Machine Learning Methods of Remote Sensing Data Processing for Mapping Salt Pan Crust Dynamics in Sebkha de Ndrhamcha, Mauritania

LEMENKOVA, Polina

doi:10.2478/arsa-2025-0003

Figures & Tables

Topographic map of Mauritania showing the extent of the study area of the Sebkha de Ndrhamcha salt pan (rotated yellow square). Data: General Bathymetric Chart of the Oceans (GEBCO)/Shuttle Radar Topography Mission (SRTM). Software: Generic Mapping Tools (GMT).
Map source: author. — Topographic map of Mauritania showing the extent of the study area of the Sebkha de Ndrhamcha salt pan (rotated yellow square). Data: General Bathymetric Chart of the Oceans (GEBCO)/Shuttle Radar Topography Mission (SRTM). Software: Generic Mapping Tools (GMT). Map source: author.

Surficial geology and lithologic units of Mauritania. Data: USGS, GEBCO. Software: QGIS. Map source: author.

Geological provinces around Mauritania. Data: USGS, GEBCO. Software: QGIS. Map source: author.

Land cover types in Mauritania. Data: FAO, OpenStreetMaps. Software: QGIS. Map source: author.

Landsat 8-9 satellite images covering salt pan Sebkha de Ndrhamcha in western Mauritania in natural colours showing floodplain for 6 years (always April): (a) 2014, (b) 2017, (c) 2018, (d) 2020, (e) 2022, (f) 2023. Brief explanations of the colours: black colour indicates water areas of the Atlantic ocean, light beige colour represents sandy areas, brown colour indicates desert and bare lands, dark brown colour shows wet areas, grey colour indicates artificial surfaces (roads and urban areas) and cyan colour indicates salt pans on the original images. The colours are in Red Green Blue (RGB). Data source: US Geological Survey (USGS), downloaded from the EarthExplorer repository.

Classified Landsat 8-9 OLI/TIRS images using clustering: (a) 2014, (b) 2017, (c) 2018, (d) 2020, (e) 2022, (f) 2023. Software: GRASS GIS. Source of maps: author.

Maps of rejection threshold probability for accuracy analysis of image classification by chi-squared test: (a) 2014, (b) 2017, (c) 2018, (d) 2020, (e) 2022, (f) 2023. Software: GRASS GIS. Source of maps: author.

Random Forest Classifier of Machine Learning (ML) classification methods applied for processing of satellite images: (a) 2014, (b) 2017, (c) 2018, (d) 2020, (e) 2022, (f) 2023.
Software: GRASS GIS. Source of maps: author. — Random Forest Classifier of Machine Learning (ML) classification methods applied for processing of satellite images: (a) 2014, (b) 2017, (c) 2018, (d) 2020, (e) 2022, (f) 2023. Software: GRASS GIS. Source of maps: author.

Decision Tree Classifier of Machine Learning (ML) classification methods applied for processing of satellite images: (a) 2014, (b) 2017, (c) 2018, (d) 2020, (e) 2022, (f) 2023.
Software: GRASS GIS. Source of maps: author. — Decision Tree Classifier of Machine Learning (ML) classification methods applied for processing of satellite images: (a) 2014, (b) 2017, (c) 2018, (d) 2020, (e) 2022, (f) 2023. Software: GRASS GIS. Source of maps: author.

Gradient Boosting Classifier of Machine Learning (ML) classification methods applied for processing of satellite images: (a) 2014, (b) 2017, (c) 2018, (d) 2020, (e) 2022, (f) 2023. Software: GRASS GIS. Source of maps: author.

Support Vector Machine (SVM) Classifier of Machine Learning (ML) classification methods applied for processing of satellite images: (a) 2014, (b) 2017, (c) 2018, (d) 2020, (e) 2022, (f) 2023. Software: GRASS GIS. Source of maps: author.

