modelling and mapping global soil information · why modelling at global scale ... modified from de...
TRANSCRIPT
![Page 1: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/1.jpg)
Modelling and mapping global soil information
Laura Poggio, Luis Moreira de Sousa, Gerard Heuvelink, Bas Kempen, Zhanguo Bai, Ulan Turdukulov, Maria
Ruiperez-Gonzalez, Eloi Ribeiro, Niels Batjes, Rik van den Bosch
![Page 2: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/2.jpg)
Why modelling at global scale?
![Page 3: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/3.jpg)
Examples of Global SOC maps
![Page 4: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/4.jpg)
Challenges● Data
● Covariates
● Modelling
![Page 5: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/5.jpg)
Challenges● Quality control
● Computational resources
![Page 6: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/6.jpg)
SoilGrids 2.0
Uncertainties
Modified from de Sousa et al, 2020
![Page 7: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/7.jpg)
Modified from Ribeiro,Batjes, 2019
Data
![Page 8: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/8.jpg)
Model tuning for global modelling
● Spatially balanced 10 fold cross-validation
● Covariates selection for parsimonious model
● Model tuning: performances and uncertainty
![Page 9: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/9.jpg)
Covariates
![Page 10: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/10.jpg)
Tuning: parsimonious modelCovariates selection:
- De-correlation
- Successive removal by performances
maximisation
- Covariates grouping
![Page 11: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/11.jpg)
Covariates RFERecursive feature elimination:
- RandomForest (machine learning) broad equivalent of
backward stepwise regression
- Optimised for large sets of covariates
![Page 12: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/12.jpg)
Covariates RFE
![Page 13: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/13.jpg)
Covariates RFE
![Page 14: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/14.jpg)
Covariates reduction
![Page 15: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/15.jpg)
Covariates reduction
![Page 16: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/16.jpg)
Parsimonious modelLess covariates similar performances
![Page 17: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/17.jpg)
Tuning: hyperparameters1. Covariates sub-set
2. Model depending ---> QuantilesRandomForest (in this
example)
![Page 18: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/18.jpg)
Tuning - hyperparameters
![Page 19: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/19.jpg)
MTRY
Wid
th c
onfid
ence
inte
rval
s
Tuning - Uncertainty and intervals
![Page 20: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/20.jpg)
Validation points included in 90% intervals
Correlation 0.928
Correlation + RFE 0.927
Correlation + RFE + Grouping 0.923
Tuning - Uncertainty and intervals
![Page 21: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/21.jpg)
Open issues● Depth:
○ 2D,2.5D,3D approach?
○ Spline?
○ Depth as covariate?
○ Something else?
![Page 22: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/22.jpg)
Open issues● Model:
○ Is ML (especially RandomForest) the best
approach?
○ Are the predictions intervals created realistic?
○ Integration with national/regional
models/results?
![Page 23: Modelling and mapping global soil information · Why modelling at global scale ... Modified from de Sousa et al, 2020. Modified from Ribeiro,Batjes, 2019 Data. Model tuning for global](https://reader033.vdocuments.site/reader033/viewer/2022050406/5f8369a6f5d450337270e677/html5/thumbnails/23.jpg)
Concluding remarks● Quality controlled and standardised input data
● Spatially aware cross-validation
● Parsimonious models
● Uncertainty assessment
● Open questions!