Neural networks are an extremely simplified model of how the brain works. Basically, the brain consists of numerous nerve cells called neurons, which can be thought of as information processors. Information from other neurons enters each neuron through hair-like structures called dendrites. This is then processed and the output is relayed to other neurons via the axon. The neuron only produces an output when the collective effect of its inputs reaches a certain threshold.
Artificial neural networks usually refer to models applied in cognitive psychology and artificial intelligence. They copy the structure of the brain (neurons with connections between neurons) consisting of a set of nodes with connections between the nodes. A single, artificial node has a number of input connections, each connected with an associated weight; a processing element which processes these inputs and calculates the output; and the output connection itself. These nodes are arranged in layers: an input layer of nodes, an output layer of nodes and may be several hidden layers between.
Actual biological neuronal activity is much more complicated than artificial neural networks, but both are similar in that they can learn and remember. In fact, artificial neural networks never forget and they are faster than biological ones. Artificial networks are always created for a specific purpose and the weights are the most important factor in determining the learning process. The adaptive weights are, conceptually, the connection strengths between the neurons. Training a network is the way it learns from experimental data and results in appropriate values being found for the weights. Numerous algorithms are available for training neural network models, which can be used for predictive purposes once it has been trained.
Artificial neural networks are extremely useful for such things as computer vision and speech recognition. However, their pattern distinguishing qualities are not just limited to object recognition by robots, but can also be used to find the ‘best fit shape’ to experimental data such as the best curve to describe the growth of trees in a forest.
Forests and humans have been interlinked from the beginning– for thousands of years a balance between environmental requirements and needs was found in hunter-gatherer societies. When a resource ran out locally, the tribe moved on. This equilibrium was broken with the advent of agriculture and the formation of the first settlements. The once tiny, global human population has since boomed into a huge, demanding and technological society that drains Earth’s resources with disastrous effects on the other species with which we share the planet.
We have learned how to control our environment, reaching a situation where the last areas of open wilderness will become little more than giant, safari-like parks as the human race strives for total control over nature.
Resources are sources of economic wealth. Human greed and the prospect of great riches coupled with the ‘herd’ effect can have a dramatic affect on the human psyche: during the great Gold Rush, fevered humans swarmed to mine the precious, rare earth metal– but is gold really more important than our global carbon store? You don’t need a lecture on the importance of rainforests. A day hardly passes without a TV program, newspaper or magazine article, telling us about the 10 million birds, mammals, reptiles, insects and plants that are under threat and how humans are destroying rainforests at about 1.5 acres a second, leaving behind large, barren deserts.
If deforestation occurs at its current rate, the world’s tropical forests will be gone in about 40 years and millions of unique life forms will be lost, including numerous vital medicines.
The rainforests near the center of the globe (in Africa, Australia, Asia, and Central and South America) consistently receive warm and wet weather throughout the year to provide thriving, yet competitive, environments for over half of the world’s plant and animal species.
The commercial carbon trade is an emerging market where both people and businesses are keen to buy up carbon offsets to reduce their carbon footprints and promote ‘green images’. The carbon market is currently booming with forestry offsets, especially popular because they represent a real, visible improvement. For example, it has been reported that the band Coldplay bought up 10,000 mango trees in India to offset the production of one of their albums.
With the advent of Google Earth and GIS software (ArcGIS, for example), it is now possible to map and monitor global carbon resources: scientists use a mixture of ground sampling, GIS software and mathematical methods to map areas of interest, such as Bangladesh and South Africa. Modeling annual growth increment in trees is very complex, depends on many factors and vary from site to site. This can be reduced to a problem of curve fitting for which artificial neural networks are ideally suited. Information that the neural network has collected during the training period is then used to generate predictions. Small pockets of forest can be chosen throughout a huge forest range and data–tree type, tree height and tree girth–is recorded through a process known as sampling. This information can be recorded over intervals of time, such as a year. A neural algorithm can then be trained by the data until the correct growth curve is obtained for each species of tree.
The forest can then be analyzed using Google Earth and ArcGIS software. Google Earth is a virtual globe, map, and geographical information program that maps the Earth by superposing images obtained from satellite imagery, aerial photography and GIS 3D globe. Esri’s ArcGIS is a geographic information system (GIS) for working with maps and geographic information. The system can be used to compile geographic data and analyze mapped information and the programs combined to superimpose a picture outlining different forest factors on the actual forest. Different areas are shaded according to tree type, size, age, etc. until the forest map looks like a huge, multicolored, patchwork quilt.
The neural algorithm can be used to forecast the future growth of trees in the different shaded areas; there are well-known equations relating the girth and height of a tree species to the carbon it contains. The ArcGIS map can then be used to predict an accurate estimate of the total future carbon content of the forest and so, the amount of carbon sequestered from the atmosphere is known.
Trees can affect both climate and weather: global temperature is positively related to the concentration of atmospheric carbon dioxide sequestered by forests. They also have a direct cooling effect on temperatures, absorbing heat from the sun which would otherwise be reflected back into the atmosphere. During transpiration, water is drawn up by the roots in the ground and evaporates through pores in the leaves increasing cloud cover. Both of these processes help turn down the global thermostat. If we were to lose the rainforest’s global wind, rainfall patterns could switch leading to droughts throughout the Americas as well as other parts of the world. In addition to modeling carbon sequestration in Bangladesh and South Africa, the same scientists are analyzing the radiation spectra emitted from the forests.
The radiation spectra relates to other factors (such as tree growth, tree type, and soil and canopy structure) and is modeled by neural networks. Once the algorithm is trained, future radiation scans of forested areas can be used to predict conditions in the forest; those predictions can be used to estimate future forest radiation emissions and their effects on the global temperature. The temperature of the Earth works similarly to an automated heating system. Like many biological processes, a self-regulating feedback mechanism keeps global conditions within narrow bounds. A small change could plunge the world into an ice age or heat it up, converting large areas into deserts.
The use of neural networks need not be limited to world forests. Some years ago, a study of sunspot cycles and climates showed that they can be used to predict heavy rains, flooding and subsequent disease outbreaks in East Africa. The sunspot cycle is now also being modeled using neural networks as its understanding could have profound implications for planning humanitarian aid to Africa.
Neural networks have allowed scientists to model many complicated, real-life situations, but they are very simple compared to the brain. The human brain is an extremely complex network of interconnecting neurons, consisting of millions of neurons, each with thousands of connections to other neurons. In some areas, such as the cerebellum, the neurons are arranged in regular patterns whereas other regions, the arrangement is more random. This degree of complexity cannot, at the present time, be reproduced by artificial neural networks, so they cannot exactly mimic the behavior of biological systems. Algorithms may be limited for now, but if China were to realize its current dream of designing and building a quantum computer, it could mean that artificial intelligence could one day surpass the human brain.
Joe recently helped develop the British Woodlands food webs educational simulation for Newbyte and is donating his share of The Last Tiger (available Amazon kindle) children’s fantasy novel profits to the Animals on the Edge conservation project.