TY - JOUR
T1 - Estimation of water-infiltration rate in Mediterranean sandy soils using airborne hyperspectral sensors
AU - Francos, Nicolas
AU - Chabrillat, Sabine
AU - Tziolas, Nikolaos
AU - Milewski, Robert
AU - Brell, Maximilian
AU - Samarinas, Nikiforos
AU - Angelopoulou, Theodora
AU - Tsakiridis, Nikolaos
AU - Liakopoulos, Vasillis
AU - Ruhtz, Thomas
AU - Ben-Dor, Eyal
N1 - Publisher Copyright:
© 2023 Elsevier B.V.
PY - 2023/12
Y1 - 2023/12
N2 - The efficiency of spectral-based assessments of soil attributes using soil spectral libraries (SSLs) covering the visible–near-infrared–shortwave-infrared (VNIR–SWIR: 400–2500 nm) region has been proven in many studies. Nevertheless, as traditional SSLs are commonly developed under laboratory conditions, their application is limited for the assessment of soil surface-dependent properties such as water-infiltration rate (WIR) into the soil profile due to the sampling procedure. Currently, few studies are based on field SSLs for the prediction of physical soil properties. This study used a field-based protocol to measure soil reflectance data and WIR simultaneously in the field, and generate spectral-based decision tree models to predict WIR solely from field spectral measurements using the SoilPRO® assembly. The obtained models were applied to both airborne hyperspectral (HySpex) and satellite multispectral (Sentinel 2) data on a pixel-by-pixel basis to generate raster maps of WIR. The study areas were located in Macedonia (Greece), and were optimal for mapping WIR because the soil crust was well developed, and sites were characterized by bare soils (no vegetation coverage) with a sandy structure. Whereas the WIR map generated with the satellite data was poor due to the low spatial and spectral resolution of Sentinel 2 (20 m, 9 bands), the results obtained with the airborne hyperspectral HySpex sensor (5 m, 408 bands) were satisfactorily validated in the ground-truth stage with good prediction accuracy due to high spatial and spectral resolution. Validation accuracy of the HySpex observations using all field samples gave R2 = 0.68, whereas the predictions of the ground-truth samples that were not part of the calibration stage (field validation group) of the model gave R2 = 0.59. We concluded that these results are favourable for rapid estimation of soil surface conditions and pave the way for a wider spatial view from orbital hyperspectral remote-sensing sensors.
AB - The efficiency of spectral-based assessments of soil attributes using soil spectral libraries (SSLs) covering the visible–near-infrared–shortwave-infrared (VNIR–SWIR: 400–2500 nm) region has been proven in many studies. Nevertheless, as traditional SSLs are commonly developed under laboratory conditions, their application is limited for the assessment of soil surface-dependent properties such as water-infiltration rate (WIR) into the soil profile due to the sampling procedure. Currently, few studies are based on field SSLs for the prediction of physical soil properties. This study used a field-based protocol to measure soil reflectance data and WIR simultaneously in the field, and generate spectral-based decision tree models to predict WIR solely from field spectral measurements using the SoilPRO® assembly. The obtained models were applied to both airborne hyperspectral (HySpex) and satellite multispectral (Sentinel 2) data on a pixel-by-pixel basis to generate raster maps of WIR. The study areas were located in Macedonia (Greece), and were optimal for mapping WIR because the soil crust was well developed, and sites were characterized by bare soils (no vegetation coverage) with a sandy structure. Whereas the WIR map generated with the satellite data was poor due to the low spatial and spectral resolution of Sentinel 2 (20 m, 9 bands), the results obtained with the airborne hyperspectral HySpex sensor (5 m, 408 bands) were satisfactorily validated in the ground-truth stage with good prediction accuracy due to high spatial and spectral resolution. Validation accuracy of the HySpex observations using all field samples gave R2 = 0.68, whereas the predictions of the ground-truth samples that were not part of the calibration stage (field validation group) of the model gave R2 = 0.59. We concluded that these results are favourable for rapid estimation of soil surface conditions and pave the way for a wider spatial view from orbital hyperspectral remote-sensing sensors.
KW - Decision tree
KW - HySpex
KW - Hyperspectral remote sensing
KW - Mediterranean soil
KW - Sandy soil
KW - Sentinel 2
KW - Soil spectroscopy
KW - Water-infiltration rate
UR - http://www.scopus.com/inward/record.url?scp=85169844450&partnerID=8YFLogxK
U2 - 10.1016/j.catena.2023.107476
DO - 10.1016/j.catena.2023.107476
M3 - ???researchoutput.researchoutputtypes.contributiontojournal.article???
AN - SCOPUS:85169844450
SN - 0341-8162
VL - 233
JO - Catena
JF - Catena
M1 - 107476
ER -