*3.3. Statistical Analysis*

Based on the descriptive statistics of the simulated time-series of the main hydrometeorological factors at Spercheios river basin, the mean annual precipitation of the entire catchment for the period 1960/61–2004/05 was 542.5 mm, and the mean annual air temperature was 13.2 ◦C for the same period. Mean annual river discharge to Maliakos Gulf ranged from 5.1 m<sup>3</sup>/s in LC1960, through 5.7 m<sup>3</sup>/s in LC1990, to 5.4 m<sup>3</sup>/s in LC2018, while annual actual basin-averaged evapotranspiration ranged from 406.1 mm in LC1960, through 384.7 mm in LC1990, to 395.0 mm in LC2018 (Table 7; Figure 7). Hence, in comparison with LC1960, LC1990 and LC2018 case studies estimated 11.8% and 5.9% higher mean annual river discharge to Maliakos Gulf, respectively. On the other hand, they estimated 5.3% and 2.5% lower basin-averaged annual actual evapotranspiration. These results can be attributed to water balance which force discharge and actual evapotranspiration to be "communicating vessels", given the same meteorological forcing in the three land cover case studies examined. It is interesting to note that the results showcased the role of richly-vegetated area variabilities on the hydrological characteristics of the catchment. Deforestation as well as intertemporal increase of artificial surfaces have negative effects on evapotranspiration while increasing discharge. For example, the reduced forested area of LC1990 in comparison with both LC1960 and LC2018, resulted in minimum basin-averaged annual actual evapotranspiration and maximum mean annual river discharge.



**Figure 7.** Basin-averaged annual actual evapotranspiration (mm) (**a**) and river discharge (m<sup>3</sup>/s) to Maliakos Gulf (**b**) reconstructions. LC1960, LC1990 and LC2018 are represented by blue, orange and gray lines.

As far as mean annual actual evapotranspiration is concerned, the three land cover case studies present spatial differences. LC1960 is characterized by more inhomogeneous patterns than in LC1990 and LC2018 (Figure 8). It resulted in values exceeding 520 mm/yr and lower than 280 mm/yr at many areas. A possible explanation for this difference is the increased scattering of areas covered by forests, agricultural land and pastures in LC1960 which have effects of different magnitude on evaporation and transpiration (see Figure 2). Forests and agricultural land increased evapotranspiration in comparison with pastures and artificial surfaces [67]. In contrast to LC1960, LC1990 and LC2018 case studies are characterized by wider continuous areas covered by the same land cover. Hence, the pattern of mean annual actual evapotranspiration is smoother in LC1990 and LC2018 than in LC1960.

Comparing the results of the three land cover case studies, annual actual evapotranspiration is almost the same at areas covered by artificial surfaces over time, for example at the city of Lamia (not shown). On the other hand, the transition from pastures to agricultural land or forest increased mean annual actual evapotranspiration, while the inverse transition had the opposite effects. In order to quantitively estimate the impact of each land cover transition, case studies are compared in pairs e.g. LC1960 vs LC 1990, regarding spatially averaged actual evapotranspiration only at areas characterized by a specific transition. As far as deforestation is concerned, the transition from LC1960 to pastures in LC1990 decreased annual actual evapotranspiration with a mean rate of about 33 mm/yr (Figure 9a,b). It is important to note that the reduction is also evidenced for the entire simulation period which means that land cover change effects can locally outflank the impact of climatic variability [11,12]. However, the transition from pastures in LC1960 to agricultural land in LC2018 increased annual actual evapotranspiration (Figure 9c,d) with a mean rate of about 24 mm/yr. On the other hand, the transition from forest in LC1990 to agricultural land in LC2018 caused a reduction in annual actual evapotranspiration with a mean rate of about 26 mm/yr. Although, differences exist at areas with the same land cover in all three cases examined, these are quite a bit smaller than those which appeared at areas where land cover changed. These smaller differences may be attributed to the horizontal propagation of land cover effects. The local differences in land cover introduce complex forcing in parameters such as run-off water, infiltration, evaporation, and transpiration which can sharply affect in a non-linear way the spatial distribution of water balance, yielding local differences in evapotranspiration.

**Figure 8.** Simulated mean annual actual evapotranspiration (mm/yr) in (**a**) 1960, (**b**) 1990 and (**c**) 2018.

**Figure 9.** Timeseries, regarding areas characterized by transition from forest in 1960 to pastures in 1990, of (**a**) annual actual evapotranspiration (mm/yr) simulated by LC1990 (orange line) and LC1960 (blue line) as well as (**b**) their differences (red for positive and blue for negative). The same for (**<sup>c</sup>**,**d**) as well as (**<sup>e</sup>**,**f**) regarding transition from pastures in 1960 to agricultural land in 2018 and transition from forest in 1990 to agricultural land in 2018, respectively.

