Unexpected contribution of cloud seeding to NPP increase during drought

The net primary production (NPP) is defined as the net accumulation of organic matter per unit of time and space due to the photosynthesis of green plants (Yu et al. 2009). The NPP is calculated by simply subtracting the ecosystem respiration (RE) from the gross primary production (GPP). The GPP is the total productivity of plants' photosynthesis, while the RE represents the total energy consumption during the process of respiration. That is, the NPP can also be explained as the total energy accumulated in plants by photosynthesis, subtracted by the energy used for their growth and maintenance. The NPP is also used to explain the exchange of CO₂ between the atmosphere and ecosystem (Prentice et al. 2001). Traditionally, the NPP is an important indicator to find the sink in the carbon cycle of an ecosystem, and is also used as an indicator to represent food productivity (Zhang et al. 2016; Jiao et al. 2018; Li et al. 2019; Wang et al. 2019). The change in the NPP is also frequently used to evaluate the effect of climate change (Zhang et al. 2018), where the data generated based on several IPCC scenarios are used to estimate the possible change of carbon balance in the future ecosystem (Ito 2010).

The NPP is basically a part of the ecosystem's carbon balance, which has been analyzed by applying various methodologies. The biological measurement method (Son et al. 2004), carbon-flux measurement method based on the Eddy-covariance technology (Saigusa et al. 2005), a method based on bio-physical modeling (Ito & Oikawa 2002; Sasai et al. 2007), and method based on the analysis of remote sensing (RS) data, including satellite observations (Running et al. 2004), are among those methods. Each method has its own advantages and disadvantages, along with its unique characteristics, like spatial-temporal resolution, data structure, and data availability. For this reason, it is not easy to conclude the superiority of any one method over others. As the measurement of NPP is a difficult, expensive and time-consuming job, it is not surprising that the use of RS data is becoming preferable (Hadian et al. 2019).

Several methods are available for NPP estimation. Based on Pei et al. (2018), these models may be divided into three groups: the climate relative model, process model, and light use efficiency (LUE) model. The climate relative model includes the Miami Model (Lieth & Whittaker 1975), the Thornthwaite Memorial Model (Zhu 1993), and the Chikugo model (Zhou et al. 1998), etc. The process model includes the Integrated Terrestrial Ecosystem model (Raich et al. 1991), the Biosphere Simulator model (Foley et al. 1996), the Boreal Ecosystem Productivity Simulator (Liu et al. 1997), the Biome–BGC model (White et al. 2000), etc. The LUE models include the Carnegie–Ames–Stanford Approach (CASA) (Potter et al. 1993; Field et al. 1995), carbon fixation model (C − Fix model; Veroustraete et al. 1994, 2002), and MODIS NPP (Running et al. 2004). Among these, the CASA model simulates the carbon exchange processes in the terrestrial biosphere and atmosphere (Cramer et al. 1999). This model has been successfully applied to estimate the regional, continental, and global scale NPP (Jay et al. 2016). For example, Potter et al. (2006) applied the CASA model to estimate and evaluate the NPP in the US with the available RS data.

Different from the RE, photosynthesis is very sensitive to light intensity, CO₂ concentration, and temperature. Generally, photosynthesis is proportional to the light intensity and CO₂ concentration. However, there must be some threshold values. Above these thresholds, photosynthesis stagnates. For example, photosynthesis increases much up to the temperature of 40 °C; after that, it also decreases much (Crafts-Brandner & Salvucci 2002). It is also important to understand that each factor works as a limiting factor. For example, photosynthesis is not active in low-temperature conditions, even though the CO₂ concentration is high. It is also the same in the opposite situation.

Along with the factors mentioned above, the soil moisture content is also an important factor in controlling photosynthesis (Jones 2004; Pereira et al. 2006). The effect of rainfall, as a source of soil moisture, on photosynthesis has also been evaluated in many studies (Hao et al. 2010; Xu & Wang 2016; Linger et al. 2020). For example, Hao et al. (2010) showed that the CO₂ exchange in the temperate steppe is very dependent on the rainfall amount and timing. Xu & Wang (2016) confirmed the positive effect of rainfall on the increase of NPP. The effect of rainfall seasonality on the temporal variability of NPP was also found in this study. A similar result was also derived by Linger et al. (2020) for the estimation of NPP in tropical forests.

