Bio-economic evaluation of greenhouse designs for seasonal tomato production in Norway

* Corresponding author. E-mail address: Na.seer@hotmail.com (M https://doi.org/10.1016/j.biosystemseng.2021 1537-5110/© 2021 The Author(s). Published (http://creativecommons.org/licenses/by/4.0 Please cite this article as: Naseer, M et al., B Biosystems Engineering, https://doi.org/10.101 Greenhouses are complex systems whose size, shape, construction material, and equipment for climate control, lighting and heating can vary largely. The greenhouse design can, together with the outdoor weather conditions, have a large impact on the economic performance and the environmental consequences of the production. The aim of this study was to identify a greenhouse design out of several feasible designs that generated the highest net financial return (NFR) and lowest energy use for seasonal tomato production across Norway. A model-based greenhouse design method, which includes a module for greenhouse indoor climate, a crop growth module for yield prediction, and an economic module, was applied to predict the NFR and energy use. Observed indoor climate and tomato yield were predicted using the climate and growth modules in a commercial greenhouse in southwestern Norway (SW) with rail and grow heating pipes, glass cover, energy screens, and CO2-enrichment. Subsequently, the NFR and fossil fuel use of five combinations of these elements relevant to Norwegian conditions were determined for four locations: Kise in eastern Norway (E), Mære in midwestern Norway (MW), Orre in southwestern Norway (SW) and Tromsø in northern Norway (N). Across designs and locations, the highest NFR was 47.6 NOK m 2 for the greenhouse design with a night energy screen. The greenhouse design with day and night energy screens, fogging and mechanical cooling and heating having the lowest fossil energy used per m in all locations had an NFR of 94.8 NOK m . The model can be adapted for different climatic conditions using a variation in the design elements. The study is useful at the practical and policy level since it combines the economic module with the environmental impact to measure CO2 emissions. © 2021 The Author(s). Published by Elsevier Ltd on behalf of IAgrE. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Introduction
The agriculture sector is one of the most energy intensive industries in the world (Diakosavvas, 2017) and can also result in environmental impacts including soil degradation, groundwater depletion and rise in greenhouse gas emissions etc. (Lamb et al., 2016;Longo, Mistretta, Guarino, & Cellura, 2017;Notarnicola et al., 2015;Tamburini, Pedrini, Marchetti, Fano, & Castaldelli, 2015). Expanding food production to high latitude regions, where cold climate, short growing seasons and light conditions limit production, could be one way of alleviating the pressures on global food production. One way to reach such an expansion in food production is to use protected cultivation techniques, which mitigate the effects of unfavourable weather conditions. Such systems can include protection against wind, rain and sun as well as heating, cooling, humidity control, CO 2 -enrichment, lighting and irrigation, and can help to increase the yield, optimise the resource use, improve food production and extend the growing season (Tap, 2000). Greenhouses are one of the main methods of protected cultivation that shield crops against unfavourable outdoor conditions. They are complex systems whose size, shape, construction material, and equipment for climate control, lighting and heating can vary greatly. The greenhouse design can, together with the outdoor weather conditions, have a large impact on the economic performance and the environmental consequences of the production process (Hemming, Sapounas, de Zwart, Ruijs, & Maaswinkel, 2010;. From seed to fruit, there are multiple drivers (temperature, light intensity, light spectrum and day length, humidity, CO 2 -concentration and fertigation) that can be modified under controlled environmental conditions to increase the biomass production (Incrocci, Stanghellini, & Kempkes, 2008;Moe, Grimstad, & Gislerod, 2005).
Several studies have used modelling techniques to simulate and optimise different subsystems within the greenhouse system to improve the performance of various aspects of production (Joudi & Farhan, 2015;Pakari & Ghani, 2019;Verheul, Grimstad, & Maessen, 2012;Ahamed, Guo, Taylor, & Tanino, 2019;Singh & Tiwari, 2010;Von Elsner et al., 2000). These studies included evaluations of the effect of the shape of greenhouse on energy consumption and thereby optimum productivity (Ç akır & S ‚ ahin, 2015), and of the effects of greenhouse designs on productivity (Vanthoor et al., 2012a). Kondili and Kaldellis (2006) presented an analytical model to estimate optimal dimensions of a geothermal fluid transportation network, resulting in the minimisation of heat loss and energy consumption within a greenhouse in Greece. Flores-Vel azquez et al. (2009) andFlores-Velazquez, Montero, Baeza, andLopez (2014) studied the effects of greenhouse spans, and ventilation system on the temperature exchange and distribution using computational fluid dynamics. Likewise, Roy, Fatnassi, Boulard, Pouillard, and Grisey (2015) simulated the distribution of temperature and air humidity in a semi-closed greenhouse, measuring around 960 m 2 , for tomato production and furnished with several air cooling and dehumidifying ducts. Flores-Vel azquez and Vega-García (2019) showed that, in regions with mild summers, the combined use of mechanical and natural ventilation can lower the costs related to temperature regulation and energy use. Dynamic modelling techniques have also been used to simulate the greenhouse indoor climate for different climate conditions, crops and variables (De Zwart, 1996;Impron, Hemming, & Bot, 2007;Luo et al., 2005aLuo et al., , 2005b, including predictions of indoor air temperature