Spatial Cross-Correlation of GOSAT CO₂ Concentration with Repeated Heat Wave-Induced Photosynthetic Inhibition in Europe from 2009 to 2017

Abstract

In recent decades, European countries have faced repeated heat waves. Traditionally, atmospheric CO₂ concentration linked to repeated heat wave-induced photosynthetic inhibition has been explored based on local-specific in-situ observations. However, previous research based on field surveys has limitations in exploring area-wide atmospheric CO₂ concentrations linked to repeated heat wave-induced photosynthetic inhibition. The present study aimed to evaluate the spatial cross-correlation of Greenhouse gases Observing SATellite (GOSAT) CO₂ concentrations with repeated heat wave-induced photosynthetic inhibition in Europe from 2009 to 2017 by applying geographically weighted regression (GWR). The local standardized coefficient of a fraction of photosynthetically active radiation (FPAR: −0.24) and the normalized difference vegetation index (NDVI: −0.22) indicate that photosynthetic inhibition increases atmospheric CO₂ in Europe. Furthermore, from 2009 to 2017, the heat waves in Europe contributed to CO₂ emissions (27.2–32.1%) induced by photosynthetic inhibition. This study provides realistic evidence to justify repeated heat wave-induced photosynthetic inhibition as a fundamental factor in mitigating carbon emissions in Europe.

1. Introduction

The intensity and frequency of heat waves, described as periods of days with unprecedented high temperatures, have intensified during the last decade and are anticipated to increase in the 21st century [1]. Due to continuous record-breaking high temperatures, the heat waves have expanded in amplitude and spatial extent in Europe during recent decades [2]. For instance, temperature observation records at European stations continue to be broken every year [3]. In this regard, the length of heat waves in summer has increased twofold, and the number of days reporting heat extremes has increased threefold in Europe [4]. These heat wave events are projected to become more frequent and intense in Europe.

Plants respond to heat waves by adjusting their physiological structures, such as decreased leaf area, root–shoot ratio changes, or osmolyte concentration [5]. This causes decreased CO₂ assimilation rates by reducing photosynthetic enzymes and sink strength and increasing source activity (respiration). It is thus necessary to elucidate the current status of CO₂ emitted and absorbed according to photosynthetic inhibition and net primary productivity since heat waves are a crucial regulator of rising atmospheric CO₂ concentrations [6].

It is essential to explore the vertical profile of CO₂ that changes according to the photosynthetic action of vegetation and the use of fossil fuels as it is transported from the ground to the upper atmosphere. The mechanisms of greenhouse gases, including CO₂, are monitored at approximately 530 ground stations operated by the World Meteorological Organization (WMO) [7]. However, these field-oriented surveys have considerable limitations in generalizing research outcomes because CO₂ concentrations vary significantly in time and space, and the survey area is limited to the range of the WMO station.

Satellite-based XCO₂ (column-averaged CO₂) carries large amounts of information from the bottom atmospheric layer (near-ground) to the top of the atmosphere, including background atmospheric CO₂ [8,9,10]. Hence, satellite-based XCO₂ can be a valuable indicator of atmospheric CO₂ caused by photosynthetic inhibition on a regional scale. Recently, the observation of CO₂ through advanced remote sensing technology has been suggested to overcome the spatio-temporal limitations of existing measurements. The Greenhouse gases Observing SATellite (GOSAT) is the most advanced satellite system for observing CO₂, and its usefulness has been recognized in various studies [11,12,13].

The relationship between photosynthetic inhibition induced by heat waves and its influence on the atmospheric CO₂ growth rate varies regionally, annually, and seasonally with time (including day and night) [14]. Geographically weighted regression (GWR) provides a weighting of locally correlated information and allows the building of local regression model parameters varying in space during the last ten years. Hence, GWR could help to reveal spatiotemporal variations in the empirical relationships between photosynthetic inhibition and CO₂ concentration over a more extended period. Previous studies have reported that heat waves affect the seasonal photosynthetic responses of European oak [15]. It was confirmed that these studies were the closest to the topic to be dealt with in this study. Therefore, this study aimed to evaluate the spatial cross-correlation of GOSAT CO₂ concentrations with repeated heat wave-induced photosynthetic inhibition in Europe from 2009 to 2017.

