Electromagnetic induction prediction of soil salinity and groundwater properties in a Tunisian Saharan oasis

Figures & data

Fig. 1 Study area and sampling locations.

Table 1  Summary statistics of soil properties at various soil depths, groundwater properties, and EM38 measurements collected in various seasons and years (2001–2004). N: number of observations

Parameter		N	Minimum	Maximum	Mean	Median	CV(%)
0–0.2 m	Gypsum (%)	27	40	69	57	59	11
	θ (%)	94	3	35	15	15	46
	ECe (dS m^-1)	94	3.28	37.70	7.53	5.14	94
0.2–0.4	Gypsum (%)	27	31	70	61	62	14
	θ (%)	94	4	39	18	18	45
	ECe (dS m^-1)	94	3.24	23.50	7.13	5.20	62
0.4–0.6	Gypsum (%)	27	28	78	61	66	21
	θ (%)	94	6	49	22	22	45
	ECe (dS m^-1)	94	3.13	27.70	8.04	6.18	60
0.6–0.8	Gypsum (%)	27	21	78	63	66	22
	θ (%)	94	8	46	24	23	38
	ECe (dS m^-1)	94	3.80	20.90	9.25	8.25	49
0.8–1.0	Gypsum (%)	27	39	72	64	67	13
	θ (%)	94	10	47	26	25	35
	ECe (dS m^-1)	94	3.84	25.00	10.43	9.72	46
1.0–1.2	Gypsum (%)	27	48	78	66	64	13
	θ (%)	94	8	52	27	27	35
	ECe (dS m^-1)	94	3.66	23.00	11.05	10.38	41
Groundwater	Dgw (m)	87	0.31	2.27	1.23	1.14	38
	PL (m)	87	16.08	23.09	19.36	19.17	11
	ECgw (dS m^-1)	73	4.73	41.20	15.12	15.03	41
EM38	EMh (dS m^-1)	94	0.255	2.705	0.945	0.756	59
	EMv (dS m^-1)	94	0.460	3.025	1.304	1.205	44
	EMh/EMv	94	0.480	1.107	0.690	0.653	19

Table 2  Summary statistics of soil properties for 0–0.6 and 0–1.2 m soil depths, groundwater properties, and EM38 readings collected during different March visits. N: number of observations

Date	Parameter	0–0.6 m		0–1.2 m		EM38		Groundwater
		θ	ECe	θ	ECe	EMh	EMv	Dgw	ECgw
Mar 2001	Minimum	11	3.70	12	4.34	0.26	0.50	0.35	8.44
N = 10	Maximum	34	23.18	35	17.36	2.25	2.45	2.27	41.20
	Average	20	10.17	24	10.55	1.00	1.33	1.30	20.19
	CV (%)	44	66	33	44	74	51	51	47
Mar 2002	Minimum	5	4.37	8	4.66	0.28	0.46	0.75	8.14
N = 10	Maximum	23	18.85	27	15.63	1.82	1.64	1.93	19.39
N(Dgw) = 9	Average	15	7.30	20	8.22	0.83	1.12	1.26	13.61
N(ECgw) = 7	CV (%)	40	60	31	41	55	38	34	34
Mar 2003	Minimum	6	3.85	9	4.10	0.28	0.48	0.73	8.77
N = 9	Maximum	25	10.56	26	13.90	1.46	1.83	1.94	22.60
N(Dgw) = 6	Average	13	6.80	16	8.23	0.83	1.21	1.30	14.74
N(ECgw) = 5	CV (%)	46	44	34	44	50	41	35	41
Mar 2004	Minimum	6	3.50	8	3.65	0.43	0.53	0.31	4.73
N = 24	Maximum	24	17.16	26	15.65	1.86	2.05	1.43	23.20
	Average	17	5.98	21	7.35	0.94	1.32	0.96	12.76
	CV (%)	43	47	30	35	42	34	28	46
Mar 2001 and 2004	Minimum	5	3.50	8	3.65	0.26	0.46	0.31	4.73
N = 53	Maximum	34	23.18	35	17.36	2.25	2.45	2.27	41.20
N(Dgw) = 49	Average	16	7.16	20	8.27	0.91	1.26	1.13	14.72
N(ECgw) = 46	CV (%)	44	60	33	42	53	39	39	49

Fig. 2 Adjusted R ² (R _a ²) and mean square error (MSE) of predicting lnECe observed at various soil depth intervals and seasons (12 dates from 2001 to 2004) with various models (best SLR, equations (1)(1) and (3), and MLR). MLR1: EM variables and plot coordinate as predictors; MLR2: same inputs than MLR1 plus groundwater properties.

Table 3  Performance of three models (SLR, equations (1)(1) and (3)) to predict the soil salinity (lnECe) from EM38 readings

Soil depth (m)	Model	Training ^(a)		Validation ^(b)		Total (N = 19)
		R ^2	MSE	R ^2	MSE	R ^2	MSE
0–0.2		0.943	0.022	0.953	0.025	0.925	0.024
0–0.4		0.965	0.009	0.906	0.018	0.939	0.013
0–0.6		0.971	0.005	0.838	0.037	0.897	0.020
0–0.8	SLR	0.953	0.008	0.837	0.034	0.887	0.020
0–1.0		0.893	0.015	0.750	0.051	0.807	0.032
0–1.2		0.814	0.025	0.760	0.045	0.780	0.034
0.6–1.2		0.586	0.054	0.805	0.045	0.699	0.050
0–0.6	Equation (3)(3)	0.949	0.015	0.690	0.112	0.820	0.044
0–1.2		0.820	0.040	0.776	0.075	0.792	0.041
0–1.2	Equation (1)(1) ^(c)	0.608	0.052	0.424	0.149	0.405	0.098
0–1.2	Equation (1)(1)	0.689	3.168	0.411	9.820	0.465	6.319
^(a) March 2002, N = 10.
^(b) March 2003, N = 9.
^(c) Equation (1)(1) applied to predict lnECe instead of ECe.

Fig. 3 The R _a ² and MSE of predicting lnECe at various soil depths and dates of measurement using various models (equations (1)(1) and (3), best SLR and MLR1 models).

Fig. 4 (a) Performance of the best SLR model to predict Dgw from EMv at various time measurements. (b) Observed and predicted Dgw (m) for calibration (March 2002) and validation (March 2003) subsets.