The Influence of Opencast Lignite Mining Dehydration on Plant Production—A Methodological Study
Abstract
:1. Introduction
2. External Costs in Agriculture and the Difficulties Associated with Their Estimation
- Drainage depth—with the increase of the depth of the open pit, the area of a cone of depression becomes larger,
- Drainage period—with the increase of the drainage time, the area of a cone of depression also increases,
- Location of the opencast in the catchment area, its size and the directions of inflow of groundwater—in simplified terms, if the open-pit mine is situated in a valley, the drained area increases but the water relations are restored faster. The location of an open-cast mine on a water parting reduces the drainage area, however it is much more difficult and it takes longer for the water level to get restored after the drainage is completed because the runoff of water from the areas situated higher up is restricted,
- The geological structure of drained areas such as the shape and the direction of buried valleys, abundance in water, tectonic faults, hydrogeological cracks, the thickness of geological layers which affect the conditions of supply, circulation, and drainage of groundwater—those factors are very specific for every open cast, however, the location of thicker and impermeable layers closer to the surface reduces the risk and the area of drainage of subsurface and surface water resources,
- The amount of rainfall and surface water supply—with the increase of the abundance of rain and the level of subsurface water, the impact of the mine on the areas located further away from the open cast decreases. The increase in the share of drained agricultural land reduces the permeation of water into deeper soil layers, especially in the period from late autumn to early spring. High variability of the level of precipitation, both seasonal and during individual years, affects the changes in the level of groundwater, which makes it difficult to determine the actual impact of open-pit mines on water conditions,
- Local conditions, e.g., impermeable formations that create areas of the perched water table, hydrogeological cracks that lower the level of subsurface water below the standard level of the area,
- The initial (primary) level of groundwater, which in the case of peripheral areas of the impact land means that the mine will affect the areas with higher water levels while it will not have any influence on the surrounding areas with lower water tables, even those located further away.
3. Materials and Methods
- Use of agricultural land for an open pit, an external dump, and the necessary accompanying infrastructure, e.g., a power plant, conveyor belts, access roads, etc. (the term “open pit area” will be also used later in this paper),
- The occurrence of areas with lowered groundwater level (the term “cone of depression area” will be also used later in the paper),
- Changes in animal populations in the area impacted by the open pit (it will not be analyzed in this paper).
- AoAL—stands for the average area of agricultural land excluded from agricultural production in the area of the open-pit mine (ha AL),
- Ac—stands for the surface allocated for the open-pit mine, the external dump, and the necessary infrastructure, in particular years (ha),
- S—stands for the share of agricultural land in the total area of the analyzed territory (%),
- t—stands for the period of the impact of the open-pit mine, covering the period from the first exclusion of agricultural land until the completion of reclamation, or the entire period of the open-pit mine exerting its impact (years).
- Act—stands for total surface allocated for the open-pit mine, the external dump, and the necessary infrastructure (ha),
- EAL—is an indicator of the average exclusion of area for open-pit mining (%).
- Eco—stands for the external cost in the area of the open-pit mine, the external dump, and the necessary infrastructure ($, €),
- Spi—stands for the average share of the i-th crop in the structure of agricultural land (%),
- Yoi—stands for the yield of the i-th crop in the area of the open-pit mine (t × ha−1),
- pi—stands for the average selling price of the i-th crop ($, € × t−1),
- Pi—stands for the profitability of the production of the i-th crop (%). The average profitability of the whole crop production can also be used in the calculations, but in such a case, it is necessary to use this value for all analyzed crops.
- AdAL—stands for the average area of agricultural land (UR) within the area of the cone of depression (ha UR),
- Af—stands for the area of the cone of depression in subsequent years.
- Ecf—stands for the external cost in the area of the cone of depression,
- Sfi—stands for the share of the i-th crop in the structure of agricultural land in the area of the cone of depression (%),
- Yfi—stands for the yield of the i-th crop in the area of the cone of depression, in the period where the cone of depression does not exert impact (t × ha−1),
- Cli—stands for the estimated average loss of yield for i-th crop (%). For crops where it is not possible to estimate losses, one may use an average weighted level of losses, calculated from losses incurred in crops for which the parameter is known. The average loss in yield for the entire crop production can also be used in calculations, but in such a case, it is recommended that the value be used for all analyzed crops. The amount of lost yield when calculations are based on the level of yield that does not take into account decreased yield due to the cone of depression (mainly ex-ante analyses) can be calculated from the following formula:
- Yfdi—stands for the yield of the i-th crop in the area of the cone of depression, in the period where the cone of depression exerts impact.
- Eco—stands for the external cost in the area of the cone of depression,
- Sfo—stands for the average share of other crops in the structure of agricultural land (%),
- V—the average value of crop production sales ($, € × ha−1).
- –
- The first group, “up to 20 km”, includes the districts of Konin and Turek, where brown lignite open-pit mines are located,
- –
- The second group, 4 districts located at an average distance of 21–40 km away from the open pits,
- –
- The third group, 6 districts located at an average distance of 41–60 km away from the open pits,
- –
- The fourth group, 10 districts located at an average distance of 61–80 km away from the mines,
- –
- The fifth group, 16 districts located at an average distance of 81–100 km away from the mines.
- –
- Group I, the district of Konin Voivodship, where lignite open-cast mines are located,
- –
- Group II, the Bydgoszcz and Włocławek Voivodship, which are located closest to the open-cast mines. In this group, it was also possible to include the district of Sieradz, however, after 1980, the southern part of it was located within the range of strong influence of Bełchatów open-cast mine, which could cause discrepancies in the calculations of the yield level in this group in that period,
- –
- Group III, the remaining 6 voivodships located at an average distance of up to 100 km away from the nearest open pits, i.e., Leszno, Kalisz, Płock, Poznań, Sieradz, and Toruń Voivodships.
- H0: Ѳ1 = Ѳ2=…= Ѳk
- H1: Ǝi, jϵ{1,…,k} Ѳi ≠ Ѳj, where
- Ѳ1, Ѳ2,… Ѳk is the median of the tested variable x for the i-th group.
- n = n1+ n2+…+ nk;
- T1 (i- 1,2,…,k) denotes the sum of ranks in each trial;
- C—correction for bind ranks
- if p ≤ α ⇒ we reject H0 and accept H1
- if p > α ⇒ there are no grounds to reject H0
- H0: μ1 = μ2 = μ3 = … = μk, where
- μ(1, 2, …, k) denotes the mean of the dependent variable in the k-th group,
- towards the alternative hypothesis:
- H1: at least two group means differ.