The relative role of temperature and rainfall has also been an interesting issue. For example, Sun et al. (2018) compared the effect of land cover change, riding CO₂, temperature increase, and water condition change on the GPP. Their results show that, even though the GPP change at middle and high latitudes is dominated by the temperature rise, the NPP change in the low latitude is more governed by the water condition. A similar study in India (Bejagam & Sharma 2022) also shows that the NPP is more dependent on the increase of rainfall rather than the temperature under climate change situations. Zhang et al. (2022) compared the relative roles of temperature and rainfall on the NPP in the Dajiuhu Basin, China. Their finding indicates that the temperature was the more dominant factor on the NPP during the wet summer season, but the rainfall played a key role in the rather dry spring and fall seasons. They also mentioned that the overall contribution of rainfall on the NPP has recently become more significant than the temperature.

Water stress must have a negative impact on the NPP (Zhang et al. 2021), and thus the soil moisture content shows a strong positive correlation to the NPP (Yue et al. 2022). As a result, in the drought period, the NPP becomes significantly smaller (Arneth et al. 1998; Nanzad et al. 2021). Churchill et al. (2015) also showed that the drought has more impact on the vegetation abundance in the peatland than sustained flooding. MODIS GPP/NPP products are also generated by considering the soil water stress (Mu et al. 2007). Obviously, after the drought is over, the NPP recovers fast. Proper water supply to the plants during a drought can thus be very helpful to increase the NPP.

This study focuses on the possible increase of the NPP due to the increased rainfall by cloud seeding. Cloud seeding is generally tried to secure additional water resources (Silverman & Sukarnjanaset 2000; Maryadi et al. 2015; Pokharel et al. 2021). However, as the affecting range of cloud seeding is wide, increased rain by cloud seeding cannot easily be collected. On the other hand, it simply moistens the neighboring forest. This study focuses on the unexpected contribution of cloud seeding to the increase in the NPP. The dependency of the NPP on soil moisture must be primary information to derive the estimation of the NPP increase. The Boryeong Dam basin is considered a study area, which is a small dam known to be very vulnerable to drought. The CASA model is used in the estimation of the NPP. As a result, it may also be possible to estimate the increase of CO₂ reduction, as the NPP can directly be converted into carbon accumulation.

The CASA is a model for estimating the NPP in a region. As the CASA model is an algorithm based on the physical theory of carbon circulation processes on the land surface soil and vegetation, it is easy to apply for the estimation of the NPP and carbon accumulation. The input data of the CASA model can also be easily prepared with some satellite information. The CASA model is composed of three sub-models: soil moisture, NPP, and soil carbon and nitrogen. In particular, the sub-model NPP is designed to use the LUE to calculate the NPP (Monteith 1972). The LUE represents the ratio of absorbed solar radiation to be used in plant photosynthesis to produce organic matter. The LUE is also a key factor in estimating the light absorption and carbon absorption of plants.

To apply the CASA model, it is required to use data, such as the total solar radiation (SOL), the water stress coefficient (WSC), and the fraction of absorbed photosynthetically active radiation (FPAR). Temperature stress factors are also important, which include the temperature at which photosynthesis is most active (temperature 1; ⁠), the temperature at which the light use is most efficient (temperature 2; ⁠), and the maximum possible efficiency of photosynthesis. Among those data, the FPAR and the maximum possible efficiency can be derived by RS data analysis. T_ɛ1 and T_ɛ2 can be derived by interpolating the temperature data on the land surface. The SOL, the most important driving force of the CASA model, can be estimated by the Angstrom–Prescott equation (Prescott 1940), or the solar radiation flux (SolarFlux) model (Rich et al. 1995). It is also possible to use the measured data on the Earth.

