1 / 70

The role of students in Digital Soil Mapping in BC

The role of students in Digital Soil Mapping in BC. presented by Chuck Bulmer. Pacific Region Soil Science Society March 29 2014. Outline. Introduction Students and learners The need for soils information Then... Now... And in the future

taini
Download Presentation

The role of students in Digital Soil Mapping in BC

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. The role of students in Digital Soil Mapping in BC presented by Chuck Bulmer Pacific Region Soil Science Society March 29 2014

  2. Outline • Introduction • Students and learners • The need for soils information • Then... Now... And in the future • Digital soil maps as a way to meet the need for soils info • the nuts and bolts of making a digital map • Cool stuff we could be doing together

  3. Students and learners • Student (Wikepedia) • A student is a learner, or someone who attendsan educational institution. • In its widest use, student is used for anyone who is learning, including mid-career adults who are taking vocational education or returning to university • Community (Wikepedia) • Community can refer to a usually small, social unitof any size that shares common values. The term can also refer to the national community or international community. • Society (Wikepedia) • .... more broadly, a society may be illustrated as an economic, social, or industrial infrastructure

  4. Students and learners • Student (Wikepedia) • A student is a learner, or someone who attendsan educational institution. • In its widest use, student is used for anyone who is learning, including mid-career adults who are taking vocational education or returning to university • Community (Wikepedia) • Community can refer to a usually small, social unitof any size that shares common values. The term can also refer to the national community or international community. • Society (Wikepedia) • .... more broadly, a society may be illustrated as an economic, social, or industrial infrastructure

  5. Students and learners • The Soils Community • soils and natural resource students • researchers professors and instructors • Practitioners (gov’t industry private)

  6. Students and learners • The Soils Community • soils and natural resource students • researchers professors and instructors • Practitioners (gov’t industry private) Sectors Academia Government Industry / Private Forestry / Agr

  7. Students and learners • The Soils Community • soils and natural resource students • researchers professors and instructors • Practitioners (gov’t industry private) Reality Money Time Sectors Academia Government Industry / Private Forestry / Agr The rest of the world (ROW) Government priorities and initiatives Family Friends Pets “a good life”

  8. The (expanded) soils community Soil scientists have always brought a unique perspective to their work, And have been successful at branching into areas beyond soil science The internet

  9. The (expanded) soils community “Digital soil mapping is one wayto expand the reach of our soils community” Soil scientists have always brought a unique perspective to their work, And have been successful at branching into areas beyond soil science The internet

  10. The (expanded) soils community Soil scientists have always brought a unique perspective to their work, And have been successful at branching into areas beyond soil science “Digital soil mapping is one wayto expand the reach of our soils community” The internet

  11. The need for soils information • BC’s soil surveys • More than 65 years of soil survey work • Federal and Provincial Gov’t • Still provide a wealth of information on: • General description of the area • Parent materials, climate, topographyphysiography, vegetation • Soil development • Soil descriptions & Soil properties • Capability & Management • And of course,,, MAPS Photos: Soil Publications Archive, Agriculture and Agri-Food Canada, http://sis.agr.gc.ca/cansis/publications/index.html

  12. The need for soils information • GIS datasets (polygons) • CanSIS Canadian Soil Information System • TEIS (Terrestrial Ecosystem Information System) Soils of BC • TEIS Terrain Mapping http://www.env.gov.bc.ca/soils/ • Soil Landscapes of Canada Digitized toPolygons Database

  13. The need for soils information • GIS datasets (polygons) • CanSIS Canadian Soil Information System • TEIS (Terrestrial Ecosystem Information System) Soils of BC • TEIS Terrain Mapping http://www.env.gov.bc.ca/soils/ • Soil Landscapes of Canada Digitized toPolygons Database

