Comparing the impacts of wildfire and meteorological variability on hydrological and erosion responses in a Mediterranean catchment

Land degradation and water resources pollution caused by catastrophic wildfires is of growing concern in fire‐prone regions. Studies on the effects of wildfire on hydrology and erosion have mostly been conducted at plot or hillslope scale, while relatively few studies investigated post‐wildfire hydrological responses and erosion at the meso‐catchment scale (~ > 10 km2) in the Mediterranean. This study used measured discharge and suspended sediment at the outlet of a burnt catchment in southern Portugal, before and after a wildfire, to investigate post‐wildfire changes in hydrological and erosion responses to rainfall. Hydrological and sediment connectivity patterns were derived to investigate changing dynamics induced by the fire within the catchment. The main findings were: (a) although a large part of the catchment experienced moderate to high severity burning, post‐wildfire hydro‐sedimentary response was considerably limited; (b) meteorological variability determined hydrological responses and erosion more strongly than wildfire effects; and (c) during the post‐wildfire vegetation recovery period, only rainfall events with a high return period (~ 2 years) enhanced the hydrological and erosion responses. This can be explained by the spatial scale dependency of these processes and limited fine sediment supply, or relatively low connectivity in the study catchment. While connectivity is only an indicator, this implies that, at the meso‐catchment scale, pollution of downstream water bodies by contaminated soil and ash may not occur immediately. Rather, because sediments and associated ashes and contaminants are first being transported to the areas around the stream networks, they only reach the outlet during heavy events which do connect the entire catchment. Thus, dynamic indices of connectivity that take rainfall event characteristics into consideration need to be further tested to assess and manage post‐wildfire soil and water contamination risk.