in the greenhouse by studying six greenhouse types with different orientations related to energy consumption in the Iranian region of Tabriz (Mobtaker, Ajabshirchi, Ranjbar, & Matloobi, 2016). De Visser (2011a, 2011b) developed and applied a model to simulate tomato production, and its design elements can be adjusted to represent those suitable to different climate conditions. The model has been used in conjugation with an economic module (Vanthoor et al., 2012a) to evaluate the effect of greenhouse construction types on the economic performance of the production as determined by its annual net financial return (NFR). Hence, in this combined greenhouse design and economic module, the NFR is a function of yield, variable costs, construction costs, depreciation and costs for maintenance of equipment that is used in greenhouse production. Previously this model has been applied to identify suitable greenhouse construction types under a range of warm climates and lower latitude countries such as Spain, Netherlands etc. (Vanthoor et al., 2012a). However, previous studies of greenhouses and greenhouse subsystems have mostly excluded high latitude regions. The few studies that did include high latitude or otherwise cold regions did not consider renewable energy (Ahamed, Guo, & Tanino, 2018;Ahamed et al., 2019;Torrellas et al., 2012). The climate and light conditions in these regions differ considerably from those in lower latitude regions. Moreover, overall there is a considerable production of renewable energy in these regions, especially in comparison with other regions with significant greenhouse production (IRENA, 2021). Hence, in total, findings about greenhouse performance, energy use and related environmental impact from previous simulation and optimisation studies cannot be directly extrapolated to these regions.
Norway is suitable as a case for evaluating greenhouse economic and energy performance under high latitude regions. Its greenhouse vegetable production is small compared to the vegetable consumption but nevertheless growing (Rebnes & Angelsen, 2019). The production of tomatoes in Norway, its economically most important greenhouse vegetable, increased by, on average, 3.5% per year from 2009 to 2018. This increase is also in line with great preference for locally produced tomatoes in Norwegian markets over imported ones (Bremnes, Hansen, Slimestad, & Verheul, 2019). The growing season and the area for agricultural production in the field are short with an average temperature of 5e6 C and low outdoor light conditions. Most of the greenhouse production takes place during the summer season which is from May to October and a little with some artificial lighting in the months from February to November. Heating in greenhouses is primarily obtained from boilers by burning gas and is supplied through pipes. There is potential to further decrease the CO 2 emissions from the greenhouse sector (Verheul & Thorsen, 2010), which is needed to meet national goals to reduce carbon emissions as outlined by 'Klimakur 2030' (lit. climate cure 2030) (Miljødirektoratet, 2019) towards which attempts are being made by both the agriculture sector and the Norwegian government (Fremstad, 2020). Norway has the highest share of electricity produced from renewable sources, mainly hydropower, in Europe along with the lowest carbon emissions from the power sector (Ministry of Petroleum and Energy, 2020) and the large hydroelectric energy production in Norway provides the possibility to replace fossil energy in the greenhouse sector with renewable energy.
Energy costs, of which heating is a major component and lighting, account for about 44% of total production costs in Norwegian greenhouse vegetable production (Verheul, Maessen, & Grimstad, 2012). This is a high percentage in comparison with production in other countries (Raviv, Lieth, & Bar-Tal, 2019). However, the efficiency of the use of gas, electricity and other inputs and thus their costs may vary between greenhouses with different designs for insulation and shading equipment, heating and cooling system, artificial lighting and system for CO 2 supply (Hatirli, Ozkan, & Fert, 2006). Labour costs, depreciation of the structure and equipment, and costs for plant material, substrate, fertilisers and plant protection agents also have great impact on the total production costs (Moe et al., 2005;Vanthoor et al., 2012a). Production designs, which reduce the use of energy, water or CO 2 emissions per unit of product, could increase the profitability for the grower and the tomato production sector as a whole (Verheul et al., 2012), and hence encourage growers to use environmentally friendly methods. There is a growing understanding that an agreement between the government and the growers is fundamental in order for policy decisions regarding environmentally sustainable production methods to be practised by growers, something that is only possible if they are also economically profitable (www.climplement.no;Pretty, Ball, Xiaoyun, & Ravindranath, 2013;Fremstad, 2020).
Suitable greenhouse designs may also vary considerably between regions in Norway with different climate conditions. Moreover, the effect of the greenhouse design on the profitability may not always be correlated with the environmental impact. The objective of this study was to identify the greenhouse design, out of a number of feasible designs, that generated the highest NFR and the lowest fossil fuel use for seasonal tomato production from mid-March to mid-October in Norway. Therefore, we adjusted and evaluated the greenhouse production model of Vanthoor (2011) against observed climate conditions and seasonal tomato yield in a commercial greenhouse in Norway. Subsequently, tomato production for a set of combinations of outdoor climate and light conditions and greenhouse designs was simulated, and the economic performance and fossil use associated with these combinations were evaluated.

