Integração de dados PALSAR-2/ALOS-2 e OLI/LANDSAT-8 para mapeamento de uso e cobertura da terra no Norte da Amazônia brasileira

Resumo:Na porção norte da Amazônia brasileira, as savanas, florestas estacionais e terras agropecuárias formam uma paisagem complexa, onde o mapeamento de uso e cobertura da terra é uma tarefa desafiadora. Nesse trabalho, dados Landsat-8/OLI e ALOS-2/PALSAR-2 foram combinados para mapeamento de 17 classes de uso e cobertura da terra usando o algoritmo Random Forest. O potencial de cada conjunto de dados foi analisado separadamente e em comparação ao modelo híbrido. Os resultados mostraram que o modelo híbrido com as polarizações PALSAR-2 HH/HV e seis bandas de reflectância do OLI produziu os melhores resultados, com acurácia global de 83% e Kappa de 0,81. Isto representou um aumento de 6% em relação à classificação das bandas do OLI somente. Os modelos usando os dados ópticos produziram resultados melhores do que os do SAR. Entretanto, a maior contribuição do PALSAR-2 foi melhorar a discriminação de classes de savana com menor biomassa, como os campos limpos e campos cerrados.

Palavras-chave:
Random Forest; uso e cobertura da terra; classificação híbrida

Autoria

João Arthur Pompeu Pavanelli

Instituto Nacional de Pesquisas Espaciais, Divisão de Sensoriamento Remoto, São José dos Campos - São Paulo, Brasil. E-mail: joao.pompeu@inpe.br; joao.roberto@inpe.br; lenio.galvao@inpe.br Instituto Nacional de Pesquisas EspaciaisBrazilSão José dos Campos, São Paulo, Brazil Instituto Nacional de Pesquisas Espaciais, Divisão de Sensoriamento Remoto, São José dos Campos - São Paulo, Brasil. E-mail: joao.pompeu@inpe.br; joao.roberto@inpe.br; lenio.galvao@inpe.br

https://orcid.org/0000-0001-9126-8229

João Roberto dos Santos

https://orcid.org/0000-0002-1139-9577

Lênio Soares Galvão

https://orcid.org/0000-0002-8313-0497

Maristela Xaud

Empresa Brasileira de Pesquisa Agropecuária, EMBRAPA Roraima, Boa Vista - Roraima, Brasil. E-mail: maristela.xaud@embrapa.br; haron.xaud@embrapa.brEmpresa Brasileira de Pesquisa AgropecuáriaBrazilBoa Vista, Roraima, BrazilEmpresa Brasileira de Pesquisa Agropecuária, EMBRAPA Roraima, Boa Vista - Roraima, Brasil. E-mail: maristela.xaud@embrapa.br; haron.xaud@embrapa.br

https://orcid.org/0000-0001-5536-2736

Haron Abrahim Magalhães Xaud

https://orcid.org/0000-0002-5195-3966

SCIMAGO INSTITUTIONS RANKINGS

Figuras | Tabelas | Fórmulas

Figuras (6)
Tabelas (4)
Fórmulas (1)

Thumbnail

Figure 1
Location of the study area in Roraima state, Brazil. The colour composite (February 6, 2015) includes the OLI/Landsat-8 bands 6, 5 and 7 in red, green and blue, respectively.

Thumbnail

Figure 2
Flow diagram for Random Forest (RF) classification of the LULC classes using optical (OLI/Landsat-8) and SAR (PALSAR/ALOS-2) attributes and the combination of metrics of both datasets.

Thumbnail

Figure 3
LULC map derived from the best Random Forest classification (model 15) using the HH and HV PALSAR/ALOS-2 amplitude and the surface reflectance of OLI/Landsat-8 bands 2 to 7. From top to bottom in the map, red rectangles are zoomed below showing the result of the RF classification, the OLI colour composition (R6G5B7) with examples of training polygons and the PALSAR-2 HH Amplitude band with examples of validation polygons.

Thumbnail

Figure 4
The most important variables in the hybrid Random Forest model 15, expressed by the normalized mean decrease in classification accuracy, which ranges from zero (low importance) to one (high importance).

Thumbnail

Figure 5
Producer's accuracy (PA) and User's accuracy (UA), indicating gains (positive values) and losses (negative values) in Random Forest classification from the inclusion of the PALSAR-2 HH and HV polarization into the OLI dataset.