The WSC, another driving force of the CASA model, indicates the available moisture content. The WSC can be estimated by the ratio of the actual to potential evapotranspiration amount, factors which are included in the soil moisture sub-model of the CASA model. This sub-model requires soil information like soil type and soil depth, as well as meteorological data like temperature and precipitation amount. However, it is also possible to use the FAO Penman–Monteith equation (Allen et al. 1998) to directly estimate the potential evapotranspiration amount, and the complementary relationship of evapotranspiration of Bouchet (1963) or the Pike equation (Pike 1964) to calculate the actual evapotranspiration amount. The choice is dependent upon the available ground observation data. As the CASA model requires continuous data in time and space, it generally requires interpolation, which can be another source of error.

The maximum, minimum, and optimal temperatures of some plant types can be found in Raich et al. (1991). In particular, for the deciduous as the representative plant in the study area, the maximum, minimum and optimal temperatures are 48, 0, and 26 °C, respectively. Also, the land surface temperature (LST) available in the MODIS data was used for the temperature T in the above equation.

The LSWI in this equation has the range of −1 to 1, while indicates the maximum during the growing period.

This study collected the MODIS product (https://modis.gsfc.nasa.gov/data/dataprod/) and daily radiation data at the weather station near the Boryeong Dam from the Korea Meteorological Administration (KMA; https://data.kma.go.kr/). These data were all applied to the CASA model to calculate the NPP. MODIS sensors are installed in the Terra and Aqua satellites launched for the Earth Observing System (EOS). MODIS sensors include a total of 36 spectral bands, which provide detailed information about the earth's environment, including the land, ocean and atmosphere (Vermote & Vermeulen 1999; Wan 1999). The MODIS data used in this study are the surface reflectance (SR), LST, and fraction of photosynthetically active radiation (FPAR). The SR data have a resolution of 500 m, which is provided at intervals of 8 days (d) from January 1. The FPAR data are also provided with the same spatial resolution and time interval. However, the LST data have a spatial resolution of 1 km and are provided in the same way.

The calculation of the 8-day NPP was also repeated for the years 2019 and 2020. However, in these 2 years, the annual sums of the 8-day NPPs were found to be far smaller than the annual MODIS NPPs. This underestimation was mainly due to the very small daily radiation values, missing LSTs and very small FPARs during rainy days. It is true that while rainfall may decrease the daily radiation, LST and FPAR, the rainy day does not lead to extremely small radiation, LST or FPAR.

Based on the findings in Figure 4, the somewhat abnormal values of daily radiation, LST, and FPAR were corrected. First, the extremely small or zero daily radiation values were replaced by the neighboring normal values. The mean decrease of daily radiation in Figure 4(a) was used to correct those small or zero values. Second, the missing LST values on rainy days were filled by the normal values in neighboring cells. Finally, abnormally small FPAR values were also replaced by the normal value observed before or after the rainy day in the same cell. However, even after correcting these factors, the annual sum of the calculated 8-day NPPs was still much smaller than those annual MODIS NPP values. In the year 2020, the difference was more than 20%. However, this study used these calculated 8-day NPPs for further analysis. At this time, this calculated 8-day NPP is the maximum information the authors could obtain.

It is also possible to find the effect of rainfall on the GPP and NPP. When the rainfall amount is large, both the GPP and NPP drop significantly. It is normal that the daily radiation decreases during rainfall, as it is reflected by the rain cloud. Decreased daily radiation results in decreased photosynthesis. However, it recovers fast after the rain stops.

However, in the years 2019 and 2020, there were many abnormal values, which were all corrected by the method explained in the previous section (Figures 5(b) and 5(c)). The original NPP values are marked by empty triangles, and the corrected ones by solid triangles. After correcting these abnormal values, the overall pattern of the 8-day NPPs was found to become more similar to that of the 8-day GPPs. The effect of rainfall was also found to be more reasonable, as in the case of the year 2018.

