et al. 2014). For example, the PML area in China has grown from 4200 ha in 1981,
10,620,000 ha in 2000, to 18,140,255 ha in 2014, which is about 15% of the total
farmland. Such a large-scale land-cover change must have impacted on climate,
ecosystem, and environment regionally and globally because it alters the energy
balance and water cycles on the land surfaces, deteriorates the soil structure, and
reduces the biodiversity (Lu et al. 2014). The first step toward understanding the
overall impact is to map and monitor PML in a large geographic area. Therefore,
mapping and monitoring the plasticulture in a large geographic area is important
both scientifically and socio-economically.
Remote sensing is the only feasible approach for monitoring plasticulture and
understanding its impacts on climate and eco-environment in a large geographic area
(e.g., whole China or whole East Asia). Over the last two decades, plasticulture
detection methods have been developed mainly to extract PML/greenhouse distribution from high (at meter level) spatial resolution imagery. For example, Levin
et al. (2007) claimed that there were three absorption bands (centered at 1218 nm,
1732 nm, and 2313 nm, respectively) for clear plastic and used 1 m resolution AISAES hyper-spectral image data to achieve an above 90% detection accuracy for clear
plastic cover and about 70% for black plastic cover. Agüera and Liu (2009) used the
maximum likelihood classification method to extract greenhouse locations from both
Quickbird and IKONOS images and achieved satisfactory detection accuracy.
Takakura and Fang 2002successfully mapped rural areas with widespread plasticcovered (tunnel/greenhouse) vineyards using both spectral information and spatial
texture from very high-spatial-resolution true color aerial images. Carvajal et al.
(2006) developed an artificial intelligence neural network to identify greenhouse
using 2.44 m resolution QuickBird imagery. Hasituya et al. (2017) used a support
vector machine (SVM) algorithm to detect PML from GaoFen-1 satellite imagery.
A number of studies on mapping plasticulture based on medium resolution
(at tens meter level) imagery have also been performed. For example, Lu et al.
(2014) developed a decision-tree classifier for extracting the transparent PML
information from Landsat-5 TM images. Picuno et al. (2011) applied the parallelepiped method to detect PML from multi-temporal Landsat TM images. Noveli and
Tarantifo (2015) combined the green normalized difference vegetation index
(NDVI) with three other ad hoc spectral indices (rescaled brightness temperature,
plastic surface index, and normalized difference sandy index) to extract plastic
mulched vineyards from Landsat-8 operational land imager (OLI) data. Hasituya
et al. (2016) proposed a scheme of combing spectral and textural features for
monitoring PML from Landsat-8 OLI images. Novelli et al. (2016) and Wu et al.
(2016) combined object-based and random forest techniques for mapping greenhouse from Landsat-8 OLI data. Aguilar et al. (2016) combined an object-based
technique and a decision-tree classifier for mapping greenhouse through the combined use of WorldView-2 and Land 8 OLI data time series.
Since PML is by far the largest type of plasticulture in terms of the area it covers
and in China 95% of plasticulture exists in mulch form (Zhou 2010), this study only
deals with PML. The time for applying plastic film as mulch is normally concentrated on a short period (e.g., within 1 week) during the planting stage of a growing
17 Remote Sensing–Based Mapping of Plastic-Mulched Land Cover
353
10,620,000 ha in 2000, to 18,140,255 ha in 2014, which is about 15% of the total
farmland. Such a large-scale land-cover change must have impacted on climate,
ecosystem, and environment regionally and globally because it alters the energy
balance and water cycles on the land surfaces, deteriorates the soil structure, and
reduces the biodiversity (Lu et al. 2014). The first step toward understanding the
overall impact is to map and monitor PML in a large geographic area. Therefore,
mapping and monitoring the plasticulture in a large geographic area is important
both scientifically and socio-economically.
Remote sensing is the only feasible approach for monitoring plasticulture and
understanding its impacts on climate and eco-environment in a large geographic area
(e.g., whole China or whole East Asia). Over the last two decades, plasticulture
detection methods have been developed mainly to extract PML/greenhouse distribution from high (at meter level) spatial resolution imagery. For example, Levin
et al. (2007) claimed that there were three absorption bands (centered at 1218 nm,
1732 nm, and 2313 nm, respectively) for clear plastic and used 1 m resolution AISAES hyper-spectral image data to achieve an above 90% detection accuracy for clear
plastic cover and about 70% for black plastic cover. Agüera and Liu (2009) used the
maximum likelihood classification method to extract greenhouse locations from both
Quickbird and IKONOS images and achieved satisfactory detection accuracy.
Takakura and Fang 2002successfully mapped rural areas with widespread plasticcovered (tunnel/greenhouse) vineyards using both spectral information and spatial
texture from very high-spatial-resolution true color aerial images. Carvajal et al.
(2006) developed an artificial intelligence neural network to identify greenhouse
using 2.44 m resolution QuickBird imagery. Hasituya et al. (2017) used a support
vector machine (SVM) algorithm to detect PML from GaoFen-1 satellite imagery.
A number of studies on mapping plasticulture based on medium resolution
(at tens meter level) imagery have also been performed. For example, Lu et al.
(2014) developed a decision-tree classifier for extracting the transparent PML
information from Landsat-5 TM images. Picuno et al. (2011) applied the parallelepiped method to detect PML from multi-temporal Landsat TM images. Noveli and
Tarantifo (2015) combined the green normalized difference vegetation index
(NDVI) with three other ad hoc spectral indices (rescaled brightness temperature,
plastic surface index, and normalized difference sandy index) to extract plastic
mulched vineyards from Landsat-8 operational land imager (OLI) data. Hasituya
et al. (2016) proposed a scheme of combing spectral and textural features for
monitoring PML from Landsat-8 OLI images. Novelli et al. (2016) and Wu et al.
(2016) combined object-based and random forest techniques for mapping greenhouse from Landsat-8 OLI data. Aguilar et al. (2016) combined an object-based
technique and a decision-tree classifier for mapping greenhouse through the combined use of WorldView-2 and Land 8 OLI data time series.
Since PML is by far the largest type of plasticulture in terms of the area it covers
and in China 95% of plasticulture exists in mulch form (Zhou 2010), this study only
deals with PML. The time for applying plastic film as mulch is normally concentrated on a short period (e.g., within 1 week) during the planting stage of a growing
17 Remote Sensing–Based Mapping of Plastic-Mulched Land Cover
353