| INTRODUCTION
Wildfire is a natural component of many ecosystems. However, in recent years, the increase in the frequency, size and severity of wildfires is of growing concern in fire-prone regions around the world (Fill, Davis, & Crandall, 2019;Nunes et al., 2019;San-Miguel-Ayanz et al., 2012). As one of the most fire-prone regions in the world, European countries have experienced the worst fires in decades, such as two large fires in 2017 across central Portugal, a series of wildfires in 2018 in Greece and throughout much of Sweden, leading to loss of vegetation, property and life. With global warming, the severity and frequency of wildfires are projected to continue to increase, especially in the European Mediterranean and the latter region is experencing some of the most significant impacts of global warming (Liu, Stanturf, & Goodrick, 2010;Turco et al., 2018).
Wildfires cause land degradation and downstream water pollution. Depending on the fire severity that represents the intensity of the fire or the released energy, wildfires can, partly or completely, consume surface litter and ground vegetation, and through heating and combustion processes alter physical, chemical and biological soil properties, for example, clogging of pores by ash, decreasing organic matter, increasing soil water repellency and changing soil aggregation (Mataix-Solera, Cerdà, Arcenegui, Jordán, & Zavala, 2011;Mataix-Solera, Gómez, Navarro-Pedreño, Guerrero, & Moral, 2002;Zavala, de Celis Silvia, & López, 2014). After a wildfire, quicker and enhanced hydro-sedimentary response to rainfall including more effective rainfall, quicker and higher runoff with more sediment delivery is observed in burnt areas, which is attributed to a decrease in ground interception, infiltration rates and shorter time to runoff, resulting in soil erosion and land degradation (Moody, Shakesby, Robichaud, Cannon, & Martin, 2013;Shakesby, 2011;Shakesby & Doerr, 2006). In addition, together with sediments, associated ashes and contaminants could be transported downstream, posing a risk to water resources Robinne et al., 2018;Smith, Sheridan, Lane, Nyman, & Haydon, 2011).
This study investigated how hydrological and sediment connectivity modified by fire controls runoff and sediment delivery. Connectivity representing the linkage between system components stems from geomorphology and was applied in hydrology (Wohl et al., 2019). Hydrological and sediment connectivity can be divided into structural connectivity and functional connectivity, which are described as the degree of spatial linkage and the actual flux of water and sediment between system components, respectively (Heckmann et al., 2018). The index of connectivity (IC) proposed by Borselli, Cassi, and Torri (2008) is one of the most widely used indices of structural connectivity (Cavalli, Marchi, Goldin, Schenato, & Crema, 2015), which is calculated based on digital elevation model (DEM) and its' interaction with properties of the local land use and soil surface that affects sediment transport assigned as weighting factor. IC provides an estimate of the potential for sediment sources to reach specific targets areas. In this study, the IC was determined for both pre-and post-wildfire conditions and used as a spatial approach to understand the impacts of wildfire on hydrology and erosion processes.
Although the impacts of wildfire on hydrological and erosion responses have been extensively studied in fire-prone areas around the world at the plot or hillslope scale, only a few studies investigating postwildfire hydrological and erosion responses at the meso-scale catchment scale ( > 10 km 2 ) exist, especially for the European Mediterranean area (Estrany et al., 2019;García-Comendador, Fortesa, Calsamiglia, Calvo-Cases, & Estrany, 2017;García-Comendador, Fortesa, Calsamiglia, Garcias, & Estrany, 2017;Inbar, Tamir, & Wittenberg, 1998;Mayor, Bautista, Llovet, & Bellot, 2007;Moody et al., 2013;Shakesby, 2011;Van Eck, Nunes, Vieira, Keesstra, & Keizer, 2016). This stems partly from wildfire contingency and destruction resulting in a scarcity of measurement data, but mostly from complex non-linear hydrological and erosion responses when scaling up (Moody et al., 2013). Studies at smaller catchment scale in the Mediterranean, typically draining an area of less than 1 km 2 , observed that first-year runoff and sediment delivery following wildfire increased by more than 100-fold, and even 10,000-fold (García-Comendador, Fortesa, Calsamiglia, Calvo-Cases, & Estrany, 2017;Inbar et al., 1998;Mayor et al., 2007). However, due to the scale dependency of hydrological and erosion processes, the impacts of wildfire at larger catchment scale tends to be overestimated by plot-and hillslope-scale studies (Ferreira et al., 2008;Stoof et al., 2012). Consequently, accurate predictive assessments of fire-induced risk for water resources are hindered by the limited availability of databases on postfire hydrological and erosion effects at the larger scale (Robinne et al., 2018).
Given few catchment-scale studies and the importance of catchment dynamics for assessment of risks to soil and downstream water resources, this study was conducted at the meso-catchment scale to investigate changes in hydrological and erosion responses to rainfall before and after wildfire, and the causes of any such changes. While this topic is of importance wherever wildfire occurs, it is of particular concern in the Mediterranean region due to wildfire frequency, limited water resources and already degraded soils. Hence the study site selected is in the Mediterranean's Iberian peninsula. Our main hypothesis follows the general conclusion of the reviewed literature, that significantly increased runoff and sediment delivery would be observed after wildfires, especially for catchments experiencing severe burning.
To test this hypothesis and improve understanding of fire-induced hydrological changes and soil erosion at the scale of the catchment, runoff events were analysed for the Odeáxere catchment (18.53 km 2 ) in southern Portugal where a wildfire in early August 2003 burnt almost the entire catchment with moderate-to-high severity, and for which there was hydrological and sediment data available before and after the fire.
Multivariate statistical analysis was used to investigate hydrological and erosion responses to rainfall before and after the fire, and IC was calculated to detect the potential for the sediment to reach the outlet. The results can contribute to understanding the impacts of wildfire on hydrology and erosion processes at the meso-catchment scale, and related risks for soil and water resources.  (CNR, 2005). Few of these plans were implemented; in eucalypt forest plantations, most owners recovered burned logs and eucalypts were replanted or allowed to resprout, while there was little intervention in Mediterranean oaks and shrubland areas and vegetation was allowed to recover naturally.

| Vegetation dynamics monitoring
The normalized difference vegetation index (NDVI) is a commonly used index to reflect vegetation dynamics from satellite imagery, but it has a strong seasonal oscillation. In order to eliminate NDVI's seasonal oscillation, DNDVI following Gouveia, Páscoa, and DaCamara (2018), was employed as the indicator to monitor vegetation dynamics for the analyzed period (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010). NDVI values were retrieved from MOD13Q1 included in the MODIS Terra V6 product, which provides an adequate temporal and spatial resolution of, respectively, 250 m and 16 days (Didan, Munoz, Solano, & Huete, 2015). Subsequently, monthly NDVI time series were generated based on the pixel reliability index with good quality or marginal quality.

