MODELLING FOREST FIRE AND POST-FIRE MANAGEMENT IN A CATCHMENT

33 Forest fires change soil surface properties, alter the hydrological processes, and increase soil erosion. Post34 fire rehabilitation measures are useful to mitigate the effect of fire on soil erosion. This work aims to predict 35 the effects of forest fires and post-fire mitigation measures on runoff and specific sediment yield (SSY) in a 36 river basin (Celone, S-E Italy). The Soil and Water Assessment Tool model, calibrated with field 37 observations, was used to evaluate runoff and SSY for the current land use (baseline) and six post-fire 38 scenarios. From 1990 to 2011, at the basin scale, the average annual SSY was 5.60 t ha y (SD = 3.47 t ha 39 y). 20% of the total drainage area showed a critical value of SSY (>10 t ha y). The effects of different 40 fire-severity levels were analysed for one year after the fire, acting on a limited area (2.3% of the total basin 41 area). At the basin scale, the post-fire effect on surface runoff was negligible for all scenarios (< 0.4%), and 42 the impact on SSY increased from 5.86 t ha y up to 12.05 t ha y. At the subbasin scale, the post-fire 43 logging scenario showed the highest increase of soil loss (SSY increased from 9.48 t ha y to 57.40 t ha y 44 ). Post-fire mitigation treatments like straw mulching and erosion barriers effectively reduced soil erosion 45 in highand moderate-severity fires (19.12 t hay and 20.93 t ha y, respectively). At the hydrological 46 response unit level, the SSY estimated for the forest in the baseline ranged from 1.18 t ha y to 2.04 t ha y 47 . It increased more than one order of magnitude for the high-severity fire scenarios and ranged from 4.33 to 48 6.74 t ha y in the very low-severity fire scenario, underlining the scale effect from the HRU to the basin 49 scale. 50


52
The Mediterranean European Region is a high fire risk area due to a combination of several factors. The high decades.  The soil erosion by water in the basin is both distributed (sheet erosion) and localised (rill erosion) (De 139 Girolamo et al., 2015). It is favoured by agricultural practices such as conventional tillage (multiple 140 operations with chisel plough and disks). The prevalent land use is for cereal growth (mostly winter and 141 durum wheat; 45% of the catchment area). Other land use includes sunflowers (9%), natural degraded areas 142 (6%), olive groves (8%), vineyards and vegetables (2%), and urban areas (1%). Forests, primarily oaks and 143 conifers (29%), cover the mountainous part of the basin. 2015) and the impact of best management practices (BMPs) on water resources (Ricci et al., 2020). 161 In SWAT, the basin is divided into subbasins that are further subdivided into HRUs, which are characterised 162 by homogeneous land use, soil, and slope. The water cycle is divided into the land phase and routing phase. 163 The components of the land phase (i.e. runoff, evapotranspiration, crop growth, soil erosion, nutrient and 164 pesticides loads entering into the main channel) and the methods used for their computation are described in 165    In the present study, the model SWAT2012 version was used. The basin was divided into 74 subbasins, 188 further partitioned into 200 HRUs. Conservation practices were not adopted in the study area (Panagos et al.,189 2015a; Wischmeier and Smith, 1978). The conservation practice factor (USLE_P) was assumed to be equal 190 to 1 for all land uses, except for forested areas where the P factor was set to 0.8. According to the crop 191 systems, the crop management factor (USLE_C) was set within 0.0019 to 0.2, as suggested by Panagos et al. 192 (2015b). The model was calibrated for the streamflow at the SV gauge over 1994-1996 and at the MP gauge over was carried out for streamflow at the SV gauge (1992). Manual calibration was performed, including the 196 above-mentioned parameters for hydrology. For sediment load calibration, the following parameters were 197 for the main channel (CH_N2), the maximum amount of sediment that can be transported from a river reach 199 (SPCON), and the exponent for calculating sediment that can be transported in the channel (SPEXP). Table  200 II shows the parameter values corresponding to the best fit for the most sensitive parameters and their range 201 of variability. 202 The model's performance was evaluated by using the coefficient of determination (R 2 ), the Nash-Sutcliffe 203 efficiency (NSE), and the observation standard deviation ratio (RSR). The simulations were considered good 204 if 0.65 < NSE < 0.75, 0.5 < RSR < 0.6 and R 2 > 0.8 and satisfactory if 0.5 < NSE < 0.65, 0.65 < RSR < 0.7 205 and R 2 > 0.5 (Moriasi et al., 2007). 206   The following six scenarios were simulated to provide a wide range of potential impacts on hydro-229 sedimentary response to support post-fire management. The model parameters influencing runoff and soil 230 erosion were properly modified for each scenario using literature values. Table III  It was assumed that "high-severity fire" was ground and canopy fire (all shrubs and herbaceous plants killed) 236 with high soil heating and alteration of soil structure (decreased infiltration and increased water repellency). 237 This scenario analysed the potential effect of removing fire-killed trees from burnt areas (logging) and the 238 successive tillage operation (chisel plough) on those areas. 239 The fire effect on soil characteristics was simulated by modifying the USLE erodibility factor (USLE_K). 240 The effect of fire on soil water repellency (Sol_K) was incorporated into the USLE_K by adopting the   (Table III). 246 USLE_P was set to 1, and USLE_C was fixed to 0.13 to mimic the effect of the regrowth of vegetation.  (Table III). In this scenario, the moderate-severity fire was hypothesised, and erosion barriers were simulated as a post-264 fire mitigation measure to reduce surface runoff and soil losses. The assumption was ground fire and burning 265 of lower tree limbs, moderate soil heating, increased water repellency and decreased infiltration. The 266 baseline value of USLE_K was assumed to increase (Table III) Table III. OV_N was assumed to be 0.3, and the baseline value of CN2 was 275 increased (+5), USLE_K was assumed to increase to a lesser extent than moderate-and high-severity fire. 276 277 Scenario Fr6: very low-severity fire and natural regeneration 278 In this scenario, it was assumed that fire had very lightly charred only fine fuel and litter on the ground and 279 soil properties (i.e. hydraulic saturated conductivity, water repellency) were unchanged. The baseline value 280 of CN2 was slightly increased (+3), and USLE_K was unchanged (Table III). 281

