Applications of Gaussian process regression for predicting blue water footprint: Case study in Ad Daqahliyah, Egypt
Graphical Abstract
Introduction
Environmental challenges have attracted more attention in most countries all over the world as a result of unsustainable and inappropriate exploitation of environmental resources (Engström et al., 2008, Lenschow et al., 2016, Fan et al., 2018). The accelerated resource consumption and the constant abuse of nature, associated with rapid climate change have negatively affected ecosystems recovery potentials (Chowdhary et al., 2018, Duckett et al., 2016). On the other hand, acceleration of global economic growth led to environmental problems and retrogression of ecosystem services and functions (Han et al., 2021). In this context, the United Nations (UN) launched the UN-2030 Agenda and its sustainable development goals (SDGs) to meet the eminent earth’s challenges (UN, 2015). The SDGs are an ambitious plan to achieve earths sustainability within various dimensions (i.e. 17 SDGs) with a special attention to land (SDG-15), climate change (SDG-13), and water (SDG-6) (UN, 2015; Di Baldassarre et al., 2019). In a changing climate, water availability and its efficient management (SDG-6) is one of the urgent challenges that need to be addressed, due to its direct or indirect role in achieving other SDGs for insuring sustainability (El-mageed et al., 2017; Di Baldassarre et al., 2019; Jerin et al., 2021; Elbeltagi et al., 2020e, 2021; Kumar et al., 2021; Mokhtar et al., 2021; Bajirao et al., 2021; Zerouali et al., 2021; Adnan et al., 2021; Suryakant et al., 2021). Hence, proper management of water resources by using new approaches is a vital issue for achieving SDGs goals especially in arid and semi-arid regions. In this sense, water footprint (WF) (Hoekstra, 2008) could be one of the tools for meeting water security challenges, especially in the agricultural sector (Elbeltagi et al., 2020a).
The term WF, firstly introduced, is a method for quantifying the amount water that is used for producing one unit of a target product through the whole supply chain. Scientifically, the WF is a combination between green-WF (rainfall (R) (m3); blue-WF (evapotranspiration (ET) – R, when ET ≥ R) (m3); and gray-WF (water pollution/ quality – neglected-) (m3)) (Lovarelli et al., 2016, Jamshidi et al., 2020). Since, 85% of the global freshwater is consumed by agricultural sector (Mekonnen and Hoekstra, 2011, Bhat et al., 2017a, Bhat et al., 2017b, Elbeltagi et al., 2020a, Elbeltagi et al., 2020b, Elbeltagi et al., 2020c, Elbeltagi et al., 2020d, Elbeltagi et al., 2020e); the WF was broadly employed in the agriculture sector to calculate water consumption, environmental impact, and investigate the better water management (Novoa et al., 2019, Nezamoleslami and Hosseinian, 2020). In this sense, water footprint assessment (WFA) was previously applied in a global scale for maize (Liu et al., 2017), wheat (Mekonnen and Hoekstra, 2010), crop production (Babaee et al., 2021, Mekonnen and Hoekstra, 2011), hydropower, animal products (Mekonnen and Hoekstra, 2012), and humanity (Hoekstra and Mekonnen, 2012). Notably, WFA research was carried out in many parts of the world. For instance, in Morocco; China (Zhai et al., 2019); Romania (Ene et al., 2013); Italy (De Girolamo et al., 2019); South Africa (Pahlow et al., 2015); Iran (Ababaei and Etedali, 2017, Karandish and Hoekstra, 2017); Turkey (Muratoglu, 2019), and Spain (Pellicer-Martínez and Martínez-Paz, 2016).
So far, many models were developed for WFA such as Crop Water Use (CWU) Model (De Miguel et al., 2015); WFA approach (Hoekstra et al., 2011), WF model (Nezamoleslami and Hosseinian, 2020); and WF- HYDRUS (Karandish and Šimůnek, 2018). Recently, WFA was integrated with machine learning (ML) technique, however, holistic approach between WFA and machine learning is still emerging. For example; Dai et al. (2021) proposed a fuzzy modeling approach, which enhanced WF under different levels of credibility. Mokarram et al. (2021) used artificial neural networks (ANNs) for calculating WF, and reported that radial basis function (RBF) has better estimation than other models. Interestingly, Elbeltagi et al. (2020a) developed an ANN model for calculating WF in Egyptian Delta for maize, where the result showed a high trustworthy outcome with low deviation, even if the climate data were limited. Their findings observed that the combination of solar radiation, vapor pressure deficit, and humidity (H) was the best variables for predicting blue WF. Moreover, Elbeltagi et al. (2020b) studied the impact of climate changes on the future water footprint of two agro-nomically important crops by applying deep neural networks (DNN) under three scenarios of climate changes. In the proposed work, we have presented new techniques in modeling WF easily without conducting field experiments, no cost, simple and achieve satisfactory result. Gaussian processes under four kernels as a rapid decision tools were developed to model blue WF in the study area as a new tool for water users and developers. Furthermore, to our knowledge, there are no studies available in the literature that model and simulate the WF in the semi-arid regions using the processes of Gaussian kernels.
