Abstract
Using plastic film mulch on cropland improves crop yield in water-deficient areas, but the use of plastic film on cropland leads to soil pollution. The accurate mapping of plastic-mulched land (PML) is valuable for monitoring the environmental problems caused by the use of plastic film. The drawback of PML mapping is that the detectable period of PML changes among the fields, which causes uncertainty when supervised classification methods are used to identify PML. In this study, a new workflow which merging PML of multiple temporal phases (MTPML) is proposed. For each temporal phase, the “possible PML” is firstly generated, these “temporal possible PML” layers are then combined to generate the “possible PML” layer. Finally, the maximum normalized difference vegetation index (NDVI) of the growing season is used to remove the non-cropland pixels from the “possible PML layer,” and then generate PML images. When generating “temporal possible PML layers,” three new PML indices (PMLI with near-infrared bands known as PMLI_NIR, PMLI with shortwave infrared bands known as PMLI_SWIR, and Normalized Difference PMLI known as PMLI_ND) are proposed to separate PML from bare land at plastic film cover stage; and the “temporal possible PML layer” are identified by the threshold based method. To estimate the performance of the three PML indices, two other approaches, PMLI threshold and Random Forest (RF) are used to generate “temporal possible PML layer.” Finally, PML images generated from the five MTPML approaches are compared with the image time series supervised classification (SUPML) result. Two study regions, Hengshui (HS) and Guyuan (GY), are used in this study. PML identification models are generated using training samples in HS and the models are used for PML mapping in both study regions. The results showed that MTPML workflow outperformed SUPML with 3%–5% higher classification accuracy. The three proposed PML indices had higher separability and importance score for bare land and PML discrimination. Among the five approaches used to generate the “temporal possible PML layer,” PMLI_SWIR is the recommended approach because the PMLI_SWIR threshold approach is easy to implement and the accuracy is only slightly lower than the RF approach. It is notable that no training sample was used in GY and the accuracy of the MTPML approach was higher than 85%, which indicated that the rules proposed in this study are suitable for other study regions.
Highlights
Plastic mulch film has been widely used over soil surface to suppress weeds, conserve water, and regulate temperature during crop growth [1,2,3] and to improve the crop yield in arid/semi-arid regions [4,5,6,7]
normalized difference vegetation index (NDVI) and multi-spectral bands led to a poor performance when discriminating bare land from plastic-mulched land (PML)
This study proposes a new workflow (MTPML) for PML identification, and the performance of the threshold-based approaches using three new PML indices with multiple temporal phases (MTPML) are tested
Summary
Plastic mulch film has been widely used over soil surface to suppress weeds, conserve water, and regulate temperature during crop growth [1,2,3] and to improve the crop yield in arid/semi-arid regions [4,5,6,7]. Accurate data of the spatial distribution of plastic-mulched land (PML) at the local and regional scale is an important data source for both agricultural and environmental analyses. Satellite observations can be used for land cover mapping even from regional to global scales [12,13]. The identification of PML is more challenging than PG because, unlike greenhouses, which are not covered by the canopy during the growing season, PML is not detectable when the plants grow above the soil. Only a few studies have concentrated on mapping PML using satellite data. Hasituya and Chen [18] used multi-temporal Landsat-8 images and machine learning algorithms (Support Vector Machine and Random Forest) to identify PML, and found that the best temporal period for PML detection is between the time of planting and green-up [19,20]. When only using the SAR features to identify PML, the classification accuracies are low, and when adding SAR feature into optical data, the PML identification accuracies just improve by 1%~3% [2,22], so that the optical features still have high contribution to PML identification
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
