TY - JOUR
T1 - Capability of Sentinel-2 MSI data for monitoring and mapping of soil salinity in dry and wet seasons in the Ebinur Lake region, Xinjiang, China
AU - Wang, Jingzhe
AU - Ding, Jianli
AU - Yu, Danlin
AU - Ma, Xuankai
AU - Zhang, Zipeng
AU - Ge, Xiangyu
AU - Teng, Dexiong
AU - Li, Xiaohang
AU - Liang, Jing
AU - Lizaga, Ivan
AU - Chen, Xiangyue
AU - Yuan, Lin
AU - Guo, Yahui
N1 - Publisher Copyright:
© 2019 Elsevier B.V.
PY - 2019/11/1
Y1 - 2019/11/1
N2 - Soil salinization is one of the most important causes for land degradation and desertification and is an important threat to land management, farming activities, water quality, and sustainable development in arid and semi-arid areas. Soil salinization is often characterized with significant spatiotemporal dynamics. The salt-affected soil is predominant in the Ebinur Lake region in the Northwestern China. However, detailed local soil salinity information is ambiguous at the best due to limited monitoring techniques. Nowadays, the availability of Multi-Spectral Instrument (MSI) onboard Sentinel-2, offers unprecedented perspectives for the monitoring and mapping of soil salinity. The use of MSI data is an innovative attempt for salinity detection in arid land. We hypothesize that field observations and MSI data and MSI data-derived spectral indices using the partial least square regression (PLSR) approach will yield fairly accurate regional salinity map. Based on electrical conductivity of 1:5 soil:water extract (EC) of 72 ground-truth measurements (out of 116 sample sites) and various spectral parameters, such as satellite band reflectance, published satellite salinity indices, red-edge indices, newly constructed two-band indices, and three-band indices from MSI data, we built a few inversion models in an attempt to produce the regional salinity maps. Different algorithms including Pearson correlation coefficient method (PCC), variable importance in projection (VIP), Gray relational analysis (GRA), and random forest (RF) were applied for variable selection. The results suggest that both the newly proposed normalized difference index (NDI) [(B12 − B7) / (B12 + B7)] and three-band index (TBI4) [(B12 − B3) / (B3 − B11)] show a better correlation with validation data and could be applied to estimate the soil salinity in the Ebinur Lake region. The established models were validated using the remaining 44 independent ground-based measurements. The RF-PLSR model performed the best across the five models with R2 V, RMSEV, and RPD of 0.92, 7.58 dS m−1, and 2.36, respectively. The result from this model was then used to map the soil salinity over the study area. Our analyses suggest that soil salinization changes quite significantly in different seasons. Specifically, soil salinity in the dry season was higher than in the wet season, mostly in the lake area and nearby shores. We contend that the results from the study will be useful for soil salinization monitoring and land reclamation in arid or semi-arid regions outside the current study area.
AB - Soil salinization is one of the most important causes for land degradation and desertification and is an important threat to land management, farming activities, water quality, and sustainable development in arid and semi-arid areas. Soil salinization is often characterized with significant spatiotemporal dynamics. The salt-affected soil is predominant in the Ebinur Lake region in the Northwestern China. However, detailed local soil salinity information is ambiguous at the best due to limited monitoring techniques. Nowadays, the availability of Multi-Spectral Instrument (MSI) onboard Sentinel-2, offers unprecedented perspectives for the monitoring and mapping of soil salinity. The use of MSI data is an innovative attempt for salinity detection in arid land. We hypothesize that field observations and MSI data and MSI data-derived spectral indices using the partial least square regression (PLSR) approach will yield fairly accurate regional salinity map. Based on electrical conductivity of 1:5 soil:water extract (EC) of 72 ground-truth measurements (out of 116 sample sites) and various spectral parameters, such as satellite band reflectance, published satellite salinity indices, red-edge indices, newly constructed two-band indices, and three-band indices from MSI data, we built a few inversion models in an attempt to produce the regional salinity maps. Different algorithms including Pearson correlation coefficient method (PCC), variable importance in projection (VIP), Gray relational analysis (GRA), and random forest (RF) were applied for variable selection. The results suggest that both the newly proposed normalized difference index (NDI) [(B12 − B7) / (B12 + B7)] and three-band index (TBI4) [(B12 − B3) / (B3 − B11)] show a better correlation with validation data and could be applied to estimate the soil salinity in the Ebinur Lake region. The established models were validated using the remaining 44 independent ground-based measurements. The RF-PLSR model performed the best across the five models with R2 V, RMSEV, and RPD of 0.92, 7.58 dS m−1, and 2.36, respectively. The result from this model was then used to map the soil salinity over the study area. Our analyses suggest that soil salinization changes quite significantly in different seasons. Specifically, soil salinity in the dry season was higher than in the wet season, mostly in the lake area and nearby shores. We contend that the results from the study will be useful for soil salinization monitoring and land reclamation in arid or semi-arid regions outside the current study area.
KW - Red-edge
KW - Remote sensing
KW - Sentinel-2
KW - Soil salinity
KW - Spectral indices
UR - http://www.scopus.com/inward/record.url?scp=85068520510&partnerID=8YFLogxK
U2 - 10.1016/j.geoderma.2019.06.040
DO - 10.1016/j.geoderma.2019.06.040
M3 - Article
AN - SCOPUS:85068520510
SN - 0016-7061
VL - 353
SP - 172
EP - 187
JO - Geoderma
JF - Geoderma
ER -