  14. Filling the gaps: predictive mapping • … a series of raster maps for BC • 1 ‘layer’ each for MATL, DEVEL, DRAIN, DEPTH, TEXTURE, etc • 1ha digital elevation model (www.habc.org) • Step 1. Subdivide into 108 EcoDistricts • Stratify the province into areas with similar physiography and ecology • Each ED gets a separate training dataset, model and map • Step 2. Training datasets • ‘pure’ polygons from soil/terrain maps provide the known locations • topographic derivatives calculated in SAGA-GIS, plus climate, geology from haBC • Step 3. Modeling • build a RF model with the training data (randomForest package in R statistical software) • predict result for the entire map (display the results in QGIS) • And… a plan for continuous improvement • Open source software, collaboration, public access data • More detailed information can be quilted into the complete provincial product

  15. DSM: Step 1… Subdivide the province by EcoDistricts “relatively homogeneous biophysical and climatic conditions” Differentiated by: - regional landform - local surface form - permafrost - soil development - textural group - vegetation cover - land use - precipitation - temperature http://sis.agr.gc.ca/cansis/nsdb/ecostrat/hierarchy.html

  16. DSM: Step 1… a closer look at EcoDistricts “relatively homogeneous biophysical and climatic conditions” Okanagan Valley (1007) Thompson Plateau (1006) Shuswap Highland (1008) These areas serve as ‘rule sheds’ for modeling

  17. DSM: Step 1… ED_1007 Includes the Okanagan Valleyfrom OK Falls to Enderby

  18. DSM: Step 1… ED_1007 Includes the Okanagan Valleyfrom OK Falls to Enderby 1.1 create buffered boundary file

  19. DSM: Step 1… ED_1007 Includes the Okanagan Valleyfrom OK Falls to Enderby 1.1 create buffered boundary file 1.2 clip soils data to boundary

  20. DSM: Step 1… ED_1007 Includes the Okanagan Valleyfrom OK Falls to Enderby 1.1 create buffered boundary file 1.2 clip soils data to boundary

  21. DSM: Step 1… ED_1007 Includes the Okanagan Valleyfrom OK Falls to Enderby 1.1 create buffered boundary file 1.2 clip soils data to boundary 1.3 add topographic covariates … elevation

  22. DSM: Step 1… ED_1007 Includes the Okanagan Valleyfrom OK Falls to Enderby 1.1 create buffered boundary file 1.2 clip soils data to boundary 1.3 add topographic covariates … elevation … slope

  23. DSM: Step 1… ED_1007 Includes the Okanagan Valleyfrom OK Falls to Enderby 1.1 create buffered boundary file 1.2 clip soils data to boundary 1.3 add topographic covariates … elevation … slope … channel network base

  24. DSM: Step 1… ED_1007 Includes the Okanagan Valleyfrom OK Falls to Enderby 1.1 create buffered boundary file 1.2 clip soils data to boundary 1.3 add topographic covariates … elevation … slope … channel network base … valley bottom flatness

  25. DSM: Step 1… ED_1007 Includes the Okanagan Valleyfrom OK Falls to Enderby 1.1 create buffered boundary file 1.2 clip soils data to boundary 1.3 add topographic covariates … elevation … slope … channel network base … valley bottom flatness … topographic openness

  26. DSM: Step 1… ED_1007 Includes the Okanagan Valleyfrom OK Falls to Enderby 1.1 create buffered boundary file 1.2 clip soils data to boundary 1.3 add topographic covariates … elevation … slope … channel network base … valley bottom flatness … topographic openness … relative slope position … and 12 more.

