Water resource management has become an issue in recent times. It has been quantified that the effects of climate change are expected to decrease regional surface and ground water availability (Kioutsioukis et al. 2016), and lead to changes in seasonality and extreme floods and droughts (Sheffield et al. 2018). Thus, water has become a vulnerable resource and its sustainable use is a global concern. The situation is aggravated by the heavy use of water in the industry, energy production and agriculture (Kioutsioukis et al. 2016), the latest being the largest consumer of fresh water around the world. About 71 % of the total water withdrawals globally, 81 % in low-income countries (Amarasinghe and Smakhtin 2014) and almost 60 % of all the world’s freshwater withdrawals are used in irrigation (Hutson et al. 2004)
Accordingly, demographic explosion and the consequent food demands do not allow agricultural production without irrigation at all farming levels, water scarcity is becoming a threat to global food security (Mu, Zhao, and Running 2011)l, particularly in arid and semi-arid areas (Vyas and Subbaiah 2016) and subtropical regions (Kioutsioukis et al. 2016).
Thus, in order to achieve sustainable water resource management within the agricultural sector, improving irrigation, and increasing water productivity within a sustainable environmental management is a research priority (Kioutsioukis et al. 2016; Luo et al. 2015). Hence, the rational use of water for irrigation requires accurate planning of the amount and frequency of water, which could be achieved by analysing short-term forecasts of rainfall and crop water requirements (evapotranspiration; ET) (Luo et al. 2015; Srivastava et al. 2015; Zanetti et al. 2007).
Accurate estimation of reference evapotranspiration (ETo) is essential in agriculture, for the calculation of crop water requirements, irrigation scheduling, reduction of water and agrochemical use (Bughici et al. 2019), crop yield simulation, water resources planning and management, water balance studies, land resource planning (Adeloye, Rustum, and Kariyama 2011; Chang et al. 2010; Kumar, Raghuwanshi, and Singh 2011; Luo et al. 2015) and climate change (Nourani and Sayyah Fard 2012). ETo over-estimation leads to cause the waste of valuable water resources and its underestimation leads to the plant moisture stress and decrease in the crop yield (V
In this sense, to calculate ETo, the Penman-Monteith (PM) equation is the standard method recommended by the FAO, which uses complete meteorological data (Allen et al. 1998). However, it requires large number of climatic parameters such as temperature, humidity, wind speed, and solar radiation that are not always easily available for many locations (Luo et al. 2015; Vyas and Subbaiah 2016) and are site specific. Although there may exist many ground information to provide meteorological information for ETo calculation, these sites are sparsely distributed and their ET observations do not represent the regional ET condition.
In addition, data inaccessibility can be caused either by logistical, political, or security reasons, or simply, data is not available in real time (Sheffield et al. 2018). On the other hand, hydro-meteorological and agricultural monitoring networks are often sparse and impractical for real-time decision-making in many regions. Such is the case of Latin-America, including the Northern-Andes, a region highly dependent on agriculture and water resources, subject to diverse climatic and hydrologic regimes influenced by the cordillera. Under this scenario, a broad range of management challenges and in-situ data availability hamper the potential estimation of regional ETo estimation.
Satellite remote sensing could provide large area observations of the land surface frequently and continuously, and it is thus regarded as the most feasible approach to obtaining a regional estimation of ETo (Chen, Shi, and Zhang 2013). Therefore, for large scales of study satellite remote sensing tools offer meteorological information needed for ETo estimation (K. Wang and Dickinson 2012), or even spatially distributed regional ET information on land surface (i.e. MODIS; Mu, Zhao, and Running 2011). Wang and Dickinson (2012) provide in their review some algorithms that have produced global or regional data sets of ETo for more than 10 years using accumulation of satellite observations.
Among the satellite remotely sensed products, the Weather Research and Forecasting (WRF) is a next-generation mesoscale modelling system designed to place new and existing U.S. research and operational models under a common software architecture that has become a true community model by its long-term development through the interests and contributions of a worldwide user base (Powers et al. 2017). The initial and boundary conditions for the WRF model are built using near real-time Global Forecasting System (GFS) data of the National Centers for Environmental Protection (NCEP), together with the terrestrial data sets (Kioutsioukis et al. 2016).