Additionally, in this study, the ratio between the annual GPP and annual NPP was derived to check if the calculated 8-day NPPs in this study were acceptable or not. Based on Zhang et al. (2009), the ratio of global NPP to global GPP fluctuated around 0.51–0.52 in the years from 2000 to 2003. The annual variation was not so large. However, as can be seen in Table 1, the ratio of the annual sum of 8-day NPPs in this study to the annual sum of 8-day GPPs was far smaller than 0.5. This ratio is near 0.5 if considering the annual MODIS NPP. The estimates in this study were smaller than the MODIS NPP. In particular, the NPP in the year 2020 was much smaller.

Soil moisture is very dependent upon rainfall. A large amount of rainfall increases the soil moisture a lot. But, after several days, the soil can be drier as the water in the soil percolates into the groundwater. When calculating the infiltration amount in a rainfall–runoff analysis, the antecedent 5-day rainfall amount is generally considered to determine if the soil condition is dry, normal or wet (Kang & Yoo 2020). The maximum antecedent 3-month rainfall is sometimes considered when evaluating agricultural drought (Kim et al. 2011). In this study, the possible dependency of 8-day NPP on the soil moisture was evaluated using three different rainfall data: the first is the rainfall amount accumulated for the current 8-day period, the second accumulated for the last 8-day period, and the third accumulated for the last two 8-day periods (i.e., the last 16 days). The use of the 8-day interval was to consider the 8-day NPP and 8-day GPP. Also, as the soil depth in Korea is very shallow, generally less than 1 m (Kwon et al. 2021), and the soil is easily dried up, the last 16 days (i.e., two 8-day intervals) was assumed to be sufficient antecedent days to estimate the soil moisture condition.

The above-derived results are dependent on the time scale. In cases where the monthly data were analyzed, the vegetation condition was found to be highly correlated with the 1- or 2-month-lagged rainfall. For example, García et al. (2010) investigated the lagged effect of rainfall on the vegetation greenness in the evergreen forest in central California. They confirmed that the monthly NDVI was highly correlated with the 1-month-lagged rainfall. Davenport & Nicholson (1993), Richard & Poccard (1998) reported a similar result that the monthly NDVI data were highly correlated with the 1- or 2-month-lagged monthly rainfall in the regions of Eastern and Southern Africa. Chandrasekar et al. (2006) also showed that the monthly NDVI in India was highly correlated with the 2-month-lagged rainfall.

When considering different time scales, such as the 8-day interval in this study, a few different results were also derived. For example, Li et al. (2013) showed that there was an 8- to 16-day lag time between the rainfall event and the NDVI response in the temperate desert region of northwest China. Derived by analyzing the 8-day data, this result is very close to the result in this study. Interestingly, Li et al. (2013) also suggested that the threshold rainfall for a large NDVI response could be about 30 mm. A similar analysis but with the daily data by Fan et al. (2016) showed that there was a lag of 10.2 ± 6.5 days in the desert steppe between the NPP and rainfall. Their suggestion of a rainfall threshold of 80.2 mm is also similar to that considered in this study. Regardless of the different climate conditions, the threshold rainfall seems to be around 50 mm, such as derived in this study.

This study estimated the increase of NPP due to the increase of the last 16-day rainfall by cloud seeding. Even though the role of cloud seeding is limited, it was assumed possible to make the last 16-day rainfall be 50 or 100 mm. In 2018, 2019, and 2020, the number of cases with the last 16-day rainfall amount <50 mm was found to be 27, 22, and 29, respectively, while those within the range between 50 and 100 mm were 11, 8, and 7, respectively. However, as the NPP increase is minimal in winter, early spring, and late fall, this study considered only the growing season from May to October. As a result, in 2018, 2019, and 2020, the cases considering the cloud seeding with the target rainfall amount of 50 mm became smaller, to 11, 4, and 8, respectively. In 2018, 2019, and 2020, additional cases to be considered for cloud seeding with the target rainfall amount of 100 mm were 3, 6, and 4, respectively. Table 2 summarizes the possible number of cases for cloud seeding for NPP increase.

