Accès libre

Interpretative Machine Learning as a Key in Recognizing the Variability of Lakes Trophy Patterns

Jarosław Jasiewicz
Jarosław Jasiewicz
, 
Izabela Zawiska
Izabela Zawiska
, 
Monika Rzodkiewicz
Monika Rzodkiewicz
 et 
Michał Woszczyk
Michał Woszczyk
   | 01 avr. 2022
Quaestiones Geographicae's Cover Image
À propos de cet article
Article précédent
Article suivant
Citez
Publié en ligne: 01 avr. 2022
Pages: 127 - 146
Reçu: 14 févr. 2022
DOI: https://doi.org/10.2478/quageo-2022-0009
Mots clés
© 2022 Jarosław Jasiewicz et al., published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.

Fig. 1

Location of lakes and extent of the watershed on the study area.
Location of lakes and extent of the watershed on the study area.

Fig. 2

Relations between Ptot and selected explanatory variables, see Table 1 for details.
Relations between Ptot and selected explanatory variables, see Table 1 for details.

Fig. 3

The outline of methodology. See Section ‘Methods’ for details.
The outline of methodology. See Section ‘Methods’ for details.

Fig. 4

The concept of mapping variable values into influence. The size of dots simulates the values of the dependent variable.
The concept of mapping variable values into influence. The size of dots simulates the values of the dependent variable.

Fig. 5

Relation between actual values of dependent variables and the outcome of the model Ptot.
Relation between actual values of dependent variables and the outcome of the model Ptot.

Fig. 6

Relation between clusters and variable influence. The red-white-blue gradient denotes the influence of the given variable. The colours marking the clusters are used in the same way in the other figures.
Relation between clusters and variable influence. The red-white-blue gradient denotes the influence of the given variable. The colours marking the clusters are used in the same way in the other figures.

Fig. 7

Variation of Ptot in clusters. X-labels denote the name of the class: High trophy, Moderate trophy (Mod.-I, Mod.-II, Mod.-III), and Low trophy.
Variation of Ptot in clusters. X-labels denote the name of the class: High trophy, Moderate trophy (Mod.-I, Mod.-II, Mod.-III), and Low trophy.

Fig. 8

The influence plots for each variable. The X-axis contains the original values of the variable, Y-axis contains the influence. Colours in legend denote clusters (see Fig. 7), size of dots value of Ptot. Partial Dependency Plot (PDP) is marked by a light grey line.
The influence plots for each variable. The X-axis contains the original values of the variable, Y-axis contains the influence. Colours in legend denote clusters (see Fig. 7), size of dots value of Ptot. Partial Dependency Plot (PDP) is marked by a light grey line.

Fig. 9

The SHAP (SHapley Additive exPlanations) plots (Lundberg, Lee 2017) for the five most representative lakes for each class. X-axis presents influence in units of Ptot standard deviation scale is the same for each subplot, but ranges are different. Length and direction of arrows denote the scale and influence orientation (negative or positive) brought by a given variable.
The SHAP (SHapley Additive exPlanations) plots (Lundberg, Lee 2017) for the five most representative lakes for each class. X-axis presents influence in units of Ptot standard deviation scale is the same for each subplot, but ranges are different. Length and direction of arrows denote the scale and influence orientation (negative or positive) brought by a given variable.

Fig. 10

The variable importance estimated using multiple linear regression (ElasticNet) and random forest. See text for details.
The variable importance estimated using multiple linear regression (ElasticNet) and random forest. See text for details.

Fig. 11

Visualisation of dissimilarities between lakes in a form of MDS (multidimensional scaling). The axes of the plot have no units. Lakes morphometry presents dissimilarity for a group of morphometric features (see Table 1); Land cover presents dissimilarity for four land cover variables (urbanised, agriculture, forests, wetlands); ‘All variables’ plate uses all 25 variables; Influence presents dissimilarity between lakes in a space of variables influence. For colours see Figure 7.
Visualisation of dissimilarities between lakes in a form of MDS (multidimensional scaling). The axes of the plot have no units. Lakes morphometry presents dissimilarity for a group of morphometric features (see Table 1); Land cover presents dissimilarity for four land cover variables (urbanised, agriculture, forests, wetlands); ‘All variables’ plate uses all 25 variables; Influence presents dissimilarity between lakes in a space of variables influence. For colours see Figure 7.

Explanatory variables used in the study.

VariableAbbreviationUnitSource
ElevationELEVm a.s.l.Jańczak 1999
Lake areaLAREhaJańczak 1999
Lake capacityLCAPkm3Jańczak 1999
Lake max depthLDMXmJańczak 1999
Lake average depthLDAVmJańczak 1999
Lake max lengthLLENmJańczak 1999
Lake max widthLWIDmJańczak 1999
Lake shoreline lengthPRIMmJańczak 1999
Lake elongationLELNRatioLLEN / LW ID
Lake capacity/length ratioLVARRatioLCAP / LLEN
Lake perim developmentLPDVRatioLLEN / sqrt (2 × π × LARE)
Lake expositionLEXPRatioLARE / LDAV
Watershed areaWAREhaCalculated
Mean slopeWSLP%Calculated
Height stddevWHSDmCalculated
UrbanisedWURB%Calculated
AgricultureWAGR%Calculated
ForestsWFRS%Calculated
WetlandsWWET%Calculated
SandsWSND%Calculated
TillsWTLS%Calculated
ClayWCLS%Calculated
OrganicWORG%Calculated
Schindler ratioSRRatioWARE / LCAP
Ohle ratioORRatioWARE / LARE
eISSN:
2081-6383
Langue:
Anglais
Périodicité:
4 fois par an
Sujets de la revue:
Geosciences, Geography
RSS Feed de la revue