  27. DSM: Step 1… ED_1007 Includes the Okanagan Valleyfrom OK Falls to Enderby 1.1 create buffered boundary file 1.2 clip soils data to boundary 1.3 add topographic covariates … elevation … slope … channel network base elev … valley bottom flatness … topographic openness … relative slope position 1.4 all data stored in directories with identical file structures to aid automated processing of multiple ed’s

  28. DSM: Step 2… Training data How do we develop a training dataset? Step 2.1 Identify locations where the soil type is known

  29. DSM: Step 2… Training data How do we develop a training dataset? Step 2.1 Identify locations where the soil type is known Step 2.2 Attach topographic and other attributes to those locations

  30. DSM: Step 2… Training data How do we develop a training dataset? Step 2.1 Identify locations where the soil type is known Step 2.2 Attach topographic and other attributes to those locations Step 2.3 Clean the dataset to facilitate modeling - decide which categories to model (remove the rest) - remove points that don’t represent the condition (apply constraints)

  31. DSM: Step 2… Training data How do we develop a training dataset? Step 2.1 Identify locations where the soil type is known Step 2.2 Attach topographic and other attributes to those locations Step 2.3 Clean the dataset to facilitate modeling - decide which categories to model (remove the rest) - remove points that don’t represent the condition (apply constraints) The training dataset controls the model output …. or: The ‘model’ is really just a mathematical representation of the training dataset

  32. DSM: Step 2… Training data Step 2.1:Identify locations where the soil type is known …. Okanagan Seamless dataset themed by material

  33. DSM: Step 2… Training data Step 2.1:Identify locations where the soil type is known …. Okanagan Seamless dataset themed by material Step 2.1.1: Select only ‘pure’ polygons

  34. DSM: Step 2… Training data Step 2.1:Identify locations where the soil type is known …. Okanagan Seamless datasetthemed by material Step 2.1.1: Select only ‘pure’ polygons

  35. DSM: Step 2… Training data Step 2.1:Identify locations where the soil type is known …. Okanagan Seamless datasetthemed by material Step 2.1.1: Select only ‘pure’ polygons ..close-up at Cosens Bay

  36. DSM: Step 2… Training data Step 2.1:Identify locations where the soil type is known …. Okanagan Seamless datasetthemed by material Step 2.1.1: Select only ‘pure’ polygons ..close-up at Cosens Bay Step 2.1.2:Establish point locations

  37. DSM: Step 2… Training data Step 2.2 Attach topographic and other attributes to those locations

  38. DSM: Step 3… Modeling with randomForest “in bag” 67% used for training randomForest is a “classifier”: It sorts the training dataset into classes based on the attributes. It uses multiple decision trees (ie it creates a ‘forest’ of decision trees) At each node of each tree, only a (randomly) selected subset of the attributes is available to make the split. Each decision node is based on computing the Gini index, which is a measure of diversity Gini = 1 if all classes occur equally Gini = 0 for a perfect split Training data with 10 classes18 attributes Gini = 0.9 6 attributes (random) Test splits based on each Keep the best split (l0west Gini) “out of bag” 33% used for testing .. Continue until all boxes contain only a single class

  39. DSM: Step 3… Modeling with randomForest Default value is to run 500 trees Unknown points are predicted by running them through all trees in the forest and the most common answer wins…

  40. DSM: Step 3… Modeling with randomForest Default value is to run 500 trees Unknown points are predicted by running them through all trees in the forest and the most common answer wins… Why randomForest? …. it works RF is an important method in “machine learning” because it performs well in a variety of modeling applications … medical … financial … environmental

  41. DSM: outputs Lower mainland .. Heung et al. 2014.Geoderma 214: 141–154

  42. DSM: outputs BC parent materials First draft

  43. DSM: outputs Reliability …. Kelowna seamless

  44. DSM: outputs Reliability …. Kelowna balanced model

  45. DSM: outputs Reliability …. Kelowna seamless

  46. DSM: outputs Reliability …. Kelowna constrained

  47. DSM: incorporating change ED_1007

  48. DSM: incorporating change ED_1007 + Soil landscapes

  49. DSM: incorporating change Kelowna + Soil landscapes

  50. DSM: incorporating change Kelowna + Soil landscapes Prepare individualmap for each soil lansdscapeSame techniqueSame dataNew dataImproved model

More Related