2.
Materials and methods

Model overview
The present study uses the approach presented by Vanthoor (2011) in order to design a greenhouse which maximises the profit, as quantified by the NFR, and minimises energy use for tomato growers in Norway. The design technique consists of a greenhouse climate module, crop yield module and an economic module that are connected to each other as shown in Fig. 1. The model simulates greenhouse climate conditions, crop growth and yield with an hourly time step and provides the yearly NFR as an output. The greenhouse climate module describes the effect of the outdoor climate, internal set points for temperature, CO 2concentration , humidity as well as greenhouse design elements on the indoor climate of the greenhouse and its resource consumption. The crop yield module simulates the tomato growth and yield as a function of the indoor climate. The economic module calculates the NFR of the production, which is affected by the resource use and the crop yield. The climate model, extensively described by Vanthoor et al. (2011a), is based on the energy and mass balance of each greenhouse element. Righini et al. (2020) later added heat storage through a heat pump to the model, and the work includes a summary of all the equations, along with an updated scheme of the model. The structure of the yield model, with a common carbohydrate buffer and carbohydrate distribution to plant organs, based on the photosynthesis model of Farquhar, Von Caemmerer, and Berry (2001) is the one generally applied. Vanthoor, Stanghellini, Van Henten, and De Visser (2011b) added two lumped temperature-dependent functions inhibiting re-distribution of carbohydrates and thus growth. Both sub-and supra-optimal temperature inhibit growth, short term deviations having less impact than deviations in daily means. A temperature sum representing the development stage of the crop was modelled to define the timing of first fruit set and the time at which the carbohydrate distribution to the fruits reaches its potential. The temperature functions, which Vanthoor et al. (2011b) derived from an extensive literature survey, have not been changed. A short presentation of the components of the economic module is given in the following section.

Economic tomato yield module
The yearly net financial return P NFR (NOK m À2 year À1 ) is calculated according to: where t 0 and t f are the start and the end time of the growing seasons, C fixed ðNOK m À2 Year À1 Þ are the fixed costs for the tangible assets (greenhouse structure, climate computer, cooling system, heating system and structure), C Var ðNOK m À2 Year À1 Þ are the variable costs, and Q CropYield ðNOK m À2 Year À1 Þ is the economic value of the crop yield. Figure 2 presents details of the costs and sub-costs that are included in the economic module.
2.1.1.1. Fixed costs. The yearly fixed costs are calculated based on the interests and the total investments of the construction elements, C fixed ðNOK m À2 Year À1 Þ; which include maintenance and depreciations and are defined as: where C interest ðNOK m À2 Year À1 Þ are the interest costs of the total investments. Here, i denotes the construction elements and N is the total number of greenhouse design elements used in selected greenhouses construction. C construction ðNOK m À2 Year À1 Þ are the costs for depreciation and maintenance and C Rem ðNOK m À2 Year À1 Þ are the remaining costs of construction and equipment. For equations for construction elements, interests and remaining costs see Vanthoor et al. (2012a).
2.1.1.2. Variable costs. The variable costs are the sum of the costs for plant, water used, CO 2 , and the two types of energy used: fossil fuel and the electricity. The total variable _ C var are defined as:

Locations, greenhouse design and evaluated cases
The present study applied the model described above to identify the greenhouse design that generated the highest NFR and the lowest energy used out of several plausible greenhouse designs for tomato production at four locations ( Fig. 3) in Norway. Five combinations of alternative choices of seven greenhouse design elements, as described in the subsequent sections were evaluated.

Locations
First, to evaluate the applicability of greenhouse tomato production model to conditions that represented Norway, we tested its prediction accuracy for indoor temperature, CO 2 concentration and tomato fresh mass that was observed in a greenhouse in southwestern (SW) Norway (Orre (lat. 58.71, long. 5.56, alt. 18 m a.s.l.)) during one seasonal production cycle for one of the selected greenhouse designs (Night screen (NS) as defined in section 2.2.3). Subsequently, the greenhouse designs of the selected combinations as well as its underlying economic components were identified for tomato production from 10th (Fig. 3). These locations were included because they represent different latitudes and have varying coastal and inland climate conditions in Norway (Fig. 4), and either represent major tomato-producing regions or could, in our opinion, have the potential for greenhouse tomato production due to local demand for tomatoes.