Egypt, is located in the northern Africa, suffers from water shortage, which reaches 13.5–24 Bm3 yr−1, due to arid climate, rapid population growth (94 million), and limited water resources (90% from Nile river (55.5 Bm3 yr−1)) (Allam et al., 2015, Mohie El Din and Moussa, 2016, El-Essawy et al., 2019, Assar et al., 2020). However, water scarcity is the main challenges facing the agricultural sector and limiting crop production and national food security (Allam and Allam, 2007, El-Essawy et al., 2019). Nevertheless, 50% of the labor forces working in agricultural sector contributed to about 23% of national GDP (El-Essawy et al., 2019). Thus, defining water consumption from main produced crops (i.e. maize, wheat, and cotton) is critical for efficient water management and sustainability of agricultural system (Elbeltagi et al., 2020a, Elbeltagi et al., 2020b, Elbeltagi et al., 2020c, Elbeltagi et al., 2020d, Elbeltagi et al., 2020e).
Maize (Zea mays L. spp.) is one of the important crops in Egypt (Abdrabbo et al., 2013). According to Food and Agriculture Organization (FAO) the cultivated area by maize is nearly about 1.1 M ha, which is slightly higher than year 2019 by 2.72%. Regardless, the efficiency of maize and other C4 crops in water use comparing to C3 crops (Begg and Turner, 1976); maize cultivation in Egypt and especially in Nile Delta required a careful assessment of water balance and water use efficiency for sustainable production. Nonetheless, few studies were carried out in the Mediterranean region (Novoa et al., 2019), and in Egypt -in particular- for estimation blue WF of agricultural crops. Thus, to bridge the gap in literature, this work aims to: (1) compare between the developed GPR models under different kernels in Egyptian Nile Delta Governorate, and (2) select the best model and algorithm, which achieve a high degree of accuracy and low error for predicting blue WF of maize.
Section snippets
Study area
The study field is on Egypt's Nile Delta. It covers the northern portion (Lower Egypt) of the Mediterranean Sea. The Nile River is the world's longest river, with a total length of about 7000 kilometers. The Nile Delta accounted for approximately 2% of Egypt's total territory, with up to 63% of agricultural property. The Delta started about 20 km north of Cairo and up to 150 km north. The width of the delta is about 250 km i.e. Alexandria governorate in the west and Port Said governorate in the
Results and discussion
In this study, different meteorological data were combined in 10 scenarios as inputs into the GPR to model the blue WF of maize. Performance of four different kernels (including Poly kernel, RBF kernel, PUK kernel and Normalized poly kernel) in GPR model are presented from Table 4, Table 5, Table 6, Table 7, respectively, for training and testing periods. As Table 4 shows, by increasing the number of meteorological variables in GPR in the Poly kernel, the amount of error (in all 4 indexes of
Conclusion
In this study, we have developed four kernels in GPR models based on ten scenarios from meteorological variables including (Tmax, Tmin, Tmean, WS, SH, VPD, SR, Ra,H and Kc). In order to select the best kernel with scenario for predicting blue WF of maize in Egyptian Nile Delta Governorate during the period from 2000 to 2019. Overall, the performance of the PUK kernel in predicting blue WF is the best kernel followed by the Poly kernel, the Normalized poly kernel, and finally the RBF kernel. For
CRediT authorship contribution statement
Ahmed Elbeltagi: Conceptualization, Formal analysis, Writing - original draft, Writing - review & editing. Nasrin Azad: Writing - review & editing. Arfan Arshad: Writing - review & editing. Safwan Mohammed: Writing - review & editing. Ali Mokhtar: Writing - review & editing. Chaitanya Pande: Writing - review & editing. Hadi Ramezani Etedali: Writing - review & editing. Shakeel Ahmad Bhat: Writing - review & editing. Abu Reza Md. Towfiqul Islam: Writing - review & editing. Deng Jinsong:
Funding
This work was supported by the Ministry of Science and Technology of the People's Republic of China (2016YFC0503404) and the Natural Science Foundation of Zhejiang Province (LY18G030006).
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (79)
- et al.
Water footprint assessment of main cereals in Iran
Agric. Water Manag.
(2017) - et al.
Evaluation of the effect of climate change on maize water footprint under RCPs scenarios in Qazvin plain, Iran
Agric. Water Manag.
(2021) - et al.
A simulation-based suitability index of the quality and quantity of agricultural drainage water for reuse in irrigation
Sci. Total Environ.
(2015) - et al.
Crop water deficits
Adv. Agron.
(1976) - et al.
Evaluation of ten machine learning methods for estimating terrestrial evapotranspiration from remote sensing
Int. J. Appl. Earth Obs. Geoinf.
(2019) - et al.
Environmental pollution and health hazards from distillery wastewater and treatment approaches to combat the environmental threats: a review
Chemosphere
(2018) - et al.
A fuzzy fractional programming model for optimizing water footprint of crop planting and trading in the Hai River Basin, China
J. Clean. Prod.
(2021) - et al.
Improving grey water footprint assessment: accounting for uncertainty
Ecol. Indic.
(2019) - et al.
Tackling wicked environmental problems: the discourse and its influence on praxis in Scotland
Landsc. Urban Plan.
(2016) - et al.
Crop Water footprint estimation and modeling using an artificial neural network approach in the Nile Delta
Agric. Water Manag.
(2020)