2. Materials and Methods

2.1. Study Area

Europe is the second-smallest continent after Australia but includes 18 climate zones in small continents from Arid to Polar [16]. There are various types of plants in Europe, such as boreal tundra woodland, boreal coniferous forest, temperate steppe, temperate continental forest, subtropical dry forest, and so on [17]. The 18 climate zones show the different frequencies and characteristics of heat waves within the European continent. The diverse types of heat waves that occurred in the 18 climate zones induce photosynthetic inhibition stress in terms of diverse exposure temperatures, such as exposure duration, the ability of tolerance or acclimation, time of year, and soil moisture availability. The heat stress of terrestrial plants inhabited in 18 climates can present the quantitative influence of differentiated response patterns on photosynthetic inhibition (reductions of carbon assimilation and growth) [18]. Therefore, the European continent is ideal for studying heat wave-induced photosynthetic inhibitions from terrestrial plants.

2.2. Variables for Building the GWR Model

In this study, we utilized satellite observation data acquired by the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the Terra satellite from June 2009 to October 2017: normalized difference vegetation index (NDVI; MOD13A2), leaf area index (LAI), fraction of photosynthetically active radiation (FPAR; MOD15A2), daytime and nighttime land surface temperatures (LSTs; MOD11A2), and daily net photosynthesis (PSNet; MOD17a2) (

Table 1. Descriptive statistics for monthly anomalies of dependent and independent variables used in GWR.

Category	Resolution		Min	Max	Mean	STDEV
GOSAT level 4 XCO2	2.5°	Month	−26.2	16.69	0.23	4.71
ODIAC (tonnes C/km2)	1 km	Month	−0.385	0.747	0.009	0.201
MOD17a2 PSNnet (tonnes C/km2)	0.5 km	8 days	−508.01	649.23	−3.98	148.32
MOD11A2 LST (°C)	1 km	8 days	−53.78	45.28	−0.11	6.85
MOD13A2 NDVI	1 km	16 days	−0.55	0.44	−0.001	0.11
MOD15A2 FPAR (%)	0.5 km	8 days	−0.51	0.45	0.001	0.11
MOD15A2 LAI (m2/m2)	0.5 km	8 days	−3.44	4.24	−0.01	0.59
MOD16A2 Average Latent Heat Flux (J/m2/day)	0.5 km	8 days	−0.0002	0.0002	0.00	0.00004

). The original temporal and spatial resolutions of LAI, FPAR, and PSNet were eight days and 0.5 km; those of NDVI were 16 days and 1 km; and those of LST were eight days and 1 km, respectively. MODIS is a major observation sensor mounted on NASA’s Earth Observation Satellite, with multipurpose sensors that can be applied to the ocean, land, and atmosphere. The MODIS sensors observe the Earth’s surface once or twice a day at an altitude of 705 km, with a viewing angle of ±55°, an observation width of 2330 km, and a total of thirty-six spectral bands in the range of 0.4–14.4 μm. Two of the bands make 250-m resolution images at the nadir, five bands have a resolution of 500 m, and the remaining 29 bands have a resolution of 1 km.

The GOSAT Level 4 (L4) product comprises Level 4A (L4A) surface CO₂ flux data and Level 4B (L4B) 3D-CO₂ concentrations modeled with L4A. The L4B data contain the CO₂ concentrations at 17 vertical levels from the ground to the upper atmosphere (666 km) [19]. The closest vertical level of the L4B data to the ground is 975 hPa [20]. Because the CO₂ nearest to the ground surface provides more information on changes in the CO₂ sink and source, this study utilized the L4B CO₂ concentrations at 975 hPa to reflect near-ground CO₂. The L4B product for CO₂ was obtained between June 2009 and October 2017. The GOSAT L4B data provide the average monthly CO₂ concentrations modeled on a 2.5° × 2.5° horizontal grid in netCDF format. Therefore, we fitted the spatial and temporal scales of the MODIS indicators into GOSAT XCO₂ by averaging MODIS observations. This study used the monthly anomalies of GOSAT XCO₂ and MODIS observations to build the GWR model.