| Rainfall and outlet measurements and analysis
Hourly data for rainfall, streamflow, conductivity and turbidity between October 2001 and September 2006 were obtained from the Portuguese Environment Agency through the National Water Resource Information System (SNIRH, 2019). Hourly rainfall was measured in two rainfall gauges located in the north and south of the Odeáxere catchment ( Figure 1). Hourly water level was measured at the stream outlet and streamflow discharge was calculated with a stage-discharge curve also provided by SNIRH (Figure 1). At the same location, hourly water conductivity and turbidity were measured with an automatic sensor; suspended sediments were calculated from turbidity using a linear relationship (sediments = 2.1 x turbidity, r 2 > 0.99), built from monthly water samples taken from the same site and analysed for suspended sediment concentration, including during a period of high streamflow and turbidity; this data was also taken from SNIRH. Turbidity data was limited to values under 999 NTUs The intensity-duration-frequency (IDF) curve is a mathematical function that relates rainfall intensity with its duration and the return period. The regional IDF curves were constructed for the Monchique rainfall gauge (Brandão, Rodrigues, & Costa, 2001), and were used in this study to estimate the return period of selected pre-wildfire and post-wildfire rainfall-runoff events.

| Runoff events
Given two rainfall gauges located at the top and bottom of this catchment, hourly rainfall from the two rainfall gauges was averaged to calculate the rainfall characteristics of the analysed events. Maximum rainfall intensity in 30 min for each event was estimated based on relationships with maximum hourly rainfall (Brandão et al., 2001). All runoff events with total rainfall greater than 9 mm and available turbidity data were selected to be analysed. Based on the hydrograph of each event, streamflow discharge was separated into surface runoff and baseflow using an automated filter technique (Arnold, Allen, Muttiah, & Bernhardt, 1995).
Events were considered to begin with the first rainfall, and end when surface runoff was over, except when a new rainfall event started before discharge of the previous one was finished, in which case the first event was considered to end when the second one

| Statistical analysis
For each runoff event, the following meteorological variables were determined (Table 2): rainfall duration (P dura , h); rainfall amount (P, mm); rainfall intensity (IP, mm h −1 ); maximum rainfall intensity in 30 min (IP30 max , mm h −1 ); and relative time-to-maximum rainfall intensity (IPRT max ). IPRT max was calculated using time to maximum rainfall intensity divided by rainfall duration. IP30 max was calculated from maximum hourly rainfall, using existing relationships between hourly and 30 mins rainfall (Brandão et al., 2001).
The antecedent precipitation index (API) and initial baseflow (Q i , m 3 s −1 ) were also included to describe antecedent conditions prior to each runoff event. API was calculated using the equation proposed by Kohler and Linsley (1951) and based on the rainfall of the previous 10 days.
In addition, the following hydrological variables were determined for each event (Table 2): runoff duration (R dura , h); runoff (R, mm); runoff coefficient (RC); total discharge (Q, m 3 s −1 ); base discharge fraction (Q b , %); peak discharge (Q peak , m 3 s −1 ); and relative time-to-peak discharge (QRT peak ). QRT peak was calculated from time to peak discharge divided by runoff duration. The transport of sediment by discharge was characterized by suspended sediment (SS, g l −1 ) and sediment yields (SY, kg). SY is not represented in this study because peak SS was missing and SY was well related to R and Q (Pearson's correlation coefficient, p < 0.05).
Fire impacts (DNDVI and dNBR) were not added, as a preliminary analysis (not shown) did not reveal relationships with hydrological and sediment response; the classification of events as pre-fire or post-fire

| Index of connectivity
In this study, the Index of Connectivity (IC; Borselli et al., 2008) was estimated using the stand-alone freely available software   Additionally, rainfall had a strongly seasonal pattern typical of Mediteranean climates. Less than 6% of the total annual rainfall was recorded during summer while more than 41% was recorded during winter. Therefore, more discharge with more suspended sediment was recorded from November to April, mostly during the wet season during which more than 68% of the total annual rainfall occured.  between pre-and post-wildfire events (Wilcoxon, p < 0.05) ( Figure 5).

