Drivers and trends of global soil microbial carbon over two decades

Predictors of microbial carbon stocks

We used a machine learning modeling approach to predict soil microbial carbon from a set of environmental covariates. To account for stochastic variability, we ran a set of models to assess the importance of environmental factors, which showed that the contribution of each variable to the model fit differed between runs, with some overlap between a number of them (Fig. 2b). Mean annual temperature was always the most important variable, with soil organic carbon and soil pH following. Clay content, precipitation, land-cover type, nitrogen content, and sand content contributed roughly equally to explaining variations in microbial carbon. Finally, NDVI and elevation had the lowest variable importance. Coniferous forests had the highest and most variable predicted values of microbial carbon (Supplementary Figs. 1, 2), which can be explained by high soil organic matter and a thick litter layer²⁶. Tropical forests also had fairly high values of microbial carbon, while shrublands and croplands had the lowest values²⁶. We used partial prediction response curves to evaluate the direction and range of effect of the predictor variables (Supplementary Figs. 1, 2). In agreement with the variable importance measure, variables that scored high often showed strong effects on the predicted microbial carbon values, while variables with a low variable importance score (e.g., elevation, NDVI, and sand content) only showed smaller responses. The only exception was for precipitation, which had a relatively high variable importance, although the response curves only showed a weak effect of precipitation for forests and grasslands, with limited effect on other land-cover types (Supplementary Fig. 2). The importance of precipitation might also indicate that this relationship involves interactions with other variables^7,28. Overall, the differences in microbial carbon between land-cover types showed mostly similar patterns across the range of variables. Soil organic carbon and nitrogen content had a positive and mostly linear effect on microbial carbon (Supplementary Fig. 1). In contrast, clay content, soil pH, and mean temperature had non-linear relationships, with high microbial carbon in the low range of these variables and a rapid decrease that reached an asymptote at low microbial carbon values for the higher portion of the range. Soil pH patterns showed a decrease in microbial carbon for values between 4.1 and 5.8, and a constant pattern between 5.8 and 8.6. Contrary to our expectations, we did not find a parabolic effect of soil pH on microbial carbon²⁶. Instead, our model predicted higher values in very acidic soils with a pH below 5.2, which are rare globally and almost only found in central Amazonia. Similarly, locations with a clay content lower than 16.9% had higher values in microbial carbon, and then stabilized until 51.0%.

Mean temperature showed an interesting shift with much higher microbial carbon values with a mean annual temperature below zero, but had otherwise a limited effect on microbial carbon values in the rest of the range above zero up to 28.9 °C. Based on partial predictions (Supplementary Figs. 1–2), microbial carbon decreased monotonically with an increase in temperature (with all other variables fixed to their median), with the relationship being mostly stable for parts of the range. We observed an especially sharp decrease at around 0°C, which is in agreement with the patterns observed in the data. The reason for sites with a mean annual temperature below the freezing point to have higher microbial carbon stocks is not fully understood. This could be due to a regime shift in which microbial communities are in a semi-dormant state for a major part of the year³⁵. Moreover, it could also be in part explained by the soil organic carbon content that follows a similar trend and accumulates in higher latitude soils⁹, thus promoting higher microbial carbon stocks. Within these cold, high organic carbon soils, large microbial populations can be maintained, due to the low temperature that reduces metabolic requirements³⁵. In contrast, at higher temperatures, metabolic activity increases and requires more resources and nutrients to maintain microorganisms alive. Experimental evidence is divided about the effects of warming on microbial carbon^18,36, highlighting the strong context-dependency of this relationship, although global observations show a clear pattern, where low-temperature sites have higher soil microbial carbon stocks. Despite this uncertainty, there is a strong indication that a warming soil would tend to lose organic carbon^17,37, and subsequent patterns in microbial carbon can also be expected, because of the dependency on organic substrate^9,26,38. These dynamics were observed in Melillo et al.³⁹, where the warming of sites in a mid-latitude forest ecosystem led to a decrease in soil carbon, followed by a decrease in microbial carbon¹².