2.3. Moran’s I Analysis

When analyzing the statistical distribution of spatial data, the results differ according to the location and circumstances of the data. This is because spatial data are related to each other due to the influence of spatial interaction and spatial diffusion. Before performing the spatial regression, spatial autocorrelation tests were conducted for each explanatory and dependent variable. Spatial autocorrelation is multidimensional (i.e., two to three dimensions of space) and multidirectional; it is more complicated than one-dimensional autocorrelation. The GWR builds a regional regression model using spatial weights. Therefore, spatial autocorrelation must be investigated before building the GWR model. Hence, we utilized Moran’s I to evaluate the autocorrelations of the variables used to build the GWR model.

Moran’s I was computed from −1 to 1. The negative autocorrelation (close to −1) exhibits that nearby locations tend to have unrelated values in adjacent areas [10]. By contrast, a positive autocorrelation (close to 1) implies that similar values tend to occur in adjacent areas. The spatial arrangement is randomly distributed if there is no spatial autocorrelation (close to zero). In this study, individual variables for GWR showed strong positive autocorrelations of 0.61–0.86 (

Table 2. Descriptive statistics for Moran’s I estimated from individual variables.

Category	Max	Min	Mean	Standard Deviation
LST	0.98	0.35	0.75	0.14
NDVI	0.93	0.34	0.7	0.13
FPAR	0.93	0.29	0.69	0.13
LAI	0.99	0.31	0.7	0.14
Average latent heat flux	0.96	0.34	0.61	0.12
CO2 (GOSAT level 4)	0.99	0.61	0.86	0.09
Emission (ODIAC-PSNnet)	0.97	0.37	0.72	0.11

). The mean autocorrelation of GOSAT Level 4 XCO₂ appears the highest at 0.86, while the average latent heat flux appears the lowest at 0.61. The autocorrelation of GOSAT Level 4 XCO₂ is strongly associated with the spatial distribution of CO₂ sources and sinks. The GOSAT Level 4 XCO₂ is calculated with Priori flux, which utilizes ODIAC data. ODIAC data is a global, high-resolution monthly emission data, which disaggregates the emission from diverse CO₂ sources (fossil fuel, nuclear, hydro, and other renewable energy plants) obtained from the Carbon Monitoring for Action (CARMA) data set [21]. Average latent heat flux is estimated with diverse data such as land surface temperature (LST), the fraction of absorbed photosynthetically active radiation (FPAR), the normalized difference vegetation index (NDVI), the enhanced vegetation index (EVI), leaf area index (LAI), and albedo [22]. The parameters for estimating average latent heat flux are heterogeneous in the terrestrial landscape [23]. Thereby, it shows less autocorrelation than other indicators.

2.4. Building GWR Model

A GWR local model was employed to evaluate how the variations in carbon uptake sources and GOSAT XCO₂ regional relationships changed from 2009 to 2017. GWR is a local regression that locally differentiates the variables of a regression estimation. Unlike traditional regression, which establishes a single global regression among explanatory and response variables, GWR considers spatial variation in a model and provides maps to explain spatial non-stationarity [24,25]. The GWR is estimated by multiplying the geographically weighted matrix $w\left(g\right)$ composed of geo-references [24,26]. $w\left(g\right)$ is calibrated using the geographically neighboring spatial relations between points. In this study, we presumed that the degree of influence had an inverse ratio to the square distance of GOSAT Level 4. This means that the greater the $w\left(g\right)$ is, the closer the points of geographical data and the stronger their influence on one another [27]. This study examined the spatial variability of a locally computed coefficient to identify the underlying process that presents spatial heterogeneity [24]. The regression model can be defined as follows (Equation (1)):

$$\begin{gathered} {\mathrm{GOSAT}\mathrm{XCO}}_{2i}\left(g\right)={\mathsf{\beta }}_{0i}\left(g\right)+{\mathsf{\beta }}_{1i}\mathrm{LST}\left(g\right)+{\mathsf{\beta }}_{2i}\mathrm{NDVI}\left(g\right)+{\mathsf{\beta }}_{3i}\mathrm{FPAR}\left(g\right)+{\mathsf{\beta }}_{4i}\mathrm{LAI}\left(g\right) \\ +{\mathsf{\beta }}_{5i}\mathrm{Average}\mathrm{latent}\mathrm{heat}\mathrm{flux}\left(g\right)+{\mathsf{\beta }}_{6i}\mathrm{Emission}\left(g\right)+{\mathsf{\epsilon }}_i \end{gathered}$$

(1)

where $\left(g\right)$ represents the parameters estimated at each independent variable in which the coordinates are presented by vector $g$. ${\mathsf{\beta }}_{1i}\cdots {\mathsf{\beta }}_{6i}$ denotes the regression coefficient for the ith datum of independent variables (LST, NDVI, FPAR, LAI, Average latent heat flux, emission). ${\mathsf{\epsilon }}_i$ is a residual [28]. Our analysis was implemented utilizing ArcMap 10.3 with a significance level of 0.05. Because the distribution of variables in the study area was non-uniform, an adaptive kernel was selected.

