Скачать презентацию Self-organising maps for integrating data across multiple scales Скачать презентацию Self-organising maps for integrating data across multiple scales

943d869c49fadfd8c1eb05551daf91c2.ppt

  • Количество слайдов: 28

Self-organising maps for integrating data across multiple scales S Shanmuganathana, P Sallisa and J Self-organising maps for integrating data across multiple scales S Shanmuganathana, P Sallisa and J Buckeridgeb Environmental and Oceanic Sciences Centre Auckland University of Technology, New Zealand subana. [email protected] ac. nz b a

Summary of slides Global environmental issues n Need for improved human-environment relationship n n Summary of slides Global environmental issues n Need for improved human-environment relationship n n Conventional data analysis methods n n n drawbacks digital data Artificial neural networks (ANNs) n n Challenges in ecosystem understanding Need for new tools Self-organising maps (SOMs) SOMs in integrated analysis n n n local global scales across scales

Global environmental issues “We also know that 15 per cent of the world’s population Global environmental issues “We also know that 15 per cent of the world’s population accounts for 56 per cent of consumption and if everybody lived like they do we would need 2. 6 additional planets to support us all…Our assessments of the state of the environment suggest we will need to innovate a transformation to sustainable production and consumption patterns in the space of just one generation”. (Shrestha 2003)

Global environmental issues n THE STATE OF THE ENVIRONMENT: PAST, PRESENT, FUTURE - Hard Global environmental issues n THE STATE OF THE ENVIRONMENT: PAST, PRESENT, FUTURE - Hard Facts biodiversity loss n overexploitation of natural resources n deforestation … http: //www. unep. org/Geo/press. htm n

Deforestation Deforestation

Biodiversity can be managed n “We know enough about the distribution of species and Biodiversity can be managed n “We know enough about the distribution of species and ecosystems to ensure that the world's biodiversity is managed effectively…. Give nature half a chance, and it will take care of itself” http: //news. bbc. co. uk/2/hi/science/nature/2166306. stm

Grand challenges of this era areas of ‘highest priority’ for focus within environmental sciences Grand challenges of this era areas of ‘highest priority’ for focus within environmental sciences n n Biological Diversity and Ecosystem Functioning: Hydraulic Forecasting: Infectious Disease and the Environment: Land-Use Dynamics: interdisciplinary research n n involving ecologists, ethnologists, psychologists, engineers, economists, planners, landscape architects, and others. data needs, collection and synthesis of data requiring cooperation among physical, biological, and social scientists; engineers and planners; and other associated funding agencies. to protect biodiversity: Need for coordinated effort hydraulic modelling n human management institutions on ecosystems. n climate change in ecosystem n urban long-term ecological research sites. - Graedel et al. (2001) n

Need for new tools humans ---> extensive damage to the environment n sustainable environmental Need for new tools humans ---> extensive damage to the environment n sustainable environmental development - remote n lack of proper ecosystem understanding n n ecosystem functioning and biodiversity n n ecosystems are complex diverse major challenge (Graedel et al. 2001) new tools integrated approaches n cross scale n using dissimilar data sets n

Conventional methods n CM are complex rigorous, still unable to distinguish n n human Conventional methods n CM are complex rigorous, still unable to distinguish n n human induced from natural environmental variations Research of the 20 th century n n significantly contributing --- current environmental problems the research became focused in gaining in-depth knowledge with highly specialised scientific fields. encouraged a fragmented image of nature - Bowler (1992). responsible for altering the Earth’s natural flows and cycles n Digital data - experimentation of novel approaches n n n ANNs AI

Artificial neural networks Source: http: //hem. hj. se/~de 96 klda/Neural. Networks. htm Artificial neural networks Source: http: //hem. hj. se/~de 96 klda/Neural. Networks. htm

Conventional Vs ANNs n needs step by step details of the problem and solution Conventional Vs ANNs n needs step by step details of the problem and solution Solves laborious repetitive tasks applications n n n Mathematical calculations payroll, cannot solve real world problems i. e. image/ voice recognition n n stores information as patterns- based on animal brain /nerve cell structure and functioning Neurons, architecture, training and recall algorithm Train - input/ output Heuristics -> computing applications n n image/ voice recognition character recognition

