Skip to main content
ARS Home » Research » Publications at this Location » Publication #314930

Title: Regional-scale soil salinity assessment using Landsat ETM+ canopy reflectance

item SCUDIERO, ELIA - University Of California
item Skaggs, Todd
item Corwin, Dennis

Submitted to: Remote Sensing of Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/23/2015
Publication Date: 11/1/2015
Citation: Scudiero, E., Skaggs, T.H., Corwin, D.L. 2015. Regional-scale soil salinity assessment using Landsat ETM+ canopy reflectance. Remote Sensing of Environment. 169(C):335-343. doi: 10.1016/j.rse.2015.08.026.

Interpretive Summary: Despite decades of research on soil mapping using remote sensing data, the spatial-characterization of soil salinity in agricultural fields, at high spatial resolution, remains a challenge. Especially in the ranges of soil salinity that are of particular interest to farmers and policy makers, reliable maps with low prediction errors are not available. For the first time, reliable soil salinity predictions in the cultivable ranges of soil salinity are presented using a regional-scale case study, with spatial resolution of 30×30 m. Seven years of Landsat 7 canopy reflectance, joined with information on meteorological settings (yearly total rainfall and average minimum temperature) and crop type (i.e., cropped or fallow) could predict the soil salinity of 22 fields (542 ha or ~1340 acres) scattered throughout the western San Joaquin Valley, California, with low mean absolute errors (MAE) of 2.94, 2.12, 2.35, 3.23, and 5.64 dS/m over the 0-2 (non-saline), 2-4 (slightly saline), 4-8 (moderately saline), 8-16 (strongly saline), and >16 (extremely saline) dS/m1 salinity (measured as electrical conductivity of a saturated paste extract) intervals. This study shows that fairly accurate and discretely precise high resolution regional-scale remote sensing of soil salinity is possible for the critical ranges of salinity that are of great importance to farmers, land resource managers, agriculture consultants, extension specialists, and Natural Resource Conservation Service field staff when planning agronomical practices, dealing with water allocations, and managing soil quality.

Technical Abstract: Salinity is widely recognized to be one of the major threats for worldwide agriculture. Despite decades of research in soil mapping, no reliable and up-to-date maps are available for wide geographical regions, especially in agronomically and environmentally relevant salinity ranges (i.e., <20 dS/m, when measured as electrical conductivity of the saturation extract). This paper explores the potentials and constraints of assessing soil salinity with linear modeling. A case study is presented using multi-year Landsat 7 ETM+ canopy reflectance, over the western San Joaquin Valley (ca.1.5 × 10**6 ha), California, USA. In 2013, 22 fields comprising 542 ha were mapped for soil salinity using geospatial electromagnetic induction measurements calibrated (R2=0.91) with actual root-zone (0-1.2m) soil analyses at 267 locations. The salinity maps (5891 pixels, 30×30 m) returned ground-truth values in the 0-35.2 dS/m range. The multi-year maxima values of the Canopy Response Salinity Index (CRSI) were used to model soil salinity. The use of soil type, meteorological, and crop type covariates was evaluated. The best relationship with salinity was a function of CRSI, crop type (i.e., cropped or fallow), rainfall, and average minimum temperature, returning a R2=0.728. All presented assessment models were validated with a leave-one-field-out spatial cross-validation, to assure maximum independence between training and validation datasets. The best performing salinity assessment model validation returned R2=0.611, mean absolute error (MAE) of 2.94, 2.12, 2.35, 3.23, and 5.64 dS/m over the 0-2 (non-saline), 2-4 (slightly saline), 4-8 (moderately saline), 8-16 (strongly saline), and >16 (extremely saline) dS/m salinity intervals, respectively. By-field, the validation predictions showed very strong correlation coefficients and fairly low MAE with the observed field average (R2=0.79, MAE=2.46 dS/m), minimum (r=0.76, MAE=2.25 dS/m), and maximum (r=0.76, MAE=3.09 dS/m) salinity. To further understand the uncertainty of cross-validation predictions across models, an analysis of the absolute error distribution was suggested. Fairly accurate and discretely precise high resolution regional-scale remote sensing of soil salinity is possible, even over the critical ranges of 0 to 20 dS/m, where much of the attention from policy makers should focus to ameliorate future deterioration of agricultural productivity and ecosystem health through management decisions.