Capturing the Impact of 2018 European Drought and Heat Using OCO-2 Solar-Induced Fluorescence

The European heatwave of 2018 led to record-breaking temperatures and extremely dry conditions in many parts of the continent resulting in widespread decrease in agricultural yield, early tree-leaf senescence, and increase in forest fires in Northern Europe. Our study aims to capture the impact of the 2018 European heatwave on terrestrial ecosystem through the lens of a highresolution solar-induced fluorescence (SIF) data acquired from the Orbiting Carbon Observatory (OCO-2) satellite. SIF is proposed to be a direct proxy for gross primary productivity (GPP) and thus can be used to draw inferences about changes in photosynthetic activity in vegetation due to extreme events. We explore spatial and temporal SIF variation and anomaly during spring and summer months across different vegetation types (agriculture, broadleaved forest, coniferous forest, and mixed forest) during the European heatwave of 2018 and compare it to non-drought conditions (most of Southern Europe). About one-third of Europe’s land area experienced a consecutive spring and summer drought in 2018. Comparing 2018 to mean (2015-2017) conditions, we found a change in intra-spring season SIF dynamics for all vegetation types, with lower SIF during the start of spring followed by an increase in fluorescence from mid-April. Summer, however, showed a significant decrease in SIF. Our results show that particularly agricultural areas were severely affected by the hotter drought of 2018. Furthermore, the intense heat wave in Central Europe showed about 31% decrease in SIF values during July and August as compared to the mean over three previous years. Furthermore, our MODIS and OCO-2 comparative results indicate that especially for forests, OCO2 SIF has a quicker response and possible higher sensitivity to drought in comparison to MODIS’s fPAR and NDVI when considering shorter reference periods, which highlights the added value of remotely sensed solar-induced fluorescence for studying the impact of drought on vegetation.


Introduction
Anthropogenic global warming is estimated to currently have reached about 1°C above preindustrial levels and is estimated to further increase to 1.5°C between 2030 and 2052 under current rates of greenhouse gas (GHG) emissions [1]. This increase in global temperature is expected to lead to an increase in intensity and frequency of extreme events like heatwaves, drought and associated wildfires, [2][3][4]. In 2018, Europe faced such an extreme event characterized by record-breaking temperatures and long-lasting dryness, referred to as 'hotter drought' [5]. According to the European Drought Observatory of the European Commission, most of Northern and Central Europe was affected by such hotter drought in 2018. On the contrary, much of the Southern Mediterranean Europe experienced relatively cool and moist conditions [6,7]. For Central Europe, various news and media articles reported severe decrease in crop yields, increase in elderly mortality, and forest fires along with many other adverse impacts. According to a quick attribution study by the World Weather Attribution, [8] heat in Northern Europe, anthropogenic climate change has more than doubled the probability of occurrence of such a heatwave in the region.
Along with severe economic implications and adverse health impact on the public, such recordbreaking hotter drought events severely affect ecosystems. A direct comparison of vegetation response to drought in 2018 and 2003 European heatwaves using MODIS's vegetation indices concluded that negative impacts in 2018 were even stronger than in 2003, with strong decline in agricultural land and forests, especially in Central Europe [6]. Such drought events strongly affect the carbon cycle, by leading to plant mortality and reductions in ecosystem carbon dioxide (CO2) uptake, with the potential to convert ecosystems from carbon sinks into carbon sources [9,10].