Greenhouse design
All the greenhouse designs that were evaluated were Venlotype greenhouses (Fernandez & Bailey, 1992) as usually used in Norway, covered with standard glass and with natural ventilation (alternate roof vents on both sides that corresponded to about 15% of floor area (Fig. 5)). There was no ventilation in the side wall of the greenhouses. The greenhouses had a rectangular shape of 90 Â 64 m, i.e., a floor area of 5760 m 2 . The light transmission of the greenhouse cover including structural material (aluminium/steel) was set to 64%. No artificial lighting was used. Two types of heating systems were evaluated, with one that used fossil fuel energy and the other green energy. More specifically, a boiler heating system, using natural gas, and a heat pump, using electricity generated in a hydropower plant, were applied. The evaluation included the use of night and day energy screens. Both the boiler and heat pump were used for primary and secondary pipe heating. CO 2 was supplied to the greenhouse either by burning of natural gas in the boiler or as pure CO 2 from a tank. The heat distribution system consisted of both rail pipes and grow pipes made of steel, which were filled with hot water. The capacity of the CO 2 enrichment system was 130 kg CO 2 ha À1 h À1 . Temperature, humidity and CO 2 supply were controlled by settings for global radiation, indoor temperature and window opening (Table 4). Plants were grown in standard Rockwool slabs and irrigated by a drip irrigation system.
The tomato price trajectory ( Fig. 6) from 2016, obtained from Grøntprodusentenes Samarbeidsråd (the Green Growers' Cooperative Marked Council) (https://www.grontprodusentene.no), was applied for all greenhouse designs and locations. Likewise, the fixed and variable costs per input unit that were associated with the Norwegian construction and production conditions presented in Tables 1 and 2 were set the same for all greenhouse designs and locations. These costs were either obtained from literature or from interviews with tomato growers across Norway by advisors at The Norwegian Institute of Bio-economy Research (NIBIO).  b i o s y s t e m s e n g i n e e r i n g x x x ( x x x x ) x x x 2.2.2.1. Greenhouse climate control. For all four locations and greenhouse designs, the same greenhouse climate set points were used, as presented in Table 4. However, the period for which day and night energy screens were applied was adjusted according to the local light and temperature conditions and was thus allowed to vary between locations. The strategy for controlling the air temperature is presented in Fig. 7. 2.2.2.2. Indoor climate and tomato fresh mass predictability.
The model for tomato production was validated for an existing greenhouse in Orre in Norway for seasonal production for the year 2016 without artificial lighting in Orre. The validation was conducted with the following production conditions. Hourly outdoor weather data including average temperature, wind speed, relative humidity and global radiation that were input to the climate module also represented the year 2016 and were obtained from the Saerheim station of  Table 1 e Fixed costs used in the greenhouses. The costs associated with the greenhouse design elements and element alternatives e j represent the number for each design element option. The depreciation percentage has been derived from the consultations with the local growers. E* ¼ around 10% extra for transportation expenses and exchange rate from the Netherlands to Norway.

Design element/Fixed costs
Vermeulen ( b i o s y s t e m s e n g i n e e r i n g x x x ( x x x x ) x x x the Agroclimate Station Network (https://lmt.nibio.no/) of NIBIO. The weather station from which weather data was obtained for simulation at Orre was located 8 km northeast of the greenhouse. Weather input data for 2016 was chosen because the mean monthly outdoor air temperature and global radiation in that year adequately represented monthly mean values of these weather elements over the past 30 years at the four locations (Fig. 4). Global radiation was measured with a Kipp solarimeter, placed outside of the greenhouse. Light transmission of total photosynthetic active radiation (PAR, mol m À2 d À1 ) was estimated based on measurements in the empty greenhouse and the outdoor global radiation. CO 2 of greenhouse air was measured at 5 minute intervals with a gas analyser (Priva CO 2 monitor Guardian þ). Air temperature and relative humidity were measured by dry-and wet-bulb thermocouples placed in ventilated boxes that shielded against direct solar radiation and placed in the middle of the canopy. Thermocouples were calibrated before the start and controlled at the end of the experiment. Temperature ( C), relative humidity (%), CO 2 concentration (ppm) and window opening (%) were registered every 5 min using a Priva computer (Priva Connext). Tomato seeds were sown at the end of January 2016 in a separate greenhouse. Young plants were transplanted in the greenhouse on standard Rockwool slabs with a density of 2.60 plants m À2 and a row separation of 1.5 m on 10th March and grown until 15th October. The night, day and ventilation temperature set points were 17, 19, 23 C respectively. Light transmission of total photosynthetic active radiation (PAR, mol m À2 d À1 ) was estimated based on measurements in the empty greenhouse and the outdoor global radiation. Leaf area was estimated once a week by measuring leaf length and leaf number on 10 representative plants.
CO 2 was applied up to the maximum concentration of 1000 ppm when the temperature and global radiation matched the criteria in Table 4 for CO 2Air_ExtMax and the windows were closed, and decreased with decreasing global radiation, decreasing indoor temperature and increasing ventilation rate according to Mag an, L opez, P erez-Parra, and L opez (2008) to a minimum value of Table 2 e Variable costs that were used in the simulations. * ¼ The data was obtained from interviews with commercial tomato growers whose production is representative for Norway.   Table 4. Adapted from Vanthoor et al. (2011b).

Resource
390 ppm with 100% window opening. Greenhouse temperature, CO 2 concentration and humidity were measured every five minutes but, in the simulations, the hourly average values were used. For pollination, bumblebees were used in the greenhouse during the whole cultivation period. Fruits were harvested, twice a week, at light red ripening stage and only 1st class fruits (marketable fraction) were taken into account here. The model prediction accuracy of the indoor air temperature, CO 2 -concentration and fresh mass tomato yield was evaluated using the Relative Root Mean Squared Error (RRMSE), Mean Bias Error (MBE) and Mean Absolute Error (MAE) as defined below: where y data is the mean of measured data over the total time span, n is the number of measurements, y Mod,i is the simulated output at time instant i and y Data,i is the corresponding measured value at time instant i.