Kohonen’s SOM (1982) A feed forward neural network with unsupervised training al. n Displays Kohonen’s SOM (1982) A feed forward neural network with unsupervised training al. n Displays the input vectors on a two D grid n Preserves the topology of the input vectors n Visual analysers, enables the detection of hidden patterns n Ideal data mining tool for knowledge extraction n Teuvo Kohonen - 1 st modelled the human brain’s cortex cells n Output layer Input layer Figure : Simplified diagrams of the human brain and a self-organising map (SOM)

SOMs in integrated analysis n n n New Zealand's annual gross domestic product (GDP) SOMs in integrated analysis n n n New Zealand's annual gross domestic product (GDP) composition, with household consumption and housing patterns, from 1988 to 2001. World Bank's rural development, GDP and biodiversity for year 1980 and 2000 Ozone hole with green houses from 1979 -2001

SOMs in national data analysis n n n Annual GDP composition household composition housing SOMs in national data analysis n n n Annual GDP composition household composition housing patters n n separate house Vs flat/ apartments cause indirect pressure on the environment

Annual GDP composition Agriculture 2. forestry and logging 3. fishing 4. mining 5. manufacturing, Annual GDP composition Agriculture 2. forestry and logging 3. fishing 4. mining 5. manufacturing, 6. electricity 7. gas and water supply 8. construction 9. transport and storage other services 10. Unallocated 1.

household composition n food and beverages clothing and footwear housing household goods and services household composition n food and beverages clothing and footwear housing household goods and services health and medical goods and services transport leisure and education n hotels and restaurants and other goods and services n n n

GDP, Household & housing Figure 1 a & b: SOM created with New Zealand's GDP, Household & housing Figure 1 a & b: SOM created with New Zealand's national GDP, household consumption and housing pattern data.

SOM results - national data n n n cluster 1 (1988 -1991), cluster 2 SOM results - national data n n n cluster 1 (1988 -1991), cluster 2 (19921996) and cluster 3 (1997 -2001). GDP from Agriculture, mining and other services increase manufacturing and unallocated categories decrease; unless measures to preserve habitat biodiversity is taken this is not a good trend household goods and services, and leisure and education increase. GDP through fishing, forestry and logging, education decrease. flat/ apartment dwelling and occupancy have increased.

SOM clusters SOM clusters

SOMs in global data analysis Growth of development variables n agriculture, n industry, n SOMs in global data analysis Growth of development variables n agriculture, n industry, n manufacturing and n services development Biodiversity Threatened mammals n T. birds n T. plants n Forests area (sqk) n

SOM clusters & components SOM clusters & components

SOM clusters - WB&WRI data SOM clusters - WB&WRI data

SOM clusters SOM clusters

SOM - GD analysis • In cluster 1, China and Indonesia • highest threatened SOM - GD analysis • In cluster 1, China and Indonesia • highest threatened species rates -mammals and birds 2000. • highest GDP both periods with high industry & manufacturing • % of average annual deforestation for (1990/2000) - their GDP growth / the expense of biodiversity loss.

SOMs-ozone depletion gases c 1 (1981 -84), c 2 (1985 -88), c 3 (1989 SOMs-ozone depletion gases c 1 (1981 -84), c 2 (1985 -88), c 3 (1989 -95 and 1999) and c 4 (1996 -98 and 2000 -01). ozone hole area/ minimum ozone not proportionate to the total gases CFC 12 and 113 - contribute more Methane gas - remarkable increase since 1985.

SOMs in prediction 1. Reduction in CFC 12 & CFC 113 (512. 9, 80. SOMs in prediction 1. Reduction in CFC 12 & CFC 113 (512. 9, 80. 88 total 2750) for 20002001 2. CFC 12 & CFC 113 for 1996 -1999 (525. 7 and 82. 06 total 2689) 3. All donot contribute the same towards O 3 depletion 4. The high 2000 -2001 total release might have caused from increased methane.

Acknowledgements n For details please contact subana. shanmuganathan@aut. ac. nz Acknowledgements n For details please contact subana. [email protected] ac. nz