*3.4. Cross Validation and Comparison of Results*

Using the relationship between NBR and FVC we estimated the erosion risk directly from NBR using the established methods as presented above, the coefficient of determination (R2) from the erosion estimates between FVC and NBR reached 0.8346 (Figure 8). It implies that the commonly used NBR index in wildfire studies can be used as a substitute for FVC to estimate the C-factor values and the erosion rates along with other RUSLE factors.

Figure 9 compares the rainfall erosivity and the final erosion estimation from the BoM daily rainfall gridded data (5 km) and the radar Rainfields data (15 min, 1 km). Though the source datasets are very different in resolution and measurement, there is a high correlation between them (R<sup>2</sup> = 0.7562), implying that the radar Rainfileds data can be directly used in estimating erosivity.

**Figure 8.** Relationships of fractional vegetation cover (FVC) estimated from MODIS and normalized burn ration (NBR) from Sentinel-2: (**a**) FVC and NBR relationship; (**b**) estimated erosion rates using FVC and NBR.

**Figure 9.** Estimated erosion rates (Mg ha−<sup>1</sup> month<sup>−</sup>1) from BoM daily rainfall gridded data and radar Rainfields data for January–February 2020 over Sydney drinking water catchment.

#### **4. Discussion**

Vegetation cover (or C-factor) and rainfall (or R-factor) are the two dominant factors affecting the post-fire hillslope erosion. Vegetation cover, which is important in reducing the impact of rainfall, was not significantly lower (about 10%) after the wildfire based on the monthly FVC products [25–27]. Over 80% of the soil surface was still covered by either PV or NPV, both can protect soil from erosion. This agrees well with the study in another national park in Australia [40,41].

Compared to the mean values in February for all years (2000–2020), C-factor value only increased 1.5 times (due to wildfire) but the rainfall erosivity increased about 7 times in February 2020 across the SDWC area. This suggests that rainfall has a far larger impact than groundcover factor, on hillslope erosion after wildfires. This is in line with a similar study in the Blue Mountains catchment using the eWater toolkit (Source) model which reported six times higher sediment load at the downstream outlet after extreme wildfire in 2001 [42]. The post-fire soil erosion is mainly limited by rainfall erosivity which agrees with similar studies [43–45].

There was a prolonged dry period before the 2019–2020 wildfires with less than 300 mm rainfall over the SDWC area for 10 months from April 2019 to January 2020, leading to large quantities of fuel loads. In February 2020 the rainfall amount reached 337 mm month−1, with several intense storm events between 6–13 February. The severe hillslope erosion at SDWC was mainly caused by these extreme rainfall events in February 2020. For example, the erosion rate on a single day (9 February 2020) reached 3.2 (Mg ha−<sup>1</sup> day<sup>−</sup>1) which contributed to 65% of the total monthly erosion (4.9 Mg ha−<sup>1</sup> Month<sup>−</sup>1) in February 2020.

Despite the increased erosion risk, WaterNSW successfully maintained the supply of safe water to the treatment plants, by proactively managing the water supply system configuration (through sources selection and offtake depth changes) preventing the inflows from the fire-impacted catchment entering the supply during these storm events.

Soil erodibility (K) and slope-steepness (LS) factors are relatively stable compared to the dynamic C-factor and R-factor. Wildfires may alter the soil properties including soil structure, texture, permeability, and soil organic carbon which are all related to the K-factor as discussed in one of our previous studies [46]. The extent of fire effects on these soil properties depends on fire intensity, fire severity, and fire frequency [47] which is complicated and beyond the scope of this study.

The terrain factor also contributed significantly to the post-fire erosion. SWDC area has a rugged terrain compared with many other parts of NSW. The mean slope is about 17% over the SDWC area, and 29% in the Warragamba Dam area, much higher than the state average (6%). Because of the steep terrain, the mean RUSLE LS-factor value at SDWC is 4.7 which is about 2.6 time higher than the State average (1.8). The LS-factor value at the Warragamba Dam area is even higher (7.2) which is about 4 times of the State average. This implies that the erosion risk at our drinking water catchments are likely to be 2.6–4 times higher than the rest of the State regardless of other erosion factors. This also implies the importance of maintaining good vegetation cover and erosion control practices in the drinking water catchment area.

RUSLE, when appropriately used, can produce meaningful information on relative hillslope erosion risk at a given time. The event-based rainfall erosivity estimation relies on the availability of high-resolution rainfall data. The weather radar Rainfield data (15-min, 1 km) are adequate for estimating EI30 index at daily or sub-daily scale. Such data sets will be increasingly important because of the higher likelihood of intense storm events and fire frequency under the changing climate with warmer and drier conditions.

In a GEE environment, various remote sensing data can be searched and used to estimate the FVC and RUSLE C-factor. The widely used NBR in fire severity mapping and FVC has close correlation and can be used as a surrogate to FVC for C-factor estimation in erosion modeling.

As the erosion model has been applied at such high spatial (5–30 m) and temporal (daily or hourly) resolutions, the impacts of fires on soil erosion can be explored at finer landscape scales including individual hydrological catchments, drainage units, or even paddocks. This helps to locate the high

erosion risk areas or sediment sources to assist prioritized management practices. Accurate and effective decision-making is best enabled by linking with detailed and timely information on erosion to identify the impacts of rain events after the fires.

However, process-based studies to understand the factors controlling surface runoff and erosion, particularly in relation to aspects of the fire regime are still required to precisely predict the sediment transport and deposition, especially in the waterways.
