Presentation is loading. Please wait.

Presentation is loading. Please wait.

Digital Soil Mapping: Past, Present and Future

Similar presentations


Presentation on theme: "Digital Soil Mapping: Past, Present and Future"— Presentation transcript:

1 Digital Soil Mapping: Past, Present and Future
Phillip R. Owens Associate Professor, Soil Geomorphology/Pedology

2 Digital Soil Mapping Also called predictive soil mapping.
Computer assisted production of soils and soil properties. Digital Soil Mapping makes extensive use of: (1) technological advances, including GPS receivers, field scanners, and remote sensing, and (2) computational advances, including geostatistical interpolation and inference algorithms, GIS, digital elevation model, and data mining

3 Digital Soil Mapping These techniques are simply tools to apply your knowledge of soil patterns and distributions. The maps can only be as good as your understanding of the soils and landscapes DSM - Same type of advancement to the Soil Survey as aerial photographs and stereoscopes introduced by Tom Bushnell and others early in the Survey.

4 Key Point It is impossible to use these products and create good maps if you do not know your soil-landscape relationship.

5 Opportunities Available soil data are increasingly numerical
Tools (GIS, Scanners, GPS,… Soil Data Models Increasing soil data harmonization The spatial infrastructures are growing DEMs: Global coverage Remote Sensing Web servers Quantitative mapping methods Geostatistics (pedometrics) Data mining Expert knowledge modeling

6 Models Essential tools of science Viewing and organizing thoughts
Conceptual Models – framework to ponder thoughts Simplify reality Must generate testable hypothesis to separate cause and effect New models must be advanced before facts can be viewed differently – break ruling theories

7 Dynamic Nature of Soils
Society perceives soils as static Pedologists deal with larger time scales – soils are dynamic Many soil forming factors are active at a site – but only a few will be dominant Importance of understanding soil dynamics- better predict results of management and evolution of soils

8 Types of Models Mental and Verbal – Most pedogenic models
Mathematical – Hope for the future Simulation – Knowledge of rate transfers

9 Energy Model (Runge, 1973) Similar to Jenny’s model, but emphasizes intensity factors of water (for leaching) and O.M. production S = f(o, w, t) where: W = water available for leaching (intensity factor) O = organic matter production (renewal factor) T = time

10 Energy Model (Runge, 1973) Many researchers continue to show that infiltrating water is a source of organizational pedogenic energy. Many critics say designed for unconsolidated P.M. with prairie vegetation.

11 Factors of Soil Formation
S = (p, c, o, r, t, …) (Jenny, 1941) Soils are determined by the influence of soil-forming factors on parent materials with time. Parent material Climate Organisms Relief Time

12 Functional Factorial Model (Jenny, 1941)
Good conceptual model, but not solvable Factors are interdependent, not independent Most often used in research by holding for factors constant – i.e. topo-, clino-, bio-, litho-, chronosequences Has had the most impact on pedologic research Divide landscapes into segments along vectors of state factors for better understanding

13 Functional Factorial Model (Jenny, 1941)
Climate and organisms are active factors Relief, parent material and time are passive factors, i.e. they are being acted on by active factors and pedogenic processes Model has the most utility in field mapping – may be viewed as a field solution to the model Very useful for DSM!

14 DEM Derived Terrain Attributes
These terrain attributes quantify the relief factor in Jenny’s Model Some of the most commonly used are: Slope; Altitude Above Channel Network; Valley Bottom Flatness; Topographic Wetness Index (TWI).

15 Paradigm Shift in Pedology
S = (s, c, o, r, p, a, n, …) (McBratney, 2003) Reformulation of Jenny 1941 Soil variability is understood as: Soil attributes measured at a specific point Climate Organisms Relief Parent material Age (time) Space Soils influence each other through spatial location! GIS

16 Paradigm Shift in Pedology
PCORT (Jenny, 1941) Emphasizes soil column vertical relationships Considers soils in relative isolation Descriptive terms used for landscapes (e.g. “noseslope”) SCORPAN (McBratney, 2003) Accounts for lateral relationships and movements Examines spatial relationships between adjacent soils Terrain attributes used to quantify landscapes (“topographical wetness index”) Catena – a “chain” of related soils (Milne, 1934) Have properties that are spatially related by hydropedologic processes (Runge’s Model)