|
This indicates that although the number of runoff events was different, their characteristics were comparable.
According to the distribution of individual rainfall events on the Monchique's IDF curve (Figure 6), most of the events were medium and low-intensity events with return periods of less than 1 year in the Monchique region. Exceptions to this were two events with return periods around 2 years: pre-wildfire event E25, which had the longest duration; and post-wildfire event E38, which had the highest intensity.

| Hydro-sedimentary response
The hydrologic variables API, Q i , and Q b were similar for pre-wildfire and post-wildfire events ( Figure 5), with no statistically significant differences (p < 0.05). Unexpectedly, there was also no statistically significant difference for any of the hydro-sedimentary variables.
Nevertheless, post-wildfire events had, on average, shorter runoff duration and relatively shorter time-to-peak discharge; both can be at least partially explained by a shorter rainfall duration, although the differences were more marked for runoff (lower p values). It should especially be noted that relative time-to-peak runoff (QRT peak ) after the wildfire was lower than before the fire, despite similar values for time-to-peak rainfall (IPRT max ), indicating a change in runoff response caused by different catchment conditions. Also, post-wildfire events showed higher runoff, discharge, peak discharge and sediment concentration, despite similar rainfall amount, rainfall intensity and antecedent precipitation values. The RC was also higher for post-wildfire events compared to pre-wildfire events, even though both were low. Consequently, the distribution of individual events does not show the differences between pre-and post-wildfire on the PCA plane; most of individual events with return periods below 1 year (small events) are located on the left of the PCA plane, which led to low hydrological and erosion responses in both periods, whereas few storms with long return periods (big events) are located on the right lower quadrant of the PCA plane, which led to quicker and higher runoff generation with more suspended sediment after the fire compared with before the fire. Only these big events shows some differences between pre-and post-wildfire. As shown in Figure 7, for big events, the pre-wildfire events E2, E8, E9, E13, E18, E20, E21 and E25 are situated to the upper left side of the post-wildfire events E28, E32, E33, E38 and E39, which indicates wildfire did increase hydrological and erosion responses. The spatially explicit IC before and after the fire is shown in Figure 9, revealing clearly different patterns for pre-and postwildfire conditions. Before the fire, the relatively higher IC values occurred close to the channels and outlet, whereas, after the fire, the higher values were distributed throughout the entire catchment. In addition, the differential IC map (Figure 9c; post-wildfire IC minus pre-wildfire IC) shows that, after the fire, IC values that had increased by more than 2.5 were located in the upper parts of the catchment, while IC values increased by less than 1.5 were in the lower parts of the catchment close to the outlet. Thus, even though IC was higher post-wildfire than pre-wildfire, the largest However, compared with other Mediterranean studies at the catchment scale (García-Comendador, Fortesa, Calsamiglia, Calvo-Cases, & Estrany, 2017;Inbar et al., 1998;Mayor et al., 2007), post-wildfire hydro-sedimentary response was considerably limited in this study.

