Evapotranspiration (AETI), Land Cover Classification (LCC), Net Primary Productivity (NPP), Precipitation (PCP), Phenology (PHE), Quality layers (QUAL), Reference Evapotranspiration (RET), Soil moisture (RSM), Total Biomass Production (TBP) and Water Productivity (WP)

Relative Root Zone Soil Moisture (RSM) data components (beta product)

Soil moisture availability is one of the most important parameters governing biomass production and evapotranspiration. Lack of soil moisture (soil moisture stress) can seriously hamper biomass growth by reducing vegetation transpiration. Soil moisture is directly released to the atmosphere from the top soil through evaporation and from the vegetation cover through transpiration.
Evaporation reduces as vegetation cover increases. Soils fully covered by vegetation experience very little evaporation as nearly all of the available energy is captured by the vegetation cover and used for transpiration. Transpiration drives the transport of soil moisture from the sub soil through plant roots. The root zone may hold more water and enables the plant to continue with transpiration even when the top soil is dry.

WaPOR data components

Relative soil moisture content and soil moisture stress is produced at both levels at a dekadal temporal resolution. Soil moisture stress is an intermediate data components that is used as input to other data components and are not published through WaPOR. Relative soil moisture content is published as a beta product.
Soil moisture content varies strongly in time and place. Within the WaPOR area of interest extremes occur in northern Africa and the Middle East where relative soil moisture content is very low throughout the year (with the exception of areas close to rivers) and the equatorial region which is characterised by high relative soil moisture content throughout the year. Other areas generally show more seasonal variation in soil moisture content. Pixel values of relative soil moisture content range between 0 and 1, where 0 is equal to the soil moisture content at wilting point and 1 is equal to the soil moisture content at field capacity. Soil moisture stress values also range between 0 and 1, where 0 means no stress and 1 maximum stress.

Table 1: Overview of the (intermediate) data components related to Soil Moisture

Methodology

The methodology applied for calculating relative soil moisture content and soil moisture stress is based on the correlation between Land Surface Temperature (LST, derived from thermal infrared imagery), vegetation cover (derived from the NDVI) and soil moisture content. This is also known as the triangle method (Carlson, 2007). External input data required are visual/NIR and thermal imagery. (An alternative approach is based on the use of radar imagery from ASCAT. WaPOR data production partners apply the LST method as it has a higher resolution and therefore provides a better representation of the spatial variability of soil moisture content. It is also a better indicator for the water content in the root zone in the sub-soil than radar methods which are only able to observe soil moisture content in the top layer of the soil. The moisture content of these two soil layers is not necessarily correlated. The results based on radar also tend to be less accurate for areas with moderate to dense vegetation cover.)

The triangle method is named after the shape of the scatter plot that emerges when all pixels in an image are plotted with NDVI on one axis and temperature on the other axis. Discarding outliers, a triangle shape appears, delineated by two marked boundaries (see Figure 1). These boundaries represent two physical conditions of water availability at the land surface, called the cold edge and the warm edge. At the cold edge, water is readily available and the soil moisture content is at field capacity. Evapotranspiration takes place at maximum rate, with the latent heat flux at its maximum and the sensible heat flux at zero. In this situation, the LST is close to the ambient air temperature. At the warm edge no soil moisture is available and evapotranspiration and the latent heat flux are equal to zero.

Incoming radiation increases LST. This increase depends on the vegetation cover (NDVI). The LST increase is highest when no vegetation is present and smallest when vegetation fully covers the land surface. Therefore, the difference between the cold and the warm edge is largest for bare soil and smallest for fully vegetated surfaces. In general, LST is lower when the soil moisture content and/or the vegetation cover are higher.

Figure 1: An example of a scatter plot of NDVI versus surface radiant temperature taken from Carlson (2007). The cold edge on the left side and the warm edge on the right side of the point cloud are clearly distinguishable.

A drawback of this method is that it requires calibration by manual selection of references pixels for each thermal image. This introduces subjectivity through the selection process and makes it difficult to operationalize for a larger area. This problem was overcome by the method developed by Yang et al. (2015). The original triangle method was modified by introducing the effect of stomatal closure of vegetation under dry condition as a result of water stress (Moran et al., 1994). As a result, the temperature of the warm edge at a fully vegetated surface becomes higher than under wet conditions. This results in a trapezoid shape as depicted in Figure 2, taken from the improved trapezoid method of Yang et al. (2015).