Even with predictions being made for each grid location separately, microbial carbon values showed distinctive patterns and transitions over the globe (Fig. 2a). While temporal changes took place, broad spatial patterns were relatively constant over the range of years studied (Supplementary Movie 1). The highest microbial carbon stock values ranging from 1.50 to 7.00 t ha⁻¹ were found at high latitudes in the Northern Hemisphere in areas of coniferous forest. Tropical humid regions also showed high microbial carbon values between 0.50 and 1.50 t ha⁻¹ in the Amazon Rainforest and Central Africa. The main regions with low microbial carbon below 0.30 t ha⁻¹ were in Eastern South America, areas directly south of the Sahara Desert, East Africa, and most of Australia, all of which mostly correspond to shrublands. Cropland areas as seen in India were also predicted with low microbial carbon values ranging from 0.06 to 0.38 t ha⁻¹. A strong latitudinal gradient was visible for North America and Eurasia, with the highest microbial carbon stocks at high latitude, medium values in temperate ecosystems, and decreasing values towards the Equator. Positive coastal effects can also be observed, mostly on the Eastern South American and Australian coasts. In total, we estimated that there is 4.34 Gt of microbial carbon in the 5 to 15 cm layer for the predicted areas. Using the coefficient of variation calculated from the variability assessment set of models, we found that predictions made for the Amazon Basin, Northern Canada, and South-East Russia were more variable than for other regions (Supplementary Fig. 3a). Especially Western Europe, Central North America, and South-East Asia, however, showed high stability in the predictions between model runs.

Drivers of change

The analysis of the rate of change of microbial carbon stocks over time revealed that large regions of the globe experienced important changes in soil microbial carbon stocks between 1992 and 2013, with contrasting patterns across areas, and overall larger regions showed a decrease rather than an increase in microbial carbon stocks (Fig. 2c and Supplementary Fig. 3b). To account for spatial differences in microbial carbon stocks, we calculated the relative rate of change in percentage for each location (Fig. 2c). When considering all predictable regions together, microbial carbon stocks in the 5–15 cm layer showed a decrease of 7.09 Mt per year, summing to 148.80 Mt between 1992 and 2013, or 3.4% of the global microbial carbon pool predicted (Supplementary Fig. 4a; p = 0.038). The main regions with a microbial carbon loss higher than 0.7 kg ha⁻¹ y⁻¹ were in Northern Canada and a large continuous region in North-Eastern Europe. These northern regions accounted for an important part of the global loss in microbial carbon stocks, with large areas that had both a high soil microbial carbon stock and a fast decrease (Figs. 3 and 4). Other areas of high loss were in the Amazon basin, Western Argentina, the USA East Coast, Southern South Africa, and South-East Russia. The main continuous region of microbial carbon increase above 0.7 kg ha⁻¹ y⁻¹ was in central Russia, with smaller regions present in India, Europe, Central North America, and parts of Africa. Besides these general patterns, predictions vary at the local scale, and they consider the effects of parameters including soil properties, elevation, and land-cover type, which change between neighbor locations and affect the observed patterns. This is especially visible in the Americas, where both increases and decreases happen side-by-side.

Patterns in the relative rate of change have a lot in common with that of absolute change, with a few notable differences (Fig. 2c and Supplementary Fig. 3b). Both positive and negative stock changes in tropical and subtropical regions are more prominent in relative terms, as these regions typically have low microbial carbon stocks. Similarly, regions in Central Russia with high microbial carbon stocks show less decrease in relative terms. To assess how stable these trends are over time, we show the p values of the rate of change for the 22 years (Supplementary Fig. 3c). The largest region with low p values is associated with more significant trends in Western Russia, and corresponds to an area with a fast loss of microbial carbon. India and Central Russia show high p values, and are informative of high variability compared to the strength of the signal. Considering that only up to 22 data points are available for each grid location and that especially climatic conditions vary considerably from year to year, p values are only provided as a complementary assessment. We can summarize the global situation by combining the two maps of microbial carbon stocks and relative rate of change to categorize and define vulnerable locations that experienced a high loss of microbial carbon (Figs. 3 and 4), and where the provision of soil functions is potentially at risk.