Gross primary productivity (GPP) is a measure of ecosystem CO2 assimilation. The estimation and modeling of regional or global GPP involves various uncertainties arising from model parameterization [11,12], thus demanding a more direct and reliable approach to estimate GPP. With recent advances in satellite-based measurements of solar-induced (chlorophyll) fluorescence (SIF), it is now possible to use SIF as a direct proxy for photosynthesis or GPP worldwide [13][14][15][16][17]. SIF is an electromagnetic signal emitted in the wavelength range of 600 to 800 nm by chlorophyll molecules [18]. Since SIF is directly associated with photosynthesis, it is more physiologically-based compared to the traditionally used vegetation indices (VIs) such as the normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) [19,20]. Over the last decade, several studies have demonstrated a strong linear relationship between satellite-based SIF and GPP [13,15,[21][22][23][24][25][26][27][28], indicating that SIF might provide a useful approximation of GPP. The SIF retrievals obtained from the Japanese Greenhouse Gases Observing Satellite (GOSAT) and the European Global Ozone Monitoring Experiment-2 (GOME-2) were most widely used in the last decade (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017); [24,[28][29][30][31]). However, due to coarse spatial resolution of the SIF footprint (10 km diameter for GOSAT and 40 × 40 km 2 , for GOME-2), these satellites presently impose serious limitations to detect eco-physiological changes across different vegetation types. The recent NASA's Orbiting Carbon Observatory-2 (OCO-2) satellite (launched September 2014) provides SIF data at a much higher spatial resolution of 1.3 × 2.25 km 2 with increased SIF acquisition frequency, thereby allowing to resolve SIF responses of various vegetation types under different climatic conditions [15,32]. SIF acquired from GOSAT and GOME-2 have been previously used to study the effect of drought on vegetation. Using GOSAT's SIF [33] showed that midday fluorescence explained the water stress over the Amazonian forests during a dry season in 2010. Yoshida et al [34] used GOME-2's SIF to study the impact of the 2010 Russian drought and highlighted the drought-related SIF decrease for croplands and grasslands. Sun et al [35] studied the drought onset mechanisms and its impact on agriculture using GOME-2 SIF during the droughts of 2011 in Texas and 2012 in the central Great Plains, United States. They concluded that SIF is sensitive to both structural and physiological variation in vegetation during drought and considered it an appropriate tool for drought monitoring. Koren et al [36] developed a GOME-2 based SIF product to show that spatial SIF response was in good agreement with meteorological (temperature, soil moisture, evapotranspiration and terrestrial water storage) anomalies during the 2015/2016 El Niño Amazon drought. Recently Zhang et al [37] constructed a global gridded SIF dataset (CSIF) using a neural network trained by OCO-2 and MODIS data, which they also employed for drought monitoring. Their preliminary result showed that CSIF well-captured the spatial extent of droughts during the 2015 European heatwave.
Here, we present the impact of the 2018 European hotter drought on terrestrial ecosystem as perceived through OCO-2 SIF data. The novelty of this study is highlighted through the detection of spatial and temporal patterns of OCO-2 SIF responses to drought and heat across different vegetation types. We captured and analyzed the spring-summer seasonal variation and anomaly in chlorophyll fluorescence across different vegetation types and tracked it with VPD and temperature anomalies. Furthermore, we compare SIF responses of different vegetation types under drought and non-drought conditions in Europe to explore the ability of OCO-2 SIF for drought impact characterization. Furthermore, we also discuss and compare our results with recent MODIS-based NDVI drought impact study. Finally, we discuss the drought-induced changes in satellite-based SIF measurements and perspectives of OCO remote sensing satellite products for future studies.

Study area
The study area is restricted to the European Union. About 1/3 rd of Europe (1.8 M km², drought area, Figure 1(a-c)) was under consecutive spring (March-May) and summer (June-August) meteorological drought (drought area) according to the widely accepted standardized precipitation evapotranspiration index (SPEI) [38] (Figure 1 (a-c)). Also, from mid-July to mid-August, Central Europe was affected by heatwaves with temperatures more than 5°C higher than average for more than a week (NOAA Global Climate Report -July, August 2018; [39,40]). These drought and heat conditions resulted in a severe decline in agricultural yields and early leaf senescence and forest fires across Central and Northern Europe. At the same time, about 1/3rd of Europe (2 M km 2 , nondrought area, Figure 1 (a-c)) experienced non-drought (i.e., normal or wet) conditions during spring and summer, mostly across the Mediterranean (Figure 1(a-c)). Although during summer, Spain and Portugal did experience record-breaking temperatures, above normal precipitation prevented drought conditions to prevail. Thus, Europe in 2018 provides us the opportunity for a comparative study between productivity under consecutive spring-summer drought and under non-drought conditions.  Light energy absorbed by the leaf chlorophyll molecules has three different pathways: photochemistry, non-photochemical quenching (NPQ, i.e., heat dissipation), and a small fraction reemitted as SIF [18].Theoretically, SIF can be expressed as shown in Equation 1 [22].
