Vegetation indices derived from remote sensing data are very important for the accurate assessment of plant growing conditions, especially in the case of extreme weather events, such as droughts. Ground-based meteorological and agrometeorological drought indices only allow the evaluation of the risks for agricultural lands, while the satellite information makes it possible to identify damaged vegetation in various land types and to assess the magnitude of damage.

Remote sensing of the vegetation condition is based on the fact that healthy plants have more chlorophyll and therefore absorb more visible (VIS) radiation and reflect more near-infrared (NIR) radiation (Myneni et al., 1995). Often vegetation conditions are determined by calculating the normalized difference vegetation index (NDVI). Since 1981 this index has been provided on a global scale using the advanced very-high-resolution radiometer (AVHRR) instruments on-board NOAA satellites.

The long-term data set is a big advantage, but problems may arise in interpreting the index changes. During more than 30 years of measurements, the land use has been changed in many locations and it is difficult to determine the climatic signal in the NDVI changes. The accuracy of the growing conditions evaluation depends on the assessment of environmental and atmospheric conditions as well as peculiarities of the sensor response (Jackson and Huete, 1991). During the period without precipitation, NDVI values can decrease not only due to the deterioration of the plant but also due to the increase in dust in the air and on the surface of the plant, which is usually washed out along with the rain. For this reason, the vegetation index can have lower values than it should (Mirzaei et al., 2011).

It is necessary to emphasize that the vegetation (and hence NDVI values) response to the meteorological conditions in a given year depends on the geographical region and environmental factors such as vegetation type, soil type and land use (Usman et al., 2013). NDVI is a good indicator of vegetation–soil moisture conditions, but seasonality should be taken into account when using this index for drought monitoring (Ji and Peters, 2003). Therefore, in most cases NDVI values are complexly analysed with ground-based meteorological and agrometeorological drought indicators such as the Standardized Precipitation Index (SPI) (Ji and Peters, 2003; Bhuiyan et al., 2006; Quiring and Ganesh, 2010; Gebrehiwot et al., 2011; Gaikwad and Bhosale, 2014; Stagge et al., 2015), Standardized Precipitation Evapotranspiration Index (SPEI) (Stagge et al., 2015), Standardized Water-Level Index (SWI) (Bhuiyan et al., 2006), Palmer drought severity index (PDSI) and Moisture Anomaly Index (z index) (Quiring and Ganesh, 2010). Most commonly, the spatial and temporal variability in the drought is associated with precipitation deficit; thus, the SPI is often used due to its simplicity (Gebrehiwot et al., 2011).

Previous studies have shown that the NDVI and SPI values are correlated and this relation is the strongest in the middle of the active growing season and the weakest at the beginning and at the end (Ji and Peters, 2003). However, a negative NDVI anomaly cannot be related with low SPI values in all cases (Bhuiyan et al., 2006). The strongest relationship between SPI and NDVI was found in the areas with low soil water-holding capacity (Ji and Peters, 2003). Also, the relationship between these two indices differs in various agricultural areas: a positive correlation between SPI and NDVI was determined in the rain-fed areas, while a negative correlation was determined in the irrigated areas (Ozelkan et al., 2016). The SPI indicates moisture conditions and the vegetation reacts to the lack of precipitations with some delay. For this reason, the strongest link was established between the SPI values in spring and NDVI values in summer, which means that spring watering is critically important for the growth of the most plants (Ozelkan et al., 2016).

NDVI is frequently analysed by calculating Vegetation Condition Index (VCI), which compares the current NDVI to the observed values of this index in previous years (Gebrehiwot et al., 2011; Ozelkan et al., 2016) and has a good correlation with the SPI values (Dutta et al., 2015). In different regions of the world, the relationship between the 3- (Gebrehiwot et al., 2011), 6- or 9-month SPI (Quiring and Ganesh, 2010) and VCI values were established. Some studies showed that the impact of the short-term precipitation fluctuations on VCI values is weak (Quiring and Ganesh, 2010).