Table 3 shows the effect of cloud seeding on the increase of NPP. As explained earlier, by cloud seeding, the rainfall amount during the last two 8-day intervals was assumed to be increased to meet the target rainfall of 50 or 100 mm. On average, in the case where the target rainfall is 50 mm, the cloud seeding should increase the rainfall by 36.27 mm. This is just 29.17 mm to satisfy the target rainfall of 100 mm for those cases with a rainfall amount between 50 and 100 mm. If considering the 16-day period for cloud seeding, the 16-day rainfall of 50 mm must be an easy target to achieve.

In 2018, the possible increase of NPP with the target 16-day rainfall of 50 mm was estimated to be about 20.60 g/m² of carbon. This amount corresponds to a 3.33% increase in the annual NPP. In 2019 and 2020, it was 8.91 and 14.02 g/m² of carbon (1.47 and 2.67%), respectively. If considering the additional cases with the target rainfall of 100 mm, then the increase of NPP in 2018, 2019, and 2020 becomes 26.22, 22.28, and 21.02 g/m² of carbon (4.24, 3.66, and 4.00%), respectively. On average, by cloud seeding with the target 16-day rainfall of 50 mm, about a 3% increase in annual NPP was found to be available, while with the target 16-day rainfall of 100 mm, about a 4% increase in annual NPP was found available.

In fact, this small increase in NPP corresponds to a large amount of carbon accumulation. In the case of considering only the Boryeong Dam basin, the amount of carbon accumulation can easily be calculated by multiplying the carbon accumulation per unit area by the basin area. In 2018, the NPP per unit area was 0.62 kg/m²/year of carbon, which if considering the basin area of 163.6 km², corresponds to the carbon accumulation of 101,292,940 kg/year in the Boryeong Dam basin.

It is also possible to derive the increase of carbon accumulation in the Boryeong Dam basin. As the carbon accumulation is proportional to the NPP, it is easy to estimate the increase in carbon accumulation. In 2018, in the case that the target 16-day rainfall amount is 50 mm, the carbon accumulation in the Boryeong Dam basin can be increased up to 104,663,591 kg/a. In the case of considering additional cases with the target rainfall amount of 100 mm, the carbon accumulation becomes 105,582,859 kg/year. Similar to the results in Table 3, in the years 2019 and 2020, about a 4% increase in annual carbon accumulation could also be expected.

Similar examples confirm that about a 4% increase in NPP by cloud seeding seems fairly realistic. In fact, several studies have reported that additional rainfall can contribute much to increasing the NPP. Those results were derived through simulation methods or field experiments. As an example of a simulation study, Shi et al. (2014) showed, in their analysis with the terrestrial ecosystem model (TECO), that NPP was reduced significantly by extreme droughts (67% reduction of annual rainfall) at all four study sites located within the Great Plains, USA. The sensitivity of NPP to drought was also found to be directly attributable to rainfall amount. Wilcox et al. (2015) compared the above- and below-ground NPP after increasing the rainfall amount by 50%. They noticed that larger rainfall events contribute much to increasing the NPP. Heisler-White et al. (2008) also showed that the increase in rainfall amount by 15% in a semi-arid grassland increased the NPP by about 6% on average. Especially, a few larger rainfall events were found to contribute more than many small rainfall events.

Even though just a few cases are available, field experiments also derived similar results. For example, Chou et al. (2008) reported a 4-year experiment result at the Sierra Foothill Research and Extension Center in California. Their result shows that there was a significant positive relationship between the NPP and annual rainfall that about a 5% increase of NPP was expected with the annual rainfall increase by 10%. Also, in a long-term rainfall manipulating experiment in the northern Chihuahuan Desert, USA, Báez et al. (2013) showed that chronic drought had a higher impact than water addition. During the drought, the vegetation cover was significantly decreased, but water addition slightly decreased the vegetation coverage.

About a 4% increase in NPP by cloud seeding can have another important meaning in Korea. The national target for carbon reduction in 2030 of Korea is 40% of the total carbon emission in 2018. Based on the National Greenhouse Gas Statistics in Korea (MOE 2022), the total amount of greenhouse gas emissions in 2018 was 727,000,000 ton-eq/year of CO₂, which, if considering only the carbon, also corresponds to 198,272,727 ton/year. That is, the target amount of carbon reduction becomes 107,067,273 tons/year.