The trapezoid, corners numbered A, B, C, D, are defined by the linear relationship between LST and vegetation cover under the two extreme conditions of the cold edge and the warm edge. The top line segment (A – B) shows this relationship under completely dry conditions (no available soil moisture). Point A represents bare soil. Point B represents full vegetation cover. The bottom line segment (D – C) represents soil moisture at field capacity. Again, on the left side (D) for bare soil and on the right side (C) for full vegetation cover. This linear relationship between LST and vegetation cover (under equal soil moisture conditions) is not only true for the extreme conditions but for each value of the soil moisture content, as shown by the soil wetness isolines in below figure.

Figure 2: The trapezoidal vegetation coverage (Fc) / land surface temperature (LST) space (transposed axis). Points A, B, C and D are estimated for each separate pixel using modified Penman/Monteith equations. Source: Yang et al., 2015.

The relative soil moisture content of a specific location (e.g. point E) can be derived from its relative distance to the cold edge (a) and warm edge (b) using:

Where:

Solving these equations in order to derive the relative soil moisture content first requires calculation of the four corner points of the trapezoid (A – D) as well as information on vegetation cover and LST of point E. The NDVI intermediate data component is used to derive vegetation cover whilst LST is derived from thermal satellite imagery.

Assuming no sensible heat flux, the temperature of the cold edge (C and D) T_min is approximated by the wet-bulb temperature (T_wet) for bare soil (corner point D) and air temperature for full vegetation cover (corner point C) at around the same time as when the LST is measured. A linear relationship between the temperatures at corner point D and corner point C provide T_min for all other vegetation covers. The wet bulb temperature is defined as the minimum temperature which may be achieved by bringing an air parcel to saturation by evaporation in adiabatic conditions (Monteith & Unsworth, 2013). Thus the cold edge conditions are considered to be such that there is enough soil moisture and a sufficient evaporation rate to reach saturation of the cooling air and therefore for the temperature to approach T_wet }. Compared to the cold edge, calculating the corner points A and B of the warm edge requires more effort. This is done with the Penman-Monteith equation rewritten to yield T_max at point A and B. We provide an overview of the steps below, more detail can be found in Yang et al. (2015).

At the warm edge, a large part of the incoming radiation is used for heating the land surface, thus increasing LST. The amount of energy available depends on the incoming solar radiation (R_s) and net long wave radiation L*. The surface albedo (α) is an important factor in determining how much of this energy is retained to heat the land surface. This requires the deduction of two theoretical albedo values, one for bare soil (point A) and one for full vegetation cover (point B). Soils generally have a higher albedo, reflecting more of the incoming radiation than vegetated cover. Theoretical values can be derived from the land cover class and soil type maps. Here it is derived from the surface albedo intermediate data component.

Part of the warming of the land surface is lost again through the sensible heat flux (H). The sensible heat flux depends on the aerodynamic resistance to heat transfer determined by soil and canopy characteristics. Bare soils have a higher resistance than vegetation due to the lower surface roughness, resulting in a lower sensible heat flux. Surface roughness is derived from the land cover class.

The method to calculate the aerodynamic resistance to heat transfer is based on equations A1 (full canopy), A6 (bare) and A7 (soil) of Sánchez, Caselles and Kustas (2008). The aerodynamic resistance is strongly determined by wind speed which in WaPOR is derived from model data such as GEOS-5 and ERA5. Under certain conditions the GEOS-5 and ERA5 windspeed get unrealistically low. In reality, if the surface heat flux gets sufficiently high, it will generate turbulence which provides a negative feedback on that heat flux and surface temperature. With the WaPOR v2 model, this feedback was missing which led to conditions with very low wind (< 1 m/s) yet at the same time a very high heat flux, and unrealistically high (dry) surface temperatures.

For these situations with free convection, an alternative method to calculate the aerodynamic resistance is used. Garrat (1992) explored the theoretical upper limit of the land surface temperature and proposes to calculate the air-surface temperature difference with an alternative relation when wind speeds are low and the friction velocity is no longer a physically meaningful quantity:

with:
k = von Karman constant
g = gravitational constant
H = heat flux
ρ = air density
c_p = specific heat
z_t = temperature scaling surface length θ = potential temperature

The heat flux can be written as:

When rearranged to solve for r_a,h:

Assuming ΔT ≈ Δθ and by inserting equation 15 from Garrat (1992), the r_a,h for the case of free convection can be calculated (independently of wind or friction velocity):

Finally, the aerodynamic resistance is computed for dry, bare soil and a full canopy; we adopt the lowest out of the two values computed for forced and free convection.

