Creation of high resolution soil parameter databy use of artificial neural network technologies (advangeo®)<br />A. Knoblo...
Available Data and Knowledge<br />Training Data: Aerial Images,<br />Field Observations<br />Data:<br />DEM, <br />soil ma...
Traditional prediction methods are based mainly on the expert´s knowledge / experience supported by modern information tec...
Modern Approach Using Artificial Intelligence<br />Pre-Processing<br />The artificial neuronal network “replaces” the expe...
Model: Neuron Cell<br /><ul><li>Functionality as a biological neural system
Consists of artificial neuron cells
Simulation of biological processes of neurons by use of suitable mathematical operations
In most cases layer-like configuration of the neurons</li></ul>The Neuron Cell as a Processor<br /><ul><li>Connection betw...
Enforce or reduce the level of the input information
Are directed, can be trained
Input signals
Re-computed to a single input information: the propagation function
Output signals
Activation function computes the output status of a neuron (often used: Sigmoid function) </li></ul>Definition: Artificial...
Principle Setup of Artificial Neural Networks<br />Network Topology: MLP (Multi Layer Perceptron)<br /><ul><li>Set-up of n...
Direction and degree of connections
Amount of hidden layers and neurons</li></li></ul><li>Training of Artificial Neural Networks<br />Learning Algorithm: Back...
Modification of weights w
Reduces error between expected and actual output of the network</li></ul>MSE - Curve<br />
Advantages / Disadvantages of Artificial Neural Networks<br />Advantages:<br /><ul><li>learnable: learning from examples
generalization: able to solve similar problems that have not been trained yet
universal: prediction, classification, pattern recognition
able to analyze complex, non-linear relationships
fault-tolerant against noisy data (e.g. face recognition)
quickness</li></ul>Additional characteristics:<br /><ul><li>choice of topology and training algorithm
black box system: evaluation of weight of parameters</li></li></ul><li>Software: advangeo <br />Referenced Data  Sources<b...
Documentation of Working Steps
Capture and Management of Metadata for Geodata
Tools for Data Pre-Processing, Post-Processing and Cartographic Presentation
Integration into Standard ESRI ArcGIS-Software </li></ul>GIS Extension <br />  Data- and Model Explorer<br />
Case Study 1: Extensive Soil Erosion<br />Input Data: Derivate of the Digital Elevation Model<br /> Slope<br />DEM Saxony...
Case Study 1: Extensive Soil Erosion<br />Input Data: Soil Map<br /> Fine Soil (Clay, Silt, Sand), Skeleton Soil<br />Soi...
Case Study 1: Extensive Soil Erosion<br />Input Data: Land use (ATKIS, Biotope Type Mapping)<br /> Forest, arable land, p...
Case Study 1: Extensive Soil Erosion<br />Training Data: Based on Aerial Images<br /> Mapping of Erosion Areas<br />Aeria...
Input<br />Data /<br />Layers<br />Training Data<br />Weights <br />Hidden<br />Layers<br />Known<br />Erosion<br />Areas<...
Re-Modeled Erosion Areas:<br />Training Areas<br />	ca. 80 % of known erosion areas with p>75%<br />Test Area:<br />	ca. 9...
Case Study 1: Extensive Soil Erosion<br />
Case Study 1: Extensive Soil Erosion<br />Validation of Prediction Results in the Field<br />
Upcoming SlideShare
Loading in …5
×

Creation of high resolution soil parameter data by use of artificial neural network technologies

840 views

Published on

Published in: Technology
0 Comments
2 Likes
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
840
On SlideShare
0
From Embeds
0
Number of Embeds
2
Actions
Shares
0
Downloads
7
Comments
0
Likes
2
Embeds 0
No embeds

No notes for slide
  • artificial neural networks. The principle of ANNs is based on the interaction of biological nerve cells in the
  • AK
  • AK
  • Creation of high resolution soil parameter data by use of artificial neural network technologies