4. Characteristics of the Konin-Turek Lignite Basin
5. Results
6. Discussion
- –
- External costs associated with the emission of dust in combustion processes and the impact on human health and global warming,
- –
- External costs associated with the emission of suspended particulates as a result of mining processes,
- –
- External costs for agriculture (both crop and animal production), for agri-food industry and forestry, related to the drainage of open pits.
Author Contributions
Funding
Institutional Review Board Statement
Data Availability Statement
Conflicts of Interest
References
- Thielemann, T.; Schmidt, S.; Gerling, J.P. Lignite and hard coal: Energy suppliers for world needs until the year 2100–An outlook. Int. J. Coal Geol. 2007, 72, 1–14. [Google Scholar] [CrossRef]
- Federal Institute for Geosciences and Natural Resources. Energy Study 2014: Reserves, Resources and Availability of Energy Resources; Federal Institute for Geosciences and Natural Resources: Hanover, Germany, 2014.
- Kashiwagi, T.; Ishino, H.; Takagi, T.; Hirose, K. Mine-mouth Power Generation System Based on Upgraded Brown Coal (UBC). Kobelco Technol. Rev. 2015, 33, 28–33. [Google Scholar]
- Bielecki, A.; Ernst, S.; Skrodzka, W.; Wojnicki, I. The externalities of energy production in the context of development of clean energy generation. Environ. Sci. Pollut. Res. 2020, 27, 11506–11530. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Violidakis, I.; Drosatos, P.; Nikolopoulos, N. Critical review of current industrial scale lignite drying technologies. In Low-Rank Coals for Power Generation, Fuel and Chemical Production; Luo, Z., Agraniotis, M., Eds.; Woodhead Publishing: Cambridge, UK, 2017; pp. 41–71. [Google Scholar] [CrossRef]
- Wilczyński, M. Coal in Poland: After the Sunset, Change Partnership.org. 2016. Available online: http://www.changepartnership.org/wp-content/uploads/2016/02/Wilczynski-2016-Coal-in-Poland-after-the-sunset-translation-FINAL.pdf (accessed on 8 December 2020).
- Mohr, S.H.; Evans, G.M. Forecasting coal production until 2100. Fuel 2009, 88, 2059–2067. [Google Scholar] [CrossRef] [Green Version]
- Patzek, T.; Croft, G. A global coal production forecast with multi–Hubbert cycle analysis. Energy 2010, 35, 3109–3122. [Google Scholar] [CrossRef]
- Mukherjee, S.; Pahari, D.P. Underground and Opencast Coal Mining Methods in India: A Comparative Assessment. Space Cult. India 2019, 7, 39–55. [Google Scholar] [CrossRef]
- Braun, M. Environmental External Costs from Power Generation by Renewable Energies; Institut für Energiewirtschaft und Rationelle Energieanwendung, Universität Stuttgart: Stuttgart, Germany, 2004. [Google Scholar]
- International Institute for Sustainable Development. Financial Supports for Coal and Renewables in Indonesia; International Institute for Sustainable Development: Winnipeg, MB, Canada, 2017. [Google Scholar]
- Samadi, S. The Social Costs of Electricity Generation—Categorising Different Types of Costs and Evaluating Their Respective Relevance. Energies 2017, 10, 356. [Google Scholar] [CrossRef] [Green Version]
- Karkour, S.; Ichisugi, Y.; Abeynayaka, A.; Itsubo, N. External-Cost Estimation of Electricity Generation in G20 Countries: Case Study Using a Global Life—Cycle Impact—Assessment Method. Sustainability 2020, 12, 2002. [Google Scholar] [CrossRef] [Green Version]
- Adani’s Coal Mine Plans, and Why Coal Is Still So Lucrative. Available online: https://www.dw.com/en/adanis-coal-mine-plans-and-why-coal-is-still-so-lucrative/a-52025626 (accessed on 4 January 2021).
- Global Coal Exit List. Available online: https://coalexit.org/ (accessed on 4 January 2021).
- Kasztelewicz, Z. Report on the Condition of the Lignite Coal Industry in Poland and Germany Together with a Diagnosis of Measures for the Development of this Industry in the First Half of the 21st Century; AGH University of Science and Technology: Krakow, Poland, 2018. (In Polish) [Google Scholar]
- Kasztelewicz, Z.; Tajduś, A.; Cała, M.; Ptak, M.; Sikora, M. Strategic conditions for the future of brown coal mining in Poland. Energy Policy J. 2018, 21, 155–178. [Google Scholar] [CrossRef]
- Pietraszewski, A. Polish lignite mining in 2014. Węgiel Brunatny 2015, 90, 4–14. (In Polish) [Google Scholar]
- Widera, M.; Kasztelewicz, Z.; Ptak, M. Lignite mining and electricity generation in Poland: The current state and future prospects. Energy Policy 2016, 92, 151–157. [Google Scholar] [CrossRef]
- Ministry of Energy. Energy Policy of Poland Until 2040 (EPP2040)—Extract from Draft v. 2.1; Ministry of Energy: Warsaw, Poland, 2019.
