📄 earth science

Huge spread in CMIP6 diffuse radiation impairs carbon budget estimates

This study reveals that significant discrepancies in diffuse solar radiation estimates across 16 CMIP6 models lead to severe underestimations and multi-petagram uncertainties in global carbon sink projections, underscoring the critical need to constrain these radiation values in Earth system models.

Original authors: TC Chakraborty

Published 2026-10-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: TC Chakraborty

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Sunlight that reaches the Earth's surface is not a single, uniform beam. It arrives in two distinct forms: direct rays that travel straight from the sun, and scattered light that has bounced off clouds, dust, and gas molecules in the atmosphere. This scattered light, known as diffuse radiation, behaves differently than direct sunlight. While direct beams tend to shine brightly on the top layers of a forest canopy, often leaving the leaves below in deep shadow, diffuse light spreads out in all directions. It penetrates deeper into the vegetation, illuminating the shaded leaves that would otherwise remain dark. This extra illumination allows plants to photosynthesize more efficiently, effectively turning the entire forest into a more productive engine for absorbing carbon dioxide from the air. Because this process links the atmosphere's clouds and particles directly to the health of the world's forests, understanding how much of this scattered light reaches the ground is essential for calculating how much carbon the Earth's land can store.

A researcher at the Pacific Northwest National Laboratory recently investigated how well modern climate models capture this specific type of sunlight. They examined data from sixteen different global climate models, which are complex computer simulations used by scientists to project future climate changes and understand past environmental shifts. These models are the primary tools for estimating the global carbon budget—the balance between carbon released into the atmosphere and carbon absorbed by plants and oceans. The researcher focused on the historical period from 2000 to 2014, comparing what the models predicted against actual measurements taken from weather stations around the world. Their goal was to see if the models agreed with each other and with reality regarding the amount of diffuse radiation reaching the surface.

The findings revealed a startling lack of agreement among the models. When the researcher looked at the global average, the estimates for diffuse radiation varied by nearly three times. Some models predicted very low amounts of scattered light, while others predicted amounts almost three times higher. This divergence was even more pronounced over land, where the difference between the highest and lowest estimates reached a factor of three. The problem was not just that the models disagreed with each other; they also disagreed with real-world observations. When the team compared the average of all the models against thousands of ground measurements, they found that the models consistently underestimated the amount of diffuse light, particularly during the growing season when plants are most active. In the Northern Hemisphere, for instance, the models missed the peak amount of scattered light by a significant margin during the summer months, suggesting a systematic error in how these simulations handle the interaction between the atmosphere and the surface.

This discrepancy matters because the amount of scattered light directly influences how much carbon plants can absorb. The researcher calculated that the wide variation in the models' estimates of diffuse radiation could translate into massive uncertainties in predictions of plant growth. If the models are wrong about how much scattered light reaches the forests, they are likely wrong about how much carbon those forests are storing. The study suggests that this uncertainty could amount to several billion tons of carbon per year, a figure large enough to rival the total amount of carbon currently absorbed by the world's land ecosystems. In tropical regions, where the spread in model estimates is particularly wide, the uncertainty in carbon absorption could exceed thirteen percent of the total uptake. This means that current projections of future carbon sinks, which are critical for climate policy, may be built on a shaky foundation regarding how plants respond to sunlight.

The root of the problem appears to lie in how these models are constructed. The way they calculate the split between direct and scattered sunlight relies on complex formulas for clouds and atmospheric particles, which are not yet fully understood or accurately represented. Furthermore, many of the land models used within these systems simplify the way light moves through a forest, often treating the canopy as a single layer rather than a complex structure with sunlit and shaded leaves. This simplification fails to capture the "fertilization effect" of diffuse light, where the extra illumination boosts the productivity of the entire plant community. The researcher noted that even models built by the same institutions, using similar core atmospheric components, produced vastly different results for the same region, highlighting that the issue is not just a matter of different data but of fundamental differences in how the physics are simulated.

The study concludes that to improve our understanding of the Earth's carbon cycle, scientists must prioritize getting the representation of diffuse radiation right. This involves better observations of scattered light, especially in remote tropical forests where data is currently scarce, and more sophisticated ways of modeling how light interacts with vegetation. As the scientific community prepares for the next generation of climate models, the author argues that standardizing how these models record and evaluate diffuse radiation is no longer optional. Without fixing this blind spot, the uncertainty in global carbon budgets will remain too large to provide the reliable guidance needed for climate action. The path forward requires a coordinated effort to ensure that the next generation of simulations can accurately reflect the subtle but powerful role of scattered sunlight in sustaining the planet's life.

