Monitoring drought impacts on crop productivity of the U.S. Midwest with solar-induced fluorescence: GOSIF outperforms GOME-2 SIF and MODIS NDVI, EVI, and NIRv

Highlights

• SIF with finer resolution improves the monitoring of drought impacts on crop.

• Finer resolution SIF can more accurate assessment of crop anomalies and yield.

• SIF is more sensitive to water and heat stress conmpare with VIs and NIRv.

• SIF enables more accurate assessment of crop seasonal cycles.

• C4 plants are better adapted to drought stress than C3 plants.

Abstract

The frequency and severity of drought are increasing in the context of global warming. Elucidating the responses of crop productivity to drought is essential for informing agricultural management and ensuring food security. Here we used satellite-derived solar-induced chlorophyll fluorescence (SIF) data and vegetation indices to evaluate the impacts of the 2012 drought on crop productivity in the U.S. Midwest. We used SIF from the global, OCO-2 based SIF product (GOSIF; 0.05°, 8-day), GOME-2 SIF product (0.5°, monthly), and three MODIS-derived vegetation indices (NDVI, EVI, and NIRv). We compared the seasonal cycles and anomalies of SIF and VIs from 2008 to 2018. We also examined to what extent these proxies could capture the variations of gross primary production (GPP) for eddy covariance flux sites. SIF and VIs were able to capture the seasonal cycle in drought and normal years. SIF better captured the photosynthesis changes due to water and heat stresses than vegetation indices. In particular, GOSIF data with the finer spatio-temporal resolution was a good monitor of crop response to drought. Crop yield decreased by 25% in the 2012 drought relative to the multi-year mean, while GOSIF, NDVI, EVI, and NIRv reduced by 22%, 4%, 10%, and 8%, respectively. GOSIF had the strongest relationship with crop yield (R² = 0.91), followed by NIRv (R² = 0.89), EVI (R² = 0.68) and NDVI (R² = 0.48). Compared to the crop yield data, the mean difference of the yield estimates based on GOSIF, EVI, and NIRv were 379.32, 328.43, and 503.67 kg/ha, respectively. For both corn and soybeans, yield anomalies were better correlated with GOSIF anomalies than with NIRv and EVI anomalies. Our study demonstrated that SIF with finer spatio-temporal resolution has great potential for monitoring the responses of crop productivity to drought.

Introduction

Drought is one of the most widespread disasters in the world that endanger natural and managed ecosystems. Drought can substantially reduce gross primary production (GPP) and thereby the carbon sink of terrestrial ecosystems. Droughts can bring irreversible loss to the fragile and sensitive agro-ecosystems (Lobell et al., 2014; Wang et al., 2016; Xu et al., 2019), and can also cause widespread crop mortality and an increase in the frequency of pests and diseases, further reducing crop yield. For example, the 2012 flash drought in the U.S. Central Great Plains, the most severe drought since 1930 (Otkin et al., 2016; Wolf et al., 2016), led to agricultural losses amounting to $20 billion (Kam et al., 2014). What's more, global warming also increases the frequency and severity of drought (Boisier et al., 2015; Cook et al., 2015). Therefore, it is essential to monitor the responses of crop growth condition and crop yield to drought at regional to global scales (Wang et al., 2019).

The eddy covariance (EC) technique is generally considered as the most accurate method for estimating GPP at the ecosystem scale (Baldocchi et al., 2001). The EC flux towers directly observe NEE (Net Ecosystem Exchange), and GPP can be derived from NEE using nighttime-based or daytime-based partitioning method (Reichstein et al., 2005). However, the EC techniquet can only estimate GPP at the ecosystem scale and reveal the effects of drought on vegetation within the footprint of the EC tower. Uneven and sparse distributions of EC flux towers make it challenging to monitor drought effects on ecosystem productivity at large spatial scales (Qiu et al., 2020b). In contrast, the remote sensing technique is well suited for monitoring the effects of drought on ecosystems over broad spatial domains (West et al., 2019). The occurrence of drought is generally accompanied by a decrease in precipitation and an increase in temperature. When plants are subjected to water and/or heat stress, photosynthesis or productivity usually declines. The changes in plant productivity can be detected by satellite-derived vegetation indices (VIs). Numerous studies have shown that VIs such as the normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) are able to monitor the growth status of vegetation and estimate crop yield (Huang et al., 2019; Labus et al., 2002; Lee et al., 2017; Potgieter et al., 2007; Satir and Berberoglu, 2016). However, VIs usually indicate the greenness or chlorophyll content of vegetation canopy and do not accurately capture rapid changes in vegetation photosynthesis due to water and heat stresses (Smith et al., 2018). A recently proposed new VI, near-infrared reflectance of terrestrial vegetation (NIRv, the product of NDVI and NIR reflectance) (Badgley et al., 2017) was found to be strongly correlated with GPP over different spatial and temporal scales (Wang et al., 2021; Wu et al., 2020). Several studies have also demonstrated the ability of NIRv to estimate crop yield. In contrast to NDVI, both EVI and NIRv do not saturate in areas with high density vegetation cover (Baldocchi et al., 2020). It is worth noting that NIRv estimates the global GPP with high accuracy without the input of additional environmental variables (Badgley et al., 2019). Hence, NIRv can be used as a proxy for GPP to explore photosynthesis of vegetation.