The WRF derived data has been used worldwide for ETo calculation. For example, in England WRF was used for ETo estimation using machine learning algorithms (Srivastava et al. 2015). Also, Bughici et al. (2019) used precipitation forecasts and simulations from the WRF Model to calculate Eto using the PM equation over Israel. In the Iberian Peninsula, Rios-Entenza and Miguez-Macho (2014) analyzed the role of ET fluxes on precipitation regimes within the region interior using WRF- ARW. In California, Falk et al. (2014) and L. Xu et al. (2017) applied the Advanced Canopy-Atmosphere-Soil Algorithm (ACASA)-WRF model using the MODIS leaf area index product. Also, D. Wang et al. (2020) proposed a scheme to estimate ET using WRF-simulated surface skin temperature and then modifying the deviation using the normalized difference vegetation index (NDVI) in the Heihe River basin in China. In the same region, Zhang et al. (2020), completed simulations with the WRF model, and incorporated an irrigation scheme to measure the influence of irrigation on local potential ET. In addition, Xu et al. ( 2020) compared the performance of the WRF-CLM4.0 model, WRF–Noah model and Complementary Relationship with in-situ data methods to estimate daily ET, showing that the first one exhibited the best performance in the Agricultural-Pastoral Ecotone in Northwest China.
Finally, the effect of WRF climate data resolution was tested by Kioutsioukis et al. (2016), who found that that WRF can downscale global forecast data into finer resolutions in space and time for hydro-meteorological applications such as ET and irrigation water needs over Europe; and Strong et al. (2017) analyzed whether WRF downscaled climate data would provide physically meaningful additional information for ETo calculation compared to GridET framework that uses interpolation from coarser-scale simulations.
Although the advantages WRF derived data for simulations and forecast, the variables recorded by the WRF model, particularly wind estimates, may be highly influenced by topography, which constitutes a limitation for the increasing use of the model (Cheng and Steenburgh 2005; Jiménez and Dudhia 2012). In this sense, surface winds result from the interaction between mesoscale circulation and other more local factors, many of which are related to topographic characteristics (elevation, aspect, and slope) and terrain morphology (hills, valleys, and others) (Santos-Alamillos et al. 2013). These distortions may be accentuated by vegetation. If their effects are not considered, within WRF data, overestimation of the wind speed can arise (Jiménez and Dudhia 2012) and therefore unrealistic Eto estimations. This fact could potentially influence WRF derived data gathered for the Northern-Andes region, so the opportunity arises, as pointed out by Xing et al. (2016), for the selection of as few dominant meteorological variables as possible affecting ETo, and to develop an application of an established model for the computation of ETo in more regions.
It is important to indicate that global and regional studies require large amounts of data with multiple variables under irregular patterns, therefore, applicable procedures with good prediction and organization capabilities are required (Kumar, Raghuwanshi, and Singh 2011). Hence, artificial intelligence technology (e.g. Artificial Neural Networks; ANNs) has allowed the application of algorithms in hydro-meteorological fields, including evaporation and ETo estimation.
As ETo can reflect a nonlinear physical-based complex process, and ANNs are nonlinear intelligent platforms that adapt the neural weight to solve the nonlinear problem due to their generalized and probabilistic estimation properties (Luo et al. 2015). Thus, the performance of different types of ANNs has been widely applied for ETo estimation, including Multilayer Perceptron Networks (Chen, Shi, and Zhang 2013; Luo et al. 2015; Nourani and Sayyah Fard 2012; Vyas and Subbaiah 2016; Zanetti et al. 2007) . Linear Regression, Probabilistic Neural Network, Generalized Feedforward (Luo et al. 2015), Radial Basis Neural Network, and Elman network (Nourani and Sayyah Fard 2012).
Kumar, Raghuwanshi, and Singh (2011) assert that locally trained ANN models in ETo quantification present a limitation, since require validation at several locations under varying sets of conditions. Only then, the evaluation of the generalization potential of ANNs in ETo modelling could be performed. Accordingly, the authors suggest that this could be achieved either by developing an ANN model to map ETo process both in space and time or by incorporating data from several locations to train the ANN. Thus, the Self-Organizing Map (SOM) ANN, first introduced by Kohonen (1984) appears as an important option that displays the modelled process and presents its results as a topology map (Chang et al. 2010). The algorithm is capable of clustering, classification, prediction, and data mining in large datasets (Kohonen and Somervuo 2002), and it is based on unsupervised learning, which means that the training is entirely data-driven (ALHONIEMI et al. 1999). It allows visualization of entire datasets in terms of the original variables while detecting nonlinear relationship between them (Belkhiri et al. 2018).
Several studies have used the SOM to address water resource problems. For instance, Hsu et al. (2002) presented the self-organizing linear output map (SOLO) for hydrologic modeling and analysis. Richardson et al. (2003) claimed the SOM-algorithm succeeds in providing insight in the groundwater quality data set, highlighting the main differences between groups of samples and pointing out anomalous wells and well screens. Tadeusz et al. (2006) applied the SOM maps to revealing variation in non-obligatory riverine fish. Grieu et al. (2006) developed a data exploration technique for wastewater treatment monitoring based on SOM maps. Chang et al. (2007) presented an SOM network for flood forecasting and stated that it has great efficiency for clustering, especially for peak flow, and super capability of modeling flood forecasts. However, to the author´s knowledge, limited is the existent literature that applies SOM to Eto quantification and forecasting, and includes a research study in South Taiwan (Chang et al. 2010); and an investigation in two small experimental catchments the United Kingdom and in India (Adeloye, Rustum, and Kariyama 2011).