2.5. Mediation Analysis

Mediation analysis helps interpret the causality between dependent and independent variables by adding a third mediator, as there is a hidden connection between these variables [29]. Baron and Kenny (1986) found that four criteria have to be satisfied to implement the mediation analysis [30], as shown below (

Table 3. Results of Pearson’s r among mediator (LST), independent (NDVI, FPAR, LAI) and dependent variables (GOSAT Level4 XCO₂).

Paths	X→Y			X→M			M→Y
Category	NDVI→GOSAT XCO2	FPAR→GOSAT XCO2	LAI→GOSAT XCO2	NDVI→LST	FPAR→LST	LAI→LST	LST→GOSAT XCO2
Pearson’s r	−0.40 **	−0.47 **	−0.40 **	0.32 **	0.64 **	0.373 **	−0.45 **

** p-value < 0.01.

); (1) $X$ significantly accounting for $M$, (2) $X$ significantly accounting for $Y$, (3) $M$ significantly accounting for $Y$, (4) decreases in the effect of $X$ on $Y$ with $M$ entered simultaneously with $X$. Table 3 presents the evidence that the results of this study satisfy the four criteria suggested by Baron and Kenny (1986) [30]. Based on this result, we assume that LST ($M$) directly or indirectly involves the changes in the effects of photosynthetic activities ($X$), preceding changes in GOSAT XCO₂ ($Y$). The total effects of MODIS photosynthetic indicators on GOSAT XCO₂ could be apportioned individually into direct and indirect effects from an LST (mediator). Preacher and Hayes bootstrapping [31] with 5000 bootstrap samples were utilized to investigate mediation effects that can be caused by LSTs.

3. Results

Figure 1 and

Table 4. Results of the GWR between GOSAT XCO₂ (dependent variables) and heat wave-induced photosynthetic factor, including CO₂ emissions (ODIAC-PSnet, independent variables).

Category		LST	NDVI	FPAR	LAI	ALHF	Emi.
Standardized local coefficient	Min	0.04	−0.58	−0.61	−0.62	−0.34	−0.32
	Max	0.77	0.09	0.21	0.46	0.45	0.64
	Mean	0.35	−0.22	−0.24	−0.16	0.03	0.14
	Standard deviation	0.17	0.14	0.20	0.26	0.25	0.23
t-statistics	Min	0.20	−3.41	−4.28	−2.69	−1.15	−1.61
	Max	4.56	0.28	1.05	0.89	2.46	2.93
	Mean	1.99	−1.44	−1.63	−0.84	0.18	0.53
	Standard deviation	1.20	0.70	1.45	0.92	1.08	1.08
Local p-value	Min	0.00	0.00	0.00	0.00	0.00	0.00
	Max	0.11	0.22	0.18	0.17	0.21	0.16
	Mean	0.01	0.03	0.03	0.02	0.03	0.01
	Standard deviation	0.05	0.08	0.09	0.08	0.09	0.09

Local R²: 0.56–0.89 R²: 0.60–0.96 p-value: 0.02, LST: land surface temperature, ALHF: average latent heat flux, Emi.: ODIAC-PSNnet.