Thumbnail

Figure 6
(a) Spectral confusion between the savannah grassland and the wooded savannah in the OLI/Landsat-8 bands 5 and 7. (b) Increase of discrimination between the two physiognomies with the inclusion of the PALSAR-2/ALOS-2 HV amplitude. Values of OLI data are in reflectance and PALSAR-2 in amplitude, as used for inputs in the RF classifier.

Thumbnail

Table 1:
Main attributes from optical and SAR datasets.

Thumbnail

Table 2:
Combination of attributes (marked with X) used in the different Random Forest (RF) models. Models 1 to 6 were processed only with PALSAR-2 metrics, while models 7 to 12 were generated only with OLI attributes. Models 13 to 18 were hybrid when combining optical and SAR attributes.

Thumbnail

Table 3:
Accuracy assessment of each of the 18 Random Forest (RF) classification models. Values in bold highlight the best models using either PALSAR-2 or OLI data, or the combination of both datasets (hybrid models).

Thumbnail

Table 4:
Confusion matrix derived from Random Forest classification using the validation set of pixels and the combination of optical (OLI/Landsat-8) and SAR (PALSAR-2/ALOS-2) attributes (RF model 15).

Thumbnail

(1)

Figure 1 Location of the study area in Roraima state, Brazil. The colour composite (February 6, 2015) includes the OLI/Landsat-8 bands 6, 5 and 7 in red, green and blue, respectively.

Figure 2 Flow diagram for Random Forest (RF) classification of the LULC classes using optical (OLI/Landsat-8) and SAR (PALSAR/ALOS-2) attributes and the combination of metrics of both datasets.

Figure 3 LULC map derived from the best Random Forest classification (model 15) using the HH and HV PALSAR/ALOS-2 amplitude and the surface reflectance of OLI/Landsat-8 bands 2 to 7. From top to bottom in the map, red rectangles are zoomed below showing the result of the RF classification, the OLI colour composition (R6G5B7) with examples of training polygons and the PALSAR-2 HH Amplitude band with examples of validation polygons.

Figure 4 The most important variables in the hybrid Random Forest model 15, expressed by the normalized mean decrease in classification accuracy, which ranges from zero (low importance) to one (high importance).

Figure 5 Producer's accuracy (PA) and User's accuracy (UA), indicating gains (positive values) and losses (negative values) in Random Forest classification from the inclusion of the PALSAR-2 HH and HV polarization into the OLI dataset.

Figure 6 (a) Spectral confusion between the savannah grassland and the wooded savannah in the OLI/Landsat-8 bands 5 and 7. (b) Increase of discrimination between the two physiognomies with the inclusion of the PALSAR-2/ALOS-2 HV amplitude. Values of OLI data are in reflectance and PALSAR-2 in amplitude, as used for inputs in the RF classifier.

Table 1: Main attributes from optical and SAR datasets.

	Landsat-8/OLI (CDR product)	ALOS-2/PALSAR-2 (Stripmap Fine Bean Dual Level 1.5 product)
Pixel size	30 m	10 m
Temporal resolution	16 days	On demand
Radiometric resolution	16 bits	16 bits
Spectral	11 bands (435 - 12510 nm)	L-band (23 cm)
Imaging date	February 6th, 2015	February 23th, 2015
Swath	185 km	70 km
Cloud cover of the scene	5.2 %	-
Polarization	-	Dual (HH + HV)
Pass	Descending	Descending (Right)
View angle	0	32.5 degrees
Path/row	232/58	-

Table 2: Combination of attributes (marked with X) used in the different Random Forest (RF) models. Models 1 to 6 were processed only with PALSAR-2 metrics, while models 7 to 12 were generated only with OLI attributes. Models 13 to 18 were hybrid when combining optical and SAR attributes.

Attributes

Random Forest models

HH Amplitude

HV Amplitude

HH GLCM textures (10 images)

HV GLCM textures (10 images)

HH+HV

HH-HV

HH/HV

HV/HH

(HH*HV)/(HH+HV)

OLI Band 2

OLI Band 3

OLI Band 4

OLI Band 5

OLI Band 6

OLI Band 7

NDVI

EVI

Number of variables (N)

Table 3: Accuracy assessment of each of the 18 Random Forest (RF) classification models. Values in bold highlight the best models using either PALSAR-2 or OLI data, or the combination of both datasets (hybrid models).