Accuracy assessment for ML models in GRASS GIS: 1) Random Forest (RF); 2) Support Vector Machine (SVM); 3) Decision Tree Classifier (DTC); 4) Gradient Boosting Classifier (GBC)_ Estimated classes of land cover types for 2014–2023 in West Mauritania_

Year	Producer’s accuracy, %				User’s accuracy, %				Kappa statistics
Year	RF	SVM	DTC	GBC	RF	SVM	DTC	GBC	RF	SVM	DTC	GBC
Land Cover Class 1: Water bodies
2014	77	83	74	71	78	88	62	62	0.80	0.74	0.65	0.65
2017	78	84	72	70	77	85	65	67	0.81	0.93	0.53	0.59
2018	76	79	75	69	71	79	66	68	1.00	0.90	0.69	0.61
2020	81	81	73	72	89	89	68	72	0.93	0.95	0.58	0.61
2022	82	80	65	73	82	73	75	74	0.95	0.70	0.84	1.00
2023	79	82	66	68	77	74	74	70	0.74	0.79	0.83	0.67
Land Cover Class 2: Shelf and coastal plains
2014	89	78	72	67	100	98	56	76	0.82	0.91	0.67	0.69
2017	95	91	68	72	92	65	64	65	0.73	1.00	0.56	0.74
2018	100	92	87	91	87	77	63	68	0.78	0.94	0.88	0.68
2020	88	100	86	65	89	87	69	62	0.89	0.84	0.83	0.85
2022	100	95	96	59	78	77	58	59	0.91	0.78	0.72	0.73
2023	84	83	75	71	73	75	74	72	0.92	0.75	0.75	0.56
Land Cover Class 3: Sebkha
2014	95	78	67	74	92	90	66	67	0.88	0.74	0.68	0.72
2017	85	93	88	72	78	87	69	81	0.78	0.81	0.55	0.56
2018	88	89	63	64	77	94	71	69	0.91	0.88	0.71	0.54
2020	94	75	64	69	93	85	72	65	1.00	0.64	0.75	0.78
2022	73	92	64	81	73	95	59	71	0.84	1.00	0.73	0.78
2023	69	68	76	83	67	89	73	72	0.91	0.90	0.74	0.84
Land Cover Class 4: Urban areas
2014	87	78	64	68	88	86	67	66	1.00	0.85	0.76	0.81
2017	94	79	69	64	92	75	61	72	0.95	1.00	0.82	0.73
2018	74	91	74	72	91	79	72	74	0.85	0.87	0.75	0.80
2020	69	80	65	76	84	64	68	68	1.00	0.79	0.81	0.69
2022	71	66	83	81	74	62	54	65	0.92	1.00	0.65	0.75
2023	83	65	72	83	79	75	59	71	0.87	0.72	0.69	0.73
Land Cover Class 5: Sahelian grassland
2014	87	81	65	67	82	90	83	81	0.93	0.74	0.95	0.91
2017	64	65	71	68	76	79	74	78	0.78	0.68	0.76	0.58
2018	78	72	73	57	79	91	68	63	1.00	0.83	0.73	0.79
2020	74	79	72	71	67	77	64	68	0.91	0.65	0.88	0.63
2022	88	82	71	82	81	80	72	69	0.53	1.00	0.54	0.77
2023	91	75	65	64	91	82	71	90	1.00	0.91	0.67	0.82
Land Cover Class 6: Salty sands
2014	88	72	67	63	91	79	68	67	0.98	0.73	0.73	0.78
2017	72	79	72	76	82	74	65	78	1.00	0.75	0.81	0.75
2018	91	90	64	78	90	78	71	81	1.00	0.81	0.76	0.69
2020	89	85	69	74	85	81	78	82	0.81	0.86	0.88	0.65
2022	90	84	56	71	78	90	64	89	0.84	0.91	0.72	0.67
2023	77	71	68	80	74	72	73	75	0.78	1.00	0.63	0.68
Land Cover Class 7: Compact soil
2014	89	78	67	63	89	81	69	63	0.88	0.73	0.65	0.61
2017	73	77	63	69	73	72	73	71	1.00	0.61	0.69	0.83
2018	78	69	72	65	85	85	74	70	0.85	0.68	0.81	0.95
2020	88	81	79	71	69	74	65	65	0.97	0.74	0.85	0.74
2022	91	90	80	75	70	78	69	69	0.83	1.00	0.72	0.61
2023	74	75	71	61	66	64	81	74	0.74	0.92	0.78	0.89
Land Cover Class 8: Stony desert and yellow dunes
2014	85	83	75	67	89	73	68	74	0.84	0.81	0.71	0.86
2017	78	88	78	69	93	84	69	82	0.75	1.00	0.82	0.92
2018	89	81	74	78	78	86	89	76	1.00	0.63	0.70	0.84
2020	92	79	91	71	95	72	81	81	0.72	0.79	0.62	0.70
2022	91	84	73	74	81	89	90	89	0.68	0.71	0.59	0.69
2023	84	75	75	73	73	90	83	73	0.66	0.65	0.60	0.76
Land Cover Class 9: Sandy desert and white dunes
2014	92	81	67	65	81	82	65	71	0.77	0.82	0.77	0.71
2017	83	89	72	81	84	78	78	78	0.81	0.73	0.81	0.55
2018	93	70	68	72	95	73	71	73	0.64	0.67	0.80	1.00
2020	78	93	80	68	88	89	83	67	1.00	0.54	0.68	0.75
2022	77	94	71	73	89	71	69	81	0.98	0.91	0.72	0.67
2023	74	85	81	79	73	69	62	66	0.90	1.00	0.74	0.79
Land Cover Class 10: Bare soil and rocks
2014	91	83	68	78	84	83	78	74	0.87	1.00	0.73	0.69
2017	89	82	63	72	83	78	73	75	0.81	0.74	0.77	0.57
2018	88	75	65	77	79	69	69	81	1.00	0.73	0.81	0.61
2020	79	69	73	64	74	71	74	67	0.74	0.69	0.80	0.74
2022	82	91	79	69	75	73	81	83	0.63	0.81	1.00	0.83
2023	77	80	75	71	82	95	83	74	0.84	0.68	0.65	0.80