present the results of the 10-year GWR with the monthly anomaly of GOSAT level 4 XCO₂ versus photosynthetic activity. NDVI, FPAR, and LAI have negative mean values of the standardized local coefficient of −0.22, −0.24, and −0.16, respectively. A decrease in chlorophyll levels affects the ability of plants to reflect incident solar radiation. Hence, stressed plants with low photosynthetic capability have reduced NDVI values. The distribution of LAI is another crucial determinant of photosynthesis because canopy leaf area rather than vegetation cover indicated by NDVI is often chosen as a base reference for the growth index of plants [32,33,34]. Therefore, the negative local coefficients of NDVI, FPAR, and LAI on GOSAT level 4 XCO₂ show that the capability of photosynthetic activities in Europe has been reduced from 2009 to 2017. Changes in NDVI, FPAR and LAI possibly contributed to increasing atmospheric CO₂ because of the reduced capability of photosynthetic activities in terrestrial ecosystems. Temperature showed the most decisive positive influence (0.35) on the geographical variations in CO₂ concentrations during June 2009–October 2017. In this study, LST and latent heat flux increased sharply, whereas FPAR decreased. This has been caused by the scarcity of nutrients, humidity, and water stress due to the rapid increase in temperature and latent heat flux. This possibly has impeded the growth of carbon stocks through photosynthetic inhibition.

During one specific heat wave, the local R² and R² plunged simultaneously (Figure 2). It was the second hottest year without an El Niño since 1850 [35]. Heat stress causes reductions in enzymatic activity [36], mesophyll/chlorophyll, and stomatal conductance [37]. The structural changes that occurred by leaf wilting and rolling decreased NDVI and LAI, reducing FPAR [38]. However, according to Shekhar et al. (2020), FPAR from European forest areas during the heat wave and drought events was not lower than FPAR in the past three years. While FPAR did not decrease, the proportion of FPAR decreased in transporting electrons for carbon assimilation, causing a surplus of photosynthetic energy [39]. It is considered that a similar pattern has occurred in this study. In August 2013 (heat wave events happened), the NDVI and LAI were relatively low, but the FPAR was the highest in the study period (Figure 3). Generally, the relations between NDVI, LAI, and FPAR are positively linear [40]. This abnormal negative correlation between NDVI, LAI, and FPAR involves the stiff decline of R² and local R² in August 2013.

The indirect effects of sudden climatic events (higher heating demands, drought, etc.) have influenced terrestrial carbon uptake activities, leading to decreased local R² and R². According to Bezak and Mikoš (2020) [41], heat waves have shown an increased probability of occurrence with compound events (hot and dry) across Europe during recent decades. Moreover, the heat waves with high surface temperature tend to cause the soil moisture deficit reducing the evaporative cooling and increasing heat flux. In turn, this exacerbates the prevailing drought condition in Europe [42]. Therefore, Europe’s photosynthetic inhibition resulted from heat waves and coincident droughts.

Interestingly, as we explored the monthly local R² and R² values during June 2009–July 2017, there were specific patterns between local R² and R². Generally, the R² was higher than the local R² value. This means that regionally differentiated patterns of photosynthetic inhibition influence the temperature rise. In Figure 2, the grey boxes denote the months showing the larger discrepancy between local R² and R² of 0.15–0.28. The local R² and R² discrepancies increase during the summer and winter (Figure 2). The abnormal heat wave or temperature rise appears more clearly in the summer and winter (Figure 3). Thus, this discrepancy between local R² and R² in summer and winter indicates that abnormal photosynthetic inhibition has occurred regionally in Europe from 2009 to 2017.

The trends of MODIS photosynthetic indicators and GOSAT XCO₂ support the influence of photosynthetic inhibition resulting from heat waves. LST continues to increase by 0.29 °C per year in Europe. This indicates that Europe had hotter summers and milder winters from June 2009 to July 2017. Simultaneously, NDVI and LAI also increased to 0.007 and 0.018 m²/m², respectively, because of the milder winters. However, FPAR, which represents the photosynthetic activities of the terrestrial ecosystem, decreased annually over Europe. This indicates that droughts induced by repeated heat waves increase water stress in plants [43].

The trends of MODIS photosynthetic indicators and GOSAT XCO₂ support the influence of photosynthetic inhibition resulting from heat waves. LST continues to increase by 1.81% per year in Europe. This indicates that Europe had hotter summers and milder winters from June 2009 to July 2017. Simultaneously, NDVI and LAI also increased to 1.66% and 2.16%, respectively, because of the milder winters. However, FPAR, which represents the photosynthetic activities of the terrestrial ecosystem, decreased by 10.19% annually in Europe. This indicates that droughts induced by repeated heat waves increase water stress in plants [43]. To verify the linear trends of variables utilized for GWR, this study computes the confidence interval/confidence bands for the slope parameter $\widehat{b}$ of the linear regression. This interval reads for an $\alpha$ error of 5% as follows:

$$\left[\widehat{b}-t_{n-2;0.95}\cdot \sqrt{\frac{1}{n-2}\frac{{\displaystyle \sum }{\left(y_i-\widehat{y}\right)}^2}{{\displaystyle \sum }{\left(x_i-\overline{x}\right)}^2}}; \widehat{b}+t_{n-2;0.95}\cdot \sqrt{\frac{1}{n-2}\frac{{\displaystyle \sum }{\left(y_i-\widehat{y}\right)}^2}{{\displaystyle \sum }{\left(x_i-\overline{x}\right)}^2}}\right]$$

