We found that, among the four models examined in this work, XGBoost reached the highest skill with the S2S prediction as inputs, scoring R² of 0.85 and RMSE of 0.78 t/ha 3–4 months, leading the winter wheat harvest. Moreover, the results demonstrated that crop yield forecasting with S2S dynamical predictions generally outperforms that with observational climate data. Our findings highlighted that the coupling of ML and S2S dynamical atmospheric prediction provided a useful tool for yield forecasting, which could guide agricultural practices, policy-making and agricultural insurance.

We evaluate the ability of the latest generation atmospheric general circulation model from State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics, Institute of Atmospheric Physics, Chinese Academy of Sciences (namely, FAMIL2) in simulating some key characteristics (genesis location, track, number, and intensity) of tropical cyclones (TCs) in terms of their climatology and seasonal to interannual variability. A standard 1° × 1° atmospheric model intercomparison project experiment is carried out for the period 1979–2002, and the last 20 years of outputs are used for analysis. The same period from International Best Track Archive for Climate Stewardship (IBTrACS) is used as the observation for comparison purposes. The evaluations focus on TC activity at the global scale as well as in the three key regions of the northern Indian Ocean (NIO), western Pacific (WP) and northern Atlantic (NA). With respect to the simulated TC climatology, FAMIL2 shows notable ability in correctly reproducing the main characteristics of the genesis locations, tracks, and numbers of TC, particularly over the key regions of TC activity in the Northern Hemisphere; whereas, it underestimates the intensities of TC, as is the case with many state‐of‐the‐art climate models operating at a medium resolution. On seasonal‐to‐interannual time scales, meanwhile, FAMIL2 successfully reproduces the seasonal cycles of TC numbers over the NIO and WP regions, the former being characterized by double TC peaks (in May and October) and the latter by a maximum peak season in August. However, the model only captures these features approximately. For the simulated interannual variability of TC activity, the correlation coefficients of 20‐year TC numbers between FAMIL2 and IBTrACS are 0.22, 0.51 (95% confidence interval), and 0.49 (95% confidence interval) for the NIO, WP, and NA, respectively. We also examine the possible reasons behind the performance of FAMIL2 by investigating its subseasonal signs related to the Madden‐Julian Oscillation (MJO) and convectively coupled equatorial waves. The TC genesis potential index is employed to investigate the possible impacts of the large‐scale dynamic fields on the simulation of TC activity. Finally, the biases of simulated TC activity, as well as possible solutions for these biases, are discussed with respect to the horizontal resolution of the model. A TC forecasting case study is introduced as a first step in applying FAMIL2 to a TC forecasting system.

The effects of horizontal resolution on the simulation of tropical cyclones were studied using the Chinese Academy of Sciences Flexible Global Ocean–Atmosphere–Land System Finite-Volume version 3 (FGOALS-f3) climate system model from the High-Resolution Model Intercomparison Project (HighResMIP) for the Coupled Model Intercomparison Project phase 6 (CMIP6). Both the low-resolution (about 100 km resolution) FGOALS-f3 model (FGOALS-f3-L) and the high-resolution (about 25 km resolution) FGOALS-f3 (FGOALS-f3-H) models were used to achieve the standard Tier 1 experiment required by HighResMIP. FGOALS-f3-L and FGOALS-f3-H have the same model parameterizations with the exactly the same parameters. The only differences between the two models are the horizontal resolution and the time step. The performance of FGOALS-f3-H and FGOALS-f3-L in simulating tropical cyclones was evaluated using observations. FGOALS-f3-H (25 km resolution) simulated more realistic distributions of the formation, movement and intensity of the climatology of tropical cyclones than FGOALS-f3-L at 100 km resolution. Although the number of tropical cyclones increased by about 50 % at the higher resolution and better matched the observed values in the peak month, both FGOALS-f3-L and FGOALS-f3-H appear to replicate the timing of the seasonal cycle of tropical cyclones. The simulated average and interannual variabilities of the number of tropical cyclones and the accumulated cyclone energy were both significantly improved from FGOALS-f3-L to FGOALS-f3-H over most of the ocean basins. The characteristics of tropical cyclones (e.g., the average lifetime, the wind–pressure relationship and the horizontal structure) were more realistic in the simulation using the high-resolution model. The possible physical linkage between the performance of the tropical cyclone simulation and the horizontal resolution were revealed by further analyses. The improvement in the response between the El Niño–Southern Oscillation and the number of tropical cyclones and the accumulated cyclone energy in FGOALS-f3 contributed to the realistic simulation of tropical cyclones. The genesis potential index and the vorticity, relative humidity, maximum potential intensity and the wind shear terms were used to diagnose the effects of resolution. We discuss the current insufficiencies and future directions of improvement for the simulation of tropical cyclones and the potential applications of the FGOALS-f3-H model in the subseasonal to seasonal prediction of tropical cyclones.