17

18

19 Digital Elevation Model Dillon Creek, Dubois County, Indiana
This is the 5 m resolution DEM re-sampled from the 1.5 m Indian DEM. We are fortunate in Indian to have such a high resolution DEM, however the temptation to use the higher resolution need to be weighted carefully against processing time and power and too much information. So purpose and scale should always be the first guidance to selecting the right pixel size. This watershed is about 30 km2 and the 5 m dem seems to do well. Elevation m m

20 Aerial Photo draped over 3-d view
A careful examination of the aerial photography shows the mimicking of the depth to the limiting layer and slope by land use, with pastures located on flat ridge tops and valley bottoms and forested areas on side slopes. This is a validation from the user perspective, however as scientist we are also interested in validation of the predicted property by comparing the predicted soil properties with measured ones. I am still working on this.

21 Altitude Above Channel Dillon Creek, Dubois County, Indiana
Here are some examples of DEM derived terrain attributes that a soil scientist trained to read landscapes and identify soils with landscape positions would see the benefit. As we will see later there are certain soils that have very high values for AACH. AACH

22 Topographic Wetness Index Dillon Creek, Dubois County, Indiana
However, one terrain attribute alone may not be adequate to distinguish two soils that are share adjecent similar landscape positions like Gilpin and Wellston (referring back to the OSD) , so the other terrain attributes are used. TWI shows the potential of the certain landscape positions to accumulate water. It has been very useful for northern Indiana glaciated landscapes but not so much for the southern ones perhaps do to the high gradient slopes characteristic for these older landscapes compared to the relatively new ones in northern Indiana. TWI

23 Multi Resolution Ridge Top Flatness
Dillon Creek, Dubois County, Indiana We found so far that for the southern Indiana landscapes the MRRTF and MRVBF are better attributes for separating soils on the landscape. It works very well for these landscapes due to the more pronounced boundaries between ridgetops, side slopes and flood plains. So while MRRTF separates soils of ridgetop the MRVBF separates better the ones on the flood plaines. MRRTF

24 Multi Resolution Valley Bottom Flatness
Dillon Creek, Dubois County, Indiana MRVBF clearly shows the flood plains in a better way that TWI. What is useful about these Terrain attributes is the values that they all come with, values that can now be used to numerically quantify the soil landscape relationships. This is similar to the traditional techniques used in the past by the soil survey meaning stereoscopes, aerial photography and 7.5 minute USGS topographic images, besides the field works that will still remain part of the soil survey. MRVBF

25 Numerical Soil-Landscape Relationships, Indiana Site
No Soil Series MRRTF MRVBF Slope AACH TWI_29 1 Tilsit, Bedford, Apallona, Johnsburg (TBAJ) > 2.4 < 2.9 < 2 2 Tilsit, Bedford, Apallona (TBA) 2-6 3 Zanesville, Apallona, Wellston (ZAW) 6-12 4 Gilpin, Wellston, Adyeville, Ebal (GWAE) < 2.4 12-18 5 Gilpin, Ebal, Berks (GEB) 18-50 6 Pekin , Bartle (PB) > 2.0 2-12 7 Cuba, (C) > 2.9 0-2 > 0.09 < 12 8 Steff, Stendal, Burnside, Wakeland (SSBW) 0-1 <0.09 > 12 9 Rock Outcrops, Steep Slopes > 50 After going through several terrain attributes certain threshold values are identified as the ones to quantitatively separate soils or group of soils on the landscape. One of the issues that we had to address in this watershed was the presence of two counties and for those familiar with the soils survey it means we had to deal with the Soil County lines where soils often change abruptly.

26 Hardened SoLIM Map SOLIM map
Finally the last step in SoLIM is hardening the map which means assigning a fuzzy membership value for each soils at each pixel based on the followed rules. So far we have only demonstrated the power of combination between expert tacit knowledge and high resolution data and powerful geospatial tools. The map is far from perfect and as other digital mappers have realized it does not look as pretty as the colorplath maps with smooth lines (ironic because there are no lines in nature, everything is fuzzy depending on the scale). However, the current digital map has some very potential uses, especially for soil property maps. And this leads us to the last step the property map.