- Pactwa, K.; Woźniak, J.; Strempski, A. Sustainable mining—Challenge of Polish mines. Resour. Policy 2018. [Google Scholar] [CrossRef]
- Faaij, A.; Meuleman, B.; Turkenburg, W.; van Wijk, A.; Bauen, A.; Rosillo-Calle, F.; Hall, D. Externalities of biomass based electricity production compared with power generation from coal in the Netherlands. Biomass Bioenergy 1998, 14, 125–147. [Google Scholar] [CrossRef]
- Rowe, R.D.; Lang, C.M.; Chestnut, L.G.; Latimer, D.A.; Rae, D.A.; Bernow, S.M.; White, D.E. New York State Environmental Externalities Cost Study; Empire State Electric Energy Research Corporation, Oceana: New York, NY, USA, 1995. [Google Scholar]
- Lee, R.; Krupnik, A.; Burtraw, D. Estimating Externalities of Electric Fuel Cycles: Analytical Methods and Issues and Estimating Externalities of Coal Fuel Cycles; Techical Report; McGraw—Hill/Utility Data Institute: Washington, DC, USA, 1995. [Google Scholar]
- Krewitt, W.; Heck, T.; Trukenmüller, A.; Friedrich, R. Environmental damage costs from fossil electricity generation in Germany and Europe. Energy Policy 1999, 27, 173–183. [Google Scholar] [CrossRef]
- EC. ExternE—Externalities of Energy, Methodology 2005 Update; Office for Official Publications of the European Communities: Luxembourg, 2005. [Google Scholar]
- Rečka, L.; Ščasný, M. Brown coal and nuclear energy deployment: Effects on fuel-mix, carbon targets, and external costs in the Czech Republic up to 2050. Fuel 2018, 216, 494–502. [Google Scholar] [CrossRef]
- Jorli, M.; Van Passel, S.; Saghdel, H.S. External costs from fossil electricity generation: A review of the applied impact pathway approach. Energy Environ. 2018, 29, 635–648. [Google Scholar] [CrossRef] [Green Version]
- Ščasný, M.; Massetti, E.; Melichar, J.; Carrara, S. Quantifying the Ancillary Benefits of the Representative Concentration Pathways on Air Quality in Europe; Nota di Lavoro, No. 84.2015; Fondazione Eni Enrico Mattei (FEEM): Milan, Italy, 2015. [Google Scholar]
- Preiss, P.; Klotz, V. EcoSenceWeb V1-3; Insitute of Energy Economics and the Rational Use of Energy: Stuttgart, Germany, 2008. [Google Scholar]
- Wang, L.; Watanabe, T.; Xu, Z. Monetization of External Costs Using Lifecycle Analysis—A Comparative Case Study of Coal–Fired and Biomass Power Plants in Northeast China. Energies 2015, 8, 1440–1467. [Google Scholar] [CrossRef] [Green Version]
- Máca, V.; Melichar, J.; Ščasný, M. Internalization of External Costs of Energy Generation in Central and Eastern European Countries. J. Environ. Dev. 2012, 21, 181–197. [Google Scholar] [CrossRef] [Green Version]
- Pepliński, B. External costs of opencast brown coal mining in agriculture and agri–food industry (on the example of Wielkopolska). Cent. Eur. Econ. J. 2018, 5, 154–171. [Google Scholar] [CrossRef] [Green Version]
- Pepliński, B. External costs of open-cast lignite mining for the agri–food industry (as illustrated by the example of the Oczkowice mine). J. Agribus. Rural Dev. 2018, 2, 213–223. [Google Scholar] [CrossRef]
- Poczta, W.; Pepliński, B.; Sadowski, A.; Czubak, W. Impact of the Planned Oczkowice Mine on the Economic, Productive and Social Potential of Agriculture and its Surroundings in the South–Western Functional Area of the Wielkopolskie Voivodeship; Poznań University of Life Sciences: Poznan, Poland, 2017. (In Polish) [Google Scholar]
- Georgescu-Roegen, N. The Entropy Law and the Economic Process; Harvard University Press: Cambridge, MA, USA, 1971. [Google Scholar]
- Dal, E.; Koksal, M.A. Modeling the use of renewable energy sources for electricity generation based on technological, political, and environmental constraints. Energy Procedia 2017, 142, 3129–3134. [Google Scholar] [CrossRef]
- Rafaj, P.; Kypreos, S. Internalisation of external cost in the power generation sector: Analysis with global multi-regional MARKAL model. Energy Policy 2007, 35, 828–843. [Google Scholar] [CrossRef]
- Coester, A.; Hofkes, M.; Papyrakis, E. Economics of renewable energy expansion and security of supply: A dynamic simulation of the German electricity market. Appl. Energy 2018, 231, 1268–1284. [Google Scholar] [CrossRef]
- Redlinger, R.Y.; Andersen, P.; Morthorst, P. Wind Energy in the 21st Century—Economics, Policy, Technology and the Changing Electricity Industry; Palgrave Macmillan: Basingstoke, UK, 2002. [Google Scholar] [CrossRef]
- Becker, N.; Soloveitchik, D.; Olshansky, M. A weighted average incorporation of pollution costs into the electrical expansion planning. Energy Environ. 2012, 23, 1–15. [Google Scholar] [CrossRef]
- Sakulniyomporn, S.; Kubaha, K.; Chullabodhi, C. External costs of fossil electricity generation: Health–based assessment in Thailand. Renew. Sustain. Energy Rev. 2011, 15, 3470–3479. [Google Scholar] [CrossRef]
- Bozicevic Vrhovcak, M.; Tomsic, Z.; Debrecin, N. External costs of electricity production: Case study Croatia. Energy Policy 2005, 33, 1385–1395. [Google Scholar] [CrossRef]
- Maxim, A. Sustainability assessment of electricity generation technologies using weighted multi-criteria decision analysis. Energy Policy 2014, 65, 284–297. [Google Scholar] [CrossRef]
- Przybyłek, J. Current problems of the lignite open cast mines dewatering in the Wielkopolska region. Górnictwo Odkrywkowe 2018, 2, 5–14. (In Polish) [Google Scholar]
- Przybyłek, J.; Górski, J. The Oczkowice lignite deposit. Voicing concern for proper hydrogeological recognition. Przegląd Geologiczny 2016, 64, 183–191. (In Polish) [Google Scholar]
- Adamczyk, A.F.; Motyka, J. Methodical problems of determination of depression cone boundaries in the areas affected by mining drainage. Współczesne Problemy Hydrogeologii 1995, 7, 209–212. (In Polish) [Google Scholar]
- Pusch, M.; Hoffmann, A. Conservation concept for a river ecosystem (River Spree, Germany) impacted by flow abstraction in a large post-mining area. Landsc. Urban Plan. 2000, 51, 165–176. [Google Scholar] [CrossRef]
