Head-cut gully erosion susceptibility modelling based on ensemble Random Forest with oblique decision trees in Fareghan watershed, Iran

Figures & data

Figure 1. Location of the study area in Fareghan watershed, Hormozgan province, Iran, as representative of the hot and dry climatic conditions. Source: Author.

Figure 2. View of the gully erosion in the Fareghan watershed: (a) gully erosion in agriculture area, (b) measurement of morphometric characteristics of gully erosion, and (c) gully erosion in rangeland area.Source: Author.

Figure 3. Methodological flow chart. Source: Author.

Table 1. Result of the screen test based on deriving sensitivity, specificity, and positive and negative predictive values (Banoo et al. 2007).

	Has the condition	Does not the condition
Positve	a True positive	b False positive	Row entries for determining PPV
Negative	c False negative	d True negative	Row entries for determining NPV
	Columns entries for determining sensitivity	Columns entries for determining specificity

Table 2. Physical and chemical properties of soil in the gullies of Fareghan watershed.

Features	Soil depth (cm)
	0–30	30–75	75–130	130–180	180–250	250–290
PH	7.9	8.2	8.1	8.4	7.7	7.9
EC (mmhos/cm)	67.3	35.1	29.6	55	61.7	49.6
Na (Meq/lit)	434	211	185	397	419	220
Ca + Mg (Meq/lit)	200	126	100	150	160	230
SAR	43.39	26.57	26.16	45.83	46.83	20.512
Clay (%)	72	70	64	62	17	72
Silt (%)	19	17	35	33	80	21
Sand (%)	9	13	1	5	3	7
Soil texture	Silt-Loam	Silt-Loam	Silt-Loam	Silt-Loam	Sand-Loam	Silt-Loam

Table 3. Multi-colinearity analysis to determine the linearity of the independent variables.

Variables	VIF	Tolerance
Altitude	4.11	0.24
Slope	2.98	0.34
Aspect	1.08	0.93
Plan curvature	3.12	0.32
Profile curvature	1.96	0.51
Distance from river	2.94	0.34
Drainage density	2.44	0.41
Distance from road	3.05	0.33
Rainfall	2.23	0.45
Land use	3.91	0.26
Lithology	1.19	0.84
Soil	2.13	0.47
TPI	1.08	0.93
TWI	1.31	0.76

Figure 5. Head-cut gully erosion map using the five models: SVM, RF, and Oblique RF. Source: Author.

Table 4. Head-cut gully susceptibility classes’ area.

Models	Area	Susceptibility class
		Very low	Low	Moderate	High	Very high
SVM	Km^2	69.74	33.27	12.37	10.41	12.37
	%	50.48	24.08	8.95	7.53	8.95
RF	Km^2	87.25	22.63	7.13	6.83	14.32
	%	63.15	16.38	5.16	4.94	10.36
Oblique RF	Km^2	83.23	25.66	7.87	9.34	12.06
	%	60.24	18.57	5.70	6.76	8.73

Figure 6. The ROC curve analysis for five head-cut gully erosion models using the testing dataset. Source: Author.

Table 5. Predictive capability of head-cut gully erosion models using train and test dataset.

Models		Stage	Parameters
			Sensitivity	Specificity	PPV	NPV	AUC
SVM		Train	0.93	0.91	0.92	0.91	0.95
		Validation	0.87	0.90	0.89	0.88	0.91
RF		Train	0.98	0.94	0.95	0.98	0.98
		Validation	0.96	0.88	0.88	0.96	0.94
Oblique RF	Logistic Regression	Train	0.99	0.96	0.96	0.99	0.99
		Validation	0.97	0.90	0.90	0.97	0.947
	Ridge Regression	Train	0.99	0.96	0.96	0.99	0.99
		Validation	0.97	0.90	0.90	0.97	0.954
	PLS	Train	0.99	0.96	0.96	0.99	0.99
		Validation	0.97	0.90	0.90	0.97	0.952
	SVM	Train	0.99	0.96	0.96	0.99	0.99
		Validation	0.97	0.90	0.90	0.97	0.946
	Ensemble	Train	0.99	0.96	0.96	0.99	0.99
		Validation	0.97	0.90	0.90	0.97	0.96

Table 6. Variable importance analysis in RF, oRF, and SVM models for HGES modelling.

Variables	RF	Oblique RF	SVM
Altitude	100	100	100
Aspect	0.02	2.19	1.53
Slope	25.21	25.94	46.021
Plane curvature	2.36	13.2	7.75
Profile curvature	0	0	7.15
Drainage density	14.89	12.89	32.3
Distance from river	20.68	14.39	40.85
Distance from road	61.82	63.72	77.83
Land use	36.95	43.83	65.17
Lithology	21.29	15.52	9.65
Soil	43.51	38.09	70.45
Rainfall	15.89	16.96	30.99
SPI	3.53	12.57	0
TWI	4.98	33.94	6.92

Data availability statement

Data sharing is not applicable to this article as no new data were created or analyzed in this study.