305
At the basin scale, from 1990 to 2011, the average annual rainfall was 777 mm (SD = 179 mm), mainly 306 concentrated from November to April (wet season), the surface runoff was 114 mm (SD = 66 mm), and the 307 total water yield was 288 mm (SD = 140 mm). Most of the rainfall (61%) was lost via actual 308 evapotranspiration (471 mm; SD = 41 mm), and the potential evapotranspiration was 954 mm (SD = 30 309 mm). The average annual SSY (sediment yield per unit of catchment area and unit of time; t ha -1 y -1 ) was 310 5.60 t ha -1 y -1 (SD = 3.47 t ha -1 y -1 ). A high inter-annual variability characterised all the water balance 311 components and the SSY due to differences in climate conditions. In the driest year (2000), the total annual 312 rainfall was 471 mm, surface runoff (SR) was about 26 mm, and the SSY was 3.03 t ha 1 y -1 . In the wettest 313 year (2009), the total annual rainfall was 1217 mm, SR was 300 mm, and SSY was 13.82 t ha -1 y -1 .

317
At the subbasin scale (Figure 4), over the period 1990-2011, the mean annual SSY was < 1.4 t ha -1 y -1 in the 318 subbasins located in the plain area (14% of total drainage area). Most of the subbasins showed values 319 between 1.4 to 10 t ha -1 y -1 , and some mountainous subbasins (20% of total drainage area)-characterised by 320

337
The reach-scale analysis for identifying the river segments where sediment deposition occurs showed that 338 most first-order river segments were under erosion. Meanwhile, sediment deposition was predicted in some 339 intermediate reaches and those located in the alluvial plains ( Figure 6). In the latter, if fire events occur in the 340 upstream areas, pollutants such as Fe, Mn, As, Cr, Al, Ba, and Pb could be deposited along the river bed, and 341 the water quality could be impaired (Smith et al., 2011).  SSY was modelled, ranging from 5.86 t ha -1 y -1 (baseline) to 12.05 t ha -1 y -1 (Fr1) ( Figure 7A). The severity 353 of the fire played an essential role in SSY. A massive difference was predicted between high-severity fire 354 (Fr1 and Fr2) and low-severity fire scenarios (Fr5, Fr6, Figure 7A). Fr5 and Fr6 showed limited increases in 355 SSY (6.4 and 6.3 t ha -1 y -1 ) compared to the baseline. The post-fire management decreased SSY compared to 356 Fr1 (8.9 t ha -1 y -1 and 7.7 t ha -1 y -1 for Fr3 and Fr4, respectively), although it was still higher than the baseline 357