It is informative to look at regional trends, by grouping grid locations using the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) sub-regions, and assessing regional-scale changes in microbial carbon stocks (Fig. 5, Supplementary Table 1). The main regions that contributed to microbial carbon loss were North America with a decrease of 62.49 Mt of microbial carbon and Eastern Europe with 60.88 Mt over the studied period, although both trends had high yearly variability and were non-significant. The region with the highest increase was North-East Asia with a gain of 4.49 Mt, but this change was also non-significant. The Caribbean was the only region to show a significant increase in soil microbial carbon stocks over time (+2.1% over 22 y, p = 0.017), while significant decreases in stocks were found in North Africa (−4.1%, p < 0.001), South America (−1.7%, p = 0.010), Southern Africa (−2.6%, p = 0.017), and Central and Western Europe (−2.7%, p = 0.034; Supplementary Fig. 4a). Marginally significant decreases over the studied period were found in Western Asia (−2.5%, p = 0.086) and North America (−7.2%, p = 0.093). The 10 other regions showed no significant change in microbial carbon stocks, namely in Central Africa, Central Asia, East Africa and adjacent islands, Eastern Europe, Mesoamerica, North-East Asia, Oceania, South-East Asia, South Asia, and West Africa.

Climatic conditions and land-cover type are important aspects that affect soil microbial carbon stock dynamics. As we expected and has been found for other soil organisms (e.g., refs. ^40,41,42), climatic changes tended to have a stronger effect on microbial carbon stocks than those related to land cover (Fig. 5). Temperature patterns showed overall long-term warming in most regions, despite yearly variability, with a mean increase of 0.28 °C globally, promoting microbial carbon losses (Supplementary Figs. 5, 6). When looking at the separate effects of changes in climatic or land-cover variables (obtained by fixing the other set of variables), we found that globally, the decrease in microbial carbon was driven by climate change, with little effect on land-cover change (Fig. 5). Regionally, however, different resulting scenarios emerged from this analysis (Fig. 5 and Supplementary Fig. 4). The two groups of global change drivers had different influences on the predicted microbial carbon stocks depending on the region, with some being mostly affected by climate and others by land-cover change. The results of the interaction of both groups of variables either reinforced or masked the effect of each other. In many regions, one group of drivers was more prominent than the other and was mostly driving the general pattern. In a few cases, regions with non-significant effects in the full dynamic model showed significant effects of land-cover changes only (where climate variables are fixed), either with a positive effect (Central Asia and North America) or negative effect (South-East Asia). In addition, there were cases where the non-significant effects of climate and land-cover change combined to produce a significant overall effect (e.g., South America) and a special case in Western Asia (and less strongly in Oceania), where the significant effects of both driver types went in opposite directions, leading to overall non-significant effects. In these cases, the negative effect of one driver is compensated by the positive effect of the other. Regions with a negative effect on at least one of the global drivers of change are especially vulnerable to being affected by the functions provided by the soil microbial community, especially in combination with high soil microbial carbon stocks¹⁸. These areas of vulnerability also often coincide with those most affected by environmental changes (Supplementary Figs. 5, 6). As global changes are expected to continue, and potentially accelerate³¹, areas of vulnerability are likely to experience a continued decrease in soil microbial carbon stocks and a potential reduction or change of soil ecosystem functions. With our approach, we can look into region-specific drivers that led to modeled changes in microbial carbon stocks (Supplementary Figs. 3, 4). For example, Central and Western Europe experienced a decrease of 2.9% between 1992 and 2013, almost entirely driven by an increase in temperature of 0.64 °C. In this case, despite the increase in NDVI values over the period, land cover changes did not have much of an effect on microbial carbon dynamics. This region as a whole also experienced yearly fluctuations in precipitation, but showed no general trend over the studied period.