Under this context, the main objective of this study is to develop forecasted ETo maps based on WRF revised data clustered by SOM ANN for the Northern-Andes region. For this, we will compare the forecasting models with the results of the PM method calculated with data gathered from in-situ meteorological stations. The model performance will be tested considering the changes on wind parameters that may be caused by the influence of the Andes cordillera. Hence, the maps would become an interactive tool for farmers, researchers or anyone interested in water management with agricultural, forestry, conservation and land management applications under an accessible open-source environment.
Adeloye, A J, R Rustum, and I D Kariyama. 2011. “Kohonen Self-Organizing Map Estimator for the Reference Crop Evapotranspiration.” Water Resources Research 47(8). https://doi.org/10.1029/2011WR010690.
ALHONIEMI, E S A, JAAKKO HOLLMÉN, OLLI SIMULA, and JUHA VESANTO. 1999. “Process Monitoring and Modeling Using the Self-Organizing Map.” Integrated Computer-Aided Engineering 6(1): 3. http://10.0.12.161/ICA-1999-6102.
Allen, R G, L S Pereira, D Raes, and M Smith. 1998. “Crop Evapotranspiration-Guidelines for Computing Crop Water Requirements-FAO Irrigation and Drainage Paper 56.” Fao, Rome 300(9): D05109.
Belkhiri, Lazhar et al. 2018. “Spatial Analysis of Groundwater Quality Using Self-Organizing Maps.” Groundwater for Sustainable Development 7: 121–32. http://www.sciencedirect.com/science/article/pii/S2352801X1730070X.
Bughici, Theodor, Naftali Lazarovitch, Erick Fredj, and Eran Tas. 2019. “Evaluation and Bias Correction in WRF Model Forecasting of Precipitation and Potential Evapotranspiration.” Journal of Hydrometeorology 20(5): 965–83. https://doi.org/10.1175/JHM-D-18-0160.1.
Chang, Fi-John, Li-Chiu Chang, Huey-Shan Kao, and Gwo-Ru Wu. 2010. “Assessing the Effort of Meteorological Variables for Evaporation Estimation by Self-Organizing Map Neural Network.” Journal of Hydrology 384(1): 118–29. http://www.sciencedirect.com/science/article/pii/S0022169410000399.
Chen, Zhuoqi, Runhe Shi, and Shupeng Zhang. 2013. “An Artificial Neural Network Approach to Estimate Evapotranspiration from Remote Sensing and AmeriFlux Data.” Frontiers of Earth Science 7(1): 103–11. https://doi.org/10.1007/s11707-012-0346-7.
Cheng, William Y Y, and W James Steenburgh. 2005. “Evaluation of Surface Sensible Weather Forecasts by the WRF and the Eta Models over the Western United States.” Weather and Forecasting 20(5): 812–21. https://search.proquest.com/docview/196334117?accountid=14681.
Demartines, Pierre, and Blayo, Francois, 1992. "Kohonen Self-Organizing Maps: Is the Normalization Necessary?" Complex Systems 6 (1992) 105-1 23.
Falk, Matthias et al. 2014. “Evaluated Crop Evapotranspiration over a Region of Irrigated Orchards with the Improved ACASA–WRF Model.” Journal of Hydrometeorology 15(2): 744–58. https://doi.org/10.1175/JHM-D-12-0183.1.
Jiménez, Pedro A, and Jimy Dudhia. 2012. “Improving the Representation of Resolved and Unresolved Topographic Effects on Surface Wind in the WRF Model.” Journal of Applied Meteorology & Climatology 51(2): 300–316. http://10.0.4.151/JAMC-D-11-084.1.
Kioutsioukis, Ioannis et al. 2016. “High Resolution WRF Ensemble Forecasting for Irrigation: Multi-Variable Evaluation.” Atmospheric Research 167: 156–74. http://www.sciencedirect.com/science/article/pii/S0169809515002276.
Kohonen, Teuvo. 1984. “Phonotopics Maps Insightful Representation of Phonological Features of Speech Recognition.” In Proceedings of Seventh International Conference on Pattern Recognition, Montreal, 1984, , 182–85.
Kohonen, Teuvo, and Panu Somervuo. 2002. “How to Make Large Self-Organizing Maps for Nonvectorial Data.” Neural Networks 15(8): 945–52. http://www.sciencedirect.com/science/article/pii/S0893608002000692.
Kumar, M, N S Raghuwanshi, and R Singh. 2011. “Artificial Neural Networks Approach in Evapotranspiration Modeling: A Review.” Irrigation Science 29(1): 11–25. https://doi.org/10.1007/s00271-010-0230-8.
