**4. Discussion**

Anthropogenic land cover changes and interventions on catchment's characteristics can be leading factors a ffecting the hydrological cycle components and, in some cases, the impacts can be of the same order of magnitude, or even larger than those attributed to climatic variabilities [11,12]. In order to investigate the e ffects of land cover changes on the main hydrometeorological factors of a regional river basin in Central Greece, a physically-based hydrological model (MIKE SHE) and gridded observational meteorological data (Copernicus Climate Change Service E-OBS) were employed, and three land cover case studies were adopted.

Before the simulations, the reliability of the E-OBS dataset including precipitation and daily temperature (average, minimum and maximum) was evaluated by comparing against time-series of in-situ observations from meteorological stations at the basin. Based on the results, E-OBS dataset systematically underestimated precipitation in Spercheios river basin for the entire period of evaluation. This may be attributed to issues arising in the comparison of in-situ measurements with area-averaged estimates [68], such as the identification of the most representative grid-point for each meteorological station, the insu fficient density of the weather stations network in Spercheios river basin or possible uncertainties concerning the accuracy of observational measurements [69]. Moreover, the coarse horizontal resolution of E-OBS prevented to accurately describe the influence of topography on precipitation and to adequately resolve the atmospheric mesoscale processes; 10–15 km grid spacing of meteorological variables generally improves the realism of the results but does not necessarily significantly improve the objectively scored accuracy of the forecasts [70]. Additionally, the coarse network of Greek meteorological stations used in the E-OBS development that are not evenly distributed and do not cover higher altitude su fficiently, eventually does not allow the accurate representation of area-averaged estimates. More specifically, the spatially-averaged annual precipitation calculated at the present study for the period 1960/61–2004/05 was 542.5 mm, which is close to the mean annual precipitation of Lamia meteorological station (585.5 mm for the period 1970–2000 [71]). In other studies, the spatially-averaged annual precipitation of Spercheios river basin was estimated to be 836 mm for the period 2008/09–2010/11 (precipitation estimated based on Thiessen polygons method [72]) and 1,077 mm for the wet hydrological years 2013/14–2014/15 [73] (simulated precipitation provided by Poseidon Monitoring, Forecasting and Information System [74]) [75], while for the period 1949/50–1989/90, the spatially-averaged annual precipitation for Spercheios river basin was estimated to be 904.6 mm [76]. Nevertheless, the main scope of the present study was the trend analysis of the time-series of the main hydrometeorological factors and, therefore, these discrepancies were considered to be acceptable, since no other meteorological data except from the low-altitude Lamia meteorological station (Hellenic National Meteorological Service, WMO 16675) were available for the entire simulation period (1960/61–2004/05).

As far as the results of hydrological simulations are concerned, average annual actual evapotranspiration and river discharge were the main parameters of the hydrological cycle which were analyzed in this study. First, the average annual actual evapotranspiration at Spercheios river basin was −5.3% and −2.5% decrease in LC1990 and LC2018 respectively, in comparison to LC1960. These variations can be attributed to the presence of the larger areas covered by vegetation (forest and pastures) in LC1960 (70% in comparison to 66% in LC1990 and LC2018), and especially to the larger extent of areas classified as pastures that also include shrubs, transitional woodland—shrub areas or areas with dense vegetation (38% in LC1960), that led to increased actual evapotranspiration. The higher value of actual evapotranspiration in LC2018 in comparison to LC1990 can be attributed to the increased forested land (34% in LC2018 in comparison to 30% in LC1990). The simulations also presented high spatial di fferences in average annual actual evapotranspiration. Land cover in 1960 was characterized by a more inhomogeneous pattern than in 1990 and in 2018 due to the increased distribution patterns of areas covered by forests, agricultural land and pastures in 1960 which have di fferent e ffects on evaporation and transpiration. Moreover, mean annual actual evapotranspiration was almost the same at areas covered by artificial surfaces over time, for example Lamia city, but presents variations where land cover changed. The transition from pastures to agricultural land or forest increased evapotranspiration, while the inverse transition had the opposite e ffects for the entire simulation period which means that land cover e ffects can locally outflank the impact of climatic variability.

Second, average annual river discharge to Maliakos Gulf was +11.8% and +5.9% increased in LC1990 and LC2018 respectively, in comparison to LC1960. This can partially be attributed to the contribution of the baseflow to river, that ranged from 16.7% in LC1960, through 19.2% in LC1990, to 18.1% in LC2018, following the same pattern. Additionally, the high forested land covering the area of Spercheios river watershed in the case of LC1960 (31%) combined with the lowest irrigation demands during the same period and led to the smallest river discharge. Although in 1990 the forested land slightly decreased (30%), the irrigation demand was almost double, leading to higher exploitation of underground waters, o ffering residual water in the rivers' flow and leading eventually to the highest river discharge. Finally, the increase of forested areas in 2018 (34%) and the additional high irrigation demand in 2018 led to the small decrease of river discharge.

Regarding trend analysis, the e ffect of land cover change on the trend magnitude was evident. Concerning precipitation and river discharge, the trend change points identified were almost identical. Additionally, the trend change points of actual evapotranspiration identified coincide with those of precipitation, verifying the fact that precipitation is a major factor a ffecting actual evapotranspiration in dry areas, in contrast to wet areas that evapotranspiration is energy-limited (radiation and air temperature) (for example [77–79]). On the contrary, the trend change points of actual evapotranspiration and air temperature were not the same, indicating that actual evapotranspiration is affected in a more complicated way and also by other factors except air temperature as expected, such as land cover and water availability. This was also evident during the trend magnitude analysis of each trend period, where the e ffect of land cover was noticeable. More specifically, in the case of LC1960, where mean annual actual evapotranspiration was the highest in comparison to the other land cover cases examined, and forested land and pastures (that also include natural grasslands, sclerophyllous vegetation, transitional woodland-shrub, moors and heathland and sparsely vegetated areas) consisted of 70% of the total watershed area, the trend magnitude of each trend period examined was higher. Additionally, highly vegetated watersheds showed smaller tolerance to changes of hydrometeorological factors regarding actual evapotranspiration. On the contrary, the small trend magnitude of river discharge in LC1960 in comparison to LC1990 and LC2018 indicated that in the case of a highly vegetated river basin, the response of the system to changes of hydrometeorological factors regarding river discharge was milder. It is an important finding because land cover of LC1960 could play a relaxing role on the consequences of extreme weather phenomena, either droughts or floods, which will possibly increase in the future.

It should be noted that some uncertainties arise due to the fact that during the present study precipitation and air temperature were considered to be una ffected by land coverage. This is a weakness of the present methodological approach since the current version of the hydrological model MIKE SHE does not provide the option of a two-way dynamically coupled atmospheric-hydrological modeling. The use of an uncoupled system can lead to overprediction of the change in evapotranspiration caused by land cover use changes in comparison to the use of a coupled model results [80].
