Fire and Water in the Banksia Woodland

Pyrohydrology? Fire ecohydrologists? The feedbacks between water, ecosystems and fire are complex and fascinating, and were enough to keep me, Gabrielle Boisrame, Scott Stephens, Katya Rakhmatulina and Octavia Crompton busy in California for many years. So in coming home to Australia - a fairly pyrophilic sort of a place - I was keen that our group should investigate how fire disturbances impact the hydrology of local ecosystems. This blog post unwraps a new paper that does just that.

New Holland Honey-Eater on a Banksia menziesii - one of the main species in Perth's Banksia Woodlands.

Enter Long Nguyen, who has spent his PhD determined to reveal the implications of fire for the water use of Banksia woodlands. This is an interesting place to start, because the Banksia woodlands of the Swan Coastal Plain are the major water consumers over Perth city's Gnangara Mound - the city's most important groundwater resource, and one which has been significantly depleted in the past 50 years. They are also an ecosystem that is regularly burned with existing fire management policies aiming to achieve a fire return interval of approximately 6 years. What does this regular pattern of disturbance mean for water use?

There are two possible sources of information that are sufficiently distributed within the Banksias to answer this question. One is a network of monitoring bores in the shallow aquifer that can provide insights into whether groundwater trends are altered by fire occurrence. This data source is valuable because it addresses groundwater responses directly, but it is sparse relative to the scale of the woodlands (some 400,000 ha in area). Because the unsaturated zone can also be very deep in the Banksia woodlands, and these trees have very deep roots, it is also possible that non-trivial changes in water use by the ecosystem might affect soil moisture without having notable impacts on groundwater.

The other source of information is remotely sensed estimates of evapotranspiration (ET). These are attractive because they are spatially continuous and scalable, and because we anticipate that changes in ET are likely to be the major hydrological impact of fire. However, the validity of existing ET products for Banksia woodlands was not well understood - under either normal conditions or after fires.

Fortuitously, the TERN Supersite at Gingin Banksia Woodland experienced a fire within 50% of the measured area of the eddy covariance tower in 2016. This provided us with a chance to see what the changes in evaporation were in the burned and unburned areas after the fire, and then to test whether any remote sensing ET products were able to reproduce these changes.

Panels (a,b) show the location of the study area, (c) shows the wind rose, indicating the bias towards southerly winds, (d) shows the monthly timeseries of ET from the eddy covariance tower at Gingin, (e) illustrates the changes in NDVI following the 2016 fire, and (f) illustrates the footprint of fluxes at the tower under typical northerly and southerly wind conditions.

The challenge in comparing the burned and unburned ET is that the burn was south of the tower, while the wind at Gingin blows from the south about 70% of the time. This meant we couldn't simply compare ET under northerly and southerly wind conditions, because it was very likely that there would be bias in those conditions. For example, if the northerly winds blew mostly at night, we would definitely see less ET under those conditions - not because there was no fire in the north, but because there is less evaporation at night.

To get around this problem, Long split the ET measurements into northerly/southerly wind conditions, and then used machine learning based on ambient environmental conditions to gap fill these timeseries so that we had estimated complete ET timeseries from each side of the tower. This ensured we had no more biases and we could compare them directly.

Conceptual illustration of how biased sampling could cause incorrect estimations of ET differences between the north and south footprints, making estimation of the fire impact difficult.

Illustration of the monthly timeseries of ET estimated from the Northern (no fire) and Southern (fire affected) footprints following gapfilling of the eddy covariance dataset.

Long firstly used the raw ET dataset from the tower to compare annual ET estimates against multiple remotely sensed ET products, finding that the best performing option was the PML data product.

He also used a differences in differences approach to isolate the effect of the fire on ET which of the ET products was best able to replicate these effects - again finding the PML was the best performing product. PML estimated that after 1 year the fire had reduced ET by 61 mm, which was similar to the 56 mm estimated from the eddy covariance.

Overall, Long strategically made use of a ‘natural experiment’ to learn that fire induced a 10% reduction in ET for one year, that ET was indistinguishable from pre-fire conditions thereafter, and that the PML product was the most robust remote sensing option for estimating fire impacts on ET. Clearly, we need more exploration of fire impacts on ET in the Banksias for different burn severities, occurring in different climatic contexts (wet/dry years) and landscape settings (shallow/deep groundwater) to better understand the impacts of fire on the hydrology of the Banksia woodlands - and with a suitable remote sensing product in hand, this goal is much more achievable.