    1. 1. Creation of high resolution soil parameter databy use of artificial neural network technologies (advangeo®)<br />A. Knobloch1, F. Schmidt1, M.K. Zeidler1, A. Barth1<br />1 Beak Consultants GmbH, Freiberg / Deutschland, postmaster@beak.de<br />
    2. 2. Available Data and Knowledge<br />Training Data: Aerial Images,<br />Field Observations<br />Data:<br />DEM, <br />soil map,<br />land use, <br />yield map <br />Knowledge: Natural Processes<br />
    3. 3. Traditional prediction methods are based mainly on the expert´s knowledge / experience supported by modern information technology<br />Data Analysis and Interpretation<br />Traditional Approach<br />
    4. 4. Modern Approach Using Artificial Intelligence<br />Pre-Processing<br />The artificial neuronal network “replaces” the experts empirical data analysis<br />Validation<br />
    5. 5. Model: Neuron Cell<br /><ul><li>Functionality as a biological neural system
    6. 6. Consists of artificial neuron cells
    7. 7. Simulation of biological processes of neurons by use of suitable mathematical operations
    8. 8. In most cases layer-like configuration of the neurons</li></ul>The Neuron Cell as a Processor<br /><ul><li>Connection between the neurons by weights w
    9. 9. Enforce or reduce the level of the input information
    10. 10. Are directed, can be trained
    11. 11. Input signals
    12. 12. Re-computed to a single input information: the propagation function
    13. 13. Output signals
    14. 14. Activation function computes the output status of a neuron (often used: Sigmoid function) </li></ul>Definition: Artificial Neural Network<br />
    15. 15. Principle Setup of Artificial Neural Networks<br />Network Topology: MLP (Multi Layer Perceptron)<br /><ul><li>Set-up of neurons in layers
    16. 16. Direction and degree of connections
    17. 17. Amount of hidden layers and neurons</li></li></ul><li>Training of Artificial Neural Networks<br />Learning Algorithm: Back-Propagation<br /><ul><li>Repeated input of training data
    18. 18. Modification of weights w
    19. 19. Reduces error between expected and actual output of the network</li></ul>MSE - Curve<br />
    20. 20. Advantages / Disadvantages of Artificial Neural Networks<br />Advantages:<br /><ul><li>learnable: learning from examples
    21. 21. generalization: able to solve similar problems that have not been trained yet
    22. 22. universal: prediction, classification, pattern recognition
    23. 23. able to analyze complex, non-linear relationships
    24. 24. fault-tolerant against noisy data (e.g. face recognition)
    25. 25. quickness</li></ul>Additional characteristics:<br /><ul><li>choice of topology and training algorithm
    26. 26. black box system: evaluation of weight of parameters</li></li></ul><li>Software: advangeo <br />Referenced Data Sources<br />Application <br />Metadata<br />Spatial<br />Data<br /><ul><li>Easy Access to Methods of Artificial Intelligence for Spatial Prediction
    27. 27. Documentation of Working Steps
    28. 28. Capture and Management of Metadata for Geodata
    29. 29. Tools for Data Pre-Processing, Post-Processing and Cartographic Presentation
    30. 30. Integration into Standard ESRI ArcGIS-Software </li></ul>GIS Extension <br /> Data- and Model Explorer<br />
    31. 31. Case Study 1: Extensive Soil Erosion<br />Input Data: Derivate of the Digital Elevation Model<br /> Slope<br />DEM Saxony 5m RESAMPLED<br />Slope [°]<br />Log Flow Accumulation<br />
    32. 32. Case Study 1: Extensive Soil Erosion<br />Input Data: Soil Map<br /> Fine Soil (Clay, Silt, Sand), Skeleton Soil<br />Soil Map<br />Fine Soil<br />(Clay, Silt, Sand), Skeleton Soil<br />
    33. 33. Case Study 1: Extensive Soil Erosion<br />Input Data: Land use (ATKIS, Biotope Type Mapping)<br /> Forest, arable land, pastures, wetland, etc.<br />Land use (ATKIS, Biotope Types)<br />Single raster for each land use class<br />(Forest, arable land, pastures, wetland., etc.)<br />
    34. 34. Case Study 1: Extensive Soil Erosion<br />Training Data: Based on Aerial Images<br /> Mapping of Erosion Areas<br />Aerial Images – intense rainfall/flood 2002<br />Mapped Erosion Areas<br />
    35. 35. Input<br />Data /<br />Layers<br />Training Data<br />Weights <br />Hidden<br />Layers<br />Known<br />Erosion<br />Areas<br />Weights<br />Output<br />Layer<br />Result:<br />Probability <br />Validation <br />Case Study 1: Extensive Soil Erosion<br />
    36. 36. Re-Modeled Erosion Areas:<br />Training Areas<br /> ca. 80 % of known erosion areas with p>75%<br />Test Area:<br /> ca. 90 % of known erosion areas with p>75%<br />Case Study 1: Extensive Soil Erosion<br />Input Data:<br />Slope, Silt, Clay, Sand, Landuse, Flow Accumulation<br />+ Horizontal curvature<br />Probability:<br />
    37. 37. Case Study 1: Extensive Soil Erosion<br />
    38. 38. Case Study 1: Extensive Soil Erosion<br />Validation of Prediction Results in the Field<br />
