Differential programming is a technique that enables the automatic computation of derivatives of the output of models with respect to model parameters. Applying these techniques to Earth system modeling leverages the increasing availability of high-quality data to improve the models themselves. This can be done by either using calibration techniques that use gradient-based optimization or incorporating machine learning methods that can learn previously unresolved influences directly from data.

Water and pesticide transfer models are complex and should be simplified to be used in decision support. Indeed, these models simulate many spatial processes in interaction, involving a large number of parameters. Sensitivity analysis allows us to select the most influential input parameters, but it has to be adapted to spatial modelling. This study will identify relevant methods that can be transposed to any hydrological and water quality model and improve the fate of pesticide knowledge.

Regional-level applications of dynamic vegetation models are challenging because they need to accommodate the variation in plant functional diversity. This can be done by estimating parameters from available plant trait databases while adopting alternative solutions for missing data. Here we present the design, parameterization and evaluation of MEDFATE (version 2.9.3), a novel model of forest dynamics for its application over a region in the western Mediterranean Basin.

Ice sheets gain mass via snowfall. However, snowfall is redistributed by the wind, resulting in accumulation differences of up to a factor of 5 over distances as short as 5 km. These differences complicate estimates of ice sheet contribution to sea level rise. For this reason, we have developed a new model for estimating wind-driven snow redistribution on ice sheets. We show that, over Pine Island Glacier in West Antarctica, the model improves estimates of snow accumulation variability.

Due to the increasing availability of high-performance computing over the past few decades, numerical models have become an important tool for research. Here we test two geodynamic codes that produce such models: ASPECT, a newer code, and CitcomS, an older one. We show that they produce solutions that are extremely close. As methods and codes become more complex over time, showing reproducibility allows us to seamlessly link previously known information to modern methodologies.

Climate and weather models suffer from uncertainties resulting from approximated processes. Solar and thermal radiation is one example, as it is computationally too costly to simulate precisely. This has led to attempts to replace radiation codes based on physical equations with neural networks (NNs) that are faster but uncertain. In this paper we use global weather simulations to demonstrate that a middle-ground approach of using NNs only to predict optical properties is accurate and reliable.

The National Centers for Environmental Prediction Global Forecast System version 16 experienced model instability failures in real-time runs resolved by increasing the minimum thickness depth parameter. Further investigation revealed that the issue was caused by the advection of geopotential heights at the model's layer interfaces. By replacing high-order boundary conditions with zero-gradient boundary conditions for interface-wind reconstruction, the instability was effectively addressed.

Simultaneously simulating food production and the requirements and availability of water resources in a spatially explicit manner within a single framework remains challenging on a global scale. Here, we successfully enhanced the global hydrological model H08 that considers human water use and management to simulate the yields of four major staple crops: maize, wheat, rice, and soybean. Our improved model will be beneficial for advancing global food–water nexus studies in the future.

A novel subgrid channel (SGC) model is developed for river–floodplain modelling, allowing utilization of subgrid-scale bathymetric information while performing computations on relatively coarse grids. By including adaptive artificial diffusion, potential numerical instability, which the original SGC solver had, in low-friction regions such as urban areas is addressed. Evaluation of the new SGC model through structured tests confirmed that the accuracy and stability have improved.

This paper describes a newly developed submodel ACCF V1.0 based on the MESSy 2.53.0 infrastructure. The ACCF V1.0 is based on the prototype algorithmic climate change functions (aCCFs) v1.0 to enable climate-optimized flight trajectories. One highlight of this paper is that we describe a consistent full set of aCCFs formulas with respect to fuel scenario and metrics. We demonstrate the usage of the ACCF submodel using AirTraf V2.0 to optimize trajectories for cost and climate impact.

Effects of sea ice are typically neglected in operational flood forecast systems. In this work, we capture these effects via the addition of a parameterized ice–ocean stress. The parameterization takes advantage of forecast fields from an advanced ice–ocean model and features a novel, consistent representation of the tidal relative ice–ocean velocity. The new parameterization leads to improved forecasts of tides and storm surges in polar regions. Associated physical processes are discussed.

Simultaneous modelling of ocean, sea ice, and atmosphere in coupled models is critical for understanding all of the processes that happen in the Antarctic. Here we have developed a coupled model for the Ross Sea, P-SKRIPS, that conserves heat and mass between the ocean and sea ice model (MITgcm) and the atmosphere model (PWRF). We have shown that our developments reduce the model drift, which is important for long-term simulations. P-SKRIPS shows good results in modelling coastal polynyas.

We present a new toolbox for generating input datasets for terrestrial ecosystem models from diverse and partially conflicting data sources. The toolbox documents the sources and processing of data and is designed to make inconsistencies between source datasets transparent so that users can make their own decisions on how to resolve these should they not be content with our default assumptions. As an example, we use the toolbox to create input datasets at two different spatial resolutions.