363
Results at the subbasin scale showed negligible variations in surface runoff (ranging from 129.01 mm to 364 129.17 mm) for all the analysed scenarios in the subbasin 55 ( Figure 7B) compared with the baseline (129.00 365 mm). Similarly, the increase in surface runoff simulated for the subbasin 63 was negligible, ranging from 366 98.78 mm (baseline) to 98.83 mm (Fr1), and for low-severity fire simulated in the Fr5 and Fr6 scenarios, it 367 was 98.79 mm. The SSY simulated for the baseline (9.5 t ha -1 y -1 and 9.7 t ha -1 y -1 , for sub 55 and sub 63, 368 respectively) increased up to 57.4 t ha -1 y -1 (sub 55, Fr1) and up to 26.1 t ha -1 y -1 (sub 63, Fr1), confirming 369 that the high severity of fire events and the post-fire logging may produce a dramatic increase in soil loss 370 ( Figure 7B). The extension of the burnt area within the basin played an essential role in SSY variations. 371 Indeed, as a result of the larger burnt area in the subbasin 55 (56%), the SSY predicted in this subbasin was 372 much more than SSY simulated in the subbasin 63 (burnt area 19% of subbasin area), especially in the high-373 to be high and moderate, respectively. As expected, due to the lower severity of fire represented by the Fr5 379 and Fr6, SSY increased to a lesser extent in these scenarios, ranging from 11.4 t ha -1 y -1 (Fr6) to 13.3 t ha -1 y -380 1 (Fr5) in the subbasin 55, and from 10.62 t ha -1 y -1 (Fr6) to 11.57 t ha -1 y -1 (Fr5) in the subbasin 63. 381 The analysis of the potential impact of post-fire scenarios in terms of soil erosion was carried out also at the 382 HRU level. Figure 7C shows the results for the three HRUs. The SSY estimated for the baseline ranged from 383 1.18 t ha -1 y -1 to 2.04 t ha -1 y -1 . It increased more than one order of magnitude for the high-severity fire 384 scenarios, Fr1 ranged from 78.19 t ha -1 y -1 to 95.77 t ha -1 y -1 and Fr2 from 49.40 t ha -1 y -1 to 59.91 t ha -1 y -1 . 385 As expected, the very low-severity fire scenario presented the lower increase of SSY, ranging from 4.33 386 (HRU 2,55) to 6.74 t ha -1 y -1 (HRU 3,63)( Figure 7C). 387 mentioned studies, which were oriented to support ecological status evaluation, it was identified that a zero-399 flow threshold and time series of streamflow were appropriately modified. In the present study, taking into 400 account that the extremely low flow is characterised by negligible sediment transport, the discrepancy 401 between observed and simulated streamflow was considered insignificant for the research. 402

Discussion
The model performance in simulating SSY was satisfactory. Nevertheless, SSY was underestimated in the 403 extremely wet conditions and slightly overestimated in autumn, confirming the results obtained by resolution and problems linked to the transferability of the Modified Universal Soil Loss Equation approach 406 may have influenced model performances Williams & Berndt, 1977). 407 At the basin scale, over the period 1990 to 2011 that included both dry and wet years, SSY was 5.60 t ha -1 y -1 . 408 This estimate was comparable with the studies carried out in the same region by Ricci et al. (2018). At the 409 subbasin scale, SSY varied in the range 0.2-17.6 t ha -1 y -1 . 20% of the total drainage area presented SSY 410 values higher than the critical value (10 t ha -1 y -1 ). These results agree with the soil losses estimated by 411 y -1 ). These results were expected since it is well known that human activities such as agriculture and land-use 418 change have induced an important increase in erosion rates (Foucher et al., 2021;Poesen, 2018). In the study 419 area, soil losses are favoured by up and down ploughing, which is common, especially in mountainous areas. 420 It is important to remember that the dataset used for sediment calibration was limited and that measurements 421 taken at the outlet could be insufficient for optimal parameterisation. Hence, an uncertainty degree could 422 affect the results at the subbasin and HRU levels. In the present study, parameters such as USLE_P and 423 USLE_C were fixed on the literature basis and were not calibrated. A new monitoring plan with a nested 424 approach could be very useful for improving model parameterisation and SSY estimation. repellency, which is a key factor in post-fire erosion since it reduces infiltration rate, especially after high-460 severity fires, is highly variable spatially (Doerr et al., 2009;Shakesby and Doerr, 2006)  highlighted that the type of treatment (i.e. mulching or logging) did not influence the runoff generation in 482 their plots. Fr1 and Fr2 showed a dramatic increase in SSY for the three HRUs analysed, increasing in the 483 worst case (HRU 1, Sub. 55) from 1.26 t ha -1 y -1 (baseline) to 95.8 t ha -1 y -1 and to 59.9 t ha -1 y -1 , respectively 484 ( Figure 7). Malvar et al. (2017) and Wagenbrenner et al. (2015) evidenced that logging operations may 485 increase SSY mainly because of the trail generated by the passage of heavy machinery. 486 Fr5 and Fr6 showed a moderate increase of SSY that was quantified in 8.6 and 5 t ha -1 y -1 (HRU 1, Sub. 55), 487 respectively. These results agree with the Shakesby (2011) studies, which pointed out that from high to low-488 severity fire, the effect on erosion may vary from more than two orders of magnitude or may not show 489 differences at all. From the modelling point of view, the difference in SSY between Fr5 and Fr6 was mainly 490 attributable to the USLE_K factor and, to a very small extent, to CN2 (-2 in Fr6) since all the other 491 parameters were unchanged. This result confirmed the USLE_K factor as a very sensitive parameter in soil 492 loss modelling. The difference in SSY between the Fr1 and Fr2 resulted from the integrated effect of several 493 parameters (USLE_C, CN2, and OV-N) that were differentiated in the two scenarios (Table III).