The statistical tests applied on the time-series of the main hydrometeorological factors (precipitation, air temperature, actual evapotranspiration, and river discharge) for the trend analysis and change point detection, resulted in the following findings. Although the *u*(*t*) and *u*(*t*) curves of precipitation intersect only at one point (1981/82), the following trends were identified based on the general form of the *u*(*t*) curve. C, concerning annual precipitation: (1) three increasing periods were identified (1960/61–1973/74, 1976/77–1981/82 and 1992/93–2002/03), and (2) four decreasing periods (1973/74–1976/77, 1981/82–1992/93 and 2002/03–2004/05 respectively). It should be noted that all trends identified, either with SMK test, either with CUSUM test, were not significant at the 0.05 confidence level (Figures 10a and 11a).

**Figure 10.** SMK and CUSUM tests for (**a**) precipitation and (**b**) air temperature.

**Figure 11.** Results of Mann-Kendall test (No border in columns indicates *S* test and black border *Z* test; (\*) and (+) symbols indicate if trend is significant at α = 0.05 level and at α = 0.1 level respectively) and Sen's slope for (**a**) precipitation and (**b**) air temperature.

Regarding annual air temperature, the following trends were identified, based on the form of the *u*(*t*) curve: (a) two decreasing periods were identified (1960/61–1964/65 and 1969/70–1983/84), followed (b) by two increasing periods (1964/65–1969/70 and 1983/94–2004/05 respectively). Of the abovementioned trend periods, based on the SMK test, the periods 1960/61–1964/65 and 1983/94–2004/05 were significant at the 0.05 confidence level, while with the CUSUM test, the trends identified were not statistically significant at the 0.05 confidence level (Figures 10b and 11b).

The trend analysis and change point detection tests applied in actual evapotranspiration time-series of all land cover case studies examined, led to the identification of: (1) three increasing periods (1960/61–1967/68, 1977/78–1982/83 and 1989/90–1994/95), followed by (2) three decreasing periods (1967/68–1977/78, 1982/83–1989/90 and 1994/95–2004/05 respectively; Figure 12a–c). Nevertheless, the trend magnitude of each period was different for each land cover case study examined. More specifically, the trend magnitude in all trend periods identified was higher in the case of 1960, followed by the trend magnitude calculated for the periods 1990 and 2018, with the exception of the period 1977/78–1982/83 that the trend magnitude was greater in LC2018, followed by LC1990 and LC1960, and the period 1982/83–1989/90, where trend magnitude was practically identical in all land cover cases examined. It should be noted that all trends identified were not significant at the 0.05 confidence level

with the SMK test, while with the CUSUM test, the period 1982/83–1989/90 was statistically significant at the 0.05 confidence level in all land cover case studies examined (Figures 13a and 14a).

**Figure 12.** SMK and CUSUM tests for actual evapotranspiration for (**a**) LC1960, (**b**) LC1990, and (**c**) LC2018, and river outflow to Maliakos Gulf for (**d**) LC1960, (**e**) LC1990, and (**f**) LC2018.

**Figure 13.** Results of Mann-Kendall test for actual evapotranspiration (**a**) and river discharge (**b**) for LC1960, LC1990, and LC2018. No border in columns indicates *S* test and black border *Z* test. (\*) and (+) symbols indicate if the trend is significant at the α = 0.05 level and α = 0.1 level, respectively.

**Figure 14.** Sen's slope for actual evapotranspiration (**a**) and river discharge (**b**) for LC1960, LC1990, and LC2018.

Finally, concerning Spercheios annual river discharge and outflow to Maliakos Gulf, three decreasing periods were identified for all land cover case studies examined (1960/61–1976/77, 1981/82–1992/93 and 1995/96–2004/05), followed by two increasing periods (1976/77–1981/82 and 1992/93–1995/96; Figure 12d–f). In all trend periods identified, the trend magnitude was smaller in LC1960, followed by LC1990 and LC2018, except in the case of the period 1981/82–1992/93 that the trend magnitude for LC2018 was smaller than in the case of LC1990. These results revealed a significant impact of land cover on the formation of extreme hydrometeorological events. This finding indicates that the decrease of a richly-vegetated area, for example due to deforestation between LC1960 and 1990, increased annual river discharge while intensifying the vulnerability to extreme climatic variabilities which often provokes either droughts or floods. Of the abovementioned trend periods, based on the SMK test, 1981/82–1992/93 was significant at the 0.05 confidence level, while with the CUSUM test, the trends identified were not statistically significant at the 0.05 confidence level (Figures 13b and 14b).