Metadata of the multispectral satellite images Landsat 8-9 OLI/TIRS, used in this study, obtained from USGS1

Date	Spacecraft / ID	Path/row	Entity product ID	Scene ID	Cloud/Coverage
27/04/2014	Landsat 8	205/47	LC08_L2SP_205047_20140427_20200911_02_T1	LC82050472014117LGN01	0.00
03/04/2017	Landsat 8	205/47	LC08_L2SP_205047_20170403_20200904_02_T1	LC82050472017093LGN00	0.04
22/04/2018	Landsat 8	205/47	LC08_L2SP_205047_20180422_20201015_02_T1	LC82050472018112LGN00	0.04
11/04/2020	Landsat 8	205/47	LC08_L2SP_205047_20200411_20200822_02_T1	LC82050472020102LGN00	0.17
09/04/2022	Landsat 8	205/47	LC09_L1TP_205047_20220409_20230422_02_T1	LC92050472022099LGN01	0.00
28/04/2023	Landsat 9	205/47	LC09_L2SP_205047_20230428_20230430_02_T1	LC92050472023118LGN00	0.01

Processing time of the satellite images Landsat-8 OLI/TIRS showing the effectiveness of ML methods executed by GRASS GIS

Method	Processing time
Clustering	<1 min
Min-max discriminant analysis	ca. 25 sec
Random Forest Classifier	9 min
Decision Tree Classifier	ca. 34 sec
Gradient Boosting Classifier	23 min
Support Vector Machine Classifier	47 min

Estimated classes of land cover types in western Mauritania, Sebkha de Ndrhamcha, for April_ Map units in measurements: 30 m resolution for each pixel on the multispectral scene of Landsat 8-9 OLI/TIRS_

Year	Classes of land cover types in western Mauritania, Sebkha de Ndrhamcha
Year	1	2	3	4	5	6	7	8	9	10
2014	1410	4	30	79	202	708	1239	1581	1493	178
2017	1386	43	70	141	215	668	1056	1742	1548	144
2018	1384	21	78	110	182	598	1195	1975	1353	117
2020	1395	2	28	145	228	582	881	2502	1231	20
2022	1443	2	20	116	197	532	970	2567	1146	40
2023	1468	31	53	136	253	538	1153	1622	1577	206

Machine Learning Methods of Remote Sensing Data Processing for Mapping Salt Pan Crust Dynamics in Sebkha de Ndrhamcha, Mauritania

Figures & Tables

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

Figure 6.

Figure 7.

Figure 8.

Figure 9.

Figure 10.

Figure 11.

Accuracy assessment for ML models in GRASS GIS: 1) Random Forest (RF); 2) Support Vector Machine (SVM); 3) Decision Tree Classifier (DTC); 4) Gradient Boosting Classifier (GBC)_ Estimated classes of land cover types for 2014–2023 in West Mauritania_

Metadata of the multispectral satellite images Landsat 8-9 OLI/TIRS, used in this study, obtained from USGS1

Processing time of the satellite images Landsat-8 OLI/TIRS showing the effectiveness of ML methods executed by GRASS GIS

Estimated classes of land cover types in western Mauritania, Sebkha de Ndrhamcha, for April_ Map units in measurements: 30 m resolution for each pixel on the multispectral scene of Landsat 8-9 OLI/TIRS_

Paradigm

My account