extraction and analysis of plain afforestation using hj-1...

1
Extraction and Analysis of Plain Afforestation Using HJ-1 and Mapping Satellite-1 Images YU Xinfang, WANG Zhengxing, SHANG Ke, DIAO Huijuan (Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China) ABSTRACT The term afforestationin this paper generally represents all kinds of non-crop vegetation in crop dominated plain area, including but not limited to the vegetation in crop land (e.g., orchard, herb, nursery garden, fast-growing and high-yielding timber, farmland shelterbelt, and artificial turf), residential green, urban vegetation landscapes, trees planted alongside the roads and rivers, and the wind break and sand fixation forest. Accurate and timely information about afforestation in the plain region is useful because it reflects the degree of agricultural diversity and the environmental health. However, as afforestation are spatially scattered, temporally and spectrally overlapped with some crop lands, it is very challenging to get the information about afforestation using remote sensing. Based on multi-temporal Chinese HJ-1 A/B CCD satellite data, the plain afforestation area of Henan Provience was extracted. Aside from Xinyang, Henan afforestation area is 6909.8 km 2 . It accounts for 20.53% of the total forest in Henan Province. Then, as a case study of Fengqiu County in Henan Province, the plain afforestation including farmland shelterbelt, road shelterbelt and residence shelterbelt was extracted based on Chinese Mapping Satellite-1 (MS1) imagery. The good results demonstrated the potential of object-oriented plain afforestation information extraction based on multi-temporal and high-resolution images. Keywords: plain afforestation; farmland shelterbelt; road shelterbelt; residence shelterbelt; HJ-1 CCD; Mapping Satellite-1; object-oriented; multi-scale segmentation; Henan Province; China INTRODUCTION As a land use type, afforestation in plains often shares common spaces with other types of land use, such as croplands and roads. As a result, there is little, if any, data about afforestation in plains, let alone the afforestation change monitoring. As a follow-up to UN Millennium Ecosystem Assessment, the Chinese government is currently conducting a similar assessment at the provincial level, covering years of 2000-2005-2010, and using the traditional (FAO) land cover/land use system. It is beyond the expectation that the “forest”- as the major indicator of a good environment, in the province such as Henan, has only experienced a negligible increase. This is because the recent afforestation in plains was classified as other land cover types. With the advent of Chinese satellite HJ-1 A/B, there may be a chance to extract information about afforestation in plains since its CCD sensor has Red and NIR channels, with a 30m spatial resolution and a 4-day temporal resolution. Chinese satellite MS1 with 2m spatial resolution can provide the more detail afforestation information. STUDY AREA METHODS Three steps were taken to extract afforestation in Henan based on HJ-1 A/B data: (1) extract the plain information using 2010 Land Cover Map; (2) eliminate the double-crop (winter wheat) land using NDVI June data when all the wheat had been harvested; (3) eliminate the single-crop land using NDVI April data. Afforestation was extracted by calculating ((NDVIApril≥0.22) (NDVIJune≥0.35)). Based on MS1 imagery in Fengqiu County, object-oriented classification method was used. The spectral and spatial features of farmland shelterbelt, road shelterbelt and residence shelterbelt samples were analyzed to build membership function and develop classification rule set. The plain afforestation areas were extracted according to the optimal segmentation scales and certain classification features. RESULTS Validation was conducted using correlation with the statistics of 26 major cities in Henan Province, resulting in a significant R 2 =0.92. The algorithm performed well in the wheat region, yet it did poor in the rice and rice-wheat transition regions, which are mainly distributed in the southern part of Xinyang. Aside from Xinyang, afforestation area extracted from HJ-1 is 6909.80 km 2 . This accounts for 20.53% of the total forest in Henan Province. CONCLUSIONS The results showed that the study achieved fine classification results. And the good results demonstrated the potential of object-oriented plain afforestation information extraction based on high-resolution images. This method provides a technical support for accurate estimation of plain afforestation area. Future study should make full use of HJ-1 high temporal resolution data and MS1 high spatial resolution data by conducting zoning according to the climate and the soil. MAJOR REFERENCES [1]An K, Zhang J S, Xiao Y. Object-oriented Urban Dynamic Monitoring-A Case Study of Haidian District of Beijing. Chinese Geographical Science, 2007, 17(3): 236-242. [2] Gutman G, Huang C, Chander G, Noojipady P, Masek J G. Assessment of the NASAUSGS Global Land Survey (GLS) datasets. Remote Sensing of Environment, 2013(134) : 249265. [2] Hansen C M, Loveland T R.A review of large area monitoring of land cover change using Landsat data. Remote Sensing of Environment, 2012(122): 6674. [4] Montandon L M, Small E E. The impact of soil reflectance on the quantification of the green vegetation fraction from NDVI. Remote Sensing of Environment, 2008(112): 18351845. [5] Wang H, Chen J S, Yu X M. Feature selection and its application in object-oriented classification. Journal of Remote Sensing, 2013, 17(4): 816-829. WithinFarmland Segmentation30 Brightness 12 NDVI 0.01 WithinConstruction land Segmentation20 NDVI 0.015 WithinResidence land Segmentation10 6 Green 10 Fig.6 Afforestation spatial distribution of Fengqiu, 2010 Fig.1 Land cover of Henan Province, 2010 Fig.2 Fengqiu County and Mapping Satellite-1 false color composite image with the validation samples, 2010 SATELLITE DATA Spatial resolution of HJ-1 A/B CCD: 30m, Data: 2009-2011 Spatial resolution of Mapping Satellite-1: 2m, Date: 2010-10-15 The results showed that the plain afforestation area of Fengqiu is 152.51 km 2 . More specifically, the farmland shelterbelt area is 36.09 km 2 , the road shelterbelt area is 21.29 km 2 , the residence shelterbelt area is 71.56 km 2 , and the patched shelterbelt area is 23.57 km 2 . The classification accuracy is 93.50% and the Kappa coefficient is 0.92. afforestation Xinyang Fig.5 Afforestation spatial distribution of Henan, circa 2010 Fig.4 Afforestation correlation for 26 major cities: statistics vs. HJ-1 A/B CCD statistics area Interpretation area from HJ-1 A/B (km 2 ) Statistics area (km 2 ) Fig.3 Extraction of farmland shelterbelt, road shelterbelt and residence shelterbelt