Evaluated cases
An overview of the greenhouse designs evaluated for the four locations in Norway is presented in Table 3 and details are explained below. Standard greenhouse (without additions) (0S): A gas boiler with 1.16 MW capacity was used for heating. There were no indoor day energy screens or night energy screen included in this greenhouse design. Moreover, there was no artificial cooling or fogging system used.
Night energy screen (NS): This greenhouse design is like the existing greenhouse in Orre that was used to validate the climate and yield module. It had the same design elements as 0S except for the addition of a night energy screen consisting of 50% aluminium and 50% polyethylene, which was used for energy saving purposes whenever the temperature was below 14 C at night (See Table 4 for an explanation about how day and night settings were initiated.).
Day and night energy screens (DNS): This greenhouse design was the same as the design NS except for the use of a day energy screen consisting of 100% polyethylene (PE) during the day when outside global radiation was less than 150 Wm -2 and temperature was below 10 C to save energy while also allowing more light to pass through during the day time as compared to the night energy screen.
Day and night energy screens with fogging for cooling (DNSF): The design DNSF was the same as the DNS except that a fogging system for cooling and humidification purposes was activated when the air temperature exceeded 24 C and the relative humidity was below 84%.
Day and night energy screens with fogging and mechanical cooling and heating (DNSFM): This design represents a production system in which the fossil fuel is partly substituted by hydroelectric energy. The design of DNSFM differed from DNSF in the following ways: An electrical heat pump with a coefficient of performance (COP) of 3 was used for heating i.e. 1 kWh energy consumed would provide 3 kWh of output heat. There was an activation of mechanical cooling and heat harvest during the day when the temperature in the greenhouse exceeded 25 C. In addition, CO 2 -enrichment was provided by pure CO 2 . All electricity was assumed to be from a hydroelectrical power plant representing the energy supply conditions in Norway (The Norwegian Water Resources and Energy Directorate, 2020). This design can be considered to be a relatively closed design as compared to the others and is expected to have lower fossil fuel use.

2.3.
The effect of tomato price and energy costs on the NFR Economic performance of the simulated cases depends on the tomato price and the energy cost in the production seasons. The sensitivity of the economic performance of the evaluated greenhouse designs to the seasonal tomato price was analysed by varying the tomato price and energy costs within the range of 14.5 NOK kg À1 to 19.5 NOK kg-1 using a 1 NOK stepsize and 0.14 NOK kWh À1 to 0.64 NOK kWh À1 with a 0.05 NOK step-size from the original energy cost respectively.
Table 3 e The different greenhouse technological design packages. The NS represents the greenhouse in SW Norway (Orre), for which the indoor climate and tomato yield prediction accuracy was evaluated. The greenhouse design with two energy screens was extended with various combinations of CO 2 -enrichment and with heat buffer technology. Numbers in table are explained in the e j column in Table 1. The columns 1e4 represent traditional production using fossil energy, while column 5 represents a production based on hydro-electrical energy. b i o s y s t e m s e n g i n e e r i n g x x x ( x x x x ) x x x 3.

Prediction accuracy of observed indoor greenhouse climate and tomato yield in Orre
The Relative Root Mean Squared Error (RRMSE), Mean Bias Error (MBE) and Mean Absolute Error (MAE) for temperature, CO 2concentration and fresh mass tomato yield are shown in Table 5. While the RRMSE for the three variables is less than 10%, pointing towards the model being relatively accurate, the results from MBE show that the model prediction, especially for CO 2 , is negatively biased. The MAE results show that the model's prediction of CO 2 values differs on average by 40 ppm from the measured values. This implies that the lower predictions of CO 2 could also have affected the predicted values of yield negatively.
Generally, throughout the production period, the simulated temperature varied from 2 to 3 below the measured values to 1e2 above the measured temperature with lower differences during most of the period (Fig. 8). Notably, the model under-predicted the measured temperature in the beginning of the growing season as exemplified by the period from 20th March to 26th March (day of year 80e86) whereas during mid-production the difference between predicted and measured temperature was lower. During the last period of the growing season, the model tended to over-predict the measured temperature growing season as exemplified in the period from 17th to 25th September (day of year 260e268) in Fig. 10. Also, the accuracy of the predictions of CO 2 -concentration varied during the growing season. During the first period of the growing season, as exemplified by period from 20th March to 26th March, the prediction accuracy varied (Fig. 8).
During the mid-season, as exemplified by the period from 30th June to 6th July (day of year 181e187), the prediction accuracy of the CO 2 -concentration was lower during the day than during the night (Fig. 9).
At the end of the season, the measured CO 2 -concentration was generally over-predicted during the day and underpredicted at night (Fig. 10).
Overall, the simulated yield was close to the measured freshmass yield (Fig. 11). The model, however, under-predicted the measured yield at the beginning of the season, which may be due to the lower temperature prediction at the beginning of the growing season (Fig. 8). The over-prediction of the yield at the end of the season may be due to the higher temperature predicted by the model at the end of the season (Fig. 10).
There was a clear decrease in ventilation in the DNSFM greenhouse due to the mechanical heating and cooling. For instance, the percentage ventilation for the DNSFM design decreased by 0.9% as compared to the other four designs not having the mechanical heating and cooling equipment and that had average ventilation for the entire growing season of about 0.24%.

