Precipitation process is generally considered to be poorly represented in numerical weather/climate models. Statistical downscaling (SD) methods, which relate precipitation with model resolved dynamics, often provide more accurate precipitation estimates compared to model's raw precipitation products. We introduce the convolutional neural network model to foster this aspect of SD for daily precipitation prediction. Specifically, we restrict the predictors to the variables that are directly resolved by discretizing the atmospheric dynamics equations. In this sense, our model works as an alternative to the existing precipitation‐related parameterization schemes for numerical precipitation estimation. We train the model to learn precipitation‐related dynamical features from the surrounding dynamical fields by optimizing a hierarchical set of spatial convolution kernels. We test the model at 14 geogrid points across the contiguous United States. Results show that provided with enough data, precipitation estimates from the convolutional neural network model outperform the reanalysis precipitation products, as well as SD products using linear regression, nearest neighbor, random forest, or fully connected deep neural network. Evaluation for the test set suggests that the improvements can be seamlessly transferred to numerical weather modeling for improving precipitation prediction. Based on the default network, we examine the impact of the network architectures on model performance. Also, we offer simple visualization and analyzing approaches to interpret the models and their results. Our study contributes to the following two aspects: First, we offer a novel approach to enhance numerical precipitation estimation; second, the proposed model provides important implications for improving precipitation‐related parameterization schemes using a data‐driven approach.

Precipitation downscaling is widely employed for enhancing the resolution and accuracy of precipitation products from general circulation models (GCMs). In this study, we propose a novel statistical downscaling method to foster GCMs’ precipitation prediction resolution and accuracy for the monsoon region. We develop a deep neural network composed of a convolution and Long Short Term Memory (LSTM) recurrent module to estimate precipitation based on well-resolved atmospheric dynamical fields. The proposed model is compared against the GCM precipitation product and classical downscaling methods in the Xiangjiang River Basin in South China. Results show considerable improvement compared to the European Centre for Medium-Range Weather Forecasts (ECMWF)-Interim reanalysis precipitation. Also, the model outperforms benchmark downscaling approaches, including (1) quantile mapping, (2) the support vector machine, and (3) the convolutional neural network. To test the robustness of the model and its applicability in practical forecasting, we apply the trained network for precipitation prediction forced by retrospective forecasts from the ECMWF model. Compared to the ECMWF precipitation forecast, our model makes better use of the resolved dynamical field for more accurate precipitation prediction at lead times from 1 day up to 2 weeks. This superiority decreases with the forecast lead time, as the GCM’s skill in predicting atmospheric dynamics is diminished by the chaotic effect. Finally, we build a distributed hydrological model and force it with different sources of precipitation inputs. Hydrological simulation forced with the neural network precipitation estimation shows significant advantage over simulation forced with the original ERA-Interim precipitation (with NSE value increases from 0.06 to 0.64), and the performance is only slightly worse than the observed precipitation forced simulation (NSE = 0.82). This further proves the value of the proposed downscaling method, and suggests its potential for hydrological forecasts.

The fidelity of climate projections is often undermined by biases in climate models due to their simplification or misrepresentation of unresolved climate processes. While various bias correction methods have been developed to post‐process model outputs to match observations, existing approaches usually focus on limited, low‐order statistics, or break either the spatiotemporal consistency of the target variable, or its dependency upon model resolved dynamics. We develop a Regularized Adversarial Domain Adaptation (RADA) methodology to overcome these deficiencies, and enhance efficient identification and correction of climate model biases. Instead of pre‐assuming the spatiotemporal characteristics of model biases, we apply discriminative neural networks to distinguish historical climate simulation samples and observation samples. The evidences based on which the discriminative neural networks make distinctions are applied to train the domain adaptation neural networks to bias correct climate simulations. We regularize the domain adaptation neural networks using cycle‐consistent statistical and dynamical constraints. An application to daily precipitation projection over the contiguous United States shows that our methodology can correct all the considered moments of daily precipitation at approximately resolution, ensures spatiotemporal consistency and inter‐field correlations, and can discriminate between different dynamical conditions. Our methodology offers a powerful tool for disentangling model parameterization biases from their interactions with the chaotic evolution of climate dynamics, opening a novel avenue toward big‐data enhanced climate predictions.