27 Dillion Creek – Dubois County Indiana Depth to Limiting Layer
As an example, we generated a map of the depth to the limiting layer (bedrock, weathered shale and/or sandstone) for the Dillon Creek. The property map is showing a characteristic pattern that is best captured at the landscape scale and describes the energy of the system to transport materials from higher elevations. The Dillon watershed represents an old landscape once covered in loess from different after several glaciers melts. With time the loess has moved off the slopes and deposited on the floodplains. In addition, loess caps are still present on relatively flat ridgetops for lack of energy to transport them further down the slope. cm

28 Low relief Landscape in the Glaciated Portion of Indiana

29 Slope Slope in Radians

30 Altitude above channel network (m)
Olaf Conrad 2005 methodology

31 Multi-resolution index of valley-bottom flatness
Valley Bottom Flattness Gallant, J.C., Dowling, T.I. (2003): 'A multiresolution index of valley bottom flatness for mapping depositional areas', Water Resources Research, 39/12:

32 TWI: 9 Topographic Wetness Index

33 Soils in Howard County 5 soils cover 80% of the land on Howard County
Are there relationships between these 5 soils and terrain attributes? Can we use those relationships to improve the survey in an update context? Provide predicted properties?

34 Shaded Relief Elevation Model, 242 to 248 meters
Wetness Index, 8 to 20 Slope, 0 to 4% SSURGO Brookston Fincastle

35 Frequency distributions
Terrain attribute: Altitude above channel network Terrain attribute: Curvature Frequency Frequency Fincastle Brookston Fincastle Frequency Brookston ABCN Curvature *Data extracted with Knowledge Miner Software

36 Frequency, Wetness Index
Terrain attribute: Wetness Index Fincastle Brookston Frequency Wetness index *Data extracted with Knowledge Miner Software

37 Formalize the Relationship
Example: If the TWI = 14 then assign Brookston If TWI = 10 then assign Fincastle Other related terrain attributes (or other spatial data with unique numbers) can be used. That provides a membership probability to each pixel

38 Terrain-Soil Matching for Brookston
Fuzzy membership values (from 0 to 100%) 2% 100% *Information derived from Soil landscape Interface Model (SoLIM)

39 Terrain-Soil Matching for Fincastle
Fuzzy membership values (from 0 to 100%) 97% 5% *Information derived from Soil landscape Interface Model (SoLIM)

40 Create Property Map with SoLIM
To estimate the soil property SoLIM uses: We already have Skij – the fuzzy membership value used to make the hardened soil map. So we only need to specify Dk, the representative values of the property of interest for each soil Dij: the estimated soil property value at (i, j); Skij: the fuzzy membership value for kth soil at (i, j); Dk: the representative property value for kth soil. In this case, let’s assign values to carbonate depth for Fincastle and Brookston in the east section of the county. Fincastle: 100 cm (low range of OSD) Brookston: 170 cm (high range of OSD)

41 Predicted depth to carbonates
100 to 170 cm 100 to 170 cm

42 Fuzzy vs. Crisp Soil Maps
Imagine a heap of sand… The Heap Paradox from 4th Century BCE, more than 2,000 years ago posed a problem that can be addressed by fuzzy logic Take away 1 sand grain. Is it still a heap? Take away 1 more and keep doing it. When is it not a heap? And what is it? Is it a pile, a mound? How many grains of sand does a mound have, a pile, a heap? Fuzzy logic offers a continuous spectrum of logical states represented in the unit interval of real numbers [0,1]—it is a many-valued logic with infinitely-many truth-values, and thus the sand moves smoothly from "definitely heap" to "definitely not heap", with shades in the intermediate region. Fuzzy hedges are used to divide the continuum into regions corresponding to classes like definitely heap, mostly heap, partly heap, slightly heap, and not heap. (from Wikipedia)

43 Heap of Sand vs. Pile of Sand
How many grains of sand do you need to remove from a heap to get a pile? How many grains of sand do you need to add to make your pile of sand into a heap?