where APAR is the absorbed photosynthetically active radiation in Watts/m 2 , which is the product of photosynthetically active radiation (PAR) and the fraction of photosynthetically active radiation absorbed (fPAR) by the vegetation/canopy. SIFyield is the emitted SIF per photon absorbed [17] and is also defined as the effective fluorescence yield of the canopy. SIFyield is the product of fluorescence yield in the wavelength band of measurement (ΦF) and an escape probability term (ΩC). SIFyield is expected to vary with the plant's photosynthetic light use efficiency [41,42] and canopy structural characteristics like leaf area index (LAI). However, with a satellite-based SIF measurement it is challenging to separate the canopy physiological (ΦF) and structural components (ΩC) from SIFyield [35,43] and it is still an active area of research.
For this study, the SIF product from NASA's OCO-2 satellite was used since other satellite-based SIF retrievals were of coarser spatial resolution (GOSAT and GOME-2). OCO-2 SIF is made available by the CO2 Data Portal managed by the Jet Propulsion Laboratory of the California Institute of Technology, United States (https://co2.jpl.nasa.gov/). The OCO-2 spectrometer measures spectra in the O2-A band (757-775 nm, full width at half maximum = 0.042 nm), with far-red SIF, retrieved at 757 and 771 nm based on the infilling of Fraunhofer lines [32,44]. The SIF retrieved at 771 nm is typically about 1.5 times smaller than obtained at 757 nm [44], therefore we utilized SIF retrieved at 757 nm in our study. The major advantage of OCO-2 SIF include an about 100-fold increase in data acquisition frequency over GOSAT and finer spatial resolution (1.3×2.25 km 2 ). This enables OCO-2 to acquire more than 10 5 clear-sky soundings on land per day, thus providing the opportunity to perform in-depth SIF based analysis like regional eco-physiological change detection [45,46]. However, OCO-2 SIF soundings do not have full spatial coverage ( Figure 2). In this study, the SIF retrievals from the Level-2 OCO-2 Lite product, data version SIF v8 (version 8) were used from their first availability (6 th September 2014) to 1 st October 2018 for the whole European continent. The OCO-2 Lite product is bias-corrected and contains only good quality SIF sounding [47]. The OCO-2 takes measurements in three different observation modes, namely nadir, glint viewing, and a special target mode with different viewing zenith angle (VZA), typically alternatively with a repeat cycle of 16 days. Although SIF values are affected by VZA [48], grouping all the SIF sounding over a large area results in no significant SIF difference in the mean SIF between nadir and the combined modes [13,16] (please see Appendix A, table A1). In this study, we used SIF sounding from only nadir and glint mode, which each mode contributing about half of the total number of soundings. The percentage of nadir and glint mode data was consistent throughout the study period (Table S2). Figure S1 illustrates the trajectory and spatial distribution of the SIF soundings for Europe in 2018. The OCO-2 is a sun-synchronous satellite with a local overpass time at 1:30 PM compared to 9:30 AM for GOME-2 and 1:00 PM for GOSAT, thus compared to GOME-2, it better captures the sensitivity of fluorescence yield to water stress which is higher in the afternoon when plant water stress tends to peak [49]. Two important environmental factors that regulate plant photosynthesis are temperature and vapor pressure deficit (VPD). The SIF Lite data files also included VPD and temperature data from ECMWF for each SIF measurement.