Net financial return (NFR)
The present simulation study showed clear region-dependent differences in NFR and its underlying components as well as  b i o s y s t e m s e n g i n e e r i n g x x x ( x x x x ) x x x in fossil energy use between greenhouse types with different energy saving and temperature regulation elements. Of the four locations studied, it was found that the NFR was highest for Kise, and lowest for Tromsø for all investigated greenhouse designs. Moreover, for both Maere and Tromsø, the NFR was negative for all designs. This was primarily due to the low temperature and low solar radiation at these locations, which necessitated high costs for energy and resulted in low crop yield. The effect of the greenhouse structure on NFR differed between locations. Applying a night energy screen in the NS design increased the NFR at all locations. When a day energy screen was added (DNS design), the NFR declined compared to the greenhouse with just a night energy screen (NS) at all locations and also compared to the greenhouse with no screen (0S). One possible explanation for this result could be that, while there was no significant increase in energy saving, there was a high increase in the installation costs. This makes 0S the design with the second highest NFR for all locations (Table 6). When mechanical heating and cooling was introduced in the greenhouse design DNSFM, the NFR decreased as compared to all other designs with the lowest NFR for all locations except Tromsø, which had an almost equal NFR for the DNSF and DNSFM designs. Moreover, the fact that the difference in NFR among regions followed the same pattern for all greenhouses with negative economic performance in Maere and Tromsø, gives a clear indication of the regions of Norway where traditional March to October seasonal greenhouse tomato production is economically viable for a rather wide range of greenhouse constructions. The decrease in energy use associated with the application of a day energy screen and mechanical heating and cooling equipment clearly illustrates that there is a

Fixed and variable cost analysis
With the increase in energy saving equipment across the greenhouse designs from the one with no screen (0S) to the one with mechanical heating and cooling (DNSFM), there was a gradual decline in the energy costs resulting in decreased variable costs for all locations. The decrease in variable costs ranged from 58.8 (in Kise) to 74.0 (in Tromsø) NOK m À2 year À1 in all locations for DNSFM as compared to the greenhouse with no energy screen (0S). By using energy screens and mechanical heating and cooling, less heating was required and thus a smaller sized boiler was needed. Using a boiler with smaller capacity, i.e. 0.75 MW, also reduced fixed costs (Table  6). However, the overall fixed costs increased with the increase in investments in equipment for regulation of temperature and energy use for all locations. The results show that energy-saving equipment, with the exception of the night screen, is not particularly profitable for seasonal production due to the differences between their associated costs per m 2 and the increase in yield or decrease in energy use as compared to the design with the night screen. Likewise, it was found that fogging can be omitted under the investigated production regimes, since it had negligible impact on energy saving and potential crop yield.

Prediction of crop yield
There was a slight decrease in the simulated yield for all locations when going from 0S to NS, which can be explained by the shading effect of the structure added for the night energy screen. There was a further decline in the potential crop yield when going from NS to DNS in all locations, which might be explained by the shading effect of the day energy screen. At all locations, adding mechanical heating and cooling equipment (DNSFM) had a slightly positive effect on the crop yield value (Table 6). These results indicate that a more closed system with less variability in the indoor climate is positive for the tomato growth and production. This can be explained by the observation that a closed greenhouse design prevents heat loss and CO 2 loss, which in turn has a positive effect on the photosynthesis process during the day.

Effects on energy and CO 2 use
The changes in the profit notwithstanding, the increase in investments in energy screens and mechanical heating and cooling equipment had the added benefit of lowering the use of fossil energy. These results are linked to the lower  ventilation in the greenhouses with a more advanced design than in those without mechanical heating and cooling, curtailing energy losses and water losses through transpiration. For instance, as shown in Table 7, for Kise, the fossil fuel consumption decreased with the investment in energy screen and adding mechanical heating and cooling (DNSFM) by 198.6 kWh m À2 as compared to the design with no screen (0S). The same tendency for reduced energy use can be seen for the other locations, with the highest decrease in fossil fuel use recorded in Tromsø (236.2 kWh m À2 ). Likewise, the DNSFM design had a lower CO 2 use due to shorter periods with open windows. Nonetheless, the model predicted an increase in the use of pure CO 2 of about 1.2 kg m À2 from 0S to DNSFM for all locations, with the highest pure CO 2 use in Kise. The reason for the highest usage in Kise was the low fossil fuel use as compared to the other locations. The total CO 2 use is shown in Table 7, which includes pure CO 2 and CO 2 from gas. The CO 2 from gas decreases with the increase of investments in energy screens, fogging and mechanical heating and cooling equipment.

