An appraisal of the development and testing of SPLASH v2.0: Water and energy fluxes

David Sandoval 

SPLASH v2.0 (Simple process-led algorithms for simulations habitats) (SPLASH v2.0) builds on SPLASHv1.0 to allow robust calculations of water and energy fluxes in complex terrain. It is a simple, parsimonious model based on first principles, which does not require local calibration and uses minimum meteorological inputs (precipitation, air temperature and solar radiation). It works using analytical solutions for the integrals of the daily cycles and soil hydraulic transmissivity allowing the model to perform calculations at high spatial resolution without inflating the need for computational power. Here is an appraisal of the development and testing of SPLASH v2.0 and how it compares to other predictive modelling tools on various aspects of ecohydrology: 

Net Radiation 

Using analytical integrals of the theoretical daily cycles of energy fluxes plus newly optimized empirical equations to calculate longwave radiation components we were able to accurately predict the daytime accumulated net radiation in 212 flux towers from the FLUXNET database. SPLASHv2.0 compares very well with more sophisticated models like VIC, which uses a pre-defined Leaf Area Index (LAI), plant functional types and many other parameters. 

Figure 1. Net radiation calculation. (a) Conceptualization of the net radiation flux between solar noon (i.e., h = 0) and solar midnight (i.e., h = π). (b) Correlation of observed and simulated values of daytime net radiation with data from all the FLUXNET sites pooled. 

Evapotranspiration 

There are a lot of uncertainties in calculating evapotranspiration, especially regarding soil moisture. To calculate potential evapotranspiration, we updated the classic Priestley & Taylor (1972) model by using the correction proposed by Yang & Roderick (2019), which accounts for feedback between surface temperature and net radiation, and we were able to remove the alpha constant (assumed to be 1.26). We used a new formulation for the water supply and demand, which in SPLASH are used to calculate instantaneous evapotranspiration as the minimum between those two. In this latest version we consider water may exceed field capacity, i.e., if there is extra water, it goes to drainage. Whilst SPLASHv2.0 includes terrain, which is an excellent feature, especially for mountain regions, it is not doing so well predicting evapotranspiration, compared to the observations from flux towers at dry sites, overall, it produces reasonable simulations. 

Figure 2. Spatial and temporal patterns of evapotranspiration in a small wet temperate watershed, Rietholzbach – Switzerland. (a) Mean annual simulated evapotranspiration 1994-2007. (b) Time series of monthly evapotranspiration simulated and lysimeter-based observations. 

Snowpack 

We are using classic empirical relationships between the saturation of snow cover and albedo to calculate snowpack. The difference now is that we re-optimized those functions using MODIS retrievals of albedo and snow cover. We have over 15 years of data with a daily timescale from multiple sites. We also corrected for snow ageing and used binary observations to model the probability of snowfall. For the observed snowfall against simulated, it is doing well against more sophisticated models. Now that we are factoring in terrain and corrections for net radiation, we can simulate spatial patterns at a high spatial resolution. SPLASH v2.0 is capturing very well the beginning of spring and the snow melt and the magnitude of the melt, but the results are variable. But overall, it is doing well.  

Figure 3. Spatial patterns of mean monthly maximum SWE for the period 2010-2016 over North America at 1km resolution, along with site-simulation examples from the mountains. The black line is the observed mean seasonal cycle, the red line SPLASH v2.0 and the red line VIC-3L. The grey areas show one SD from the observed mean.  

Surface runoff and lateral flow 

The SPLASH v2.0 model uses upslope area and transmissivity (the integral of the hydraulic conductivity distribution and the soil moisture profile) to produce patterns for runoff and lateral flow. Using the ratio of the two gives you the baseflow index. The patterns we get with SPLASHv2.0 are what we would expect: higher runoff near streams and where there is more saturation and higher lateral flow in steeper slopes. For the relationship between soil moisture and lateral flow, there was a significant difference in the results between the VIC model and SPLASH v2.0. With VIC, there are high lateral flow values when the soil moisture level is low. The water should surpass the fuel capacity to drain, and the VIC model does not show that, however, the SPLASHv2.0 model does more accurately reproduce that pattern.  

Figure 4. Examples of patterns for runoff and lateral flow produced by SPLASHv2.0. (a) Spatial patterns of mean annual surface runoff during 1994-2007 in Rietholzbach – Switzerland. (b) Spatial patterns of mean annual lateral flow during 1994-2007 in Rietholzbach – Switzerland. (c) Emergent response of the lateral flow to soil moisture from an ENF in SPLASHv2.0 and VIC -3L(d) 

Soil moisture 

We compared the accumulated soil moisture profile (in millimetres) with the simulations from SPLASHv2.0, obtaining reasonable approximations for all the SNOTEL sites in the United States, and some few sites from FLUXNET which report the depth of the sensors. SPLASHv2.0 was able to reproduce the expected spatial patterns of soil moisture as well when it was tested over mountain watersheds. However, there is still a lot of uncertainties regarding the water holding capacity and the hydraulic conductivity, so some simulations still produce systematic errors, although the overall dynamics is still captured. 

Figure 5. Correlation of observed and simulated values of soil water content (a) FLUXNET sites pooled. (b) SNOTEL sites pooled. (c) Mean annual simulated soil water content in the whole column during 1994-2007 in Rietholzbach – Switzerland. 
 

Condensation 

SPLASHv2.0 computes condensation assuming that a fraction of the outgoing night-time net radiation is lost as latent heat. Nevertheless, this simple assumption returned better approximations than the more complex calculation from VIC-3L. SPLASHv2.0 accurately predicts the seasonality and the annual magnitude of this input, which is marginal for most of the biomes, but it has ecological importance in very dry ecosystems. 

Figure 6. Mean seasonal cycle of daily condensation per climate zone. The grey areas show one SD from the observed mean. 

Soil water storage and conductivity 

We use empirical functions called pedotransfer functions (PTF) to calculate water storage and hydraulic conductivity. We tested the most widely used PTFs in global applications and optimized the best-performing one. We also included the effects of viscosity on hydraulic conductivity. Currently, we are testing how SPLASHv2.0 is capturing the soil water potential using in-situ measurements. 

Figure 7. Measurements that were used to calibrate the pedotransfer functions. (a) Probability density of the soil samples according to their textural classes. (b) Average soil organic matter content of the samples per textural class. (c)  Theoretical effects of viscosity on the saturated hydraulic conductivity using a hypothetical soil with 10% SOM, 30% Silt, and varying Sand and Clay