MODIS data
Theoretically, SIF is directly related to fPAR (Equation 1). Thus, for improving the interpretation of SIF variation and anomalies, we used the MODIS Aqua fPAR product (MYD15A2H) version 6 [50,51], which is an 8-day composite data set with a spatial resolution of 500 meters covering the same period as OCO-2 SIF dataset. The fPAR calculation is based on 3D radiative transfer equation (Look-Up-Table based approach, [52]), which takes 'best' quality spectral information of red and near-infrared band within the 8-day period as an input, and a back-up algorithm uses empirical relationship based on NDVI. In addition, we also used the standard MODIS 16-days composite NDVI product [53] with a spatial resolution of 250m, from Terra (MOD13Q1) and Aqua (MYD13Q1) for comparison with SIF. These composite products are constructed using the 'best' quality pixel (low clouds, low view angle and highest NDVI) over the 16 days period of Terra and Aqua measurements [53]. The MODIS products were obtained from the NASA Land Processes Distributed Active Archive Center (LP DAAC, https://lpdaac.usgs.gov/products). Moreover, only good quality MODIS pixels (VI Quality variable = 0, indicating good quality as per [53]) for fPAR and NDVI were chosen for subsequent analysis.

Corine Land Cover data
To study the SIF variation across different vegetation types (agriculture, broadleaved forest, coniferous forest, and mixed forest), we used the most recent Corine Land Cover (CLC) 2018, Version 20b2 provided by the European Environmental Agency (EEA) ( Figure S2). The CLC 2018 uses Sentinel-2 and Landsat-8 imageries for its construction, thereby achieving a minimum mapping width (resolution) of 100*100 m (CLC2018 Technical Guidelines, 2017; [54]). Agricultural land comprising arable lands, pastures, and heterogeneous agricultural area covered most of the parts of drought area (53%) and non-drought area (57%), followed by coniferous forest of about 37% in the drought area (Table S1).

Data analysis
The gridded SPEI data for spring (March, April, and May, MAM) and summer (June, July, and August, JJA) was obtained from the SPEI Global Drought Monitor (http://spei.csic.es/index.html).
The drought areas were defined as the areas which showed SPEI less than or equal to -1 (meteorological drought conditions) for both spring and summer season [55,56], whereas the nondrought areas were the ones with SPEI > -1 (near normal and wet conditions, or non-drought conditions) for both spring and summer.
Standardized meteorological based drought indices can sometimes misrepresent actual climatic water balance (CWB) [57]. The SPEI<-1 values for the study area in 2018 completely indicated water deficit conditions with negative CWB values during the summer season ( Figure S3). Figure 1c shows the drought and non-drought area used in this study. Thereafter, OCO-2 SIF sounding data for the drought and non-drought area were spatially joined to the CLC based on geographic intersection (intersection of SIF sounding and the CLC polygons). To study the variation of SIF across the drought and non-drought areas in spring and summer for Europe, we spatially aggregated all the daily SIF soundings for each of the above described vegetation types separately for drought and non-drought areas over the spring and summer season.
SIF changes because of two reasons; firstly due to natural seasonal vegetation change during the growing season (e.g. change in chlorophyll content during the seasonal cycle) and secondly due to stress (e.g., heat and drought stress). To study the inter-and intra-seasonal SIF variation and anomalies during the drought in spring and summer, we compared SIF, temperature and VPD of 2018 (SIF_2018, Temp_2018, VPD_2018) with their average (mean of 2015-2017; SIF_Mean, Temp_Mean, VPD_Mean) on a 5-day running average [9]. These anomalies of the average SIF at a given date would eliminate the SIF changes due to seasonality. It is important to note that Central Europe suffered from a heat wave during summer of 2015, with record high temperatures and precipitation deficit [58,59], which may lower our detected impact of 2018. For each OCO-2 footprint, co-located MODIS's NDVI and FPAR was interpolated in time and space so that the same sample is used for each data set. Finally, the OCO-2 SIF was spatially (Drought area) and temporally (8-day) aggregated to compare with NDVI and fPAR.
