Search

Soil salinization is one of the main problems hindering the agricultural development in the arid and semi-arid regions. However, many previous studies of soil salinization monitoring failed to consider the impact of irrigation on soil salinity. To explore such impact, we used a UAV with a portable spectrometer to monitor the dynamic change of soil salinity before and after spring irrigation. In this study, 120 soil samples at the surface (0–10 cm) were taken in Shahaoqu Irrigation Area, Inner Mongolia, China, in mid-April (before spring irrigation) and mid-June (after spring irrigation), and the images of the four study zones A, B, C and D in this area were obtained from the UAV system. Based on these images, 25 spectral covariates (6 spectral bands, 16 spectral indices, and 3 two-dimensional indices) were calculated. Then, the sensitive spectral covariates were selected with such different variable selection methods as variable importance in projection (VIP), competitive adaptive reweighted sampling (CARS), and genetic algorithm (GA). Finally, Multiple Linear Regression (MLR) and BP neural network algorithm (BPNN) was used to establish models for soil salinity inversion, based on which salinity maps of the study area were plotted. The results showed that the soil salinity of the study zone A (covered with vegetation) increased while that of the other three zones decreased after the spring irrigation. The general tendency was: the higher the original soil salinity was, the more obvious decrease the soil salinity had after the irrigation. All the three variable selection methods improved the inversion model accuracy, but the GA-BPNN model had the best performance (RP2 = 0.78, RMSEP = 0.16, and RPD = 2.13 before spring irrigation; and RP2 = 0.80, RMSEP = 0.14, and RPD = 2.26 after spring irrigation). It indicated that relatively accurate regional salinity maps could be plotted based on the spectral indices selected by GA and the salinity inversion model built on BPNN. These results have certain reference for soil salinization monitoring and farmland irrigation using UAV multispectral remote sensing. [ABSTRACT FROM AUTHOR]

Soil moisture and salinity are both important environmental variables for crop growth in agricultural production areas. Optical remote-sensing datasets from different sensors are available for estimating soil moisture and salinity from different spatial-temporal scales. Given the co-regulation of soil spectral reflectance (SR) by soil moisture and salinity, the simultaneous estimation of moisture and salinity in saline soil may result in great bias and uncertainty. To address this problem, soil samples were collected in the salinized area during irrigation. Synchronously, processed multi-spectral images were acquired from Sentinel-2 satellite. The spectrum mechanism responsive to soil moisture and salinity was verified by statistical tests, and its corresponding mathematical model (MSS model) was developed to identify the dominant factors affecting SR and to inverse moisture and salinity. The result showed that the effects of moisture and salinity were temporally constant (facilitation) and changing (from inhibition to facilitation), respectively, during the irrigation stages. The dominant factors in the variation of SR shifted from salinity and moisture-salinity interaction to moisture. Reliable accuracy was achieved in the moisture and salinity estimation using inverse MSS model. The profile from the series of estimations can further reveal the dynamic changes of soil moisture and salinity content during irrigation, and provide guidance for local irrigation management. • MSS model simulates response of soil spectrum to soil moisture and salinity. • Inverse MSS model achieves the simultaneous estimation at regional scale. • Spatial-temporal change of soil moisture and salinity is observed during irrigation. • Land-vegetation alternation causes soil salt aggregation in crop field. [ABSTRACT FROM AUTHOR]