Luo, Yufeng et al. 2015. “Medium Range Daily Reference Evapotranspiration Forecasting by Using ANN and Public Weather Forecasts.” Water Resources Management 29(10): 3863–76. https://search.proquest.com/docview/1693688744?accountid=14681.
Mu, Qiaozhen, Maosheng Zhao, and Steven W Running. 2011. “Improvements to a MODIS Global Terrestrial Evapotranspiration Algorithm.” Remote Sensing of Environment 115(8): 1781–1800. http://www.sciencedirect.com/science/article/pii/S0034425711000691.
Nourani, Vahid, and Mina Sayyah Fard. 2012. “Sensitivity Analysis of the Artificial Neural Network Outputs in Simulation of the Evaporation Process at Different Climatologic Regimes.” Advances in Engineering Software 47(1): 127–46. http://www.sciencedirect.com/science/article/pii/S096599781200004X.
Powers, Jordan G et al. 2017. “THE WEATHER RESEARCH AND FORECASTING MODEL.” Bulletin of the American Meteorological Society 98(8): 1717–38. https://www-jstor-org.ezproxy.library.uwa.edu.au/stable/26396194.
Rios-Entenza, A., and G. Miguez-Macho. 2014. “Moisture Recycling and the Maximum of Precipitation in Spring in the Iberian Peninsula.” Climate Dynamics 42(11/12): 3207–31. http://10.0.3.239/s00382-013-1971-x.
Santos-Alamillos, F J et al. 2013. “Analysis of WRF Model Wind Estimate Sensitivity to Physics Parameterization Choice and Terrain Representation in Andalusia (Southern Spain).” Journal of Applied Meteorology and Climatology 52(7): 1592–1609. http://www.jstor.org.ezproxy.library.uwa.edu.au/stable/26175852.
Sheffield, J et al. 2018. “Satellite Remote Sensing for Water Resources Management: Potential for Supporting Sustainable Development in Data-Poor Regions.” Water Resources Research 54(12): 9724–58. https://doi.org/10.1029/2017WR022437.
Srivastava, Prashant K et al. 2015. “WRF Dynamical Downscaling and Bias Correction Schemes for NCEP Estimated Hydro-Meteorological Variables.” Water Resources Management 29(7): 2267–84. https://search.proquest.com/docview/1672348646?accountid=14681.
Strong, C. et al. 2017. “Reference Evapotranspiration from Coarse-Scale and Dynamically Downscaled Data in Complex Terrain: Sensitivity to Interpolation and Resolution.” Journal of Hydrology 548: 406–18. http://www.sciencedirect.com/science/article/pii/S0022169417301245.
Vyas, Khyati N, and R Subbaiah. 2016. “Application of Artificial Neural Network Approach for Estimating Reference Evapotranspiration.” Current World Environment 11(2): 637–47. https://search.proquest.com/docview/1868286749?accountid=14681.
Wang, Dakang et al. 2020. “A Method of Using WRF-Simulated Surface Temperature to Estimate Daily Evapotranspiration.” Journal of Applied Meteorology and Climatology 59(5): 901–14. https://doi.org/10.1175/JAMC-D-19-0287.1.
Wang, Kaicun, and Robert E Dickinson. 2012. “A Review of Global Terrestrial Evapotranspiration: Observation, Modeling, Climatology, and Climatic Variability.” Reviews of Geophysics 50(2). https://search.proquest.com/docview/1018271717?accountid=14681.
Xu, L. et al. 2017. “Impact of Canopy Representations on Regional Modeling of Evapotranspiration Using the WRF-ACASA Coupled Model.” Agricultural and Forest Meteorology 247: 79–92. http://www.sciencedirect.com/science/article/pii/S0168192317302198.
Xu, Xuefeng et al. 2020. “Estimating Daily Evapotranspiration in the Agricultural-Pastoral Ecotone in Northwest China: A Comparative Analysis of the Complementary Relationship, WRF-CLM4.0, and WRF-Noah Methods.” Science of The Total Environment 729: 138635. http://www.sciencedirect.com/science/article/pii/S0048969720321513.
Zanetti, S S et al. 2007. “Estimating Evapotranspiration Using Artificial Neural Network and Minimum Climatological Data.” Journal of Irrigation & Drainage Engineering 133(2): 83–89. http://10.0.4.37/(ASCE)0733-9437(2007)133:2(83).
Zhang, Xuezhen, Nana Ding, Songjun Han, and Qiuhong Tang. 2020. “Irrigation-Induced Potential Evapotranspiration Decrease in the Heihe River Basin, Northwest China, as Simulated by the WRF Model.” Journal of Geophysical Research: Atmospheres 125(2): e2019JD031058. https://doi.org/10.1029/2019JD031058.