ASD | Soil moisture content estimate with drying process segmentation using shortwave infrared bands

Soil moisture content estimate with drying process segmentation using shortwave infrared bands

Soil moisture is an essential factor directly impacting a variety of environmental processes, including evaporation, infiltration, and runoff, etc. Moreover, soil moisture also plays an important role in related fields of studies, such as evapotranspiration and food security for agriculture, wetland degradation for ecosystem balance, drought for climate change, and even energy exchange between land-atmosphere interface.

Ground-based measurements can provide data that are easily calibrated and acquired continuously over an extended time. However, each measurement is only targeting a single, small area, making it difficult to support spatial variation studies, or field studies. Microwave remote sensing is widely used for soil moisture monitoring over large spatial scales, based on the large contrast between the dielectric properties of water and soil. But it is not suitable for various researches such as precision agriculture. Thermal remote sensing may also be used to estimate soil moisture based on the land surface temperature, but the signal from thermal remote sensing is not solely impacted by soil moisture content; other parameters like humidity, wind speed, or atmospheric condition also influence the estimated result. Optical remote sensing draws a lot of attention due to its balance of fine spatial resolution and potential for large-scale monitoring with satellite data such as MODIS, the Landsat series and Sentinel missions. Many indices and models have been proposed to illustrate the reflectance features changing with SMC, and to estimate SMC from the reflectance of narrow bands and broadbands, using lab, field, airborne and satellite data. These methods/indices were mostly developed for all levels of SMCs, from saturated to air-dry; however, the authors found that a single relationship mapping saturated to air-dry can lead to a false impression of accurate estimation.

Based on this, to fill these knowledge gaps, in this article, a group of international scientists from Nanjing University, Cornell University and Henan Agricultural University proposed a segmentation method with the goal of a more accurate SMC estimate. The whole drying process of three soil samples, representing varied soil characteristics, is monitored, and the transition points are determined by evaporation rate change (as stage-1 drying with high SMC and stage-2 drying with low SMC). An index, the shortwave normalized index (SNI), is built for SMC estimation, and the SNI index trend during drying is supported by a radiative transfer-based model.

Fig. 1. Sketch of experimental setup. An ASD Fieldspec® Pro spectrometer was used for spectral radiance collection.

[Results]

Fig. 2. a) Evaporation rate change vs. elapsed time of drying for three soil samples, and b) reflectance at 2150 nm of three soils changes during drying. c) The maximum value of derivative of evaporation rate for three samples determines the drying stage segmentation point.

Fig. 3. R² of the linear and logarithmic regression between sand/soil moisture content and spectral reflectance for three samples, a) for quartz sand, b) for masonry sand, and c) for Ithaca soil, and d) simulated atmospheric transmittance. In a), b) and c), the black dashed lines mark 1680 nm and 2150 nm.

Fig. 4. a) Shows the validation result of SMC estimation. b), c), and d) shows the modeled curve (solid lines), regression curve (dashed lines), and validation dataset (open circles) for three samples.

Fig. 5. a) Estimated SMC vs. measured SMC, where the estimated SMC is calculated using SNI2 in linear regression with B_water evaluated at 1980 nm. Figures b), c), and d) show the modeled curve (solid lines), regression curve (dashed lines), and validation dataset (open circles) for the three samples.

[Conclusions]

1. Drying process segmentation is necessary for accurate SMC estimates from spectral reflectance data, especially for soils with long stage-2 drying, e.g. Ithaca soil in this work. SMC estimation based on the whole drying process for a soil similar to the Ithaca soil may lead bias in either stage-1 or stage-2 drying, depending on in which stage is represented by more training.

2. Due to the relatively long path length through water for quartz sand, SNI has unique features when SMC is high. The half-logistic- like SNI curve is different from linear relation in masonry sand or Ithaca soil. When the optical path length is long, the fitting relation should be changed from linear to logarithmic regression.

3. In stage-2 drying, SMC is difficult to estimate accurately using spectral band combination generally available on current satellite systems; higher accuracy is achievable using hyperspectral data which can provide data near to strong water absorption bands, e.g., 1930 nm. Although 1930 nm cannot be used effectively outside of the laboratory due to atmospheric water vapor absorption, wavelengths slightly off the center (such as 1980 nm) can still perform better than wavelengths outside of the water absorption band range.