- Mwadzingeni, L.; Mugandani, R.; Mafongoya, P. Factors affecting the performance of Tshiombo Irrigation Scheme in Limpopo Province, South Africa. J. Agribus. Rural Dev. 2020, 3, 269–277. [Google Scholar] [CrossRef]
- Shoko, R.R.; Chaminuka, P.; Belete, A. Maize yield sensitivity to climate variability in South Africa: Application of the ARDL-ECM approach. J. Agribus. Rural Dev. 2019, 4, 363–371. [Google Scholar] [CrossRef]
- Rybka, K.; Nitka, Z. Modern phenotypes of cereals for growing in areas endangered with drought. Biuletyn Instytutu Hodowli Aklimatyzacji Roślin 2014, 273, 55–72. (In Polish) [Google Scholar]
- Zawora, T.; Ziernicka–Wojtaszek, A. Extreme values of precipitation deficiencies and excesses in the light of contemporary climate change based on the example from the Podkarpackie province. Infrastruktura Ekologia Terenów Wiejskich 2008, 5, 23–29. (In Polish) [Google Scholar]
- Sillmann, J.; Roeckner, E. Indices for extreme events in projections of anthropogenic climate change. Clim. Chang. 2008, 86, 83–104. [Google Scholar] [CrossRef] [Green Version]
- Trenberth, K.E.; Dai, A.; Rasmussen, R.M.; Parsons, D.B. The Changing character of precipitation. Bull. Am. Meteorol. Soc. 2003, 84, 1205–1217. [Google Scholar] [CrossRef]
- Easterling, D.R.; Horton, B.; Jones, P.D.; Peterson, T.C.; Karl, T.R.; Parker, D.E.; Salinger, M.J.; Razuvayev, V.; Plummer, N.; Jamason, P.; et al. Maximum and Minimum Temperature Trends for the Globe. Science 1997, 277, 364–367. [Google Scholar] [CrossRef] [Green Version]
- Tomassini, L.; Jacob, D. Spatial analysis of trends in extreme precipitation events in high-resolution climate model results and observations for Germany. J. Geophys. Res. Atmos. 2009, 114, D12113. [Google Scholar] [CrossRef]
- Pinke, Z.; Decsi, B.; Kozma, Z.; Vári, Á.; Lövei, G.L. A spatially explicit analysis of wheat and maize yield sensitivity to changing groundwater levels in Hungary, 1961–2010. Sci. Total Environ. 2020, 715, 136555. [Google Scholar] [CrossRef]
- Jalota, S.K.; Vashisht, B.B.; Sharma, S.; Kaur, S. Climate change and groundwater. In Understanding Climate Change: Impacts on Crop Productivity and Water Balance; Academic Press: Cambridge, UK, 2018. [Google Scholar]
- Ray, D.K.; Ramankutty, N.; Mueller, N.D.; West, P.C.; Foley, J.A. Recent patterns of crop yield growth and stagnation. Nat. Commun. 2012, 3, 1293. [Google Scholar] [CrossRef] [Green Version]
- Olesen, J.E.; Bindi, M. Consequences of climate change for European agricultural productivity, land use and policy. Eur. J. Agron. 2002, 16, 239–262. [Google Scholar] [CrossRef]
- Zipper, S.C.; Soylu, M.E.; Booth, E.G.; Loheide, S.P. Untangling the effects of shallow groundwater and soil texture as drivers of subfield–scale yield variability. Water Resour. Res. 2015, 51, 6338–6358. [Google Scholar] [CrossRef] [Green Version]
- Chai, Q.; Gan, Y.; Zhao, C.; Xu, H.; Waskom, R.M.; Niu, Y.; Siddique, K.H.M. Regulated deficit irrigation for crop production under drought stress. A review. Agron. Sustain. Dev. 2016, 36, 3. [Google Scholar] [CrossRef] [Green Version]
- Zipper, S.C.; Qiu, J.; Kucharik, C.J. Drought effects on US maize and soybean production: Spatiotemporal patterns and historical changes. Environ. Res. Lett. 2016, 11, 094021. [Google Scholar] [CrossRef]
- Cattivelli, L.; Rizza, F.; Badeck, F.-W.; Mazzucotelli, E.; Mastrangelo, A.M.; Francia, E.; Marè, C.; Tondelli, A.; Stanca, A.M. Drought tolerance improvement in crop plants: An integrated view from breeding to genomics. Field Crop. Res. 2008, 105, 1–14. [Google Scholar] [CrossRef]
- Vereecken, H.; Weynants, M.; Javaux, M.; Pachepsky, Y.; Schaap, M.G.; van Genuchten, M.T. Using pedotransfer functions to estimate the van Genuchten–Mualem soil hydraulic properties: A review. Vadose Zone J. 2010, 9, 795–820. [Google Scholar] [CrossRef]
- Connolly, R.D. Modelling effects of soil structure on the water balance of soil-crop systems: A review. Soil Tillage Res. 1998, 48, 1–19. [Google Scholar] [CrossRef]
- Talebnejad, R.; Sepaskhah, A.R. Effect of different saline groundwater depths and irrigation water salinities on yield and water use of Quinoa in Lysimeter. Agric. Water Manag. 2015, 148, 177–188. [Google Scholar] [CrossRef]
- Wu, Y.; Liu, T.; Paredes, P.; Duan, L.; Pereira, L.S. Water use by a groundwater dependent maize in a semi–arid region of Inner Mongolia: Evapotranspiration partitioning and capillary rise. Agric. Water Manag. 2015, 152, 222–232. [Google Scholar] [CrossRef] [Green Version]
- Karimov, A.K.; Šimůnek, J.; Hanjra, M.A.; Avliyakulov, M.; Forkutsa, I. Effects of the shallow water table on water use of winter wheat and ecosystem health: Implications for unlocking the potential of groundwater in the Fergana Valley (Central Asia). Agric. Water Manag. 2014, 131, 57–69. [Google Scholar] [CrossRef]
- Ayars, J.E.; Shouse, P.; Lesch, S.M. In situ use of groundwater by alfalfa. Agric. Water Manag. 2009, 96, 1579–1586. [Google Scholar] [CrossRef]
- Yang, J.; Wan, W.; Deng, W.; Zhang, G. Water fluxes at a fluctuating water table and groundwater contributions to wheat water use in the lower Yellow River flood plain, China. Hydrol. Proc. 2007, 21, 717–724. [Google Scholar] [CrossRef]
- Kahlown, M.A.; Ashraf, M.; Zia-ul-Haq. Effect of shallow groundwater table on crop water requirements and crop yields. Agric. Water Manag. 2005, 76, 24–35. [Google Scholar] [CrossRef]
- Hutmacher, R.B.; Ayars, J.E.; Vail, S.S.; Bravo, A.D.; Dettinger, D.; Schoneman, R.A. Uptake of shallow groundwater by cotton: Growth stage, groundwater salinity effects in column lysimeters. Agric. Water Manag. 1996, 31, 205–223. [Google Scholar] [CrossRef]
- Rydałowski, M. The influence of groundwater level and year conditions on the evapotranspiration of salix viminalis l. in view of lysimetric analyses. Woda Środowisko Obszary Wiejskie 2016, 16, 73–84. (In Polish) [Google Scholar]