Problem
Diffuse solar radiation ( $K_{\downarrow,d}$ ) plays a critical mechanistic role in the Earth system by penetrating plant canopies to illuminate shaded leaves and reducing light saturation in sunlit leaves, a phenomenon known as the "diffuse radiation fertilization effect." This process enhances terrestrial gross primary productivity (GPP), alters transpiration, and cools land surfaces. However, in Earth System Models (ESMs), the partitioning of total solar radiation into direct and diffuse components relies on atmospheric radiative transfer calculations heavily influenced by uncertain cloud and aerosol parameterizations. Despite its biospheric importance, the accuracy of $K_{\downarrow,d}$ in ESMs has remained largely unbenchmarked. It is rarely archived as a standard output in the Coupled Model Intercomparison Project version 6 (CMIP6), and many dynamic global vegetation models (DGVMs) do not treat diffuse and direct forcings separately, often assuming a fixed fraction of total radiation instead.

Methodology
The study evaluates historical $K_{\downarrow,d}$ estimates across 21 CMIP6 model configurations from 16 modeling centers. The analysis covers the period from 2000 to 2014. Model outputs were extracted from the Earth System Grid Federation (ESGF), focusing on the first realization, initialization, and parametrization (r1i1p1f1) where available. These model estimates were compared against each other to assess inter-model spread and against ground-based observations from the Global Energy Balance Archive (GEBA), utilizing 16,590 collocated monthly records.

Statistical evaluations included:

Global and Regional Analysis: Calculation of multi-model means, inter-model spreads, and biases across Köppen-Geiger climate zones.
Station Benchmarking: Point-to-grid comparisons using nearest-neighbor indexing to calculate Ordinary Least Squares (OLS) regression parameters, Root-Mean-Square Error (RMSE), Mean Bias Error (MBE), and Mean Percentage Error (MPE).
Sensitivity Estimation: The study applied previously estimated linear sensitivities of GPP to $K_{\downarrow,d}$ from the Community Land Model (CLM) and the Joint UK Land Environment Simulator (JULES) to quantify potential uncertainties in the global carbon budget resulting from the observed radiation spread.

Key Results

Extraordinary Inter-Model Divergence: Across the 16 models, global mean $K_{\downarrow,d}$ estimates diverge by a factor of 2.8 (full range of 66.2 W m⁻²). Over land, this spread widens to approximately 3-fold (33.9–100.9 W m⁻²). Even models sharing the same atmospheric core (e.g., UK Met Office Unified Model) exhibit severe discrepancies, with values ranging from 63.1 W m⁻² to 103.1 W m⁻².
Systematic Underestimation: The multi-model mean exhibits a persistent negative bias against GEBA observations, with a mean percentage error of -11.6% and an RMSE of 19 W m⁻².
Seasonal and Regional Biases: Discrepancies peak during the growing seasons in both hemispheres. In the Northern Hemisphere, the multi-model mean underestimates the peak growing-season $K_{\downarrow,d}$ by 15–18 W m⁻² (May–August). Individual model biases span nearly 95 W m⁻². The inter-model spread is particularly high over productive tropical carbon sinks like the Amazon and Congo Basin, where observational data is sparse.
Carbon Budget Implications: Applying CLM sensitivity estimates to the observed $K_{\downarrow,d}$ spread yields a GPP uncertainty of approximately 13 Pg C yr⁻¹ globally (roughly 11% of global terrestrial GPP). For tropical climates, this uncertainty rises to ~8 Pg C yr⁻¹ (over 13% of tropical GPP). These uncertainties are several times larger than the estimated global land and forest carbon sinks.

Significance and Claims
The paper argues that the "huge spread" in CMIP6 diffuse radiation estimates fundamentally impairs the reliability of carbon budget estimates and future projections. The author claims that current ESMs likely underestimate the diffuse fertilization effect, particularly in temperate and boreal forests during peak carbon uptake, due to the systematic underestimation of growing-season $K_{\downarrow,d}$ .

The study highlights that this issue extends beyond atmospheric forcing differences; it also stems from oversimplified canopy radiative transfer and upscaling schemes in land surface models (e.g., "one-big-leaf" schemes) that fail to resolve $K_{\downarrow,d}$ fertilization. Furthermore, current CMIP6 MIPs (such as RFMIP, LS3MIP, C4MIP, and LUMIP) systematically overlook $K_{\downarrow,d}$ , leaving it to unbenchmarked internal algorithms.

The author concludes that constraining $K_{\downarrow,d}$ across both atmospheric and land surface components is critical for reducing multi-petagram uncertainties in the global carbon budget. They assert that standardizing the archival, evaluation, and process-based representation of diffuse radiation—ideally through a dedicated MIP in the upcoming CMIP7 design—is necessary to address a critical blind spot in 21st-century climate projections and to accurately assess trade-offs related to solar radiation management and air-quality improvements.

More like this