Site description
Warra Supersite, (Lat: 43° 5′ 42ʺ S; Long: 146° 39′ 16ʺ E) is located on a floodplain of the Huon River within the Warra Long Term Ecological Research site (https://warra.com/) 60 km southwest of Hobart, Tasmania. The forest at the Supersite is a Eucalyptus obliqua tall forest with a canopy height of 50–55 m, overtopping a 15–40 m tall secondary layer of rainforest and wet sclerophyll tree species. Ferns dominate the ground layer. The forest is very productive with an aboveground biomass of 790 tonnes/ha16 and a leaf area index of 5.7 m2/m247.
The Supersite is within the Tasmanian Wilderness World Heritage Area (TWWHA). That part of the TWWHA experiences infrequent, but sometimes intense, wildfire. Except for a small proportion of mature (> 250 years-old) E. obliqua trees, the current forest resulted from seedling regeneration following the last major wildfire in that part of the landscape in 1898. No timber harvesting has ever been done in the forest at the Supersite.
The climate at Warra is classified as temperate, with no dry season and a mild summer48. Mean annual rainfall measured at the nearby Warra Climate Station (Bureau of Meteorology Station 097024) is 1736 mm and the mean daily temperature is 14 °C and 5.6 °C in January and July, respectively. The soil at the site is a Kurosolic Redoxic Hydrosol16.
Analysis of historical heatwaves in southern Tasmania
Daily maximum temperature records from the Bureau of Meteorology station at Cape Bruny Lighthouse (station number 94010) were extracted from the Bureau of Meteorology’s online climate data portal (http://www.bom.gov.au/climate/data). Cape Bruny Lighthouse is one of the 112 stations in the ACORN-SAT network of Australia’s reference sites for monitoring climate change49. The station provides a record of daily maximum temperature measurements commencing in 1923 and spanning almost a century. It is the southern-most station in the ACORN-SAT network; is 59 km south-east of the Warra Flux Site; and bounds the south-eastern extent of E. obliqua tall forest in Tasmania.
Missing temperature measurements represented less than 0.6% of the 35,795 records collected at Cape Bruny Lighthouse between January 1st 1923 and December 31st 2020. The missing measurements were gap-filled using predicted values calculated from linear regression models constructed from measurements made at nearby Bureau of Meteorology stations (listed in order of proximity to Cape Bruny Lighthouse and priority for gap-filling)—Cape Bruny Automatic Weather Station (1997-present), Hastings Chalet (1947–1987) and Hobart-Ellerslie Road (1892-present).
Average, standard deviation and 90th percentiles of daily maximum temperature were calculated for each calendar day of the year. Further analysis of heatwaves was restricted to the period between the beginning of August and the end of February. This period bounds the growing season of the forest at the Warra Supersite when there is normally a net carbon gain by the forest (Wardlaw unpublished data). Heatwaves were identified as three or more consecutive days with maximum temperatures that met or exceeded the 90th percentile value sensu Perkins and Alexander9. For each heatwave event that was identified, the following three statistics were calculated: (1) average daily maximum temperature during the heatwave; (2) summed departures (as standard deviations) from average daily maximum temperature during the heatwave; (3) summed departures (as standard deviations) from average daily maximum temperature of the 21 day period centred on the middle day of the heatwave. The November 2017 heatwave, as described by these three statistics, was ranked against all the other heatwave events identified between 1923 and 2020 at Cape Bruny Lighthouse. In addition, the z-score was calculated to measure the magnitude of the departure of the average daily maximum temperatures during the November 2017 heatwave from the long-term average of this 21-day period. Those statistics were also calculated for the same period in 2016.
Weather conditions at Warra Supersite during the 2017 warm spell
Four attributes of weather were used to describe the November 2017 warm spell—air temperature, vapor pressure deficit (calculated from temperature and relative humidity), incoming shortwave radiation and soil moisture. Air temperature and relative humidity were measured using an HMP155A probe (Vaisala, Finland) and incoming shortwave radiation was measured using a CNR4 radiometer (Kipp and Zonen, The Netherlands). Both instruments were mounted 80-m above ground level at the top of the Warra Flux tower. Data was processed to 30-min averages and logged onto a CR3000 datalogger (Campbell Scientific, Logan, USA).