Effect of tomato price and energy costs on NFR
The results showed that there is a linear relationship between tomato prices and the NFR, and that with an increase in tomato prices, NFR also increases. Likewise, a tomato price of 14.5 NOK or lower resulted in net losses for all greenhouse designs across all locations. On the contrary, a price of 19.5 NOK or higher increased profit for all designs in all locations. For Kise, however, the minimum price out of the selected range of tomato price for a positive NFR for the designs 0S and NS was calculated to be 15.5 NOK. For all other locations, the same price resulted in negative NFR for all designs. On the other hand, in Tromsø the minimum price required for a positive NFR for any design was 17.5 NOK. Another trend observed from the analysis was the variation in the effects of tomato prices on NFR in different locations (Fig. 12). For instance, Kise witnessed the most positive change in NFR following a price increase, while Tromsø faced the most negative effect in NFR with a decrease in tomato prices. The main reason for this trend is the difference in potential crop yield and energy used (Fig. 13).
However, when tomato prices are considered along with the energy costs, the results show that the designs with the energy-saving elements become more profitable and economically viable and environmental friendly as compared to the standard greenhouse design prevalent in Norway.

Discussion
The results of our study emphasise the importance of considering energy-saving design elements, notably night energy screens, which had the most positive effects on the NFR, in greenhouse construction for tomato production in Norway and can be equally relevant for other countries with similar climatic conditions. The benefits of night thermal screen are similar to findings under other climate conditions (Gupta & Chandra, 2002;Shukla, Tiwari, & Sodha, 2006;Mobtaker, Ajabshirchi, Ranjbar, & Matloobi, 2016). However, there are, to our knowledge, no published scientific findings for the conditions we have studied here. That the beneficial  199.4 198.9 197.2 197.2 197.2 196.2 195.1 193.7 193.7 193.7  b i o s y s t e m s e n g i n e e r i n g x x x ( x x x x ) x x x effects of night screen under these conditions are not established knowledge is further underlined by the fact that most greenhouse tomatoes in Norway are produced without this equipment (Milford, Verheul, Sivertsen, & Kaufmann, 2021).
Our application of a model (Vanthoor et al., 2012a) to simulate greenhouse tomato production for cold-temperate conditions with a large potential supply of renewable energy for heating has revealed results that cannot be drawn with any precision from similar studies related to greenhouse energy-yield-economy modelling and which have been applied to other climate conditions. A previous application of the same model showed that a Parral, a greenhouse with a single bay, whitewash and fogging, had a higher NFR than a Parral with whitewash and heating, and a multi-tunnel design with whitewash, for economic and climate conditions in Spain using other design elements, which contrasts with the lack of effect of fogging on NFR that we found for conditions representing Norway.
Other energy-yield-economy analyses of greenhouses have largely focused on other sources of renewable energy such as wind, solar and biomass, and primarily to study yearround production (Acosta-Silva et al., 2019; Bartzanas, Tchamitchian, & Kittas, 2005;Ç akır & S ‚ ahin, 2015;Mussard, 2017;Fuller, Aye, Zahnd, & Thakuri, 2009;Campiotti et al., 2010;Henshaw, 2017;Aș;chilean, R asoi, Raboaca, Filote & Culcer, 2018). In some studies, the model used was not validated against existing conditions and instead used data from  b i o s y s t e m s e n g i n e e r i n g x x x ( x x x x ) x x x previous models, while other studies have used the model to simulate just one day or a limited number of days (Gupta & Chandra, 2002;Su & Xu, 2015). The results of our evaluation of the effect of several design elements together on NFR and on the use of fossil fuel also differ from and arguably add to the results from other greenhouse design studies that have analysed economic performance but dealt with one or two aspects of the greenhouse design but not varied other design elements, for instance energy and economic analysis for greenhouse ground insulation design (Bambara & Athienitis, 2018), cost and benefit analysis for different greenhouse covers (Lopez-Marin, Rodriguez, Del Amor, Galvez, & Brotons-Martinez, 2019), economic analysis of greenhouse energy use (Ahamed et al., 2019;Mohammadi & Omid, 2010). There are, however, some uncertainties and shortcomings associated with our study which deserve further discussion.
First, the reliability of the simulations is arguably higher for the greenhouse Night energy screen (NS) type against which the model was validated at Orre than when using the model to evaluate the other combinations of locations and greenhouses for which there was no validation data. The accuracy of the predictions of indoor temperature and CO 2 -concentration as well as tomato growth and yield could possibly have been different in other regions with different outdoor climate conditions and for other designs. Hence the simulated NFR and its underlying components are probably more reliable for greenhouse seasonal production in southwestern Norway and regions with similar climate conditions. Additional validation against data from greenhouses with artificial light in Orre and Maere (Naseer et al., submitted) have indicated that the model can produce accurate results for a wider range of conditions. Fig. 13 e The effect of tomato price and energy costs on the NFR for the greenhouse in Orre (SW Norway). The figure shows that if the energy prices increase, the design with energy-saving elements results in higher NFR as compared to the standard greenhouse in Norway.
