SoLIM: An Effort Moving DSM into the Digital Era Overview and Recent Developments A-Xing Zhu 1,2 1 Department of Geography University of Wisconsin-Madison azhu@wisc.edu 2 Institute of Geographical Sciences and Natural Resources Research Chinese Academy of Sciences Global Soil Partnership: March 20-23, Rome, Italy
OUTLINE Overview Background Components Application and assessment Recent Developments Effective sampling New covariates Relaxing the traditional constraints Up to date software implementation Current Efforts (into the digital era) Theoretical: Overcoming the tradition constraints Computational: Overcoming the digital divides Observations (what might be next for GSP?)
Overview Background SoLIM stands for Soil-Land Inference Model An approach of using geographic information processing techniques and artificial intelligence techniques to predictively and digitally map soils under fuzzy logic at pixel level Originally, it was designed to overcome the limitations of traditional soil survey (manual and its variants): The polygonsoil type (area/class) model and manual delineation. A joint effort by the United States Department of Agriculture, University of Wisconsin-Madison, the Chinese Academy of Sciences
Components Overview The similarity model - overcoming the limitations of polygon-soil type model S ij (S ij1, S ij2,, S ijk,, S ijn ) i Zhu, A.X. 1997. Geoderma, Vol. 77, pp. 217-242. j
Components Overview Inference - overcoming the limitations of manual delineation and the like Local Experts Expertise Machine Learning Case-Based Reasoning Spatial Data Mining i j Knowledge on Soil and Environment Relationships S <= f ( E ) Inference (under fuzzy logic) Covariates: cl, pm, og, tp, G.I.S./R.S. Zhu, A.X. 1999, IJGIS; Zhu et al., 2001, SSSAJ; Qi and Zhu, 2003, IJGIS; Shi et al., 2004, SSSAJ; Qi et al., 2008, Cartography and GIS; More at solim.geography.wisc.edu
Overview Applications and Assessment Products - Basic output: Fuzzy membership maps Zhu et al., 1997, SSSAJ; Zhu et al., 2010, Geoderma More at solim.geography.wisc.edu
Overview Applications and Assessment Products - Basic output: Fuzzy membership maps Derived products and accuracy: Raster soil type maps Raster Soil Type (Series) Map
Accuracy of the Raster Soil Map Comparison between SoLIM and Soil Map against field data (Raffelson) Sample Size = 99 Overall In Complexes In Single SoLIM 83.8% 89% 81% Soil Map 66.7% 73% 61% Mismatches Correct Total Mismatches Percentage SoLIM 24 30 80% Soil Map 4 30 13%
Overview Applications and Assessment Products - Basic output: Fuzzy membership maps Derived products and accuracy: Raster soil type maps; Uncertainty Map Zhu, A.X. 1997, Photogrammetric Engineering & Remote Sensing, Vol. 63, pp. 1195-1202
Overview Applications and Assessment Products - Basic output: Fuzzy membership maps Derived products and accuracy: Raster soil type maps; Uncertainty map Soil property map A-Horizon Depth from SoLIM A-Horizon Depth from the Soil Map Zhu, A.X. 1997, Photogrammetric Engineering & Remote Sensing, Vol. 63, pp. 1195-1202
Depth Based on SoLIM vs. Depth from the Field Depth From the Soil Map vs. Depth from the Field Depth based on Similarity Vector (cm) 35 30 25 20 15 10 5 R 2 = 0.602 N = 33 Depth from Soil Map (cm) 35 30 25 20 15 10 5 R 2 = 0.436 N = 33 0 0 5 10 15 20 25 30 35 Observed Depth (cm) 0 0 5 10 15 20 25 30 35 Observed Depth (cm)
Overview Applications and Assessment Products - Basic output: Fuzzy membership maps Derived products and accuracy: Raster soil type maps; Uncertainty map Soil property map; Uncertainty map
Effective sampling Recent Developments What if no soil data (no soil field experts, no soil maps, no soil samples)? Sampling!!! Random or Regular? But the question is: Can we do smart sampling? Purposive sampling: Through spatial analysis sampling locations and sampling order are prioritized in such a way to make sampling more effective (fewer in number and integral from different campaigns). No. of Samples MAE RMSE Purposive sampling 7 0.82 1.05 Linear regression model 41 1.18 1.48 Zhu et al., 2010, Geoderma; Yang et al., 2012, IJGIS
Effective sampling New covariates Recent Developments Dynamic feedback patterns from remote sensing data Stage 1 Stimulate Feedbacks Stage 2 Capture and Characterize Feedbacks Stage 3 Extract relationships with soils Rainfall 降雨 Input Multi-temporal and multi-spectral dataset Discover Spectral-temporal response patterns Land surface 地表 Produce Obtain Identify Soil data Soil data Land surface dynamic 地表反馈 feedbacks Observe MODIS sensors MODIS 传感器 Relationships 响应模式差异 between response patterns 与土壤差异的关系 and soil types Zhu et al., 2010, SSSAJ; Liu et al., 2012, Geoderma
Recent Developments Effective sampling New covariates Dynamic feedback patterns from remote sensing data Fuzzy slope positions Summit Slope Valley Qin et al., 2009, Geomorphology; Qin et al., 2012, Geoderma
Recent Developments Effective sampling New covariates Relaxing the traditional constraints How to use ad-hoc samples (few in number and spatially biased samples) Individual Representativeness Approach Each sample is representative to some region in the feature space parent material samplek precipitation elevation
Organic matter content (top soil) Uncertainty Map No data 7 6 5 推测残差 Residual 4 3 2 1 0 0.00 0.05 0.10 0.15 0.20 0.25 Uncertainty 不确定性
Recent Developments Effective sampling New covariates Relaxing the traditional constraints Up to date software implementation SoLIMSolutions2010 contains most of the above developments and also a help manual. Available at solim.geography.wisc.edu
Current Efforts (into the digital era) Theoretical: Overcoming the tradition constraints Integration of: Soil scientist knowledge Legacy data (maps and sample points) Ad-hoc field samples Estimation of uncertainty Uncertainty guided sampling Uncertainty Progressive Mapping Order of Optimal Samples
Current Efforts (into the digital era) Theoretically: Overcoming the tradition constraints Computationally: Overcoming the digital divides Geospatial Analysis & Digital Soil Mapping Specialists Single Core Multiple Cores Computing Clusters Cloud Computing
Current Efforts (into the digital era) Intuitive Model Building Assisted Model Building Cyber Sharing EASY Geographic Computing To Use To Compute Platform (easy GC) H.P. Computing Enabled Complex Computing Enabled
easygc (prototype) - for non-specialists
Observations (what might be next for GSP?) Coordinated but distributed efforts with capacity building being the focus Distributed efforts: each member country responsible for its own country Coordinated: FAO has a mandate to do that (as part of its mission statement, I believe) Capacity building: Training of the new technology Development of easy to use technology
Thank your for your attention! Contact: A-Xing Zhu Department of Geography University of Wisconsin-Madison azhu@wisc.edu Web Site: solim.geography.wisc.edu