The path toward realizing the potential of seasonal forecasting and its socioeconomic benefits relies on improving general circulation model (GCM) based dynamical forecast systems. To improve dynamical seasonal forecasts, it is crucial to set up forecast benchmarks, and clarify forecast limitations posed by model initialization errors, formulation deficiencies, and internal climate variability. With huge costs in generating large forecast ensembles, and limited observations for forecast verification, the seasonal forecast benchmarking and diagnosing task proves challenging. Here, we develop a probabilistic deep learning‐based statistical forecast methodology, drawing on a wealth of climate simulations to enhance seasonal forecast capability and forecast diagnosis. By explicitly modeling the internal climate variability and GCM formulation differences, the proposed Conditional Generative Forecasting (CGF) methodology enables bypassing crucial barriers in dynamical forecast, and offers a top‐down viewpoint to examine how complicated GCMs encode the seasonal predictability information. We apply the CGF methodology for global seasonal forecast of precipitation and 2 m air temperature, based on a unique data set consisting 52,201 years of climate simulation. Results show that the CGF methodology can faithfully represent the seasonal predictability information encoded in GCMs. We successfully apply this learned relationship in real‐world seasonal forecast, achieving competitive performance compared to dynamical forecasts. Using this CGF as benchmark, we reveal the impact of insufficient forecast spread sampling that limits the skill of the considered dynamical forecast system. Finally, we introduce different strategies for composing ensembles using the CGF methodology, highlighting the potential for leveraging the strengths of multiple GCMs to achieve advantgeous seasonal forecast.

A warming climate will intensify the water cycle, resulting in an exacerbation of water resources crises and flooding risks in the Lancang–Mekong River Basin (LMRB). The mitigation of these risks requires accurate streamflow and flood simulations. Process-based and data-driven hydrological models are the two major approaches for streamflow simulations, while a hybrid of these two methods promises advantageous prediction accuracy. In this study, we developed a hybrid physics-data (HPD) methodology for streamflow and flood prediction under the physics-guided neural network modeling framework. The HPD methodology leveraged simulation information from a process-based model (i.e., VIC-CaMa-Flood) along with the meteorological forcing information (precipitation, maximum temperature, minimum temperature, and wind speed) to simulate the daily streamflow series and flood events, using a long short-term memory (LSTM) neural network. This HPD methodology outperformed the pure process-based VIC-CaMa-Flood model or the pure observational data driven LSTM model by a large margin, suggesting the usefulness of introducing physical regularization in data-driven modeling, and the necessity of observation-informed bias correction for process-based models. We further developed a gradient boosting tree method to measure the information contribution from the process-based model simulation and the meteorological forcing data in our HPD methodology. The results show that the process-based model simulation contributes about 30% to the HPD outcome, outweighing the information contribution from each of the meteorological forcing variables (<20%). Our HPD methodology inherited the physical mechanisms of the process-based model, and the high predictability capability of the LSTM model, offering a novel way for making use of incomplete physical understanding, and insufficient data, to enhance streamflow and flood predictions.

Precipitation variability significantly influences the heavily populated West Coast of the United States, raising the need for reliable predictions. We investigate the region’s short- to extended-range precipitation prediction skill using the hindcast database of the Subseasonal-to-Seasonal Prediction Project (S2S). The prediction skill–lead time relationship is evaluated, using both deterministic and probabilistic skill scores. Results show that the S2S models display advantageous deterministic skill at week 1. For week 2, prediction is useful for the best-performing model, with a Pearson correlation coefficient larger than 0.6. Beyond week 2, predictions generally provide little useful deterministic skill. Sources of extended-range predictability are investigated, focusing on El Niño–Southern Oscillation (ENSO) and the Madden–Julian oscillation (MJO). We found that periods of heavy precipitation associated with ENSO are more predictable at the extended range period. During El Niño years, Southern California tends to receive more precipitation in late winter, and most models show better extended-range prediction skill. On the contrary, during La Niña years Oregon tends to receive more precipitation in winter, with most models showing better extended-range skill. We believe the excessive precipitation and improved extended-range prediction skill are caused by the meridional shift of baroclinic systems as modulated by ENSO. Through examining precipitation anomalies conditioned on the MJO, we verified that active MJO events systematically modulate the area’s precipitation distribution. Our results show that most models do not represent the MJO or its associated teleconnections, especially at phases 3–4. However, some models exhibit enhanced extended-range prediction skills under active MJO conditions.