Soil moisture was measured by time-domain reflectometry using two CS616 soil moisture probes (Campbell Scientific, Logan, USA) each installed at a depth of 20 cm. These probes were installed in two pits approximately 40 m west of the tower. Soil moisture data were processed to 30-min averages and logged onto a CR1000 datalogger (Campbell Scientific, Logan USA).
Turbulent fluxes at Warra Supersite during the November 2017 warm spell
Measurement of turbulent fluxes (carbon, water and energy) were done by eddy covariance (EC) using a closed-path infra-red gas analyser (Model EC155, Campbell Scientific Inc., Logan, USA) to measure CO2 and H2O concentrations and a 3-D sonic anemometer (Model CSAT3A, Campbell Scientific Inc, Logan, USA) to measure turbulent wind vectors and virtual air temperature. The sonic anemometer and infra-red gas analyser were mounted at 80-m above the ground at the top of the Warra Flux tower. Storage of CO2 and H2O beneath the forest canopy was measured by a profile system (Model AP200, Campbell Scientific Inc, Logan, USA ), with sampling heights of 2, 4, 8, 16, 30, 42, 54, 70 m. Temperature sensors in aspirated shields (Model 110-ST, Apogee Instruments, Logan, USA) were co-located with the CO2/H2O sample inlets of the profile system. High frequency (10 Hz) measurements of turbulent fluxes were processed to 30-min averages in a datalogger (Model CR3000, Campbell Scientific, Logan USA). High frequency (2 Hz) of CO2 and water concentration measurements were processed to 15-s averages sequentially for each profile sample height in a datalogger (Model CR1000, Campbell Scientific, Logan, USA). Thus, each inlet was sampled for a 15 s interval every 2 min. The rate at which sub-canopy storage of CO2 changed was calculated from changes in the quasi-instantaneous (2-min) vertical profile concentrations beneath the tower at the beginning and end of each 30-min flux averaging period using the method of McHugh50.
Soil heat flux (SHF) was measured to enable calculation of energy balance that was needed to partition energy fluxes into latent and sensible heat. SHF was measured using five SHF plates (Model HFP01SC, Hukseflux, Delft, The Netherlands) inserted in the soil at depth 8 cm adjacent to the two pits in which the soil moisture probes were installed. Each of the five SHF plates were allocated to one of the two soil pits in a 2–3 split. Changes in soil temperature was measured by an averaging thermocouple (Model TCAV, Campbell Scientific Inc, Logan, USA) inserted into the soil above each SHF plate at depths of 2 and 6 cm. Soil moisture measurements at 20 cm depth were as described previously. Heat flux, soil temperature and soil moisture data were processed to 30-min averages on a datalogger (Model CR1000, Campbell Scientific Inc, Logan, USA).
Raw 30-min flux, CO2 storage and climate data were processed by the standard OzFlux QA/QC processing stream51 using PyFluxPro Version 1.0.2 software. Fluxes (carbon and energy) adjusted for storage were computed at the mid-stage (level 3). At the final stage of data processing (level 6), gap-filled net ecosystem exchange (NEE) data were partitioned into gross primary productivity (GPP) and ecosystem respiration (ER) using the u*-filtered night-time CO2 flux records to calculate ER with the SOLO artificial neural network algorithm as described in51. The standard conventions of the global flux network were adopted in partitioning NEE as described in52.
The full period between 10 and 30th November 2017 was defined as the November 2017 warm spell. The climate and fluxes measured during this period were compared with measurements of those made during the same calendar days of the preceding year, 2016. The carbon fluxes measured in the 10 weeks before (1 September–9 November) and the month after (1–31 December) the 2017 warm spell period were also compared with the same periods in 2016. This was done to ascertain whether changes in carbon fluxes during the 2017 warm spell we not due to differences in antecedent weather conditions and, whether or not differences in carbon fluxes arising from the warm spell persisted after the warm spell.