Upload: others

Post on 07-Apr-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Extraction and Analysis of Plain Afforestation Using HJ-1 ...earth.esa.int/dragon-2015-programme/yu-extraction... · Extraction and Analysis of Plain Afforestation Using HJ-1 and

Extraction and Analysis of Plain Afforestation

Using HJ-1 and Mapping Satellite-1 Images

YU Xinfang, WANG Zhengxing, SHANG Ke, DIAO Huijuan

(Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China)

ABSTRACT The term “afforestation” in this paper generally represents all kinds

of non-crop vegetation in crop dominated plain area, including but

not limited to the vegetation in crop land (e.g., orchard, herb,

nursery garden, fast-growing and high-yielding timber, farmland

shelterbelt, and artificial turf), residential green, urban vegetation

landscapes, trees planted alongside the roads and rivers, and the

wind break and sand fixation forest. Accurate and timely

information about afforestation in the plain region is useful because

it reflects the degree of agricultural diversity and the environmental

health. However, as afforestation are spatially scattered, temporally

and spectrally overlapped with some crop lands, it is very

challenging to get the information about afforestation using remote

sensing. Based on multi-temporal Chinese HJ-1 A/B CCD satellite

data, the plain afforestation area of Henan Provience was extracted.

Aside from Xinyang, Henan afforestation area is 6909.8 km2. It

accounts for 20.53% of the total forest in Henan Province. Then, as

a case study of Fengqiu County in Henan Province, the plain

afforestation including farmland shelterbelt, road shelterbelt and

residence shelterbelt was extracted based on Chinese Mapping

Satellite-1 (MS1) imagery. The good results demonstrated the

potential of object-oriented plain afforestation information

extraction based on multi-temporal and high-resolution images.

Keywords: plain afforestation; farmland shelterbelt; road

shelterbelt; residence shelterbelt; HJ-1 CCD; Mapping Satellite-1;

object-oriented; multi-scale segmentation; Henan Province; China

INTRODUCTION As a land use type, afforestation in plains often shares common

spaces with other types of land use, such as croplands and roads. As

a result, there is little, if any, data about afforestation in plains, let

alone the afforestation change monitoring. As a follow-up to UN

Millennium Ecosystem Assessment, the Chinese government is

currently conducting a similar assessment at the provincial level,

covering years of 2000-2005-2010, and using the traditional (FAO)

land cover/land use system. It is beyond the expectation that the

“forest”- as the major indicator of a good environment, in the

province such as Henan, has only experienced a negligible increase.

