A particle-based optimization of artificial neural network for earthquake-induced landslide assessment in Ludian county, China

Figures & data

Figure 1. Location of the study area and spatial distribution of landslides.

Table 1. Spatial relationship between the landslide and its causative factors.

Factor	Classes	No. of pixels	Percentage of domain	No. of prone pixels	Percentage of prone pixels	FR
Slope degree	0–13	3,668,185	55.03	473,991	53.23	0.97
	13–33	1,038,208	15.58	120,943	13.58	0.87
	33–48	879,139	13.19	117,403	13.18	1.00
	48–62	674,754	10.12	106,925	12.01	1.19
	62–87	405,386	6.08	71,272	8.00	1.32
Slope aspect	Flat (–1)	3,112,984	46.70	412,017	46.27	0.99
	North (0–22.5°)	681,932	10.23	87,206	9.79	0.96
	Northeast (22.5–67.5°)	81,027	1.22	9078	1.02	0.84
	East (67.5–112.5°)	778,981	11.69	97,364	10.93	0.93
	Southeast (112.5–157.5°)	96,727	1.45	13,062	1.47	1.01
	South (157.5–202.5°)	821,096	12.32	116,870	13.12	1.06
	Southwest (202.5–247.5°)	101,544	1.52	14,321	1.61	1.06
	West (247.5–292.5°)	858,203	12.87	121,617	13.66	1.06
	Northwest (292.5–337.5°)	94,013	1.41	14,066	1.58	1.12
	North (337.5–360°)	39,165	0.59	4933	0.55	0.93
Elevation (m)	568–1312	402,740	6.04	58,837	6.61	1.09
	1312–1755	876,433	13.15	254,830	28.62	2.18
	1755–2074	1,680,510	25.21	260,332	29.23	1.16
	2074–2328	1,779,312	26.69	212,321	23.84	0.89
	2328–2609	1,289,797	19.35	88,680	9.96	0.51
	2609–3356	636,880	9.55	15,534	1.74	0.18
Plan Curvature	Concave	1,765,829	26.49	242,177	27.19	1.03
	Flat	3,135,853	47.04	414,656	46.56	0.99
	Convex	1,763,990	26.46	233,701	26.24	0.99
NDVI	−0.1120–-–0.0294	1,037,869	15.57	117,068	13.15	0.84
	−0.0294–-0.0080	1,058,253	15.88	147,233	16.53	1.04
	−0.0080–0.0104	1,232,530	18.49	187,262	21.03	1.14
	0.0104–0.0275	1,426,991	21.41	197,827	22.21	1.04
	0.0275–0.0460	1,265,973	18.99	161,357	18.12	0.95
	0.0460–0.2497	644,056	9.66	79,787	8.96	0.93
PGA (gal)	50–100	88,866	1.33	1935	0.22	0.17
	100–200	452,924	6.79	27,370	3.07	0.45
	200–300	1,099,290	16.49	113,031	12.69	0.77
	300–500	2,253,089	33.80	242,363	27.22	0.81
	500–1100	2,771,504	41.58	505,835	56.80	1.37
Lithology	Dongchuan and Jialing Jian group-complex	621,491	9.32	86,047	9.66	1.04
	Uknown	96,607	1.45	13,167	1.48	1.02
	Shale, sandstone (Upper Permian)	3,664,582	54.98	414,280	46.52	0.85
	Shale	536,110	8.04	121,896	13.69	1.70
	Shale, sandstone (Upper Ordovician)	298,506	4.48	72,341	8.12	1.81
	Limestone, Dolomite, Quartzose, sandstone	782,688	11.74	88,523	9.94	0.85
	Limestone, Chert Limestone	14,930	0.22	13	0.00	0.00
	Dolomite, Limestone	280,533	4.21	29,269	3.29	0.78
	Shale, Siltstone	312,926	4.69	59,165	6.64	1.42
	Dolomite, Sandstone, Shale (Upper simian)	57,299	0.86	5833	0.66	0.77
Distance from road (m)	0–713	2,603,088	39.05	431,534	48.46	1.24
	713–1818	1,728,112	25.93	191,311	21.48	0.83
	1818–2716	1,060,663	15.91	129,623	14.56	0.92
	2716–3672	728,616	10.93	70,460	7.91	0.72
	3672–4785	399,461	5.99	49,011	5.50	0.92
	4785–7566	145,732	2.19	18,595	2.09	0.95
Distance from river (m)	0–1173	2,150,603	32.26	408,218	45.84	1.42
	1173–3190	2,151,009	32.27	300,712	33.77	1.05
	3190–5400	1,230,690	18.46	125,431	14.08	0.76
	5400–8023	545,747	8.19	16,918	1.90	0.23
	8023–11,223	403,770	6.06	23,340	2.62	0.43
	11,224–16,986	183,853	2.76	15,915	1.79	0.65
Distance from fault (m)	0–1674.25	2,229,751	33.45	380,027	42.67	1.28
	1674.25–3854.43	1,796,176	26.95	262,420	29.47	1.09
	3854.43–6548.90	1,125,910	16.89	95,066	10.68	0.63
	6548.90–9661.50	745,392	11.18	51,320	5.76	0.52
	9661.50–13,498.31	459,444	6.89	63,976	7.18	1.04
	13,498.31–19,184.70	308,999	4.64	37,725	4.24	0.91
Stream power index (SPI)	0–269,359	6,642,428	99.65	885,928	99.48	1.00
	269,359–1,616,152	20,278	0.30	3987	0.45	1.50
	1,616,152–4,444,417	2396	0.04	507	0.06	1.50
	4,444,417–10,504,984	499	0.01	101	0.01	1.00
	10,504,984–34,343,216	71	0.00	11	0.00	0.00
Topographic wetness index (TWI)	−14.49–-7.07	811,239	12.17	107,422	12.06	0.99
	−7.07–-1.74	1,769,388	26.54	245,293	27.54	1.04
	−1.74–3.46	1,642,916	24.65	209,915	23.57	0.96
	3.46–8.15	812,242	12.19	103,888	11.67	0.96
	8.15–10.36	1,364,147	20.47	182,827	20.53	1.00
	10.36–18.43	265,740	3.99	41,189	4.63	1.16

Figure 3. Graphical methodology of applied procedure for landslide hazard zonation.

Figure 4. The overall structure and learning mechanism of MLP.

Figure 6. The sensitivity analysis carried out for ANN, based on the number of hidden neurons.

Figure 7. The sensitivity analysis carried out for PSO-ANN model based on different swarm size.

Figure 8. Landslide hazard map developed by (a) ANN and (b) PSO-ANN models.

Figure 9. Column chart showing the percentage of landslide hazard classes over the study area.

Figure 10. The ROC diagram, obtained for ANN (a and b) and PSO-ANN (c and d) models for training and testing landslides.