- Kravchenko, A.N.; Bullock, D.G. Spatial variability of soybean quality data as a function of field topography: I. Spatial data analysis. Crop Sci. 2002, 42, 804–815. [Google Scholar] [CrossRef]
- Kravchenko, A.N.; Bullock, D.G. Spatial variability of soybean quality data as a function of field topography: II. A proposed technique for calculating the size of the area for differential soybean harvest. Crop Sci. 2002, 42, 816–821. [Google Scholar] [CrossRef]
- Kravchenko, A.N.; Thelen, K.D.; Bullock, D.G.; Miller, N.R. Relationship among crop grain yield, topography, and soil electrical conductivity studied with cross-correlograms. Agron. J. 2003, 95, 1132–1139. [Google Scholar] [CrossRef]
- Kennedy, B.A. Surface Mining, 2nd ed.; Society for Mining, Metallurgy and Exploration Inc.: Englewood, CO, USA, 1990. [Google Scholar]
- Shah, A.M. Reclaiming Indian Mines. Cornerstone 2015, 3, 36–39. [Google Scholar]
- Pietrzykowski, M. Reclamation and reconstruction of terrestrial ecosystems on mine sites–ecological effectiveness assessment. In Energy Science and Technology, Coal Energy; Govil, J.N., Ed.; Studium Press LLC: Houston, TX, USA; New Dehli, India, 2015; Volume 2, pp. 121–151. [Google Scholar]
- Pavloudakis, F.; Roumpos, C.; Karlopoulos, E.; Koukouzas, N. Sustainable Rehabilitation of Surface Coal Mining Areas: The Case of Greek Lignite Mines. Energies 2020, 13, 3995. [Google Scholar] [CrossRef]
- Gruszczyński, S. Classification of soils in post mining areas reclamation. Przegląd Górniczy 2010, 66, 120–125. (In Polish) [Google Scholar]
- Kasztelewicz, Z. Approaches to post-mining land reclamation In Polish open-cast lignite mining. Civ. Environ. Eng. Rep. 2014, 12, 55–67. [Google Scholar] [CrossRef] [Green Version]
- Michalski, A. Methods of management of former opencast mining areas of “Adamów” S.A. in Turek and “Konin” S.A. in Kleczew. Soil Sci. Annu. 2004, 55, 281–290. [Google Scholar]
- The Photovoltaic Farm in Brudzew Will be Built by ZE PAK Subsidiary. Available online: https://www.wnp.pl/energetyka/farme-fotowoltaiczna-w-brudzewie-wybuduje-spolka-corka-ze-pak,439843.html (accessed on 17 January 2021). (In Polish).
- Szczepiński, J. The Significance of Groundwater Flow Modeling Study for Simulation of Opencast Mine Dewatering, Flooding, and the Environmental Impact. Water 2019, 11, 848. [Google Scholar] [CrossRef] [Green Version]
- Healy, R. Estimating Groundwater Recharge, 1st ed.; Cambridge University Press: Cambridge, UK, 2010. [Google Scholar]
- Stachowski, P. Assessment of the reconstruction of the water table in the areas transformed by brown coal mining. Rocznik Ochrona Środowiska 2007, 9, 225–238. (In Polish) [Google Scholar]
- Basso, B.; Kendall, A.D.; Hyndman, D.W. The future of agriculture over the Ogallala Aquifer: Solutions to grow crops more efficiently with limited water. Earths Future 2013, 1, 39–41. [Google Scholar] [CrossRef]
- Kendy, E.; Zhang, Y.; Liu, C.; Jinxia, C.; Wang, J.; Steenhuis, T. Groundwater recharge from irrigated cropland in the North China Plain: Case study of Luancheng County, Hebei Province, 1949–2000. Hydrol. Process. 2004, 18, 2289–2302. [Google Scholar] [CrossRef]
- Aw-Hassan, A.; Rida, F.; Telleria, R.; Bruggeman, A. The impact of food and agricultural policies on groundwater use in Syria. J. Hydrol. 2014, 513, 204–215. [Google Scholar] [CrossRef] [Green Version]
- Calvino, P.A.; Andrade, F.H.; Sadras, V.O. Maize yield as affected by water availability, soil depth, and crop management. Agron. J. 2003, 95, 275–281. [Google Scholar] [CrossRef]
- Leeper, R.A.; Runge, E.C.A.; Walker, W.M. Effect of plant–available stored soil moisture on corn yields: 1. Constant climatic conditions. Agron. J. 1974, 66, 723–727. [Google Scholar] [CrossRef]
- Kasztelewicz, Z.; Sikora, M.; Zajączkowski, M. The Poniec-Krobia deposit in extending the Konin lignite mining and power generation basin. Energy Policy J. 2012, 15, 135–146. (In Polish) [Google Scholar]
- European Commission. A Farm to Fork Strategy for a Fair, Healthy and Environmentally-Friendly Food System; Communication from the Commission to the European Parliament, the Council, the European Economic and Social Committee and the Committee of the Regions; 20.5.2020 COM (2020) 381 Final; European Commission: Brussels, Belgium, 2020. [Google Scholar]
- Pepliński, B. Determinants of Regional Changes in the Pig Production Sector in Poland; Poznań University of Life Sciences: Poznan, Poland, 2019. (In Polish) [Google Scholar]
- Central Statistical Office. Procurement and Prices of Agricultural Products in 2015–2019; Central Statistical Office: Warsaw, Poland, 2020. (In Polish)
- Central Statistical Office. Statistical Yearbook 1956–1997; Central Statistical Office: Warsaw, Poland, 1998. (In Polish)
- Central Statistical Office. Statistical Yearbook of Voivodships 1975–1997; Central Statistical Office: Warsaw, Poland, 1998. (In Polish)
- Central Statistical Office. Statistical Yearbook of Poviats 1970–1974; Central Statistical Office: Warsaw, Poland, 1975. (In Polish)
- Central Statistical Office. Poviats Statistics 1966–1970; Central Statistical Office: Warsaw, Poland, 1971. (In Polish)
- Central Statistical Office. Statistical Yearbook of the Bydgoszcz Voivodship 1960–1974; Central Statistical Office: Bydgoszcz, Poland, 1975. (In Polish)
- Central Statistical Office. Statistical Yearbook of the Łódź Voivodship 1960–1974; Central Statistical Office: Lodz, Poland, 1975. (In Polish)
- Central Statistical Office. Statistical Yearbook of the Poznań Voivodship 1960–1974; Central Statistical Office: Poznan, Poland, 1975. (In Polish)
- Central Statistical Office. Statistical Yearbook of the Warszawa Voivodship 1960–1974; Central Statistical Office: Warsaw, Poland, 1975. (In Polish)
- Central Statistical Office. Economic Development of Poviats in the Years 1950–1965; Central Statistical Office: Warsaw, Poland, 1966. (In Polish)
- Local Date Bank. Available online: https://bdl.stat.gov.pl/BDL/start (accessed on 5 January 2021).