The eastern coast of the Baltic Sea is in a transitional area from the maritime to the continental climate, characterized by a strong west–east gradient in the continentality of climate (Jaagus et al., 2010, 2014). The region can be divided into three geographical zones: the western coastal areas, the central agricultural zone with fertile soils, and the eastern region, which contain most lakes, swamps and forest. The spatial pattern of seasonal temperature and precipitation in the Baltic countries depends on two main large-scale factors: latitude and the Baltic Sea (Jaagus et al., 2010, 2014). In the eastern Baltic region droughts are not very common (Loyd-Hughes and Saunders, 2002), but they can cause significant losses for economy and wildlife (BACC, 2008). During the second half of the 20th century the dryness of the region remained similar or even decreased (Bordi et al., 2009; Rimkus et al., 2012, 2013), but recent studies show that in the 21st century, at least in some parts of the study area, the water availability is likely to decrease in summer and autumn (Stonevicius et al., 2017). Dry periods are related to the atmospheric circulation patterns and can affect large areas (Rimkus et al., 2014). In temperate and boreal climates the effect of water shortage on vegetation is not as significant as in arid or semiarid areas, but there is strong evidence of water deficit effect on various types of vegetation in the study area (Kulikauskas and Sprainaitiene, 2005; Vitas and Erlickyte, 2008; Ozolincius et al., 2009; Kalbarczyk, 2010; Bijak, 2011).

The NDVI and VCI have not been used for the drought analysis in the Baltic Sea region yet. In this region, especially in the southern part, the agriculture is strongly developed, and development of new methods for evaluation of drought extension and intensity are very important.

The main objectives of this study are to determine the impact of the droughts on the active plant growing conditions in the eastern part of the Baltic region, to identify other factors that may lead to negative NDVI anomalies or low VCI values during the active growing season, and to find links between SPI and VCI values. Areas of arable land and broadleaved and coniferous forests were analysed separately and differences in precipitation deficit impact on vegetation in different land use types were determined.

The analysed area covers the eastern part of the Baltic Sea region and is located between 53–60∘ N and 20–30∘ E (Fig. 1). The NDVI was used to analyse the vegetation condition in the land cells of the study area. CORINE (Coordination of Information on the Environment) land cover data were used to identify the response of vegetation in different land use types to precipitation deficit. CORINE land cover data are available only in part of study area covering Estonia, Latvia, Lithuania and Poland (Fig. 1).

The NDVI data set was obtained from the NOAA STAR-NESDIS system, which generates global and regional vegetation health data. The NDVI is derived from the radiance observed by the AVHRR on-board polar orbiting satellites: the NOAA-7, NOAA-9, NOAA-11, NOAA-14, NOAA-16, NOAA-18 and NOAA-19. NDVI is calculated as the difference between reflectance in NIR and VIS radiation by following Eq. (1): NDVI=NIR-VISNIR+VIS. The NOAA STAR-NESDIS NDVI product has 16 km spatial and 7-day composite temporal resolution and covers a period from 1981 to present. In this study, data from 1982 to 2014 were analysed. The NDVI data set is generated using the maximum-value composite (MVC) method (Holben, 1986). This method reduces the influence on NDVI from clouds, spectral properties, resolution and residual atmospheric effects that all act to reduce NDVI (Scheftic et al., 2014). The NOAA STAR-NESDIS system produces no noise NDVI. The NDVI is filtered in order to eliminate the high-frequency noise. It is also adjusted for a non-uniformity of the land surface due to the climate and ecosystem differences using multi-year NDVI and brightness temperature data. The final NDVI product is provided in the geographic grid with equal latitude and longitude intervals (0.144∘ × 0.144∘) (NOAA-NESDIS, 2013). The data set has several gaps: from week 50 of 1984 to week 8 of 1985, from week 37 of 1994 to week 3 of 1995, and from weeks 2 to 4, from weeks 11 to 24 and week 29 of 2004.

The NDVI values range from -1 to +1. The negative index value can be recorded over the water bodies while values are close to 0 over the land without vegetation. The index value equal to 1 indicates perfect growing conditions (Lillesand and Kiefer, 1994; Belal et al., 2014).

The total number of analysed cells is 2184. In the coastal areas or near the big lakes, 31 cells were unequally recognized as land or sea cells by different satellites, and in some cases, the information was missing. In such cases, data derived from these particular cells were excluded from further analysis. Also, 99 cells near the sea coast and probably partly covered by the sea, had negative NDVI values during the active growing season and thus were excluded from analysis. In total, 6 % of the initial data set was not used in the study.

The NDVI values are not only influenced by the natural variation in and health of vegetation. In long-term data sets, variability related to the satellite orbital drift, sensor degradation and satellite change is also determined (Kogan, 1997). The initial trend observed in this research was mostly related to the satellite change and this trend was removed by applying a systematic correction for each satellite datum separately.

