Increased photosynthesis during spring drought in energy-limited ecosystems

Drought can strongly affect terrestrial carbon cycling by limiting gross primary productivity (GPP), and with droughts expected to intensify under climate change, understanding GPP responses is crucial. GPP is governed by the balance between water and energy limitations, with ecosystems varying along a water–energy limitation spectrum. In water-limited systems, productivity generally increases with precipitation, whereas in energy-limited systems, reduced precipitation can coincide with higher temperature and increased light due to reduced cloud cover, potentially enhancing GPP. Prior studies have shown instances of increased GPP during droughts in tropical and high-elevation systems, and occasionally in mid- and high-latitude regions, but comprehensive, observation-based assessments across large scales and seasons are lacking. This study examines how aridity (wetness index, WI) modulates GPP sensitivity to meteorological drought across Northern Hemisphere ecosystems (>30° N) using long-term eddy covariance observations, and evaluates whether these sensitivities are captured by terrestrial biosphere models (TBMs) and satellite remote sensing products. The central research questions are: (1) how does aridity relate to GPP sensitivity to meteorological drought across seasons; (2) can we predict where and when GPP will increase or decrease during meteorological drought; and (3) how well do TBMs and satellite products reproduce observed GPP sensitivity to precipitation compared to eddy covariance observations.

Previous work shows drought often reduces GPP via soil moisture deficits and high vapor pressure deficit, leading to stomatal closure and reduced carbon assimilation. However, theory and observations indicate that in energy-limited or light-limited systems (e.g., tropical rainforests, high elevations), reduced precipitation can coincide with increased solar radiation and warmer temperatures that alleviate energy limitation and enhance photosynthesis. Several regional events (e.g., 2012 US drought; 2018 and 2022 European droughts) demonstrated increased vegetation activity in spring or at high elevations under warm, sunny conditions. Studies also document that interannual variability of GPP is largely driven by water availability locally, but the balance between water and energy limitation varies across biomes and seasons. Model assessments indicate TBMs often impose positive GPP–precipitation relationships and may be overly sensitive to precipitation relative to temperature. Remote sensing GPP products incorporate spectral proxies and meteorology but typically lack explicit soil moisture constraints, potentially underestimating sensitivities in water-limited regions.

Data: Long-term, non-cropland eddy covariance (EC) sites >30° N were compiled from FLUXNET2015, ONEFlux-Beta, ICOS Warm Winter 2020, and ICOS Drought 2018. Monthly data (gap-filled, ONEFlux processing) were aggregated to seasons: spring (MAM), summer (JJA), fall (SON). Seasons with >50% gap-filled GPP were removed; years with severe disturbance were excluded. Sites with ≥10 years per season were retained (spring n=61, summer n=62, fall n=63). Seasonal GPP sums (GPP_sum), precipitation sums (P_sum), PAR sums (PAR_sum, estimated as 0.5×SW_IN_F), and mean air temperature (Ta_mean) were computed. Where significant GPP trends existed (Mann–Kendall p<0.05), GPP was detrended per season using Sen’s slope. Meteorological variables used site measurements when available, otherwise ERA-Interim fills as per FLUXNET. Sensitivity estimation: For each site and season, simple linear regressions quantified sensitivities (slopes) of GPP_sum to P_sum, PAR_sum, and Ta_mean, respectively. Multivariate linear regressions without interactions assessed partial sensitivities: GPP_sum = βp·P_sum + βTa·Ta_mean + βPAR·PAR_sum + β0. Standard errors of slopes were calculated. Additional analysis used seasonal mean Palmer Drought Severity Index (PDSI) from TerraClimate (∼4 km) to estimate GPP sensitivity to PDSI. Aridity characterization: The wetness index (WI = mean annual precipitation / potential evapotranspiration) from TerraClimate 1981–2010 climatology at ~4 km was used to classify sites as water-limited (WI<0.65) or energy-limited (WI≥0.65). Alternative thresholds (0.5, 1.0) were tested. Sensitivities were regressed against WI to assess how aridity mediates GPP–precipitation relationships per season. TBM comparison: TRENDY v6 S2 monthly outputs (1901–2016) were regridded to 0.5°. For 1992–2016 and for grid cells containing EC sites, seasonal GPP sensitivities to precipitation, Ta, and PAR were computed per model, using consistent meteorological drivers (CRU-NCEP v8; precipitation pr, air temperature tas, rsds converted to PAR from CABLE inputs). Model SDGVM was excluded due to unrealistic behavior after 2007. A TRENDY ensemble mean sensitivity at each site was computed. Remote sensing products: Seasonal GPP from MODIS (MOD17 Terra; MYD17 Aqua), GOSIF-GPP (0.05°), and FLUXCOM (RS-only V006 at 0.083°, and RS+METEO ERA5 at 0.5°) were extracted for site locations (time windows aligned with 2000–2016 or 1992–2016 for FLUXCOM RS METEO). Sensitivities were estimated analogous to TBMs, using CABLE meteorological inputs for consistency. Upscaling and mapping: WI was aggregated to 0.5° by spatially aggregating TerraClimate precipitation and PET. The empirical EC-derived spring GPP–precipitation sensitivity vs WI regression was applied to the WI grid to generate a Northern Hemisphere map (>30° N) of expected spring sensitivities. Differences between TRENDY mean sensitivities and EC-derived sensitivities were mapped. Land cover from MODIS MCD12 (0.5°) masked non-vegetated areas; statistics summarized by IGBP classes using middle 95% of sensitivities.