- Pepliński, B. Animal production location in suburban areas of Polish metropolitan centers. Sustainability 2020, 12, 2619. [Google Scholar] [CrossRef] [Green Version]
- Stanisz, A. An Accessible Statistics Course Using STATISTICA PL with Examples from Medicine; StatSoft: Krakow, Poland, 2007. (In Polish) [Google Scholar]
- Plans for Reclamation of Post-Mining Areas as Part of Activities Carried Out by ZE PAK S.A. Available online: https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&ved=2ahUKEwiPpe6r7MruAhVNiYsKHdDzAJgQFjAEegQIBRAC&url=http%3A%2F%2Forka.sejm.gov.pl%2Fopinie9.nsf%2Fdok%3FOpenAgent%26669_20201014_2&usg=AOvVaw08GYu_jyh245tOoCSpN3-- (accessed on 2 February 2021). (In Polish)
- Gilewska, M.; Otremba, M. Revitalisation of post-mining resions in the area of the town of Konin. Zeszyty Naukowe Inżynieria Środowiska 2013, 149, 59–67. (In Polish) [Google Scholar]
- Kordowski, J.; Brykała, D.; Bartczak, A.; Kaczmarek, H.; Słowiński, M. Anthropogenic and Natural Changes of Geographical Environment in the Kuyavian-Pomeranian Voivodeship–Selected Examples, Prace Geograficzne 223; PAN: Warsaw, Poland, 2010. (In Polish) [Google Scholar]
- Paczyński, B.; Sadurski, A. Regional Hydrogeology of Poland: Mineral, Curative and Thermal Waters and Mine Waters; Państwowy Instytut Geologiczny: Warsaw, Poland, 2007; Volume II. (In Polish) [Google Scholar]
- Kasztelewicz, Z.; Czyż, J.; Dwornik, E. Mining art in the 60th anniversary of KWB “Konin”. Węgiel Brunatny 2005, 53, 23–29. (In Polish) [Google Scholar]
- Wachowiak, G.; Wachowiak, A. Reservoir at the closing excavation of the open pit “Pątnów” of the lignite opencast mine “Konin” and its water balance in 2003–2004. Badania Fizjograficzne Polska Zachodnia Seria A Geografia Fizyczna 2005, 56, 157–176. (In Polish) [Google Scholar]
- Opencasts. Available online: http://www.kwbkonin.pl/ (accessed on 25 January 2021). (In Polish).
- Maćkowiak, J.; Siepielska, T.; Gradecki, D. Impact of lignite drainage in the Konin region on the near-surface aquifer. Węgiel Brunatny 1998, 22, 26–28. (In Polish) [Google Scholar]
- Mocek, A.; Owczarzak, W. Monitoring of Soil-Soil Water Table Necessary for Delineation of Soil Degradation Zones (Mainly Grasslands) in the Area Covered by the Stage and Soil Survey in the Tomisławice Open-Pit Area; Poznań University of Life Sciences: Poznan, Poland, March 2013. (In Polish) [Google Scholar]
- Magdziarek, M.; Brusiło, A.; Waloch, P.; Magdziarek, M.; Szulczyk, J.; Marecki, M. Report on Environmental Impact of the Odkrywka Ościsłowo; Ekogeo, Przedsiębiorstwo Projektowo—Usługowe: Wargowo, Poland, 2015. (In Polish) [Google Scholar]
- Szablewski, W.; Derkowska–Sitarz, M. Filling of Lubstów reservoir of KWB Konin S.A. Górnictwo Odkrywkowe 2014, 60, 40–46. (In Polish) [Google Scholar]
- Gilecki, R.; Kacprowska, J.; Dembicz, E.; Nagrodkiewicz, L.; Stosio, M. Primary Energy Balance 2004–2019; Agencja Rozwoju Energii: Warsaw, Poland, 2020. (In Polish) [Google Scholar]
- Kasztelewicz, Z.; Ptak, M.; Sikora, M. Lignite as the optimal energy source for Poland. Zeszyty Naukowe IGSMiE PAN 2018, 106, 61–84. (In Polish) [Google Scholar]
- Kasztelewicz, Z.; Patyk, M. The Modern and Efficiently Coal Power Plants Strategic Challenge for Polish. Energy Policy J. 2015, 18, 46–59. (In Polish) [Google Scholar]
- Zespół Elektrowni Pątnów-Adamów-Konin SA, 2016 Results. Available online: https://ri.zepak.com.pl/en/presentations.html?y=2016 (accessed on 31 January 2021).
- Zespół Elektrowni Pątnów-Adamów-Konin SA, 2014 Results. Available online: https://ri.zepak.com.pl/en/presentations.html?y=2014 (accessed on 31 January 2021).
- Zespół Elektrowni Pątnów-Adamów-Konin SA, 2012 Results. Available online: https://ri.zepak.com.pl/en/presentations.html?y=2013 (accessed on 31 January 2021).
- Sadowski, A.; Czubak, W.; Poczta, W.; Rowiński, J. Area and economic structures of Polish agriculture and other EU countries. In The Structure of Polish Agriculture Compared to the European Union; Poczta, W., Rowiński, J., Eds.; CeDeWu: Warsaw, Poland, 2019; pp. 45–81. (In Polish) [Google Scholar]
- Pepliński, B.; Poczta, W.; Rowiński, J. Structure of livestock production and its evolution. In The Structure of Polish Agriculture Compared to the European Union; Poczta, W., Rowiński, J., Eds.; CeDeWu: Warsaw, Poland, 2019; pp. 83–139. (In Polish) [Google Scholar]
- Electricity Market Characteristics. 2019. Available online: https://www.ure.gov.pl/en/markets/electricity/elctricitymrket/292,2019-Electricity-Market-Characteristics.html (accessed on 3 February 2021).
