Machine learning (ML) applications in Earth and environmental sciences (EES) have gained incredible momentum in recent years. However, these ML applications have largely evolved in ‘isolation’ from the mechanistic, process-based modelling (PBM) paradigms, which have historically been the cornerstone of scientific discovery and policy support. In this perspective, we assert that the cultural barriers between the ML and PBM communities limit the potential of ML, and even its ‘hybridization’ with PBM, for EES applications. Fundamental, but often ignored, differences between ML and PBM are discussed as well as their strengths and weaknesses in light of three overarching modelling objectives in EES, (1) nowcasting and prediction, (2) scenario analysis, and (3) diagnostic learning. The paper ponders over a ‘coevolutionary’ approach to model building, shifting away from a borrowing to a co-creation culture, to develop a generation of models that leverage the unique strengths of ML such as scalability to big data and high-dimensional mapping, while remaining faithful to process-based knowledge base and principles of model explainability and interpretability, and therefore, falsifiability.

The task of quantifying spatial and temporal variations in terrestrial water, energy, and vegetation conditions is challenging due to the significant complexity and heterogeneity of these conditions, all of which are impacted by climate change and anthropogenic activities. To address this challenge, Earth Observations (EOs) of the land and their utilization within data assimilation (DA) systems are vital. Satellite EOs are particularly relevant, as they offer quasi-global coverage, are non-intrusive, and provide uniformity, rapid measurements, and continuity. The past three decades have seen unprecedented growth in the number and variety of land remote sensing technologies launched by space agencies and commercial companies around the world. There have also been significant developments in land modeling and DA systems to provide tools that can exploit these measurements. Despite these advances, several important gaps remain in current land DA research and applications. This paper discusses these gaps, particularly in the context of using DA to improve model states for short-term numerical weather and sub-seasonal to seasonal predictions. We outline an agenda for land DA priorities so that the next generation of land DA systems will be better poised to take advantage of the significant current and anticipated shifts and advancements in remote sensing, modeling, computational technologies, and hardware resources.

The scarcity of groundwater storage change data at the global scale hinders our ability to monitor groundwater resources effectively. In this study, we assimilate a state-of-the-art terrestrial water storage product derived from Gravity Recovery and Climate Experiment (GRACE) satellite observations into NASA's Catchment land surface model (CLSM) at the global scale, with the goal of generating groundwater storage time series that are useful for drought monitoring and other applications. Evaluation using in situ data from nearly 4,000 wells shows that GRACE data assimilation improves the simulation of groundwater, with estimation errors reduced by 36% and 10% and correlation improved by 16% and 22% at the regional and point scales, respectively. The biggest improvements are observed in regions with large interannual variability in precipitation, where simulated groundwater responds too strongly to changes in atmospheric forcing. The positive impacts of GRACE data assimilation are further demonstrated using observed low-flow data. CLSM and GRACE data assimilation performance is also examined across different permeability categories. The evaluation reveals that GRACE data assimilation fails to compensate for the lack of a groundwater withdrawal scheme in CLSM when it comes to simulating realistic groundwater variations in regions with intensive groundwater abstraction. CLSM-simulated groundwater correlates strongly with 12-month precipitation anomalies in low-latitude and midlatitude areas. A groundwater drought indicator based on GRACE data assimilation generally agrees with other regional-scale drought indicators, with discrepancies mainly in their estimated drought severity.

In this paper, we review the use of satellite-based remote sensing in combination with in situ data to inform Earth surface modelling. This involves verification and optimization methods that can handle both random and systematic errors and result in effective model improvement for both surface monitoring and prediction applications. The reasons for diverse remote sensing data and products include (i) their complementary areal and temporal coverage, (ii) their diverse and covariant information content, and (iii) their ability to complement in situ observations, which are often sparse and only locally representative. To improve our understanding of the complex behavior of the Earth system at the surface and sub-surface, we need large volumes of data from high-resolution modelling and remote sensing, since the Earth surface exhibits a high degree of heterogeneity and discontinuities in space and time. The spatial and temporal variability of the biosphere, hydrosphere, cryosphere and anthroposphere calls for an increased use of Earth observation (EO) data attaining volumes previously considered prohibitive. We review data availability and discuss recent examples where satellite remote sensing is used to infer observable surface quantities directly or indirectly, with particular emphasis on key parameters necessary for weather and climate prediction. Coordinated high-resolution remote-sensing and modelling/assimilation capabilities for the Earth surface are required to support an international application-focused effort.