It was shown that by applying cloud seeding, the amount of carbon accumulation can be increased. If considering the target rainfall of 50 and 100 mm during the last 16 days, the increased carbon accumulation could be 3,373 and 4,292 tons/year, respectively. These increased amounts are just 0.0017 and 0.0022% of the total carbon emissions in 2018. However, if simply extrapolating this result to the entire forest in Korea (i.e., 62,900 km²), the increased carbon accumulation can be extended up to 1,295,740 and 1,649,238 ton/a, respectively, depending on the target rainfall amount. These amounts correspond to 0.65 and 0.83% of the total carbon emissions in 2018. Or, these amounts correspond to 1.21 and 1.54% of the target amount of carbon reduction in 2030.

Cloud seeding is generally tried to secure additional water resources, but, due to its high spatial variability, it is not easy to collect the increased rain. On the other hand, the increased rain moistens the neighboring forest. This study focused on this situation and estimated the possible increase of NPP due to cloud seeding. The Boryeong Dam basin in Korea was considered a study area, which is a small dam basin that is highly vulnerable to drought. The CASA model was considered in the estimation of the NPP. The results may be summarized as follows:

First, the 8-day NPP was calculated by the CASA model. The annual sum of 8-day NPPs calculated in this study was about 10% smaller than the annual MODIS NPP in the year 2018. However, the calculated annual NPPs in this study for years 2019 and 2020 were found to be much smaller than the annual MODIS NPPs. This was mainly due to some abnormal factors observed on rainy days. These abnormal factors were corrected by considering the effect of the rainfall amount and the corrected factors were then used to re-estimate the 8-day NPPs.

Second, the current 8-day NPP was found to be increased with the rainfall amount during the last two 8-day intervals (i.e., the last 16 days). It was especially clear that the increase of the current 8-day NPP was greater when the rainfall amount during the last two 8-day intervals was ≥50 mm. The mean increase of the 8-day NPP was estimated to be about 1.873 g/m² of carbon.

Third, the possible increase of NPP with the target 16-day rainfall of 50 mm was estimated to be about 20.60 g/m² of carbon in 2018. This amount corresponds to a 3.33% increase in annual NPP. In 2019, it was rather small at 1.47%, and, in 2020, it was 2.67%. If considering additional cases with a target rainfall of 100 mm, then the increase of NPP became 26.22 g/m² (4.24%) of carbon in 2018, and 3.66 and 4.00% in 2019 and 2020, respectively. On average, about a 3% increase in annual NPP was found available by cloud seeding with the target 16-day rainfall of 50 mm, and about a 4% increase in annual NPP was found available with the target 16-day rainfall of 100 mm.

The above-mentioned result can be directly converted into the increase in carbon accumulation due to cloud seeding. If simply extrapolating the derived result to the entire forest in Korea (i.e., 62,900 km²), the increased carbon accumulation could be extended to 0.65 and 0.83% of the total carbon emission in 2018. These amounts also correspond to 1.2102 and 1.54% of the target amount of carbon reduction in 2030 in Korea. It is obvious these last estimates could not be achieved easily, mainly due to the large coverage. However, when economically evaluating cloud seeding, it is, at least, necessary to consider this unexpected benefit. It is also worth considering that if the spatial coverage is rather small, the 16-day rainfall of 50 mm is an easy target to achieve.

C.Y. designed the research, interpreted the results and prepared the draft manuscript. M.L. analyzed the data and performed the computational work. K.C. revised the draft manuscript and added comments on the major findings of this study.

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (No. NRF-2021R1A5A1032433) and the National Institute of Meteorological Sciences (NIMS) through ‘Development of Application Technology on Atmospheric Research Aircraft (KMA2018-00222)’.

All relevant data are available from an online repository or repositories. (Satellite data: https://modis.gsfc.nasa.gov/data/dataprod/ (MODIS product); Meteorological data: https://data.kma.go.kr/ (KMA)).

The authors declare there is no conflict.