(2)

where the t denotes the Student t distribution quantile and n is the number of observations. Y is the explained variable, and x is the input variable of the linear model. $\overline{x}$ is the average of the input values, and $\widehat{y}$ is the model estimate. Based on this, we can compute the difference between the real world and the model estimate (and square it) and divide it through the sum of the squared deviations of the input values from the mean/average value. If n = 7 then n − 2 = 5 and the Student t quantile is 2.571. What remains to compute are the sums in the above formula. As explained in the above formula, the estimated slope parameter minus/plus the confidence band factor gives the confidence interval at a level of 5%. If zero is not contained in this interval, we consider the estimated parameters to be significant. Estimated slopes and the confidence bands of all the variables (LST, NDVI, FPAR, LAI, GOSAT level XCO₂, ODIAC) in Figure 3 are all significant (

Table 5. Results of the computing confidence bands (confidence intervals) of variables used in Figure 3.

Category	LST	NDVI	FPAR	LAI	GOSAT XCO2	ODIAC
Slope Estimate	0.2826	0.0073	−0.0001	0.1078	2.3749	−0.26
Confidence Band	0.2129	0.0069	0.0028	0.0429	0.1451	0.1152

).

4. Discussion

The mediation analysis results show that LST had strong negative mediating effects on MODIS photosynthetic indicators from 2009 to 2017 (

Table 6. Mediation analysis results between NDVI, fraction of FPAR, LAI, and LST on GOSAT XCO₂.

Category	Mediator	R	R2 (p-Value)	Standardized Coefficient	Total Effect (p-Value)	Direct Effect (p-Value)	Indirect Effect
							Indirect (%)	LLCI	ULCI
NDVI	LST	0.53	0.28 (0.00)	−0.40	−16.96 (0.00)	−12.15 (0.00)	−28.4	−5.17	−4.45
FPAR	LST	0.47	0.22 (0.00)	−0.47	−19.47 (0.00)	−12.82 (0.00)	−34.1	−7.25	−6.05
LAI	LST	0.40	0.16 (0.00)	−0.40	−3.22 (0.00)	−2.19 (0.00)	−32.1	−1.12	−0.95

ULCI: upper limit of the bootstrap confidence interval; LLCI: lower limit of the bootstrap confidence interval.

). The indirect effects of LST on NDVI, FPAR, and LAI ranged from −32.1% to −28.4%. Even though the mediator variable (LST) did not change the direction of the relationships between the MODIS photosynthetic indicators and GOSAT XCO₂ from 2009 to 2017, LST promoted stronger photosynthetic inhibition from NDVI, FPAR, and LAI. LST operated as a mediator for NDVI, FPAR, and LAI within a confidence level of 0.05. LST had a statistically significant mediation effect on NDVI, FPAR, and LAI. LST indirectly influenced the decrease in photosynthetic indicators for all the three variables. Furthermore, it might have accelerated the decrease in photosynthetic inhibition from 2009 to 2017.

The observed local coefficient and mediation analysis results between LST, LAI, FPAR, NDVI, and GOSAT XCO₂ indicate that recent heat waves in 2009–2017 have reduced the potential of photosynthetic activities within Europe to withstand adverse heat stress. This is well presented in an in-situ survey implemented by FOREST EUROPE [44]. The survey reported the defoliation of forests submitted from 27 European countries, monitored at 5634 plots, 103,797 trees, and more than 130 species. According to this report, the condition of European forests has recently deteriorated, with increasing mean defoliation of the six most frequent tree species (Pinus pinaster, Picea abies, Pinus sylvestris, Fagus sylvatica, and Quercus ilex) particularly obvious in 18.9% of the monitoring plots from 2010 to 2018. Furthermore, the report pointed out that heat waves appear to be the primary drivers triggering changes in forest tree defoliation [44].