Active plant vegetation in the eastern Baltic region starts when the daily mean air temperature exceeds 10 ∘C. In the majority of years this happens in the first half of May. The end of the active growing season usually occurs in the second half of September. It is necessary to mention that there is a difference in growing season of up to several weeks in the study area (due to the latitude and distance from the sea), as well as quite large year-to-year variation in such dates.

Since the drought makes the greatest impact on the plants during the active growing season, data from weeks 18 to 39 of the year (May–September) were analysed. Not only absolute NDVI values were evaluated but also their deviations from the mean. For this reason, VCI (Kogan, 1995) was calculated. VCI compares the current NDVI with measured historical NDVI values. It is defined as the following Eq. (2): VCI=NDVI-NDVIminNDVImax-NDVImin×100, where NDVI is the measured monthly (weekly) value and NDVImin and NDVImax are the historical minimum and maximum values of the analysed month (week). Lower VCI values indicate a bad vegetation state, while higher values show a good vegetation state. The VCI is expressed as a percentage and varies from 0 to 100. According to Kogan (2002) low values below < 40 can be described as mild drought, < 30 as moderate drought, < 20 as severe drought and < 10 as extreme drought.

According to Jain et al. (2010), the VCI is a better indicator of the moisture deficit than NDVI because it allows the separation of the short-term climate signal from the long-term ecological signal. VCI enables the comparison of simultaneously measured NDVI values not only under the different geographic conditions but also in the different vegetation types.

CORINE land cover data were used to identify the dominant land use type in the NDVI cells (0.144∘ × 0.144∘). CORINE data sets with 100 m resolution were used. The CORINE data sets with reference years 1990 (CLC 1990) and 2012 (CLC 2012) were compared to identify the land use changes during the study period. It was considered that land use in a particular NDVI cell was stable if CORINE land use classes coincided in at least 80 % of the cell area. From the set of cells with stable land use, the cells with different dominant land use types were identified. Diverse land use is common in the analysed region. To reduce the number of CORINE land use classes, the mixed forests and transitional woodland–shrub areas were joined with broadleaved forests and the broadleaved vegetation class was formed. On average, the broadleaved vegetation class consisted of 28 % broadleaved forests, 57 % mixed forests and 15 % transitional woodland–shrub areas. A land use class is considered dominant if it covers at least 50 % of the cell area. Only three types of land use were identified as dominant in more than five cells: arable land (209 cells), broadleaved forest (80) and coniferous forest (25 cells) (Fig. 1). These three land use types were used in this study to differentiate the effect of climatic conditions on vegetation.

Four cases with strong NDVI anomalies were investigated. Winter 1987 was one of the coldest winters during the entire study period in the whole eastern Baltic region. Also, it was the only year when the mean March–April temperature averaged over the entire area was negative (-1 ∘C), and it led to a very late beginning of the vegetation season. In 1990, after one of the warmest winters, a particularly high air temperature in March–April was recorded (5.9 ∘C), and this led to a very early beginning of the vegetation season. Years 1992 and 2002 were analysed because the largest negative precipitation anomalies in May–September were recorded, 37 and 43 % respectively below the long-term average. During these years the agrometeorological droughts were observed in the substantial part of the analysed area (Valiukas, 2015).

In order to assess the impact of precipitation deficit on vegetation condition, the SPI was used in this study. The SPI calculation for any location is based on a monthly rainfall data series, first applying gamma distribution and then transforming it into a normal distribution (McKee et al., 1993; Edwards and McKee, 1997). Positive SPI values indicate precipitation amounts that are greater than average while negative values indicate lower amounts (Table 1).

High-resolution (0.5∘ × 0.5∘ latitude–longitude) monthly precipitation data from CRU TS (Climate Research Unit Time Series) data set (Harris et al., 2013) have been used in this study to calculate SPI values. The initial analysis indicated that the vegetation conditions are not strongly affected by the moisture deficit calculated for timescales above 9 months. The effect of long-term SPI might be weakened by the conditions of the cold season when water supply depends on the precipitation type, snowmelt and soil condition. The 1-, 3-, 6- and 9-month SPI values (SPI1, SPI3, SPI6) were used in this study to investigate the effect of short- and medium-term precipitation deficit.

NDVI values in the analysed area are determined by a number of climatic factors. On average, the active growing season in the Baltic states lasts from the end of April until the beginning of October. The spatial pattern of the seasonal NDVI variation is closely related to the distance from the sea, because the Baltic Sea is a major factor, determining the temperature and precipitation regime in the analysed area (Jaagus et al., 2010, 2014). The differences in NDVI values in spring and autumn in the west–east direction are larger than in the south–north direction. The south–north NDVI gradient would be more noticeable if the determining factors were the length of day and insolation.