For bare soil, the soil heat flux (G) also has to be included, assuming a fixed fraction of the net radiation of 0.35. Soil heat flux does not need to be included for a fully vegetated surface as the soil surface is not directly heated by incoming radiation.

This method is applied on a pixel-by-pixel basis with no spatial dependencies, making it possible to apply the same methodology for different regions in a consistent manner. However, parameterising the soil moisture algorithm on a continental scale is challenging, particularly at continental level where relative soil moisture content, vegetation cover and weather conditions vary greatly (e.g. the dry Saharan desert and the wet tropical rainforests present extreme opposites). A specific challenge lies in the determination of the reference values for the corner points of the warm edge. Calculation of these hypothetical values depends on a number of assumptions under extreme conditions which can be challenging to estimate. The surface albedo intermediate data component is used to provide the minimum and maximum surface albedo which is input to the Yang algorithm. The surface albedo for point A (high surface albedo) and point B (low surface albedo) has been determined with the use of the albedo time series for each pixel, obtained from the albedo intermediate data component. By using these values instead of constant values, it is ensured that the theoretical maximum LST is being derived using realistic surface albedo values.

The relative soil moisture content is determined for both the top soil and the root zone. Therefore the same relative soil moisture content is used for the determination of evaporation and transpiration, albeit in different formulations. Soil moisture stress limits transpiration by means of the canopy resistance. For evaporation the relative soil moisture content is used to model the soil resistance. The vegetation cover determines the route of the water flow, i.e. through transpiration or evaporation.

By using the relative soil moisture content the model is able to separate between evaporation and transpiration. Some studies use the triangle/trapezoid method to calculate the evaporative fraction directly, but then it is not possible to make the distinction between transpiration and evaporation. Hence the need for the ETLook model. The relative soil moisture content determines the availability of water for evaporation and transpiration. Whether this is reduced due to a shortage can be calculated with a stress factor. This stress factor for transpiration (S_m) can be derived using the following relationship as defined in American Society of Civil Engineers (ASCE, 1996):

The tenacity factor K_sf ranges from 1 for drought-sensitive plants to 3 for drought-insensitive (tenacious) plants. In version 2 a default value of 1.5 was chosen when no crop information is available but for version 3 a value of 2.0 is being tested. This will lead to an increase in transpiration at similar soil moisture content values.

This soil moisture stress factor, ranging between 1 and 0, is used as input for the E and T to reduce evapotranspiration.

Operational approach

Daily LST is used to calculate instantaneous relative soil moisture content. This time serie of relative soil moisture is smoothened and interpolated to daily relative soil moisture by applying a pixel based temporal fill algorithm: the Whittaker smoother (Eilers, 2003, Eilers et al., 2017). We use a weighted smoother, with weights depend on the land surface temperature input (viewing angle and distance to cloud) and are normalized between 0 and 1.

Challenges

The soil moisture model is very sensitive to the correct estimation of the cornerpoints of the trapezoid, which again is sensitive to the quality of meteorological and LST inputs.

Soil moisture content varies strongly in time and place. The WaPOR-ETLook model uses Land Surface Temperature as a proxy for soil moisture.

Functions and flowcharts

The soil moisture algorithm follows the procedure outlined by Yang et al. (2015) and calculates the four cornerpoints of the vegetation cover - land surface temperature (LST) trapezoid. The actual vegetation cover and LST value are then used to estimate the relative soil moisture content and the soil moisture stress. The calculation procedure is divided into the following smaller steps:
- Air density (input to the other calculations)
- Atmospheric emissivity (input to the other calculations)
- Bare soil maximum temperature (one cornerpoint)
- Fully canopy maximum temperature (one cornerpoint)
- Soil moisture

Air density (NRT processing, using GEOS-5)

Air density (final processing, using ERA5)

Atmospheric emissivity (NRT processing, using GEOS-5)

Atmospheric emissivity (final processing, using ERA5)

Bare soil maximum temperature

Full canopy maximum temperature

Soil moisture