We performed a Wilcoxon-rank sum test [60] to test for a significant location-shift (i.e. the nonparametric difference of means) between the SIF_2018 and SIF_Mean. SIF, temperature and VPD anomalies (SIF-diff, Temp-diff and VPD-diff) were defined as differences between 2018 and average (for e.g. SIF-diff = SIF_2018 -SIF_Mean). With continued drought condition during the summer, Central Europe also experienced an intense heatwave during late July to early August, 2018. The heatwave was pronounced in Belgium (BEL), Netherlands (NLD), Denmark (DEN) and Germany (DEU) according to the reports published by the Royal Meteorological Institute of Belgium [61], Royal Netherlands Meteorological Institute [62], Danish Meteorological Institute [63] and the German Meteorological Office [64], respectively. To quantify the impact of this heatwave, SIF soundings of BEL, NLD, DEN and DEU from 15 th July to 15 th August, 2018 were analyzed for all the vegetation types and compared to the mean of 2015-2017. All analyses were processed in 'R' [65], extended by the packages stats, ggplot2 [66] and dplyr [67].

Overall spring-summer SIF variation and anomaly
Overall, the whole of Europe showed average SIF_2018 values of 0.642 W/m 2 /sr/µ m compared to SIF_Mean of 0.670 (Table 1a) during spring and summer season. Although, the SIF anomaly (SIF-Diff) was not significant for spring and combined spring-summer season for drought areas (Fig S1), and the whole of Europe, the summer season showed a significant decrease (p-value < 0.001) in SIF_2018 compared to SIF_Mean in the drought area (Table 1; Figure 2). The pattern of combined spring-summer season SIF-Diff for drought and non-drought area (Figure 1c) were complementary to each other (Table 1a). Also, the drought areas showed positive SIF-Diff for spring and negative SIF-Diff for summer and combined spring-summer season, whereas the non-drought area showed positive SIF-Diff for spring and combined spring-summer season. Although most of the SIF-Diff was not significant (p-value > 0.05), it is however interesting to see a distinct pattern between seasonal variability of SIF in relation to drought.

Intra-seasonal SIF variation and anomalies for different vegetation types
The drought area was characterized by higher temperatures during the second half of both spring and summer season (Figure 3b). For the drought area, SIF-Diff was initially significantly negative at the start of the spring season for the agricultural area, whereas the difference was not significant for forests. Here, significance is considered as a departure of (SIF/fPAR/NDVI) values > 1 standard error [34,35], i.e., for Figure Figure S2) and Köppen climate zones (warm summer and cool summer). More intense and early spring warming occurred in the warm summer climate, whereas a moderate and later warming occurred in the cool summer climate zone (blue and black line in Figure S6). This seems to have caused leaf flushing (represented by the blue and green box in S6) at a difference of about 20 days between the two climate zones. Now, since our result is spatially and temporally aggregated, this shows two peaks during the spring for BLF. Thus, although there was no overall change in SIF for the spring season (section 3.1), we observed a change in the intra-spring SIF variation pattern in 2018 compared to the mean, with an initial decrease (March to April) followed by a significant increase in SIF in May. This intraspring SIF variation reflected in fPAR variations during the spring season for drought area ( Figure  A1). Thus, SIF variation in spring was strongly related to fPAR for drought areas where continuous higher temperature and VPD were observed in the second half of the spring season (April-May). The summer season of 2018 was characterized by negative SIF-Diff ( Figure 3) especially for agricultural areas which showed continuous negative values throughout the summer season ( Figure 3). For forest areas, however, we observed significantly negative SIF-Diff only in the later part of summer (from July, DOY 185-215) when the temperature values were more than 5 degrees higher than the mean (Figure 3b).