Many studies show that beginning of the active growing season is determined by the spring temperature prior to the event (Jeong et al., 2011; Shen et al., 2014) in the temperate and high latitudes of the Northern Hemisphere. In spring the soil is saturated with meltwater and excess moisture can worsen vegetation condition. Agricultural activity on arable land usually starts when soil becomes rather dry. For this reason, in the case of an abnormally wet spring, the negative NDVI anomalies can be recorded in May, which may be associated with the crop area and not with the crop conditions (Zhang et al., 2014). Conversely, in June moisture deficit in the arable soil may lead to vegetation deterioration, while this effect in conifer and broadleaved forests is almost invisible. Such differences can be explained by the fact that on the arable land mostly annual crops with relatively shallow roots are grown. Thus, a lack of moisture may occur even during the short dry period. The Northern Hemisphere boreal forests usually grow in areas of excessive moisture and tree roots are deeper, so they react much slower to the precipitation deficit (such deficit may even lead to higher VCI values). Also, due to the high initial soil moisture, the drought impact on forested areas can be minimal (Gao et al., 2016). Only during the prolonged extreme droughts (e.g. 1992), do the VCI values in the forests of the study area decrease significantly.

Many studies indicate that VCI reacts with a delay in the change of moisture conditions and this reaction is controlled by the previously accumulated soil water storage (Quiring and Ganesh, 2010). Therefore, the strongest connection between SPI and VCI was determined in the areas with low soil water-holding capacity (Ji and Peters, 2003). Other studies also showed that forests respond to the drought on long-term scales, while arable land does so on short-term scales (Li and Zhou, 2015).

In the second half of the active growing season, the positive correlation between SPI3, SPI6 and VCI has been determined in a large part of the territory. A positive and in many places statistically significant correlation was found in the first half of the season if SPI with a 1-month lead was used. However, the moisture deficit has a significant impact on vegetation condition only in the second half of the active growing season in the analysed part of the Baltic Sea region. Meanwhile, in the drier areas, SPI6 and SPI9 have a strong positive correlation with VCI throughout the year (Quiring and Ganesh, 2010).

It should be mentioned that due to the early start of the active growing season the peak of vegetation is usually reached earlier, and an early vegetation start does not always lead to an increase in aboveground production (Livensperger et al., 2016). Therefore, the low VCI values in August and September are not always related to the precipitation deficit. However, the time when the NDVI values fall close to the typical winter values (end of October to beginning of November) in arable soils and broadleaved forests is very similar during the years with different hydrothermal regimes (Fig. 3). This is related to the routine agricultural practices on the arable lands (when the land is plowed in autumn) while the fall of tree leaves is associated with the occurrence of the first intense frosts, which are usually in October. Meanwhile, in the coniferous forests, the differences in NDVI values that form in the summer months remain until the end of the calendar year (Fig. 3).

In the arid and semiarid areas, the spatial and temporal patterns of vegetation are primarily related to precipitation (Wang et al., 2001). The analysed region has a surplus precipitation during most of the years; thus, the air temperature anomalies might be the limiting factor for vegetation condition, especially in the transitional seasons (spring and autumn). The comparison of 1987 (cold spring) and 1990 (warm spring) also indicates that the interpretation of VCI values as indicators of drought severity (Kogan, 2002) may not be universal in the eastern part of the Baltic Sea region. Dabrowska-Zielinska et al. (2002) also found that in nearby Poland the VCI plays a minor role in defining vegetation condition and crop yield.

The early spring air temperature, snowmelt water storage in the soil and precipitation have the largest influence on NDVI values in the first half of the active growing season. Negative correlation between SPI1 and VCI shows that the short-term precipitation deficit leads to a better vegetation condition. With a longer SPI scale, the correlation coefficients become positive, which shows that longer periods of moisture deficit reduce vegetation condition. The correlation is usually positive between the VCI and the previous month's SPI because vegetation responds to the external forcing with a delay.

The precipitation deficit in the first part of the vegetation season has a significant impact only on the vegetation on arable land, while vegetation in forests is less sensitive to the moisture deficit in the first half of the active growing season. The positive correlation between 3- and 6-month SPI and VCI was observed on the arable land and both types of forests in the second half of the vegetation season.

The precipitation deficit is only one of the vegetation condition drivers in the eastern Baltic region and NDVI or VCI cannot be universally used to identify droughts. However, NDVI and VCI may be applied to better assess the effect of droughts on vegetation and the crop damage.