    39. 39. Optimization of Protection Measures<br />Alteration of Input Data:<br />DEM<br />Case Study 1: Extensive Soil Erosion<br />
    40. 40. Input Data<br />Elevation Model and its Derivates<br />Geology: Lithology, Foliation<br />Landuse<br />Soil Types<br />Training Data<br />Known Areas with Sliding / Creeping<br />Case Study 2: Soil Creeping<br />
    41. 41. B<br />A<br />Landslides<br />Dip of the Foliation to N<br />B<br />A<br />Case Study 2: Soil Creeping<br />
    42. 42. Input Data<br />Elevation Model and its Derivates<br />Geology: Lithology, Foliation<br />Landuse<br />Soil Types<br />Training Data<br />Known Areas with Gullies<br />Case Study 3: Erosion Gullies<br />
    43. 43. Input Data:<br />Elevation Model and its Derivates,<br />Soil Map,<br />Forest Inventory Data<br />Training Data:<br />Qualitative/ QuantitativeInfection Data<br />Investigation Area:<br />Forests in Eastern Erzgebirge,<br />Tharandter Wald<br />Case Study 4: Spread of Forest Pests<br />
    44. 44. Average Result from a Total of 15 Different Models<br />with Different Input Data<br />Training Data:<br />Infection Data 2008<br />Coded as:<br />Infected = 1<br />Rest= 0<br />Case Study 4: Spread of Forest Pests<br />No Endangerment<br />Low Endangerment<br />Medium Endangerment<br />High Endangerment<br />Very High Endangerment<br />
    45. 45. Average Result from a Total of 15 Different Models<br />with Different Input Data<br />Training Data:<br />Infection Data 2008<br />Coded as:<br />Infection Rate<br />Infection Rate<br />(m³/Infection Area)<br />Case Study 4: Spread of Forest Pests<br />
    46. 46. Case Study 5: Regionalization of Soil Parameters: Humidity Level<br />Input Data:<br />Elevation Model and its Derivates<br />Soil Map<br />Climate Data<br />Landuse<br />Training Data:<br />Sounding Rods / Auger<br />(1252 Points)<br /> feu1 – feu6<br />TK 5046, 5047, 5146, 5147<br />
    47. 47. Case Study 5: Regionalization of Soil Parameters: Humidity Level<br />Additional Input Data: Climate Data<br /> Evaporation, Rainfall, Relative Humidity<br />Relative Humidity [%]<br />Evaporation [mm]<br />Rainfall [mm]<br />
    48. 48. Case Study 5: Regionalization of Soil Parameters: Humidity Level<br />Training Data: 1252 Sounding Rods / Auger with Humidity Level<br /> feu1 – feu6<br />
    49. 49. Case Study 5: Regionalization of Soil Parameters: Humidity Level<br />
    50. 50. Case Study 5: Regionalization of Soil Parameters: Humidity Level<br />Humidity Level<br />Influence of Exposition (here: N-Hillside) and Climate (Rainfall Distribution)<br />
    51. 51. Case Study 5: Regionalization of Soil Parameters: Humidity Level<br />Humidity Level<br />Visible Gradient of Humidity Level within Biotope Types (here: Grassland)<br />
    52. 52. Case Study 6: Regionalization of Soil Parameters: Humus Level<br />Input Data:<br />Elevation Model and its Derivates<br />Soil Map<br />Climate Data<br />Landuse<br />Training Data:<br />Sounding Rods / Auger<br />(1725 Points)<br /> h0 – h7<br />TK 5046, 5047, 5146, 5147<br />
    53. 53. Case Study 6: Regionalization of Soil Parameters: Humus Level<br />Humus Level<br />
    54. 54. Case Study 7: Regionalization of Soil Parameters: TOC [%]<br />Input Data<br />Elevation Model and its Derivates<br />Soil Map<br />Landuse<br />Training Data:<br />Exploratory Soil Excavation<br />(38 Points)<br /> TOC [%]<br />TK 5046, 5047, 5146, 5147<br />
    55. 55. Case Study 7: Regionalization of Soil Parameters: TOC [%]<br />
    56. 56. Case Study 7: Regionalization of Soil Parameters: TOC [%]<br />Computed TOC<br />Known TOC<br />
    57. 57. Outlook: Vision of a „Rasterised Soilmap“<br />Humus<br />Humidity<br />TOC<br />Fine Soil<br />(Clay, Silt, Sand)<br />Fine Skeleton Soil<br />Coarse Skeleton Soil<br />VISION:<br />Raster Soil map<br />with separate raster layers for each parameter and gradient inside the original polygon<br />CURRENT:<br />Vector Soil map<br />with defined polygon boundaries with the same parameters inside a polygon (without gradient)<br />
    58. 58. Outlook: Application of Artificial Neural Networks in Precision Farming<br /><ul><li>Input Data:
    59. 59. Various soil data
    60. 60. Water balance data
    61. 61. DEM and derivations
    62. 62. Phenomena mapped from aerial images
    63. 63. ECa maps
    64. 64. Yield maps
    65. 65. Possible Results
    66. 66. Enhanced raster soil maps
    67. 67. Prediction of pests
    68. 68. All other spatial phenomena that are based on various controlling (spatial) factors</li></li></ul><li>Summary: Application of Artificial Neural Netwroks<br />Various applications are possible, e.g.:<br />Regionalization of soil parameters,<br />Time series analysis,<br />"Raster soil map“,<br />Analysis of influencing factors<br /><ul><li>We are looking forward to your comments, knowledge sharing and collaboration!</li></ul>andreas.knobloch@beak.de<br />frank.schmidt@beak.de<br /> www.advangeo.com<br />

    ×