Recent satellite observations of solar-induced chlorophyll fluorescence (SIF) have provided a promising technique to measure plant photosynthesis from space (Frankenberg et al., 2011; Frankenberg et al., 2014; Joiner et al., 2013; Joiner et al., 2011; Li and Xiao, 2022; Li et al., 2018a; Sun et al., 2017). Sunlight absorbed by plants is partly used for photosynthesis and partly dissipated as heat via non-photochemical quenching (NPQ), while a very small fraction of the absorbed energy is emitted as SIF (Baker, 2008). Therefore, SIF is a by-product of the photosynthesis process of vegetation and is physiologically related to GPP (Joiner et al., 2011; Li et al., 2018b; Yang et al., 2018). This also means that SIF may be sensitive to water and heat stresses which affect photosynthesis. Currently, several satellite missions, including SCIAMACHY (Joiner et al., 2016), GOSAT (Greenhouse gases Observing SATellite) (Frankenberg et al., 2011; Joiner et al., 2011), GOME-2 (The Global Ozone Monitoring Experiment-2) (Joiner et al., 2013), OCO-2 (Orbiting Carbon Observatory-2) (Frankenberg et al., 2014) and TROPOMI (TROPOspheric Monitoring Instrument) (Kohler et al., 2018) provide global SIF data at different spatial and temporal resolutions. SIF is increasingly used to estimate GPP, and has been shown as a stronger proxy of GPP than VIs (Damm et al., 2010; Frankenberg et al., 2011; Guanter et al., 2012; Li et al., 2018b; Parazoo et al., 2014; Smith et al., 2018). More and more studies have also shown that SIF has a high sensitivity to environmental stress, and therefore has a strong potential for detecting vegetation phenology and diagnosing the responses of ecosystems to water and heat stresses.

Previous studies have shown that satellite-derived SIF from GOSAT and GOME-2 captured the changes in photosynthesis under drought conditions for different regions such as the Amazon (Lee et al., 2013), Europe (Wang et al., 2020) and U.S. Great Plains (He et al., 2020; Sun et al., 2015). These studies have demonstrated the potential of SIF in monitoring vegetation photosynthesis and its response to drought. However, the SIF data used in these studies have coarse spatial and temporal resolutions (e.g., 0.5° and monthly), which may lead to a weak ability to capture the timely photosynthetic changes in vegetation, especially for heterogeneous regions (Qiu et al., 2020a). For example, GOME-2 provides SIF data with a spatial resolution of 40 km × 80 km (40 km ×  40 km after July 2013) (Koehler et al., 2015), while the spatial resolution of GOSAT SIF data is 10 km in diameter (Joiner et al., 2012). OCO-2 provides SIF observations with much smaller footprints (1.3 km × 2.25 km), but has sparse coverage across the globe (Qiu et al., 2020a; Shi et al., 2021). These disadvantages to some extent hinder the application of these SIF data in carbon cycle studies from the ecosystem scale to the global scale. To address this issue, researchers have developed SIF data with high temporal and spatial resolutions using machine learning method (Li and Xiao, 2019; Zhang et al., 2018). For example, the global, OCO-2 based SIF product (GOSIF) (Li and Xiao, 2019) has finer resolutions (0.05°, 8-day) compared to the original OCO-2 SIF data, and it also has continuous global coverage and a longer time period (2000 to present), which allows us to better explore the responses of vegetation photosynthesis to drought at different spatial and temporal scales (Li et al., 2020). In addition, the sensitivity of crop NDVI, EVI, NIRv, and GOSIF to drought is not clear at present. Moreover, the yield of crops is closely related to photosynthesis. When subjected to prolonged water and heat stresses, the photosynthetic capacity of crops decreases, which may also lead to lower crop yield. Previous studies have shown that coarse-resolution SIF has a great potential in detecting heat stress in wheat in a timely manner and in assessing the impact of drought on wheat yield (Song et al., 2018), while it is not clear to what extent the fine-resolution SIF data can help.

In this study, we explored the performance of the GOSIF data with finer spatial and temporal resolution in monitoring the variations of crop productivity in response to drought at different spatial scales. We selected the widely reported 2012 drought in the U.S. Midwest as the representative drought event. We compared the performance of GOSIF with satellite-derived VIs (NDVI, EVI, and NIRv) and coarse-resolution GOME-2 SIF in capturing the changes of crop productivity over the course of the seasonal cycle under drought conditions. We hypothesized that (1) SIF performs better than VIs because SIF is considered to contain environmental information related to photosynthetic light use efficiency (Li et al., 2018b; Yang et al., 2015); (2) GOSIF performs better than the GOME-2 SIF because GOSIF is based on high-quality OCO-2 SIF data and also has finer spatial and temporal resolution than GOME-2 SIF.