Dataset	RF model	Overall accuracy (%)	Kappa
PALSAR-2	1	30.46	0.238
	2	36.23	0.296
	3	38.88	0.328
	4	44.60	0.390
	5	29.95	0.232
	6	44.54	0.389
OLI	7	57.01	0.527
	8	68.94	0.658
	9	67.89	0.646
	10	76.88	0.744
	11	76.86	0.743
	12	56.57	0.523
PALSAR-2 plus OLI (hybrid models)	13	67.29	0.638
	14	67.43	0.639
	15	82.96	0.810
	16	82.61	0.806
	17	82.15	0.801
	18	82.41	0.804

Table 4: Confusion matrix derived from Random Forest classification using the validation set of pixels and the combination of optical (OLI/Landsat-8) and SAR (PALSAR-2/ALOS-2) attributes (RF model 15).

Land use and land cover (LULC) classes	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	Total (pixels)	User's accuracy (UA) (%)
1 Agriculture	1018	0	0	69	0	0	0	7	0	0	0	32	44	0	0	251	0	1421	71.63
2 Water	0	674	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	674	100
3 Campinarana	0	0	3812	19	0	4	38	14	0	57	13	0	24	0	137	0	6	4124	92.43
4 Wooded savannah	0	0	15	2100	31	502	0	0	72	0	0	104	112	0	26	9	31	3002	69.95
5 Savannah grassland	0	0	0	94	1540	82	0	0	42	0	0	120	0	0	0	43	0	1921	80.16
6 Shrub savannah	0	0	0	79	47	79	0	0	4	0	0	1	2	0	0	6	0	218	36.23
7 SS1	0	0	16	0	0	0	29	16	0	2	0	0	0	0	0	0	4	67	43.28
8 SS2	0	0	10	0	0	0	3	7	0	6	20	0	0	0	0	0	0	46	15.21
9 Burned savannah	0	0	8	17	0	0	0	0	1532	0	0	0	0	0	34	300	0	1711	79.01
10 Woodland savannah	0	0	51	15	0	0	3	0	0	223	13	0	22	0	0	2	4	333	66.96
11 Mature forest	0	0	0	0	0	0	0	60	0	133	2498	0	0	32	0	0	49	2772	90.11
12 Clean Pasture	0	0	44	84	101	59	0	0	0	0	0	166	42	0	0	0	4	500	33.20
13 Dirty Pasture	0	0	12	45	4	57	10	169	0	69	2	48	765	0	0	84	0	1265	60.47
14 Silviculture	0	0	6	0	0	0	44	0	0	1	114	0	3	2763	0	0	10	2941	93.94
15 Clear cut silviculture	0	0	18	5	13	0	0	0	14	0	0	0	0	0	3193	9	0	3252	98.67
16 Bare soil	0	0	0	2	0	0	0	0	0	0	0	85	0	0	0	168	0	255	65.88
17 Palm swamps	0	0	1	0	0	0	1	34	0	9	23	0	0	5	0	0	15	88	17.04
Total (pixels)	1018	674	3993	2529	1736	783	128	307	1484	500	2683	556	1014	2800	3390	872	123	24590
Producer’s accuracy (PA) (%)	100	100	95.46	83.03	88.70	10.08	22.65	2.28	91.10	44.60	93.10	29.85	75.44	98.67	94.18	19.26	12.19
Overall accuracy (%)	82.96
Kappa	0.810

(1)

VI (X^{j}) = \frac{1}{ntree} \sum ({OOBE}_{t}^{j} - {OOBE}_{t}^{j})

Como citar

copiar

Universidade Federal do Paraná Centro Politécnico, Jardim das Américas, 81531-990 Curitiba - Paraná - Brasil, Tel./Fax: (55 41) 3361-3637 - Curitiba - PR - Brazil
E-mail: bcg_editor@ufpr.br

Acompanhe os números deste periódico no seu leitor de RSS

Versão para download de PDF

PDF

Inglês

Versões e tradução automática

Versão original do texto

English

Tradução automática

Google Translator
Microsoft Translator

Como citar

RIS

BIBTEX

Outros formatos de citação e exportação:

SciELO - Scientific Electronic Library Online
Rua Dr. Diogo de Faria, 1087 – 9º andar – Vila Clementino 04037-003 São Paulo/SP - Brasil
E-mail: scielo@scielo.org

Leia a Declaração de Acesso Aberto

Métricas

Integração de dados PALSAR-2/ALOS-2 e OLI/LANDSAT-8 para mapeamento de uso e cobertura da terra no Norte da Amazônia brasileira

PlumX