The non-drought areas were characterized by significantly higher temperatures only during the end of April (DOY 105-120) as illustrated in Figure 4b. Here, we observed the spring variation to be similar to that of the drought area (Figure 4a and 4b), which is also visible in the fPAR data ( Figure  S4a). In contrast to the drought area, summer in the non-drought area featured positive SIF-Diff for all the vegetation types and especially for the agricultural area. For the non-drought area, fPAR in summer was higher than the mean value ( Figure S4a) and also seems to closely relate to the SIF variation.

Drought impact on SIF
Our study showed a clear impact of the European hotter drought of 2018 on overall and intraseasonal SIF variation for different vegetation types as captured by the OCO-2 SIF. Early spring of 2018 was characterized by lower than average temperatures and VPD (non-stressed stage) followed by above average temperatures and VPD in the later part of spring (April-May; Figure 3b). Such cooler and wetter conditions in early spring can result in poor crop development and lower productivity [68] and thus, lower SIF values ( [69]; Figure 3b). Moreover, the initially cooler conditions in spring might have delayed the leaf flushing in deciduous trees, resulting in lower SIF values of broadleaved forests (Figure 3b). This pattern was less evident for the coniferous and mixed forests, probably due to the presence of evergreen trees. Also, the wet conditions during early spring most likely built up enough soil moisture for the later spring season, thus moderating the sudden warming in later spring season [70,71]. This combination of ample moisture supply with a rise in temperature resulted in an increase in SIF to above average values across all vegetation types (Figures 3 and 6). The variation in SIF depends on the variation of both APAR (and fPAR) and SIFyield (Equation 1, [72]). Both, SIF and fPAR were increasing during the spring season for both drought and non-drought areas (Figures 3, 4, A1a and S4). Moreover, for agriculture areas affected by drought, the peaks of fPAR and SIF were on almost the same day of the year (DOY ~ 145; Figure 3a and A1a). However, this clear synchronicity was not found for forests. Nevertheless, we conclude that the satellite-based SIF variation was closely coupled with fPAR variation under non-stress conditions (spring for drought areas and for non-drought areas in spring and summer). Previous studies [13,34,35] have also highlighted the synchronicity of satellite-based SIF with fPAR.
The warming during late spring continued to the summer season, with the later summer season featuring intense heat. We found decreasing SIF in all investigated vegetation types in summer of 2018, which was most pronounced in the drought areas. The agricultural areas expressed an immediate decrease in SIF from early summer onwards most likely because of severe soil moisture deficits caused by early water depletion in late spring in course of dry weather conditions [73,74]. Ongoing drought conditions during the summer season showed continuous decrease in SIF for agricultural area and, with some delay in forests (Figure 3). Similar findings were reported by an NDVI-based study where agricultural areas and, with some delay, forests in central Europe faced severe decline in NDVI during the summer season [6]. Unlike agricultural crops, forests feature a buffered micro-climate [75][76][77] and given their deeper rooting system in comparison to crops, have access to deeper soil water and thus, can maintain constant photosynthesis rates for extended periods during drought conditions [78].
Our spring-summer seasonal variation of SIF in 2018 is in close agreement with the one of MODIS NDVI ( [6]) for agricultural areas. However, for the forest ecosystems, we found a less extreme SIF decline compared to NDVI-decline in Buras et al [6]. To further evaluate this difference we analyzed MODIS NDVI over 4 years (2015-2018), and our results indicate that SIF showed higher sensitivity compared to NDVI over the 4 years period (Figures A1b and 6). We observed that for the forest ecosystems in the drought area, fPAR-and NDVI-anomalies were positive during the summer season (DOY < 215) ( Figures A1b and 6), i.e. when the SIFanomaly was reportedly negative. While Buras et al [6] used a reference period spanning 2000-2018, where several normal and even moist summers were observed, the analyses presented here rely on the period 2015-2017 due to the comparably shorter mission length of OCO-2. However, in 2015 another severe drought hit central and southern Europe which might have resulted in forest legacy effects [79], i.e. low productivity and die-back in affected areas in 2016 [80]. Consequently, regarding forests, the fPAR and NDVI values from 2018 do only marginally differ to the period mean (2015-2017) since forests could have been suffering at least in 2015 and 2016, i.e. in 2 of the three years included in the reference mean. It is therefore not surprising that the extremeness of 2018 in the 4 year NDVI analysis of forests ( Figures A1  and 6) did not capture the forest decline as clear as in Buras et al [6] (with 19 years reference period) where the majority of reference years represent years with sufficient water supply. In contrast to NDVI over the four years period, SIF was considerably lower in 2018 compared to the already low mean of previous three years (2015-2017) which may indicate its higher sensitivity to both physiological and structural drought response from plants. This finding highlights the predominance of SIF over NDVI if only short reference periods are available, for example, SIF available from OCO-2, upcoming OCO-3, and TROPOMI in comparison to SENTINEL (for NDVI).
