Spatial determination of ETo supported by weather forecasts and artificial intelligence.

The amount of open data offered by different numerical weather prediction (NWP) systems is growing due to the increase in the capacity of computing systems. This rise has enabled the development of improved and user-tailored forecasting services and products. However, one key variable in agricultural systems not usually provided by the forecasting services is the reference crop evapotranspiration (ETo), which requires ad-hoc computation and proper identification of the factors that condition it.   This work develops a spatially-distributed ETo forecast in the Jucar river basin (Eastern Spain), to support crop management in agricultural plots. ETo was determined from forecasted meteorological variables using the Penman-Monteith methodology described in FAO56. Specific ETo value maps at the AP scale were generated considering the spatial variation of the meteorological parameters that drive ETo: daily average, maximum, minimum and dewpoint temperatures, net solar radiation and wind speed at 2 meters. Calculations were downscaled using an interpolation technique based on linear regression from daily weather predictions of temperatures and wind. The procedure was tested using forecasts from the Global Forecast System (GFS) of the National Centers for Environmental Prediction (NCEP) belonging to the U.S. National Oceanic and Atmospheric Administration (NOAA), for the year 2022. Raw GFS forecasts were post-processed against the ERA5 reanalysis data, available through the Copernicus Climate Change Service (CS3), with a spatial resolution of 0.25o; and against observed data from the meteorological stations of the Agroclimatic Information System for Irrigation (SIAR) of Spain. In both cases, post-processing was done using artificial intelligence (AI), in particular Fuzzy Logic. Inputs for interpolation were the geographical characteristics at each GFS location within the Jucar river basin: longitude, latitude, distance to the Mediterranean Sea, mean solar radiation, mean solar radiation at a distance of 2.5, 5 and 25km from each GFS location, elevation, elevation at a distance of 2.5, 5 and 10km from each GFS location, slope, and orientation with respect to the north. Solar radiation is obtained using the Area Solar Radiation module of ArcGIS.   Once the forecasts and solar radiation maps were generated, the difference between the interpolated and the predicted values was calculated. This difference generated a cloud of points which, which together with a Digital Elevation Model, allowed for surface interpolation (SI) using the Splines with the Tension methodology integrated in Grass (QGIS). These SI are subtracted from the forecast’s maps obtained by interpolation, already having corrected forecasts with which the ETo is determined using the Penman-Monteith methodology described in the FAO56. The difference between the interpolated ETo and the predicted ETo is also calculated by subtracting this SI from the obtained ETo, generating a corrected ETo. Furthermore, post-processed forecasts and ETo was compared with 41 meteorological stations and evaluated using the Mean Absolute Error (MAE).   Acknowledgements: This study has received funding from the European Union’s Horizon Europe research and innovation programme under the SOS-WATER project (GA no. 101059264); and from the subvencions del Programa per a la promoció de la investigación científica, el desenvolupament tecnològic i la innovació a la Comunitat Valenciana (PROMETEO) under the WATER4CAST project.