A Data Science Central Community
Imagine yourself relocating to a more industrial place for living because your beach house was washed away in the tide. In another scenario imagine yourself wearing masks throughout the year. What if I say that, all this that you just imagined could be a reality very soon?
Whether you choose to believe it or not, climate change is happening for real. Even though you might not be able to spot a lot of its impact around you, the world around is all set to change in the coming years. Be it the rise in oceans that will lead to submerging of the low lying areas in water or the global rise in temperature. The world as we know today is going to change drastically as a consequence of our actions.
Look at any newspaper around you. Not a single day goes by without spotting the news of new issues arising on account of climate change. The fact is that it has shaken the world to its core. Research at NASA suggests that the leading cause of rapid change in climate is due to the emission of greenhouse gases. This phenomenon is termed as global warming and statistics suggest that there is a 95 percent probability that human actions in the past 50 are responsible for warming up the planet.
Why did Researchers consider AI’s deep neural networks in climate change studies?
Climate change has, therefore, become more than just an impending issue. It is an existential crisis. But, all is not lost. Researchers all across the world are coming up with new ideas and concepts that harness the best of technology to battle climate change issues. One of the most advanced and resourceful technologies that are being used in climate change studies in Artificial Intelligence. AI along with its subsidiaries like machine learning are helping scientists and practitioners come up with insightful information and predictions for climate change. These are then being further studied by experts who can collaborate with the governments to take action and form plans to preserve our home planet.
Before scientists realized the potential of machine learning for climate change, a lot of physics-based models were being used to make predictions. However, most of these followed bottom-up approaches and made predictions only based on physical boundary conditions. On one hand, there were General Circular models or GCMs that were python software developed by the numerical representation of atmospherically physical conditions. On the other hand, there were Earth System models that considered features like biochemical cycling and atmospheric chemistry. Even though some of the advanced GCM models were used for climate change studies they suffered from significant errors in prediction.
That’s when researchers started considering the potential of deep neural networks in climate change studies. To understand what a deep neural network means, let’s take a look at a simple neural network.
A simple neural network can consist of two inputs just like any other model. The differentiating factor is the hidden layer, which can have, let’s say two neurons. These neurons help imitate the functioning of a human brain and consist of attached weights for computation. Now each weight of the neuron is respectively multiplied to each of the inputs of the neuron. These are then summed and activated using an activation function. It is only upon activation that they come out as an output from the neuron. That’s how you design a simple neural network.
Similarly, when it comes to deep neural networks, the only difference is in the number of hidden layers and corresponding weights attached to them. The more you increase the number of the hidden layers, the more complex or deep your model becomes. But, at this point, it is wise to wonder how any of this helps model the data and ais climate change studies.
Understand it this way. Hidden layers are the magic of deep neural networks and provide mandatory discrimination to separate your training data. Take a simple XOR function as an example. A single layer neural network does not have the capacity to provide two disjoint decision boundaries for an XOR function. However, adding a hidden layer fulfills the complexity needed. You can choose to increase the number of neurons in a particular layer or increase the number of layers itself for the task. Both ways you will be increasing the complexity of the system. While increasing the number of neurons accounts for a decrease in the training error, it also reduces the generalization of the model, which is a crucial parameter. Similarly, the more hidden units and layers you add in your model, the more complex hyperplanes you can learn.
For climate change studies this can be a miracle. Researchers have multiple factors that they’d like to add as inputs. After all, you can’t just make a decision about global warming based on an increase in the number of vehicles on the road. The decision is far more complex. When researchers say that rising seas will erase some of the greatest cities by 2050, they don’t just have one or two input parameters. There are billions of complex calculations involved behind these predictions. Deep neural networks help researchers take a wide number of parameters into account and produce a decision boundary that sufficiently encompasses these inputs.
Neural Network Examples in Climate Change Studies
Discovery of sustainable materials: One of the greatest crisis in the world is dealing with large amounts of material on the planet that is neither degradable nor recyclable. All these accounts to garbage. But, with deep neural networks in the picture, researchers have an edge to come up with highly optimized molecular structures that are more sustainable and energy-efficient. Most people think that cotton bags are the ultimate solution when it comes to climate change. But, seldom do they realize the fact that a cotton bag consumes more resources than other materials. Generative networks, when used with predictive or recurrent neural network models, can be used to design several property-based molecular designs.
Precision Monitoring of Regions: Another great field where neural networks can assist through precision modeling is the monitoring of crops and forest regions. Convolution neural networks along with computer vision can help segment the satellite imagery and make predictions about droughts, fires, destruction or negative crop outcomes. They can also help in identifying the type of pests that can be used to find a suitable pest control solution for the soil.
The impact of neural networks on climate change studies is huge. With historic and existing data, scientists are finally able to take a peek into the future. Even if the future doesn’t look very bright, AI and ML models can be used to make world leaders aware of the impending scenario. It will help form the right strategies and take control measures within whatever time there’s left for us.
Comments are closed for this blog post