Biomes are ecosystems adapted to the specific constraints of the environment they live in. Biomes have been studied extensively on land, where the existence of natural borders and obvious geographical features helps to identify them. Defining biomes is more challenging in the ocean, where identifying and separating one ocean biome from another cannot be done simply by looking at them. Several attempts to partition the ocean into regions of broadly equivalent ecosystems have already been undertaken, typically based on observations. It is important work, both since it helps improve our understanding of the underlying marine biogeochemistry, but also because it is needed by policy makers, to help with the management of marine areas by defining “biologically significant” areas, vulnerable areas, or climate change hot-spots.
In the context of UKESM1, our study has multiple aims:
First, we aim to provide a regional evaluation of the model simulated marine physics and biogeochemistry. Typically, model evaluations are done at the large scale, effectively “bulking-up” the model’s average performance everywhere. This can be a good way to characterise a model, especially when reducing the complexity of analysis, but such budgets and trends at basin scale (or larger) overlook ocean physical or biogeochemical features. Some information, which could be highlighted and improve our understanding of model behaviour, is instead diluted and lost. This is the first reason why we decided to perform a biome-based evaluation of the model.
Second, while we now know about “real” ocean biomes, based on observations, what about our models? Do they reproduce all of the observed biomes? Are they correctly located? Do they have the same ecological properties? If not, why not?
First, how do we define our biomes? There are plenty of ways to divide the ocean, most of which are tested and validated on observational data:
- Use expert knowledge-based predefined regions to apply as a mask to the observed and modeled ocean;
- Define biomes based on the intervals of specific key ocean variables;
- Use an “unsupervised or automated” objective approach based on statistical tools (like clusters) applied to one or more key ocean variables.
The way we describe this problem helps to decide how best to proceed. One of our main aims is to evaluate how well the model is able to reproduce the observed biomes, which excludes the use of observationally-predefined regions whose boundaries are unlikely to map perfectly onto those of the model. Also we want to make sure that the biomes we compare, are actually comparable. This is something that is difficult to be sure of with automated methods, since the biome definition (i.e. what it represents), as well as the number of biomes, are data-dependent, something not well suited for model-observation intercomparison. The best solution in our case seems to be to define biomes based on the intervals of key-variables. This avoids the use of fixed boundaries that likely vary and is a transparent method that is straightforward to interpret.
In our study, biomes are defined based on annual maximum surface chlorophyll (CHL) and mixed layer depth (MLD) as follows:
Table 1: Chlorophyll and mixed layer depth criteria for defining ocean biomes.
|Biome||Chlorophyll||Mixed layer depth|
|Oligotrophic (olig)||Chl < 0.075 mg m-3||–|
|Submesotrophic (sm)||0.075 <= Chl < 0.25 mg m-3||–|
|Mesotrophic (m)||0.25 <= Chl < 1.0 mg m-3||MLD > 100 m|
|Low mixing meso- (m100)||0.25 <= Chl < 1.0 mg m-3||MLD < 100 m|
|Eutrophic (e)||1.0 < Chl||MLD > 100 m|
|Low mixing eu- (e100)||1.0 < Chl||MLD < 100 m|
The use of annual maximum MLD enables us to disentangle productive biomes with high seasonal mixing (typically at high latitudes) from those with low seasonal mixing (typically low latitude upwelling regions). Applying these rules to both observational data and UKESM1 output results in the geographical pattern of biomes shown in Figure 1 (see also Table 2’s biome areas).
Figure 1: Present-day (2000-2009) biomes based on observations (left) and UKESM1 (right). Observations used are monthly average SeaWiFS ocean colour data, and de Boyer Montégut et al. (2004) climatology.
Figure 1 shows the model accurately simulated the main marine biomes, comparable in terms of location and size to those derived from observations, although it some model biases are evident. Oligotrophic biomes (dark blue) in the Atlantic of UKESM1 cover a wider area, at the expense of submesotrophic (light blue) and low mixing productive biomes (green). By contrast, in the Pacific Ocean, the oligotrophic biomes cover a smaller area, while the low mixing productive biome (which corresponds here to equatorial upwelling) covers a much larger area. Meanwhile, in the Southern Ocean, we find that a greater area of the model is productive (orange). This general bias is already known from our model evaluation at the global scale, but using biomes enables a more geographically resolved comparison, helping us understand what is happening to the model in this important region.
Table 2: The first two columns compare the biomes areas calculated from observational data and model output (in Mkm2, i.e. million km2) for the present-day (2000-2009) presented in Figure 1. The final two columns report the relative change (as a percent) of each biome area by 2090-2099 for two future scenarios whose biomes are shown in the Figure 2. Also note the model and observation-based biomes total area cannot match, because of missing data in polar regions of the observed chlorophyll fields.
|Low mixing meso- (m100)||43.5||74.5||+1.3%||-10.1%|
|Low mixing eu- (e100)||10.3||10.1||-62.4%||-79.6%|
With the biomes defined this way, ongoing work is using them in a more regional analysis of the performance of UKESM1. This includes using them to help evaluate the model’s performance across a wide range of biogeochemical properties, such as nutrient availability, productivity and carbon cycle variables, such as surface ocean pCO2. In particular, how the modelled seasonal cycles of these properties relate to those we see in nature. This analysis is helping to provide a more complete understanding of where our model’s weaknesses are and what properties of the model most require improvement in the future.
A further analysis that the biome approach can help us with is in characterizing how marine productivity may change in the future. In the same way we apply the biome criteria across both models and observations, we can apply them to model simulations at different times into the simulated future to assess how biomes may change in response to global warming. Here, we use two different ScenarioMIP simulations of the future: SSP1 2.6, a scenario with low CO2 emissions and relatively low warming and SSP5 8.5, a scenario with high, “business-as-usual” CO2 emissions and strong warming. Figure 2 shows the resulting biomes for these two extreme cases.
Figure 2: Future (2090-2099) biomes under two different scenarios: SSP1 2.6 (left; low CO2 emissions) and SSP5 8.5 (right; high CO2 emissions).
Table 2 also lists the percentage changes in the areas of the biomes across the 21st century. In both scenarios, there is a common evolution from productive / high mixing to low production / low mixing biomes. But the scenario results differ in the scale of this evolution.
In the low CO2 emission scenario (SSP1 2.6), the eutrophic (-46%; both classes) and mesotrophic (-4.5%; both classes) biomes generally shrink and are supplanted by the submesotropic (+6.9%) and oligotrophic (+16%) biomes. However, in the high CO2 emission scenario (SSP5 8.5), the changes are much greater, with the unproductive oligotrophic biome more than doubling its area (+102%), and the eutrophic biome collapsing (-73%; both classes). Under SSP5 8.5, the last remains of the eutrophic biomes are tiny regions east of Greenland. Given that the mesotropic and eutropic biomes are the source of most of the open ocean’s productivity, including its fisheries, these large changes point to the importance of reducing CO2 emissions.
Biomes are a promising tool to help us understand our model’s performance. They enable us to evaluate the model according to regions that share common biogeochemical features rather than simply a common location, and to understand whether biases in different regions have the same root cause. Furthermore, they provide a handle for quantifying how different future scenarios may change the ocean’s living communities.
More to come about this very soon in the UKESM1 special issue.