Secondly, the results show a discrepancy in temperature and CO 2 values between the measured and simulated environmental conditions, as shown in the measurement of errors, which may be related to the ventilation. Generally, growers in Rogaland region tend to open the windows in the evening so that there is a sudden drop of greenhouse air temperature. This is done so as to allow the plants to transition into the night-time mode. In addition, the model requires a long time to adapt to such a change, and the presence of a screen lengthens the time constant. Moreover, it has to be said that leakage ventilation, which may be a relevant fraction of night-time ventilation, is something that is only ''guessed'' at by any model, as it is heavily dependent on the quality, and age, of each greenhouse. This implies that the model is not particularly sensitive to CO 2 , which lowers the accuracy of outputs from the simulations, pointing towards an inherent limitation of models. This is especially important since the growth, quantity and quality of the yield is greatly affected by levels of CO 2 enrichment (Karim et al., 2020;Kl€ aring, Hauschild, Heibner, & Bar-yosef, 2007;Lanoue, 2020;Singh, Poudel, Dunn, Fontanier, & Kakani, 2020).
Thirdly, the fraction of total tomato yield that is marketable depends on the greenhouse design and has a big impact on the NFR. In practical experiments, the marketable yield can decline due to diseases and pests (G azquez et al., 2007) and can also be affected by a high relative humidity in the air inside the greenhouse, which necessitates the opening of the windows, thereby changing the indoor climate of the greenhouse. These factors, however, have not been taken into account in our simulations and may be incorporated in future modifications of the model.
Fourthly, although the considerations of NFR include the fact that the greenhouse and the equipment used in the production process have different lifespans, also depending upon re-investments etc., the return of investment and the payback period has not been considered in the present work. The pay-back period is heavily dependent on interest on capital and thus on prevailing conditions. Adding this aspect in the future works can help in an improved ability to make relevant decisions. The results of our study, which are based on the reproduction of the physics of a complex system, are probably of more general value than could be achieved in an experiment based on a few greenhouse compartments where results may be affected by issues such as crop health, greenhouse leakages, etc. Nonetheless, this simulation study arguably provides a good indication of the economic performance and energy use of greenhouses throughout Norway using design elements and existing market conditions that make the simulations close to the actual values. The alternative of obtaining such information solely from experimental studies would be very costly and therefore would not be realistic to conduct given the number of locations and greenhouse combinations that we have included in our study.
The design alternatives, outdoor conditions and economic settings that were evaluated here represent those that were considered relevant for current greenhouse tomato production in Norway. The rather small difference in NFR sensitivity to changes in energy and tomato prices between greenhouse designs and locations indicates that the possibility to reduce the risk exposure to these factors by changing the greenhouse design is limited under Norwegian production conditions. Previous studies have revealed that there is a considerable impact of climate set-points on NFR under other climate and production conditions, which will have impact on the optimal design as well (Vanthoor, Stanghellini, van Henten, & De Visser, 2008). The next step could include an analysis of NFR for different climate set-points as well as greenhouse sizes and weather conditions at the four locations. To compare the impact of greenhouse structure and climate modification techniques on NFR, costs related to the irrigation system, climate computer, emergency power and internal transport and harvesting systems were assumed to be identical for all greenhouse designs. Since these costs vary between greenhouses, notably due to greenhouse size, it could be useful to vary them in further profitability analyses. Moreover, to improve the greenhouse design for Nordic countries, where light is often the limiting factor, other climate modification techniques such as artificial lighting (light-emitting diode (LED), high pressure sodium (HPS)), an active heat buffer and a heat pump might be integrated in a model for year-round production and evaluated for different production conditions.
The results of our study show that the evaluation of feasible greenhouse types, with a special focus on energysaving elements, could be useful for local tomato growers in decisions related to construction of new greenhouses or renovation of existing ones. The combination of NFR with reduced use of fossil energy, an important indicator of environmental impact, could prove beneficial for policy-makers regarding facilitation of measures geared towards stimulating greenhouse production and the reduction of CO 2 emissions in a country.

Conclusion
This study has used a model-based greenhouse design comprising a crop growth module, greenhouse indoor climate module and an economic module to determine the economic performance of tomato production in (semi-) closed greenhouses that use different forms of energy and utilise different temperature regulation technology under Norwegian seasonal production conditions. The results reveal that, for seasonal tomato production, adding a night energy screen, the use of which is at present limited in Norway, increased the NFR at all evaluated locations, with the highest NFR of 47.6 NOK m À2 in Kise in Eastern Norway. On the other hand, investing in high-tech energy saving equipment could be beneficial in the colder regions since they reduced the energy use, despite comparatively lower economic performance. The lowest fossil fuel use was seen in Kise that of 137.4 kWh m À2 , for the design having both a day and night energy screen, fogging equipment, cooling and heat harvest equipment. The results from our sensitivity analysis show that Tromsø was the most sensitive to variations in tomato and energy prices due to the difference in potential crop yield and energy used. The study offers interesting insights into studies related to greenhouse vegetable production in high latitude regions with large potential supplies of renewable energy and can assist growers at different locations in Norway to select suitable greenhouse designs and pave the way for further