• Eddy covariance observations show that during spring meteorological drought, energy-limited ecosystems consistently increase GPP: 83% of energy-limited sites (n=38/46) had negative GPP sensitivity to precipitation.
• Water-limited ecosystems predominantly had positive spring GPP sensitivity to precipitation (80%; n=12/15).
• The wetness index (WI) explained nearly half of the cross-site variability in spring GPP sensitivity to precipitation (R²=0.47, p<0.001); relationships were weaker in summer (R²=0.16, p=0.002 excluding IL-Yat) and fall (R²=0.30, p<0.001).
• Seasonal contrasts: few energy-limited sites increased GPP with summer meteorological drought (28%, n=13/46); fall responses were mixed (54%, n=25/46). Water-limited sites nearly always increased GPP with more summer precipitation (94%, n=15/16) and often in fall (71%, n=12/17).
• Partial sensitivities in spring (controlling for PAR and Ta): 61% of energy-limited sites still had negative GPP sensitivity to precipitation, whereas 93% had positive sensitivity to air temperature, indicating warm springs strongly enhance GPP and drier springs can also contribute secondarily.
• EC mean spring GPP–precipitation sensitivity: energy-limited sites −0.22 ± 0.05 g C m−2 per mm; water-limited sites 0.57 ± 0.17 g C m−2 per mm.
• TRENDY ensemble mean energy-limited sensitivity was near zero and not significant (−0.03 ± 0.03 g C m−2 per mm; p=0.78, n=46), underestimating the negative sensitivity seen in EC. For water-limited sites, TRENDY mean (0.54 ± 0.11) did not differ significantly from EC (p=0.87, n=15).
• Remote sensing GPP products captured negative spring sensitivities in energy-limited sites (e.g., MODIS-Aqua −0.33 ± 0.05 to FLUXCOM RS METEO −0.02 ± 0.01) but underestimated sensitivities in water-limited sites (e.g., MODIS-Aqua 0.11 ± 0.07 to FLUXCOM RS 0.19 ± 0.06 g C m−2 per mm).
• Upscaling indicates 55% of vegetated Northern Hemisphere land (>30° N) likely increases GPP during spring meteorological drought, whereas TRENDY mean estimates only 36% with negative sensitivity.
• Spatial differences show TRENDY underestimates negative sensitivities in many energy-limited regions (eastern North America, parts of Europe, and eastern Asia). By PFT, EC sensitivities are more negative than TRENDY for ENF, DBF, mixed forests, woody savanna, and wetlands; grasslands show large model variability.

Findings demonstrate that the water–energy limitation framework explains diverse ecosystem GPP responses to meteorological drought across seasons. In spring, many energy-limited ecosystems increase GPP during precipitation deficits because concurrent increases in temperature and light alleviate energy limitation while plant-available water remains sufficient. This explains previously reported event-scale observations and reconciles apparent contradictions in the literature. In contrast, water-limited systems exhibit the expected positive GPP–precipitation relationship. Seasonal dynamics matter: summer and fall responses in energy-limited regions are less consistent, likely due to depleted soil moisture, carry-over effects, and potential structural overshoot after spring enhancement. The comparison with models and satellite products highlights a gap: TBMs in the TRENDY ensemble generally lack the observed negative spring GPP sensitivity to precipitation in energy-limited ecosystems, implying they may overemphasize direct precipitation controls and underrepresent co-occurring benefits of increased radiation and temperature during drought. Remote sensing products better capture energy-limited behavior but underestimate sensitivities where soil moisture constraints dominate. These insights are relevant for improving projections of carbon–water cycle interactions, seasonal carbon uptake, and drought risk propagation across the growing season.

This study provides an observation-based, mechanistic framework linking ecosystem aridity (wetness index) to GPP sensitivity to meteorological drought across seasons in the Northern Hemisphere. It shows that spring drought commonly enhances GPP in energy-limited ecosystems and quantifies where this occurs (∼55% of vegetated lands >30° N), while confirming positive GPP–precipitation sensitivities in water-limited regions. Remote sensing products partially capture these dynamics, but TBMs generally underestimate the negative spring sensitivity to precipitation in energy-limited systems. Key contributions include: (1) a robust spring relationship between WI and GPP–precipitation sensitivity (R²=0.47); (2) an upscaled map of expected sensitivities; and (3) a model–data comparison diagnosing deficiencies in TBMs. Future research should incorporate improved representation of energy–water interactions in TBMs, explicitly include soil and groundwater availability and carry-over effects, leverage emerging water-availability datasets (e.g., SMAP, GRACE, Evaporative Stress Index), and examine how warming-induced phenological changes and snowpack dynamics may shift sensitivities regionally and seasonally.

• Geographic and data scope: Analyses are restricted to non-cropland EC sites >30° N with ≥10 seasonal years, limiting representativeness for the Southern Hemisphere and croplands (irrigation excluded). Flux tower footprints may not represent surrounding heterogeneous landscapes.
• Metric simplifications: The wetness index (WI) captures broad aridity but omits many local controls (soil texture, rooting depth/bedrock water, microclimate, plant diversity/structure). Considerable unexplained variance remains in the WI–sensitivity regression.
• Temporal treatment: GPP was detrended per season when trends were significant, while meteorological variables were not detrended due to limited climatological record lengths; residual temporal autocorrelation or trends could influence sensitivities.
• Model–data mismatch: TBMs operate at 0.5° resolution versus site-scale EC footprints; years were not matched between EC and model datasets (focus on climatological sensitivity trends), potentially affecting direct comparability.
• Drought metric choices: Precipitation was the primary meteorological drought indicator; PDSI analyses showed similar but weaker relationships. Soil moisture was not directly measured at scale.
• Seasonal-to-annual inference: Spring GPP increases do not imply higher annual net carbon uptake due to potential reductions later in the season and increased respiration; lag effects and structural overshoot can reverse gains.