- Włodek, S.; Biskupski, A. Soil moisture dynamics in the Belchatow brown coal mine area. Soil Sci. Annu. 2011, 62, 420–427. (In Polish) [Google Scholar]
- Owczarzak, W.; Mocek, A.; Kaczmarek, Z. Impact of the “Drzewce” open-cast mine drainage barrier on the hydrological changes in the surrounding area. Soil Sci. Annu. 2011, 62, 311–324. (In Polish) [Google Scholar]
- Owczarzak, W.; Mocek, A.; Kaczmarek, Z.; Gajewski, P. The evaluation of natural and anthropogenic degradation of soils in the area of dehydration influence of Drzewce lignite pit. Soil Sci. Annu. 2008, 59, 177–190. (In Polish) [Google Scholar]
- Biskupski, A.; Włodek, S.; Pabin, J. Soil moisture dynamics within the depression sink of brown coal open mine at Belchatow. Soil Sci. Annu. 2008, 59, 18–24. (In Polish) [Google Scholar]
- Pabin, J.; Włodek, S.; Biskupski, A. The influence of the deep drainage of the brown coal strip mine at Belchatow on soil water content. Soil Sci. Annu. 2008, 59, 191–195. (In Polish) [Google Scholar]
- Włodek, S.; Biskupski, A.; Pabin, J. Soil moisture in the region of brown coal strip mine at Belchatow. Soil Sci. Annu. 2008, 59, 230–235. (In Polish) [Google Scholar]
- Mocek, A.; Owczarzak, W.; Kaczmarek, Z. Changes of groundwater levels in soils surrounding the “Kozmin” brown coal exploitation hollow. Roczniki Akademii Rolniczej Poznaniu Melioracje Inżynieria Środowiska 2002, 23, 331–342. (In Polish) [Google Scholar]
- Mocek, A.; Owczarzak, W.; Kaczmarek, Z. Hydrological changes in the soils of Warta Pravalley within the influance of “Koźmin” lignite pit. Contemp. Probl. Manag. Environ. Prot. 2009, 1, 109–124. [Google Scholar]
- Chodak, T.; Kaszubkiewicz, J.; Bogda, A.; Szerszeń, L.; Bogacz, A. Water conditions of soils located within the range of depression funnel of the existing water intake in Brzeg Głogowski and the planned one in the region of Czerna-Dobrzejowice. Zeszyty Problemowe Postepow Nauk Rolniczych 1995, 418, 169–174. (In Polish) [Google Scholar]
- Chodak, T.; Borkowski, J.; Bogda, A.; Szerszeń, L.; Nowiński, S. Prognosis of changes in soil environment within the range of depression funnel of “Potoczek” water intake. Zeszyty Problemowe Postepow Nauk Rolniczych 1995, 418, 175–184. (In Polish) [Google Scholar]
- Włodek, S.; Biskupski, A.; Pabin, J.; Hryńczuk, B.; Kaus, A. The dynamics of ground moisture in the region of brown coal strip mine at Bełchatów on the example of locality Łękińsko. Roczniki Akademii Rolniczej Poznaniu 1999, 160, 211–219. (In Polish) [Google Scholar]
- Hlavinka, P.; Trnka, M.; Semerádová, D.; Dubrovský, M.; Žalud, Z.; Možný, M. Effect of drought on yield variability of key crops in Czech Republic. Agric. For. Meteorol. 2009, 149, 431–442. [Google Scholar] [CrossRef]
- Ren, D.; Xu, X.; Engel, B.; Huang, G. Growth responses of crops and natural vegetation to irrigation and water table changes in an agro–ecosystem of Hetao, upper Yellow River basin: Scenario analysis on maize, sunflower, watermelon and tamarisk. Agric. Water Manag. 2018, 199, 93–104. [Google Scholar] [CrossRef]
- Nosetto, M.D.; Jobbágy, E.G.; Jackson, R.B.; Sznaider, G.A. Reciprocal influence of crops and shallow ground water in sandy landscapes of the Inland Pampas. Field Crop. Res. 2009, 113, 138–148. [Google Scholar] [CrossRef]
- Morgan, B.J.; Daughtry, C.S.; Russ, A.L.; Dulaney, W.P.; Gish, T.J.; Pachepsky, Y.A. Effect of shallow subsurface flow pathway networks on corn yield spatial variation under different weather and nutrient management. Int. Agrophys. 2019, 33, 271–276. [Google Scholar] [CrossRef]
- Georgakellos, D.A. Impact of a possible environmental externalities internalisation on energy prices: The case of the greenhouse gases from the Greek electricity sector. Energy Econ. 2010, 32, 202–209. [Google Scholar] [CrossRef]
- Büke, T.; Köne, A.C. Estimation of the health benefits of controlling air pollution from the Yatağan coal-fired power plant. Environ. Sci. Policy 2011, 14, 1113–1120. [Google Scholar] [CrossRef]
- Dimitrijević, Z.; Tatić, K.; Knežević, A.; Salihbegović, I. External costs from coal-fired thermal plants and sulphur dioxide emission limit values for new plants in Bosnia and Herzegovina. Energy Policy 2011, 39, 3036–3041. [Google Scholar] [CrossRef]
- Taranto, Y.; Köksal, M.A.; Dal, E. The External Cost of Fossil Fuel Use in Power Generation, Heating and Road Transport in Turkey; Shura Energy Transition Center: Istanbul, Turkey, 2020. [Google Scholar]
- Papagiannis, A.; Roussos, D.; Menegaki, M.; Damigos, D. Externalities from lignite mining-related dust emissions. Energy Policy 2014, 74, 414–424. [Google Scholar] [CrossRef]
- Ministry of Energy. Energy Policy of Poland Until 2040 (EPP2040); Ministry of Energy: Warsaw, Poland, 2021.