44 Fuzzy vs. Crisp Soil Maps
Fuzzy logic says that when you keep taking grains of sand away eventually you move from definitely heap, to mostly heap, partly heap, slightly heap, and not heap. You can express heapness with values from 0 to 1, with 1 being a perfect example of a heap and 0 being nothing at all like a heap. How can we define a heap? It is a similar question to how can we define a mapping unit. You can set rules like a perfect heap is 2 tons or more of sand and not heap is less than ½ a ton of sand. You might also want an upper limit to where you say that after a certain amount it becomes more of a dune or mountain than a heap. You can then set a mathematical curve for expressing the decline in heapness as a function of the removal of sand grains. Fuzzy logic offers a continuous spectrum of logical states represented in the unit interval of real numbers [0,1]—it is a many-valued logic with infinitely-many truth-values, and thus the sand moves smoothly from "definitely heap" to "definitely not heap", with shades in the intermediate region. Fuzzy hedges are used to divide the continuum into regions corresponding to classes like definitely heap, mostly heap, partly heap, slightly heap, and not heap. (from Wikipedia)

45 Crisp vs. Fuzzy Soil Maps
Black is Brookston in the map below Brown is a different soil, but similar to Brookston. Orange is very different from Brookston and dark green is fairly different. As we move away from Brookston in geographic space we cross a threshold and suddenly we are in a different soil. There is an abrupt conceptual change from one soil to another. Black is Brookston in the map below Orange is soil very different from Brookston. Here we can express Brookston as values between 1 and 0 A given spot might have a 0.7 Brookston membership value As we move up in elevation that membership value may decrease to 0.5, 0.3, 0.1, and 0 when we know we won’t find Brookston

46 Brief History Of Digital Soil Mapping
: publications of pioneer works 2003: Digital Soil Mapping as a body of soil science 2004: 1st International workshop on Digital Soil Mapping. Workshops: Rio (2006), Logan (2008), Rome (2010), Sydney (2012) 2009: GlobalSoilMap.net

47 SoLIM in the US SoLIM “soil landscape inference model” was developed at the University of Wisconsin by A-Xing Zhu and Jim Burt (late 90’s) Knowledge based inference model, fuzzy logic, rule based reasoning. What does that mean? There were Soil Survey pilot projects in Wisconsin and the Smoky Mountains

48 S <= f ( E ) Challenges in Conducting Soil Survey
Knowledge Documentation The Polygon-based Model S <= f ( E ) Soil-Landscape Model Building Photo Interpretation Manual Delineation Polygon Maps The Manual Mapping Process There are three major challenges in this way of soil survey. The first is the polygon-based model used in soil maps on which only soil bodies of certain size (scale dependent) are shown and small soil bodies are not shown on the map. So the level of details is limited by the scale of the map, not by what the soil scientists know. Also, the soils in a given soil polygon are often treated as the same and changes of soil only occur at the boundaries of polygons. The second challenge is the manual mapping process which is not only tedious and time consuming, but also error prone and inconsistent. In addition, it is very difficult for soil mappers to identify soil-landscape units using more than three different environmental data layers due to the limited capability of visually perceiving many variables simultaneously. As a result, the delineation of soil-landscape units may not be exhaustive as soil mappers hope. In fact, most of soil mappers base their soil unit delineation on visual interpretation of stereo photos. Subtle and gradual changes in environmental conditions are often difficult to be discerned via stereoscoping. It is easy to misplace the boundaries of soil polygons in the manual delineation process. The third challenges is the documentation of a soil-landscape model for a given area. The issues are: 1) to what level the soil-landscape model being documented; 2) how much experience (knowledge) for a given area being passed from one generation of soil mappers to another. In most cases, the knowledge on the soil-landscape model of an area is lost when the soil mapper retires or moves out of the area. The new soil mapper has to start from scratch. (Slide from Zhu)

49 Relationships between Soil and
Overcoming the Manual Mapping Process Local Experts’ Expertise Artificial Neural Network Case-Based Reasoning Data Mining Relationships between Soil and Its Environment Spatial Distribution Similarity Maps S <= f ( E ) Inference (under fuzzy logic) Perceived as Cl, Pm, Og, Tp G.I.S. Overcoming the manual mapping process through the use of GIS techniques and an automated inference scheme. This inference technique determine the similarity vector for the soil at each pixel position. (Zhu., 1997, Geoderma; Zhu, 2000, Water Resources Research)

50 Valton Lamoile Elbaville Dorerton Churchtown Greenridge Urne Norden Gaphill Rockbluff Boone Elevasil Hixton Council Kickapoo Orion