Future perspectives 508
Despite the limits of the present study, the results clearly indicate that the rate of soil loss for the current land 509 use and management practices is much higher than the soil rate formation that was estimated for European 510 that those BMPs, which the Apulia Region Rural Development Programme currently supports, effectively 515 reduce soil losses but have not yet been adopted at a large scale. Several barriers still exist that limit their 516 adoption (e.g., farmers' education, lack of awareness of soil erosion). Numerous actions are needed to favour 517 the adoption of BMPs, and important public economic resources are needed to support a plan for soil 518 protection. 519 In order to address these challenges, the EU's common agricultural policy may have an important role in 520 ensuring that agriculture is in line with the soil protection principles. The new European Green Deal (EGD) 521 with the "Farm to Fork" and the "zero pollution action plan" strategies will be central in preserving soil 522 systems and biodiversity (Montanarella and Panagos, 2021). Research and monitoring may play an important 523 role in reaching the EGD's goals. 524 In the next decades, increased fire risk is expected in the Mediterranean. Watershed management will need 525 fire prevention efforts and specific actions to protect and restore the river basins before disturbance occurs. 526 95% of fires are due to human activities (i.e. agricultural practices) or negligent behaviour and arson (Vilar 527 del Hoyo et al., 2009). It is, therefore, necessary to increase public perception and awareness of the risks of 528 wildfires and their impact on soil and water resources. Fire impact on soil is significant (Cerdà and 529 Robichaud, 2009), leading to an increase in soil erosion (Shakesby and Doerr, 2006). Hence, implementing 530 mitigation measures to reduce soil erosion is imperative and should be a part of every forest and soil 531 recovery strategy (Bento-Gonçalves et al., 2012). This study has shown the effectiveness of straw mulching, 532 seeding, and soil erosion barriers in reducing soil erosion. However, further studies and new monitoring 533 programs are needed to assess additional mitigation measures and adequately analyse their cost-534 effectiveness. 535

536
This paper presents a study conducted in the Celone river basin, a Mediterranean watershed with an 537 intermittent river network. The SWAT model, calibrated with field measurements, was applied for the 538 current land use and land management practices for hydrology and sediment yield. The model adequately The present work analyses six post-fire scenarios by modelling the basin's response in terms of runoff and 551 SSY. It aims to provide a tool for post-fire risk management. The results showed that SWAT-a 552 hydrological and water quality model-may contribute to selecting the mitigation options to reduce soil 553 erosion after a fire. In addition, the model is also a useful tool for the post-fire risk assessment in terms of 554 water quality since it identifies the river segments where sediment-associated pollutants transported via 555 surface runoff could accumulate on the riverbed after fire events. 556 According to the assumption, high-severity fire vastly increases SSY at the basin and subbasin scales and 557 HRU levels. This study shows that a dramatic increase in soil erosion occurs in areas sensitive to erosion, 558 demonstrating that major efforts are needed to prevent forest fires and better manage the post-fire. The 559 results showed that a small part (2%) of the catchment is enough to cause a dramatic increase in soil loss 560 quantified at the basin scale by up to 12 t ha -1 y -1 . Post-fire management is effective at mitigating fire impact 561 on soil erosion. In particular, post-fire mitigation measures such as emergency stabilisation (straw mulching 562 and seeding) and soil erosion barriers are better at reducing soil erosion than natural regeneration or logging 563