Disaggregation of legacy soil maps at regional scale in France. Comparison of DSMART and Random Forest

(1)

HAL Id: hal-02789531

https://hal.inrae.fr/hal-02789531

Submitted on 5 Jun 2020

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Distributed under a Creative Commons CC0 - Public Domain Dedication| 4.0 International License

Disaggregation of legacy soil maps at regional scale in France. Comparison of DSMART and Random Forest

Manon Caubet, Alexis Messant, Ghislain Girot, Blandine Lemercier, Guillaume Martelet, Christian Walter, Dominique Arrouays, Nicolas Saby,

Anne C Richer-De-Forges

To cite this version:

Manon Caubet, Alexis Messant, Ghislain Girot, Blandine Lemercier, Guillaume Martelet, et al.. Dis- aggregation of legacy soil maps at regional scale in France. Comparison of DSMART and Random Forest. workshop for Digital Soil Mapping and GlobalSoilMap, Mar 2019, Santiago, Chile. �hal- 02789531�

(2)

M APPING OF SOILS AND THEIR

PROPERTIES USING LEGACY SOIL MAPS

M. Caubet, A. Messant, G. Girot, B. Lemercier, G. Martelet, C. Walter, D. Arrouays, A.C. Richer-de-Forges, N.P.A. Saby

Joint workshop for Digital Soil Mapping and GlobalSoilMap 12-15 March 2019

Santiago

Comparison of DSMART and Random Forest

(3)

Context

• Cover major part of french territory

• Different scales 1:250 000 ; 1:50 000

• Rich source of information about soil

M. Caubet. et al., Disaggregation of legacy soil maps at regional scale in France ²

• Spatial support and extent is not always adapted to demand

Objective : Increase spatial resolution of soil maps and to map soil properties How ? We tested and compared two methods :

Soil polygon maps

But ..

disaggregation

scorpan regression Soil types map

M. Caubet. et al., Mapping of soils and their properties using legacy soil maps

Soil property map Soil polygon maps

(4)

3

60polygons 196soil types

Legacy soil map 435 soil profiles for calibration

1:50 000 soil maps 30 soil profiles for validation using probability sampling

+ 24 Covariates

_{4 800 km²}

60 polygons 196 soil types

Data available

(5)

4

Generation of virtual samples

Random affectation to

soil types

Model building with covariates

Soil map unit

Prediction per pixel

C5. Model

Summarize

DSMART Input

Validation

30 points Legacy soil map Composition of Soil

Map Units 24 covariates

Output

Odgers et al., 2014

435 point observations

Disaggregation with DSMART

(6)

5

Input

Validation

30 points

~ 4500 points

+

435 points

Random Forest

Precise soil map

Systematic virtual sampling

Add real observations

Model building

with covariates Prediction

Collard et al., 2014

435 point data

1:50 000

soil maps 24 covariates

Output

scorpan regression with Random Forest

(7)

6

Purity (%) Conf.Int – 90% Conc. RMSE

Dominant soil type map

33,33 [19,9 ; 46,8] 0,51 48,04

DSMART 33,33 [17,8 ; 48,8] 0,47 49,43

R.F. 26,67 [13,2 ; 40,1] 0,15 64,75

Map of soil types Derived soil depth map

Validation with 30 points based on probability sampling

Results

(8)

7

DSMART

scorpan regression Same performance than the original map

The covariates did not capture soil spatial variability

Intense computational load Efficient computational load Pedological context is a DSM challenge

new sensors ? Implementation of soil / landscapes

rules

New 50 000 map

Conclusion : Take home message

(9)

8