Collaborative validation of GlobeLand30: Methodology and practices

Figures & data

Figure 1. The conceptual framework of the GEO-led GLC validation project

Figure 2. Three-level LSI-based sampling approach

Figure 3. Sample judgment and labeling

Figure 4. Two-scale windows for sample judgment

Figure 5. GLCVal- Function design

Figure 6. The main interface of GLCVal online validation tool

Figure 7. Distribution of the LSI-based sample points in Bulgaria (left) and Sweden (right)

Table 1. Error matrix for GlobeLand30 dataset in Bulgaria

Land cover	10	20	30	40	50	60	80	90	All	UA (%)
10–cultivated land	75	10	7	10			4		106	70.8
20 – orest	5	154	4	4					167	92.2
30–grassland	2	9	26	7		1	1		46	56.5
40–shrubland	1	3	2	23					29	79.3
50–wetland				1	1				2	50.0
60–water bodies			1		1	5			7	71.4
80–artificial surfaces	1		1	1			21		24	87.5
90–bareland		1						0	1	0.0
All	84	177	41	46	2	6	26	0	382	0.0
PA(%)	89.3	87.0	63.4	50.0	50.0	83.3	80.8	0.0	0.0	OA: 79.8%

Table 2. Error matrix for GlobeLand30 dataset in Sweden

Land cover	10	20	30	40	50	60	70	80	90	100	All	UA (%)
10–cultivated land	40	6		2							48	83.3
20–forest		51		7							58	87.9
30–grassland	1		10	1							12	83.3
40–shrubland		3		49							52	94.2
50–wetland		1		2	44			1			48	91.7
60–water bodies		2			1	45					48	93.8
70–tundra		1					37		2		40	92.5
80–artificial surfaces	1	1			2			31	1		36	86.1
90–bareland								1	22		23	95.7
100–permanent snow and ice							5		4	20	29	69.0
All	42	65	10	61	47	45	42	33	29	20	394	0.0
PA (%)	95.2	78.5	100	80.3	93.6	100	88.1	93.9	75.9	100	0.0	OA: 88.58%

Data Availability Statement

The data that support the findings of this study are openly available in “Global Change Data Repository” at http://dx.doi.org/10.3974/geodb.2014.02.01.V1, http://dx.doi.org/10.3974/geodb.2014.02.02.V1, http://dx.doi.org/10.3974/geodb.2018.02.05.V1.