Additionally, the recent extraordinary warming during winter greatly enhanced the subsequent release of CO₂ due to soil organic matter’s microbial decomposition [45]. Natali et al. [46] suggested that increased soil CO₂ loss due to warmer winters may offset carbon uptake during the growing season under future climatic conditions. Heat waves are also interlinked with other factors that affect forest health, such as soil acidification and foliar nutritional imbalances. This study did not address evaporation, water stress, or precipitation caused by heat waves. The influences of these variables should also be considered in further studies to explore the overall impacts of heat waves on photosynthetic inhibition.

Heat waves are particularly relevant because climate extremes are expected to occur more often in the near future. There were reported losses of up to 0.06–0.5 PgC from terrestrial net carbon uptakes during the European heat waves in 2003 and 2018 [47]. This is equivalent to 6–50% of the annual anthropogenic CO₂ emissions of the 28 member countries of the European Union at the 2015 level. Furthermore, the record-high increment in the atmospheric CO₂ concentration during 2015–2016 was primarily due to photosynthetic inhibition (2.5 ± 0.34 PgC) from terrestrial ecosystems in response to the drier and hotter conditions related to the 2015–2016 El Niño event [48]. Therefore, the policy target of reducing anthropogenic CO₂ emissions might be difficult to accomplish due to negative carbon cycle feedback on photosynthetic activities from the land sink in a more extreme climatic regime [49]. Similarly, in this study, despite the reduction in anthropogenic CO₂ emissions (−1.58% per year) and increases in NDVI (1.66% per year) and LAI (2.16% per year) in Europe, it was found that the atmospheric CO₂ is constantly increasing (0.60% per year) and the FPAR is decreasing (−10.18% per year). Therefore, unexpected photosynthetic inhibition caused by heat waves’ indirect or direct effects can weaken the effort to reduce anthropogenic CO₂ emissions.

This study averages MODIS indicators into monthly data within individual GOSAT XCO₂ grids. In the case of the LST, PSNnet, and GOSAT level 4 XCO₂ have more significant numbers than the other variables. This can cause the overestimating the influences of these variables on GWR models. Hence, Min-Max Normalization ($\mathrm{X}-\mathrm{MIN}/\mathrm{MAX}-\mathrm{MIN})$ was applied for rescaling the values of dependent and independent variables into a range of [0, 1]. GOSAT Level 4 XCO₂ uses the National Institute for Environmental Studies (NIES) transport model (TM; collectively NIES-TM) for inversion GOSAT XCO₂ data with prior CO₂ flux of 2.5° by 2.5° spatial resolutions. Owing to relatively low spatial resolution of CO₂ monitoring satellite data, there is a growing number of literature conducting the reconstruction and simulation of atmospheric CO₂ by modeling the correlation between satellite XCO₂ and various higher spatial resolution environmental MODIS data (NDVI, NPP, LST LAI, air temperature, wind speed, and direction) [50,51]. A high-resolution inversion model (0.1° × 0.1°), named NTFVAR (NIES-TM–FLEXPART-variational), has been recently developed to overcome the limitations of coarse resolution in the existing inversion model [52]. Therefore, further study is required regarding the impact of downscaling the spatial resolution of MODIS indicators (independent variables), which are fitted with GOSAT Level 4 XCO₂, on the GWR model. The NTFVAR inversion model is expected to offer the GWR model’s better explanatory power than this study.

5. Conclusions

European countries have faced repeated heat waves, as indicated by the continuous breaking of temperature records during the most recent decade. However, there have been few studies on the concept of repeated heat wave-induced photosynthetic inhibition in Europe. This study explores the spatial cross-correlation of GOSAT CO₂ concentration with repeated heat wave-induced photosynthetic inhibition in Europe from 2009 to 2017 by utilizing GWR. It is noted that GOSAT CO₂ concentration has a significant correlation with MODIS photosynthetic indicators, such as LST, LAI, NDVI, and FPAR. The indirect effects of sudden climatic events (higher heating demands, drought, etc.) have influenced terrestrial carbon uptake activities, leading to decreases in local R² and R². Therefore, this study could serve as a valuable reference for employing repeated heat wave-induced photosynthetic inhibition as a fundamental factor for mitigating carbon emissions in Europe.