Vegetation indices (VIs), such as NDVI are sensitive to canopy structure and pigment concentration with no direct link to photosynthesis [81]. This might lead to reduced sensitivity of these VIs to capture drought impact when the vegetation, such as high density forests, still remains green (no changes in chlorophyll content) but reduce their photosynthesis initially during heavily stressed drought conditions [82] in early summer during the European hotter drought. Similar conditions might also not have resulted in significant differences of NDVI during the early summer of 2018 from the mean ( Figure A1b). Alternatively, higher sensitivity Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 14 August 2020 doi:10.20944/preprints202008.0327.v1 of SIF (with direct links to photosynthesis) to drought impact on high density European forest was evident from our study (Figures 3 and 6).
The intense heatwave in Central Europe during July and August resulted in a significant decrease in SIF across all the four vegetation types, which was particularly pronounced for agricultural areas. The heatwave occurred at the end of the already dry summer, which was preceded by a dry spring. As a consequence of persisting drought, many crops desiccated, thus not emitting any fluorescence by end of July when temperatures were extremely high in Central Europe. Water stress due to high-temperature anomalies during a drought event can further increase heat stress due to an extended period of stomatal closure and subsequent reduction in evaporative cooling [83], thus causing positive feedback resulting in even warmer conditions [84].

SIF response during drought stress
Water stress adversely impacts photosynthesis, either physiologically or structurally [85]. Physiological alterations include change in leaf biochemistry via reductions in enzymatic activity [86] or a reduction in mesophyll/chlorophyll, and stomatal conductance [87], resulting in lower CO2 uptake by the plants and thus reduced photosynthesis (i.e. lower SIF values). The structural changes include foliar changes like leaf wilting or rolling (changes in leaf inclination), thus reducing effective leaf area index (LAI), which in turn causes a reduction in fPAR and energy available for photosynthesis and fluorescence emission [88,89]. Furthermore, plants adopt different mechanism to minimize water and heat stress, like chloroplast avoidance movement during which plants move their chloroplast from the cell surface to side walls of cell [90] so as to minimize APAR [91] and thus SIF. MODIS fPAR decrease due to these protection mechanism, and physiological and structural changes was visible for agricultural Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 14 August 2020 doi:10.20944/preprints202008.0327.v1 area during the summer drought ( Figures A1 and 6). However, for the forest areas we found that MODIS fPAR for drought areas during mid-summer (DOY 185-215) was not lower than fPAR in previous 3 years (Figures 7 and 8). This might imply that reduction of SIF is likely caused by a reduction in SIFyield. SIF variation due to changes in SIFyield can be because of canopy's physiological (ΦF term of Equation 1) or structural change (ΩC term of Equation 1) or both. The PAR absorbed by leaves is partitioned into photochemistry, heat dissipation (NPQ) and fluorescence emission (SIF). Although fPAR (or APAR) did not decrease, the proportion of fPAR (or APAR) that is used in photochemistry (to drive electron transport for carbon assimilation) decreases, resulting in a surplus of photosynthetic energy. Now, two other pathways compete to de-excite the absorbed light that is not used in photochemistry (the surplus photosynthetic energy), the energy emitted as SIF or energy dissipated as heat through NPQ [72]. The competition between these two pathways depends on the environmental stress [92], and during water stress conditions (drought) the reduction in SIF is linked to higher NPQ [93,94]. This increase in NPQ may result in a weak coupling between SIF and fPAR under water stress conditions, which seems to be visible for forests during mid-summer (DOY 185-215; Figure A1). It is also possible that the MODIS fPAR could be erroneous [35] and not be the best estimate of fPAR, especially in case of high vegetation density (as in Europe). The backup algorithm of MODIS fPAR rely on its relationship with NDVI [52], which becomes non-linear with sharp drop of NDVI sensitivity to fPAR for moderate to high vegetation density for fPAR values beyond 0.7 ( Figure S5). Nevertheless, further research to differentiate the canopy escape term (ΩC term of Equation 1) from SIF (as demonstrated by [95,96]) to study changes in fluorescence yield (ΦF) during extreme large scale drought event is recommended.