Subseasonal-to-seasonal (S2S) prediction of winter wheat yields is crucial for farmers and decision-makers to reduce yield losses and ensure food security. Recently, numerous researchers have utilized machine learning (ML) methods to predict crop yield using observational climate variables and satellite data. Meanwhile, some studies also illustrate the potential of state-of-the-art dynamical atmospheric prediction in crop yield forecasting. However, the potential of coupling both methods has not been fully exploited. Herein, we aim to establish a skilled ML–dynamical hybrid model for crop yield forecasting (MHCF v1.0), which hybridizes ML and a global dynamical atmospheric prediction system, and apply it to northern China at the S2S time scale. MHCF v1.0 demonstrates that crop yield forecasting with S2S dynamical predictions generally outperforms that with observational climate data. The coupling of ML and S2S dynamical atmospheric prediction provides a useful tool for yield forecasting, which could guide agricultural practices, policy-making and agricultural insurance.

The critical relative humidity (RH_c), which approximately measures the subgrid‐scale variability of moisture, is important to cloud parameterization. Based on the diagnostics from CloudSat/CALIPSO satellite data, we propose an improved RH_c formula that incorporates geographic dependence and allows for non‐monotonic variations in the vertical. With the parameterized RH_c, a cloud macrophysics scheme is constructed in which fractional cloudiness and subgrid‐scale condensation are synergistically solved, with the latter being calculated using two different approaches. Results show the new scheme largely alleviates the underestimation of high‐ and mid‐level clouds in the default model. The performance is also superior to the simulations applying a globally uniform RH_c as conventionally used in the literature. Varying RH_c and the techniques for computing subgrid‐scale condensation leads to a marked diversity in cloud feedback, which nearly replicates the range of uncertainty found in Coupled Model Intercomparison Project Phase 6 models. Using smaller RH_c leads to larger spread than using larger ones. And the spread caused by different techniques for calculating subgrid‐scale condensation is larger than that caused by the choice of RH_c. While many previous studies have emphasized the diversity of cloud feedback due to low clouds, varying RH_c and its implementation leads to diversity of cloud feedback being mainly due to optically thick clouds. These results highlight the importance of RH_c in inducing uncertainty of cloud feedback, as well as notes of caution to modelers when using RH_c to tune models.

Large-ensemble simulations of the atmosphere-only time-slice experiments for the Polar Amplification Model Intercomparison Project (PAMIP) were carried out by the model group of the Chinese Academy of Sciences (CAS) Flexible Global Ocean-Atmosphere-Land System (FGOALS-f3-L). Eight groups of experiments forced by different combinations of the sea surface temperature (SST) and sea ice concentration (SIC) for pre-industrial, present-day, and future conditions were performed and published. The time-lag method was used to generate the 100 ensemble members, with each member integrating from 1 April 2000 to 30 June 2001 and the first two months as the spin-up period. The basic model responses of the surface air temperature (SAT) and precipitation were documented. The results indicate that Arctic amplification is mainly caused by Arctic SIC forcing changes. The SAT responses to the Arctic SIC decrease alone show an obvious increase over high latitudes, which is similar to the results from the combined forcing of SST and SIC. However, the change in global precipitation is dominated by the changes in the global SST rather than SIC, partly because tropical precipitation is mainly driven by local SST changes. The uncertainty of the model responses was also investigated through the analysis of the large-ensemble members. The relative roles of SST and SIC, together with their combined influence on Arctic amplification, are also discussed. All of these model datasets will contribute to PAMIP multi-model analysis and improve the understanding of polar amplification.