51 The Speed of Soil Survey Using SoLIM
Overall A total of 500,499 acres since May 2001 over 526 person days, about 950 acres per person per day The product is already in digital form, no need to digitize it Currently the speed of manual mapping (including Compilation and digitization) is about acres per person per day We have done a number of testing to see the accuracy of the products. All of these tests show that the quality of the products is comparable to that achieved under the research settings (eighty some percent). The next few slides will focus on the speed and the cost of soil survey using SoLIM. (Slide from Zhu)

52 Inferred vs. Field Observed
Quality of Results: Inferred vs. Field Observed Correct Incorrect Accuracy Blue Mounds NE Cross Plain SW 34 22 4 6 89% 78% Watershed24 31 9 77% (Slide from Zhu)

53 Cost Comparison Cost about $1.5 million to complete field mapping of
the County using the manual approach Cost about $0.5 million using the SoLIM approach in digital form (Slide from Zhu)

54 SoLIM There were major advances in DSM using SoLIM.
Some minor setbacks – Smoky Mountain project “If a guy who has mapped these mountains for 20 years can’t tell you what soil is on the other side of the hill, then you can’t use a computer to do it.” Bill Craddock, Former State Soil Scientist in Kentucky

55 DSM – Current State There are many options under the umbrella of DSM: geostatistics (kriging and co-kriging), clustering, decision trees, Bayesian models, and fuzzy logic with expert knowledge. There are advantages and disadvantages to all methods.

56 DSM – Current State Knowledge based inference model like ArcSIE and SoLIM allows soil scientists to utilize their understanding of soil landscape patterns Requires less data but knowledgeable soil scientists ArcSIE is easier to use because it is within ArcGIS. SoLIM requires multiple file transfers

57 DSM Current State ArcSIE used successfully in initial soil surveys in Missouri, Vermont and Texas Requires environmental covariates and depends heavily on the DEM, terrain attributes and remote sensing (in the dry climates) Explicitly describes Jenny’s state factor model by the expansion through McBratney’s SCORPAN

58 DSM - Future DSM will be instrumental in soil survey updates. Research is currently underway to determine best methods Digital delivery gives us the ability to illustrate and deliver soils in new formats (example Isee - Using the fundamentals of DSM, we can move towards predicting soil properties and incorporating other explanatory data (i.e. ecologic site descriptions, land use, etc.)

59 Dillion Creek – Dubois County Indiana Depth to Limiting Layer
As an example, we generated a map of the depth to the limiting layer (bedrock, weathered shale and/or sandstone) for the Dillon Creek. The property map is showing a characteristic pattern that is best captured at the landscape scale and describes the energy of the system to transport materials from higher elevations. The Dillon watershed represents an old landscape once covered in loess from different after several glaciers melts. With time the loess has moved off the slopes and deposited on the floodplains. In addition, loess caps are still present on relatively flat ridgetops for lack of energy to transport them further down the slope. cm

60 “Pros” to Digital Soil Mapping
Very consistent product due to the way it is created. The soil landscape model is explicit. Updates can be completed more efficiently over large areas. The variability or inclusions can be represented (in some cases)

61 “Pros” to Digital Soil Mapping
End users in the non traditional areas can more easily use some products. We can use this information to make predictions of soil properties including dynamic soil properties. All of these “pros” will increase the support and usefulness of the Soil Survey in the future.

62 “Cons” to Digital Soil Mapping
In some locations, the soil-landscape relationship is difficult to determine and represent. Examples are areas with heterogeneous parent materials. Can be misused (It makes really pretty maps and a bad map is worse than no map at all) Complications with data can stop a project. Learning new softwares can be very frustrating

63 Saturated hydraulic conductivity (ksat , micrometers per second) from gridded SSURGO (Approximately 1:24,000 map. Gridded at 30 m resolution with STATSGO). 600

64 DSM Future Harmonize the soil data Disaggregate polygons
Create true DSM maps tied to landscapes Provide alternate raster products at multiple resolutions We must embrace and use this technology and incorporate DSM into the long-term plan/vision.


Download ppt "Digital Soil Mapping: Past, Present and Future"

Similar presentations


Ads by Google