| Index of connectivity
For example, even conducting in a terraced Mediterranean catchment Estrany (2017) reported that suspended sediment were more than 100-fold higher than that in the non-burnt terraced catchment. As our study catchment is relatively large (18.53 km 2 ), the observed, considerably limited post-wildfire hydro-sedimentary response, is probably mainly caused by the scale dependency of hydrological and erosion processes, which is in agreement with Ferreira et al. (2008) andStoof et al. (2012). Indeed, enhanced erosion on the fire-affected hillslopes of the southern Monchique mountains was recorded in a photographic inventory (CMVB, 2005) and referred to in post-wildfire recovery plans (CNR, 2005), although hillslope erosion data was not collected after the 2003 wildfire. Probably, limited hillslope sediments were transported to the stream channel system, which was also found by Estrany et al. (2015) and García-Comendador, Fortesa, Calsamiglia, Garcias, and Estrany (2017) who used fallout radionuclide tracers to understand the impacts of wildfire on sediment delivery in a Mediterranean catchment. In addition, both studies suggest that wildfires can significantly enhance sediment delivery when linked with sufficient rainfall (Estrany et al., 2015;García-Comendador, Fortesa, Calsamiglia, Garcias, & Estrany, 2017), consistent with the result of our study. That is also the main reason why low impacts of wildfire on hydrosedimentary response occurred in other burnt catchments (Moody & Martin, 2009;Owens et al., 2012;Prosser & Williams, 1998).
Besides can increase infiltration by increased surface roughness, affect soil water repellency, and protect fine sediment from being transported, resulting in supply-limited conditions for erosion. By contrast, studies from the USA and Australia, also at the catchment scale, did find enhanced erosive response after wildfire (Cannon, Gartner, Wilson, Bowers, & Laber, 2008;Jackson & Roering, 2009;. However, in those areas, mostly flooding and debris flows occurred, which mainly depend on short duration, high intensity storms as drivers and the availability of fine sediment supply (Cannon et al., 2008;Smith et al., 2012), which generate sufficient transport capacity of flows . Notably, irrigation ponds in this region were not a cause of the low hydrological and erosion responses. Although the study area contains 36 ponds which, combined, account for about 10% of the drainage area of this catchment, most of them are located in the lower part of the catchment with low burn severity ( Figure 1). Keizer et al. (2015) reported low sediment yields can be explained by irrigation ponds because some of sediments were retained in ponds. However, in our case, the effect of irrigation ponds does not extend to the entire catchment. So we would refer to this as another scale issue, where the effects of irrigation ponds on smaller-scale sediment transport cannot be upscaled directly.
Finally, and arguably most importantly, as expected, limited postwildfire hydro-sedimentary response can be explained by connectivity. In our study area, both pre-and post-wildfire ICs were quite low.
Even though post-wildfire IC in our catchment increased by 20%, the fire-increased connectivity was mainly located in the upstream part of the catchment, at relatively large distances from the catchment outlet ( Figure 9c). As a result, increased overland flow that did occur post wildfire were likely absorbed in the more highly connected upper parts of the catchment and therefore did not reach the lower parts and outlets. These results suggest that IC can be a useful spatial analysis tool for assessing changes in hydrological and erosion processes, would be interesting to test more dynamic indices of connectivity which consider functional connectivity dependent on rainfall event characteristics. Alternatively, numerical modelling could help explore, in greater detail, the differences between pre-wildfire and post-wildfire small and large rainfall events in a spatiallydistributed way.

| CONCLUSIONS
We analyzed the hydrological and erosion responses to a wildfire at the meso-catchment scale ( > 10 km 2 ), for a Mediterranean catchment, using a combination of assessment of sediment connectivity before and after wildfire and PCA analysis of runoff events. This study showed that post-wildfire events had, on average, faster and higher runoff responses with a two-fold increase in suspended sediment as compared to pre-wildfire events. However, post-wildfire hydrosedimentary response were considerably limited. We also found that wildfire itself was less important for hydrological and sediment responses than meteorological variability. We conclude that, for the conditions of the region studied, wildfire-enhanced overland flow and sediment transport does occur locally on hillslopes with high burn severity, however the enhanced flow and transport will not necessarily reach the outlet of the catchment due to the scale dependency of hydrological and erosion processes and supply-limited fine soil conditions or low connectivity. Our results support the hypothesis that wildfire can enhance hydrological and sediment responses; however, the impacts of wildfire on soil and downstream water risk contamination may be overestimated when upscaling plot-or hillslope-scale studies to the catchment-scale.
That said, enhanced soil erosion on sub-catchment hillslopes during post-wildfire vegetation recovery may lead to increased chance of longer-term (slower) downstream water risk contamination, and extreme events during the same period could significantly enhance hydrological and erosion responses with more rapid downstream impact. Therefore, we suggest that testing of more dynamic indices of connectivity which consider functional connectivity dependent on rainfall event characteristics is needed. Alternatively, modelling could be used to investigate more detailed differences between the impacts of small and large pre-wildfire and post-wildfire rainfall events.
Increased recognition and consideration of multiple dynamics with catchments, for example, connectivity, can contribute to increased understanding of the likely impacts of wildfires and probable management needs. The authors would like to acknowledge Akli Benali from the University of Lisbon for providing the dNBR data, and the Portuguese Environment Agency for providing the meteorological and hydrological data via the SNIRH system.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are openly available in the Portuguese Water Resources Information System (SNIRH) at https://snirh.apambiente.pt/.