Coal Seam | Extraction Period (Years) | Production Volume Until 2019 (Mg) | Remaining to be Mined (Mg) | Depth of Deposit (m) | Completion of Filling in the End Reservoir (Year) |
---|---|---|---|---|---|
Deposits of coal in the districts of Konin and Koło | |||||
Morzysław | 1941–1953 | 1 | - | 15 | - |
Niesłusz | 1953–1961 | 4.1 | - | 27 | - |
Gosławice | 1958–1973 | 38.9 | - | 55 | - |
Pątnów | 1962–2001 | 129.8 | - | 70 | - |
Kazimierz | 1965–2011 | 131 | - | 70 | 2024 |
Jóźwin | 1971–2022 | 146 | 4.9 | 58 | 2055 |
Lubstów | 1982–2009 | 107 | - | 158 | 2026 |
Drzewce | 2005–2023 | 31.2 | 4 | 55 | 2035 |
Tomisławice | 2010–2030 | 15.1 | 26.8 | 67 | 2042 |
Deposits of coal in the district of Turek | |||||
Adamów | 1964–2020 | 109 | 0.8 | 55 | 2036 |
Bogdałów | 1975–1991 | 38 | - | 50 | - |
Władysławów | 1976–2012 | 38 | - | 55 | 2024 |
Koźmin | 1991–2016 | 31.8 | 45 | 2023 |
Group | Average Yield in 1956–1960 Years [t × ha−1] | Average Yield in 1969–1973 Years [t × ha−1] | Dynamic [%] | ||||||
---|---|---|---|---|---|---|---|---|---|
Cereal | Potato | Sugar Beet | Cereal | Potato | Sugar Beet | Cereal | Potato | Suger Beet | |
up to 20 km | 1.56 | 13.7 | 21.9 | 2.11 | 17.4 | 32.1 | 135.0 | 127.4 | 146.5 |
20–40 km | 1.67 | 13.4 | 19.4 | 2.33 | 18.6 | 32.9 | 140.0 | 138.5 | 169.6 |
40–60 km | 1.72 | 13.3 | 21.8 | 2.40 | 18.5 | 34.4 | 139.9 | 139.2 | 158.1 |
60–80 km | 1.70 | 12.8 | 18.0 | 2.37 | 17.7 | 27.4 | 139.9 | 139.0 | 152.5 |
80–100 km | 1.64 | 12.6 | 19.9 | 2.35 | 18.2 | 30.2 | 143.9 | 144.1 | 151.4 |
Group | Average Yield in 1956–1960 Years [t × ha−1] | Average Yield in 1993–1997 Years [t × ha−1] | Dynamic [%] | ||||||
---|---|---|---|---|---|---|---|---|---|
Cereal | Potato | Suger Beet | Cereal | Potato | Suger Beet | Cereal | Potato | Suger Beet | |
group I | 1.63 | 13.5 | 20.13 | 2.56 | 17.0 | 35.7 | 156.4 | 126.5 | 177.5 |
group II | 1.68 | 13.0 | 19.48 | 2.98 | 17.6 | 34.1 | 177.4 | 135.4 | 175.1 |
group III | 1.67 | 12.9 | 20.28 | 3.20 | 18.4 | 37.6 | 191.7 | 142.8 | 185.5 |
Cultivation | Time Period | Test Results | ||||
---|---|---|---|---|---|---|
Cereals | 1956–1960 | Analysis of variance | F = 0.0962 | p = 0.9096 | ||
Levene’s test for homogeneity of variance | F = 0.5161 | p = 0.6212 | ||||
Least significant differences test | {1}↔{2} p = 0.6763 | {1}↔{3} p = 0.7570 | {2}↔{3} p = 0.8262 | |||
1993–1997 | Analysis of variance | F = 0.7265 | p = 0.5217 | |||
Levene’s test for homogeneity of variance | F = 2.1508 | p = 0.1976 | ||||
Least significant differences test | {1}↔{2} p = 0.5278 | {1}↔{3} p = 0.2892 | {2}↔{3} p = 0.6133 | |||
Potatoes | 1956–1960 | Analysis of variance | F = 1.4104 | p = 0.3147 | ||
Levene’s test for homogeneity of variance | F = 0.6015 | p = 0.5780 | ||||
Least significant differences test | {1}↔{2} p = 0.3755 | {1}↔{3} p = 0.1549 | {2}↔{3} p = 0.5005 | |||
1993–1997 | Analysis of variance | F = 0.9901 | p = 0.4250 | |||
Levene’s test for homogeneity of variance | F = 3.6032 | p = 0.0938 | ||||
Least significant differences test | {1}↔{2} p = 0.6946 | {1}↔{3} p = 0.2785 | {2}↔{3} p = 0.3745 | |||
Sugar Beets | 1956–1960 | Analysis of variance | F = 0.5185 | p = 0.6199 | ||
Levene’s test for homogeneity of variance | F = 0.5371 | p = 0.6101 | ||||
Least significant differences test | {1}↔{2} p = 0.6732 | {1}↔{3} p = 0.7997 | {2}↔{3} p = 0.3490 | |||
1993–1997 | Analysis of variance | F = 0.8080 | p = 0.4889 | |||
Levene’s test for homogeneity of variance | F = 3.3950 | p = 0.1032 | ||||
Least significant differences test | {1}↔{2} p = 0.6574 | {1}↔{3} p = 0.6880 | {2}↔{3} p = 0.2554 |
Specification | Variant I | Variant II | Variant III | Average | € × MWh−1 |
---|---|---|---|---|---|
Yield decline caused by the operation of open-cast lignite mines | |||||
Open-pit mining area | 19 | 17 | 17 | 18 | 0.03 |
Group I | 3094 | 2752 | 2815 | 2887 | 4.45 |
Group II | 2863 | 2603 | 2657 | 2708 | 4.18 |
Total | 5976 | 5372 | 5489 | 5612 | 8.66 |
Study | Georgakellos [145] | Sakulniyomporn [42] | Büke, Köne [146] | Dimitrijević [147] | Coester [39] | Máca [32] | Wang [31] | Taranto [148] |
---|---|---|---|---|---|---|---|---|
Country | Greece | Thailand | Turkey | Bosnia and Herzegovina | Germany | Czech, Hungary, Poland | China | Turkey |
Year of analysis | 2003–2004 | 2006–2008 | 2007 | 2008 | 1995–2003 | 2010 | 2015 | 2018 |
Health impacts | No | Yes | Yes | No | Yes | Yes | Yes | No |
External costs | 43.9 | 6.8 | 1.8–35.2 | 2.7–19.2 | 11.1 | 58.1–77.5 | 63.8 | 36.3 |
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |
© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Pepliński, B.; Czubak, W. The Influence of Opencast Lignite Mining Dehydration on Plant Production—A Methodological Study. Energies 2021, 14, 1917. https://doi.org/10.3390/en14071917
Pepliński B, Czubak W. The Influence of Opencast Lignite Mining Dehydration on Plant Production—A Methodological Study. Energies. 2021; 14(7):1917. https://doi.org/10.3390/en14071917
Chicago/Turabian StylePepliński, Benedykt, and Wawrzyniec Czubak. 2021. "The Influence of Opencast Lignite Mining Dehydration on Plant Production—A Methodological Study" Energies 14, no. 7: 1917. https://doi.org/10.3390/en14071917
APA StylePepliński, B., & Czubak, W. (2021). The Influence of Opencast Lignite Mining Dehydration on Plant Production—A Methodological Study. Energies, 14(7), 1917. https://doi.org/10.3390/en14071917