Data analysis
For each day of the 10–30 November period, daily sums were calculated for measurements of carbon fluxes and incoming shortwave radiation (Fsd), while daily averages were calculated for air temperature, VPD and soil moisture. Quantile plots, done for Ta and VPD, used 30-min data during daytime hours (when Fsd > 0). The significance of differences in measurements during the 10–30 November period among the two years of each variable were tested by analysis of variance. Tests were first done to confirm the data for each variable were normally distributed and between-group variances were homoscedastic. Log-transformation was used to correct skewness in the VPD data. Soil moisture data were strongly skewed, and transformation was unable to correct. For this variable, the Kruskal–Wallis method was used to test the significance of differences in medians among the two years. These analyses were repeated for the 10 weeks (1st September–9th November) leading up to the warm spell and the 4 weeks (1st–31st December) following the warm spell to examine antecedent conditions and subsequent recovery from the warm spell, respectively.
The energy fluxes were examined for evidence of coupling between GPP, transpiration and canopy conductance. Closure of the energy balance was first determined for the two periods to ensure comparability of the energy fluxes for the 2017 warm spell period and the corresponding period in 2016. This was done by firstly resampling the 30-min data and calculating 2-hourly averages of latent heat flux (Fe), sensible heat flux (Fh), net radiation (Fn) and ground heat flux (Fg), then fitting linear regressions of Fe + Fh against Fn–Fg for dates encompassing the warm spell in each of two years. Peak energy storage of the biomass, Fb, in the forest at Warra was estimated as 40 W m−2 using the method described in17. That estimate used the value of LAI of 5.72 based on the average of periodic measurements of LAI at Warra reported in47 and the value of 22.0 for the quadratic mean radius at breast height (1.3 m) calculated from tree measurements in a 1-ha plot adjacent to the Warra Flux tower (detailed in47). The ratio of energy storage in the biomass and ground heat flux at their respective daily maxima was calculated, assuming their respective diurnal peaks coincided. This ratio was then applied to each 2-h average of ground heat flux measured in the warm spell period in 2017 and the corresponding period in 2016. Available energy was recalculated using the formula Fn–(Fg + Fb). Analysis of variance was used to test the significance of differences between the 2017 warm spell and the corresponding period in 2016 of each component energy fluxes (Fn, Fe, Fh and Fg) for each of the twelve, 2-h periods, in the day. Kruskal–Wallis rank test was used if a variable had a non-normal distribution or exhibited heteroscedasticity. The Bowen ratio, which is the ratio between Fh and Fe, was calculated for each 2-h period during daytime hours. The 2-h average data were non-normal and heteroscedastic so testing the significance of differences in daytime Bowen ratio between the warm spell and comparison period was done using 2-sample t-test with unequal variance.
Latent heat flux was converted to evapotranspiration by dividing the measured latent heat flux by the latent heat of vaporisation of water. Evapotranspiration was used as a proxy of transpiration on the assumption that evaporation was a minor component of evapotranspiration in the tall E. obliqua forest at Warra based on measurements of soil and litter evaporation in similar forests by23. An estimate of total canopy conductance of sunlit leaves, Gt, was calculated from transpiration (E) and vapor pressure deficit, VPD, using the Skelton et al.53 adaptation of the method developed by Hogg and Hurdle54, whereby:
$${text{G}}_{{text{t}}} = (upalpha /1000){text{E}}/{text{VPD}}$$
The atmospheric pressure of water vapor, α, is equivalent to ρwGvTk, where ρw is the density of water (c 1000 kg m−3), Gv is the universal gas constant for water vapor (0.462 m3 kPa kg−1 K−1) and Tk is air temperature (in K = Ta + 273.15). Gt (in mmol m−2 s−1) was calculated for each 2-h period during the 2017 warm spell and the same calendar days in 2016 using each period’s corresponding values of E, VPD and Ta. Records were excluded if rain fell during the 2-h period. The significance of differences in daytime canopy conductance between the 2017 warm spell and the 2016 comparison period was tested using a two-sample t-test as the data were strongly skewed.
The diurnal patterns of GPP, ER and canopy conductance were compared with incoming shortwave radiation, air temperature and vapour pressure deficit. Each 30-min record of the six variables was recoded to its corresponding 2-h time interval. Analysis of variance was used to test for significance of differences between the warm spell and comparison period for each of the twelve 2-h diurnal periods of the six variables. Kruskal–Wallis rank test was used in the data were non-normal or heteroscedastic. Time series plots of diurnal 2-hourly averages for each of the six variables were plotted and visually compared.
Source: Ecology - nature.com