This is because the recent afforestation in plains was classified as

other land cover types. With the advent of Chinese satellite HJ-1

A/B, there may be a chance to extract information about

afforestation in plains since its CCD sensor has Red and NIR

channels, with a 30m spatial resolution and a 4-day temporal

resolution. Chinese satellite MS1 with 2m spatial resolution can

provide the more detail afforestation information.

STUDY AREA

METHODS Three steps were taken to extract afforestation in Henan based on HJ-1 A/B

data: (1) extract the plain information using 2010 Land Cover Map; (2)

eliminate the double-crop (winter wheat) land using NDVI June data when all

the wheat had been harvested; (3) eliminate the single-crop land using NDVI

April data. Afforestation was extracted by calculating ((NDVIApril≥0.22) ∩

(NDVIJune≥0.35)). Based on MS1 imagery in Fengqiu County, object-oriented

classification method was used. The spectral and spatial features of farmland

shelterbelt, road shelterbelt and residence shelterbelt samples were analyzed to

build membership function and develop classification rule set. The plain

afforestation areas were extracted according to the optimal segmentation scales

and certain classification features.

RESULTS Validation was conducted using correlation with the statistics of 26 major cities

in Henan Province, resulting in a significant R2=0.92. The algorithm performed

well in the wheat region, yet it did poor in the rice and rice-wheat transition

regions, which are mainly distributed in the southern part of Xinyang. Aside

from Xinyang, afforestation area extracted from HJ-1 is 6909.80 km2. This

accounts for 20.53% of the total forest in Henan Province.

CONCLUSIONS The results showed that the study achieved fine classification results.

And the good results demonstrated the potential of object-oriented plain

afforestation information extraction based on high-resolution images.

This method provides a technical support for accurate estimation of

plain afforestation area. Future study should make full use of HJ-1 high

temporal resolution data and MS1 high spatial resolution data by

conducting zoning according to the climate and the soil.

MAJOR REFERENCES [1]An K, Zhang J S, Xiao Y. Object-oriented Urban Dynamic

Monitoring-A Case Study of Haidian District of Beijing. Chinese

Geographical Science, 2007, 17(3): 236-242.

[2] Gutman G, Huang C, Chander G, Noojipady P, Masek J G.

Assessment of the NASA–USGS Global Land Survey (GLS) datasets.

Remote Sensing of Environment, 2013(134) : 249–265.

[2] Hansen C M, Loveland T R.A review of large area monitoring of

land cover change using Landsat data. Remote Sensing of Environment,

2012(122): 66–74.

[4] Montandon L M, Small E E. The impact of soil reflectance on the

quantification of the green vegetation fraction from NDVI. Remote

Sensing of Environment, 2008(112): 1835–1845.

[5] Wang H, Chen J S, Yu X M. Feature selection and its application

in object-oriented classification. Journal of Remote Sensing, 2013,

17(4): 816-829.

Within:Farmland

Segmentation:30

Brightness ≥ 12

NDVI ≥ 0.01

Within:

Construction land

Segmentation:20

NDVI ≥ 0.015

Within:

Residence land

Segmentation:10

6 ≤ Green ≤ 10

Fig.6 Afforestation spatial distribution of Fengqiu, 2010

Fig.1 Land cover of Henan Province, 2010

Fig.2 Fengqiu County and Mapping Satellite-1 false color

composite image with the validation samples, 2010

SATELLITE DATA • Spatial resolution of HJ-1 A/B CCD: 30m, Data: 2009-2011

• Spatial resolution of Mapping Satellite-1: 2m, Date: 2010-10-15

The results showed that the plain afforestation area of Fengqiu is

152.51 km2. More specifically, the farmland shelterbelt area is 36.09

km2, the road shelterbelt area is 21.29 km2, the residence shelterbelt

area is 71.56 km2, and the patched shelterbelt area is 23.57 km2. The

classification accuracy is 93.50% and the Kappa coefficient is 0.92.

afforestation

Xinyang

Fig.5 Afforestation spatial distribution of Henan, circa 2010

Fig.4 Afforestation correlation for 26 major cities: statistics vs. HJ-1 A/B CCD

statistics area

Interp

retation area fro

m H

J-1 A

/B (k

m2)

Statistics area (km2)

Fig.3 Extraction of farmland shelterbelt, road shelterbelt and residence shelterbelt