Our results highlight one more reason why OCO-2 SIF can be useful to study drought impact on vegetation as different vegetation types have different response to drought, which was not efficiently captured for each vegetation types before because of the coarser resolution of preceding satellite-based SIF products. We would like to stress, that earlier studies used SIF at coarser spatial resolution (from GOME-2 or GOSAT) and mostly were not able to differentiate SIF responses for different vegetation types from a heterogeneous land-use area during drought conditions.

Conclusions
This study employed the satellite-based SIF measurements acquired from the recent OCO-2 satellite to study the impact of the 2018 European hotter drought on terrestrial ecosystems. Evidently, OCO-2 SIF displayed the spatial and temporal dynamics of the 2018 European drought across different vegetation types (agricultural area, broadleaved forest, coniferous forest, and mixed forest). Our results show that the agricultural areas were severely affected by the 2018 European hotter drought. While SIF of forests showed a less strong reduction, legacy effects may become visible in the next years and thus, we recommend to closely monitor their behavior. In general, the impact of the heatwave in Central Europe during July-August showed an overall 31% decrease in SIF compared to the reference mean. The SIF variation was closely explained by fPAR variation during nonstressed conditions. However, during water-stress conditions (drought), the SIF variation was attributed to variation in SIFyield. Despite few technical limitations of the OCO-2 SIF measurement such as the 16-days repeat cycle, it provides an excellent potential to study large-scale vegetation fluorescence variation at high spatial resolution and can improve our understanding of changes in ecosystem productivity during extreme events. Furthermore, owing to its direct origin from chlorophyll, OCO-2 SIF is sensitive to both physiological and structural changes in plants and can serve as a complementary dataset to MODIS's vegetation indices (NDVI/EVI). Further application of OCO-2 SIF may include process-based SCOPE modeling to understand the variation of complex fluorescence yield due to different environmental stress, as well as its combination with the XCO2 data (main product of OCO-2 satellite) to study regional carbon budgets. dataset. The authors would like to extend its gratitude to Deutscher Akademischer Austauschdienst (DAAD) and Centre of Rivers Atmosphere and Land Sciences (CORAL), IIT Kharagpur, for funding the research stay at Technical University of Munich and to Mr. Phillip Papastefanou for his discussion about different drought induced plant physiological responses.

Conflicts of Interest:
Declare conflicts of interest or state "The authors declare no conflict of interest."

Appendix A
To show that our way of sampling is robust, we have compared the most recent version of spatially downscaled GOME-2 SIF (8 day GOME-2 downscaled SIF of 0.05*0.05 resolution at 740nm; [97] ) means with the OCO-2 SIF means. The percentage change in 2018 from both downscaled GOME-2 SIF (GSIF) and OCO-2 was comparable (Table A1). This shows that although OCO-2 does not have a fixed revisit time. Aggregated mean values over a continental scale are robust for multi-year comparison.