Model evaluation and coverage

The random forest model used for temporal predictions was validated by comparing the observed and predicted values of microbial carbon concentrations. The root-mean-square error (RMSE) was 65.0 mmol kg⁻¹, and the cross-validated R² for out-of-bag predictions was 0.40, while the overall R² was 0.90. The observed microbial carbon values correlated to the fitted values, with a Pearson’s r value of 0.59 (Fig. 6; p < 0.001). We believe that the clear definition of the geographical area of applicability of results is crucial for its proper interpretation, and that this type of assessment is often lacking from global predictions⁴³, especially as strong spatial biases in sampling locations are often observed⁴⁴. To detect grid locations that could be predicted with high confidence, we performed an environmental coverage analysis based on two complementary methods that detect locations with environmental parameters (i.e., predictive variables) that are multi-dimensional outliers compared to the predictive data set. With the combined results of the two approaches to detect environmental outliers, we identified that the current knowledge of soil microbial carbon can be used to make predictions with confidence for locations representing 50.2% of terrestrial surfaces excluding glaciers (Fig. 1, Supplementary Fig. 7). Western Asia and North Africa were the regions with the lowest percentage area that could be predicted, with 5.1% and 10.6%, respectively (Supplementary Table 2). All other regions could be predicted for above 40% of their area, with Western and Central Europe having the largest proportion at 84.7%, followed by South America with 63.6% (Supplementary Table 2). As expected, highly sampled areas were included in regions that could be predicted with high confidence based on the environmental coverage assessment. Despite relatively low sampling density, South America, East Africa, and the western part of Eastern Europe were also regions with a large portion that could be predicted with confidence by both methods (Fig. 1, Supplementary Table 2). Outlier regions from both methods include North-East Russia and the Tibetan plateau, as well as a few smaller regions, mostly related to high latitude, elevation, or aridity. An important portion of the African continent consists of outlier regions, detected by both methods, but rarely in conjunction. The larger outlier areas in Africa mostly match regions of deciduous woodlands and savannah. A number of tropical rainforest regions were also excluded, including most of the Malay Archipelago, as well as central Amazonia.

The environmental coverage assessment highlights uncertainty in large regions, for which targeted sampling campaigns are needed to complement available data sets²¹. To further our understanding of microbial carbon dynamics across the globe, it is especially relevant to target regions with underrepresented environmental parameters, with high variability in microbial carbon, and that is expected to be most affected by changes in climate and land cover⁴⁵. In addition, just as exploring under-sampled areas is important, repeated sampling at the same location provides valuable information for research, monitoring, and conservation efforts^46,47. Like other global predictions based on statistical modeling^41,42,48, the results of this study represent general patterns and are informative to detect regional trends and should not be extrapolated to accurately estimate soil microbial carbon stocks at fine spatial and temporal scales, as local heterogeneity in soil properties and temporal climatic conditions can lead to variations in microbial carbon stocks.

Ecological impacts and implications

Soil microbial carbon is a crucial aspect of ecosystem health and services, and there are indicators that it is decreasing in many parts of the world. The effects of climate and land-cover change will continue to affect biological communities with potentially stronger effects in the years to come⁴⁹. Targeted microbial communities that experience a decrease in microbial carbon can be affected in their ability to provide ecosystem functions, including food and material production, nutrient cycling, and carbon cycling. Microbial carbon represents a carbon pool that contributes to carbon sequestration and mediates carbon cycling^50,51. As the amount of microorganisms declines, these ecosystem functions are at risk to be affected negatively, and the continuity and magnitude of these services cannot be guaranteed⁵². With a decrease in soil microbial carbon, dark CO₂ fixation is also likely to decrease, therefore reducing the climate mitigation effects of soil microbial communities^53,54. While some regions were more affected by climatic changes, others were mostly driven by changes in land cover. In order to limit further losses in soil microbial carbon, both sets of drivers need to be addressed in cohesion, especially given the context-dependent effects of climate change⁵⁵. Anthropogenic climate disruptions have led to regional changes in temperature and precipitation patterns that will continue to affect soil microbial communities²². While regulations and actions at the global scale are needed to slow anthropogenic climate change, local-scale management can address land degradation caused by changes in land cover that are detrimental to soil microorganisms and threaten soil functions^3,55. Changes in land cover may take place naturally, e.g., as a response to climate changes, come from unregulated actions—as often seen in deforestation by small land-owners—or be the consequence of political decisions that affect land management at a larger scale. In that regard, land management can also be leveraged as a climate change mitigation and adaptation strategy, both to preserve microbial communities and sequester carbon^3,49. In this stream, conservation, rewilding, and reforestation efforts focused on vulnerable areas can strongly contribute to supporting soil ecosystem functions and services, and soil communities should therefore be better integrated into conservation efforts⁵⁶.

While soil microbial communities continue to be studied, we can refine our mechanistic understanding of the belowground communities using diverse techniques that become increasingly accessible to describe additional aspects (e.g., diversity, community composition) and functionality, contributing to improving our understanding of this important ecosystem compartment and reduce uncertainty in global estimates^57,58,59, to complement microbial carbon measurements as the base measurement of microbial community size^9,21. Currently, major global monitoring^46,60 as well as data mobilization and synthesis efforts²¹ are taking place, that will help develop a more detailed perspective on the distribution, drivers, and trends of soil microbial communities and functions.

Source: Ecology - nature.com