Calculating reference evapotranspiration and net radiation - methods and data sources

This is quite a thorny topic that’s not helped by a frequent lack of clear definitions.

Firstly, I’ll start by talking about evapotranspiration. Specifically, a key quantity is the reference crop evapotranspiration (ET_0). The FAO recommends this as the reference standard rather than any other exactly because of the ambiguities in their definitions. In line with this, NatCap’s modules such as the annual water yield use ET_0 as a required input. Now, ET_0 is potential evapotranspiration from your model 12 cm tall grass. This is well defined and thus provides a firm starting point. This is subsequently modified for each land-use/land-cover (LULC) type, by the respective K_c for each LULC thus (leaving the spatial dependence of LULC implied):

PET = K_c.ET_0

Here, PET is the potential evapotranspiration of a specific LULC. The mechanics of inputting this is that the K_c values are entered into a biophysical parameters CSV table for each LULC. Thus, this PET is the equivalent of FAO’s ET_c.

At this juncture, I’ll raise a question on where one can obtain datasets of ET_0 as a data product. A question was posed on the forum here, which was redirected to a prior question here. The central question here is whether one can use the MODIS16 evapotranspiration data product. If you look at the available MODIS data layers here, you’ll see there’s an ET_500m, or “Total evapotranspiration” layer. I think this is definitely not what’s wanted, as this is, I believe, an “actual” value, not “potential”. Then we have the PET_500m, or “Total potential evapotranspiration”. The critical question here is what is the definition of this potential evapotranspiration product? My discomfort centres on lack of an obvious confirmation that this is for a singular LULC, namely the reference grass LULC. Looking at the user guide, it’s not obvious how the schematic in Figure 2 maps to output data products, but it does very much imply that they are conditioned on Land-Cover/LAI. In other words, I conclude that the PET data product from MODIS is akin to the PET as defined in the InVEST user guide, but where ET_0 is modified by remote-sensed LAI rather than empirical K_c. I’m led to question whether the the answer to the question above is correct. Surely, unless you set K_c to unity for all your LULC entries (and I don’t know if there might be unintended consequences of this), wouldn’t you be “double counting” the LULC modification to ET_0? Am I missing something here, @swolny ?

In short, I’m not aware of any data products that provide the required ET_0; you have to calculate it. I do note that there’s a TerraClimate data product for ET_0, albeit at a fairly course resolution of around 4 km. Does anyone use that, even for validating their ET_0?

So, now secondly, how can we calculate this reference evapotranspiration? Here, NatCap and InVEST seem to diverge from FAO somewhat. The FAO is very keen to recommend the FAO Penman-Monteith (PM) method as the sole method for determining ET_0. NatCap, on the other hand, seems to like the modified Hargreaves’ equation (MH). I can certainly understand the appeal of the simplicity of MH: extraterrestrial radiation (easily calculable for a given day of the year), temperature data, and precipitation, and you’re good to go. I imagine the FAO position is very much influenced by the fact that FAO-56 predates Droogers and Allen, 2002. The InVEST documentation states multiple times that one really should use the same precipitation data (a required input in itself to the water yield model) as was used to calculate ET_0. The assumption here seems very much to be that one is using the modified Hargreaves’ equation, with its dependence on precipitation. If one were using the Penman-Monteith, this requirement/recommendation would not seem to be valid as precipitation doesn’t feature in the PM method. Certainly, I also like the modified Hargreaves equation for one use case I have in mind: forecasting or scenario planning. It’s one thing to play about with predicted water yield as the climate warms or rainfall patterns change, it’s another to have to guess what humidity or windspeed values to also plug in.

Finally, in the event anyone (still) reading this uses the PM method, how do you tackle the data demands of calculating the net radiation? For both methods, you need extraterrestrial radiation (R_a), but that’s readily calculable for a given latitude, longitude, and day of year. Are there any data providers producing a net radiation (at the surface) data product or is it always a case of going through the grind of calculating all the intermediate radiation terms in FAO-56 method?

1 Like