Scientific American: Most of my day job involves simulating the behavior of molecules like drugs and proteins using computer models. The field is more an art than a science, partially because the systems that are being modeled are too complex and ill-understood to succumb to exact solutions. Success often depends on experience and intuition gained by working on similar systems. That does not mean there are no correct predictions, but it does mean that surprises are more common than we think and that many phenomena are impossible to model within a very precise window of accuracy. The failure of a model can sometimes be traced to a simple inability to simulate the behavior of an essential component of the system. In several cases this component is simply the water that surrounds a protein; water remains a substance that’s as enigmatic as any other. In other cases it could be the entropy of the system. The problem is that these factors are very hard to calculate even when we know that they are responsible for the limitations of our model.
A recent report on the failure of climate change models to predict the timing of major droughts in the Southwest made me think of some of the problems in my own field. Unfortunately the actual paper is not out yet so we will have to wait for the details, but the news piece in Nature has a good summary.
Sloan Coats of Columbia University’s Lamont-Doherty Earth Observatory in Palisades, New York, and his colleagues tested whether a state-of-the-art climate model could simulate the droughts known to have occurred in the southwest during the past millennium. The model incorporated realistic numbers for factors that affect temperature and rainfall, such as atmospheric carbon dioxide levels, changes in solar radiation and ash from volcanic eruptions. It also incorporated changes in the El Niño/Southern Oscillation (ENSO).
The results were puzzling. Although the simulation produced a number of pronounced droughts lasting several decades each, these did not match the timing of known megadroughts. In fact, drought occurrences were no more in agreement when the model was fed realistic values for variables that influence rainfall than when it ran control simulations in which the values were unrealistically held constant. “The model seems to miss some of the dynamics that drive large droughts,” says study participant Jason Smerdon, a researcher at Lamont-Doherty who studies historical climate patterns.
Other climate models tested by the team fared no better, he says. In particular, the models failed to reproduce a series of multi-decadal droughts that occurred in the southwest during the Medieval Climate Anomaly, a period between AD 900 and 1200 when global temperatures were about as high as they are today.
The team goes on to provide several possible explanations for the failure of the models, most likely related to their inability to account for details in the ENSO cycle. The researchers also note that the models may not capture some important features of the biosphere.
In addition to their failure to reproduce El Niño and La Niña, existing models do not fully capture other factors that influence rainfall, such as clouds and vegetation. But Smerdon adds that the atmospheric and oceanic dynamics that inhibit rainfall and favour prolonged drought may be essentially random and so almost unpredictable.
This is in fact a problem that has plagued computer models of climate since their very inception in the 1950s. The early general circulation models (GCMs) included the motion of the atmosphere and factors like wind speed, temperature and pressure. Over time these atmospheric circulation models became quite sophisticated, account for radiation transport and the opacity of various gases. The strengths and weakness of these models largely carried over into modern day climate modeling.
In general the models are quite good at simulating the motions of the atmosphere but are still inadequate in accounting for the complex processes in the biosphere, including the behavior of the soil, forests, rivers, mountains and the various plants and animals that inhabit these environments. This discrepancy between accurate atmospheric simulation and lackluster biospheric simulation may be responsible for many of the defects in climate modeling. And as the researchers say, the models are still not great at capturing fine-grained details of clouds and their influence on water. It’s striking to me that both molecular models and climate models struggle in modeling that simplest and most ubiquitous of substances – water. No wonder they have a hard time predicting droughts and precipitation. Finally, the lack of difference in the results when the key factors are held constant and when they are allowed to vary points to an independent and possibly unknown set of factors that are influencing model results.
Nonetheless as the article says, the major predictions about global precipitation seems to be clearer and are based on extensive field studies across the globe; climate change is much more than computer modeling. The problems though are in predicting local precipitation patterns and unfortunately it’s these kinds of predictions that drive public policy at local and state levels. The most important result from such modeling data of course is the knowledge it provides about the strengths and limitations of climate change models. And knowledge is always useful.
Started in year 2010, ‘Climate Himalaya’ initiative has been working on Mountains and Climate linked issues in the Himalayan region of South Asia. In the last five years this knowledge sharing portal has become one of the important references for the governments, research institutions, civil society groups and international agencies, those have work and interest in the Himalayas. The Climate Himalaya team innovates on knowledge sharing, capacity building and climatic adaptation aspects in its focus countries like Bhutan, India, Nepal and Pakistan. Climate Himalaya’s thematic areas of work are mountain ecosystem, water, forest and livelihood. Read>>