Improving the agricultural erosion management in Finland through high-resolution data

Soil erosion reduces the sustainability of agricultural sector by loss of productive soil and through negative impacts on surface waters. In Finland, considerable efforts have been made to reduce soil erosion, but the suspended sediment loads to surface waters have not markedly reduced. A major limitation has been the lack of high-resolution data on erosion risk for efficient targeting of the erosion management efforts. In this study, by using the Revised Universal Soil Loss Equation (RUSLE) a two-meter resolution erosion risk data was developed and consequently the spatial distribution of the erosion risk 10 of Finnish agricultural land was analysed. With agricultural management practices of 2019, the average erosion of agricultural land was estimated to be 430 kg ha yr, and it varied at the municipality scale from 100 to 1290 kg ha yr. At more local scales the erosion risk had even greater variability, and areas with high erosion risk were differently located in terms distances to water bodies. The results also suggest that the past erosion management efforts have not been well-targeted according to the actual erosion risk. Altogether, the results indicate that erosion mitigation measures can be improved by inclusion of high15 resolution data in the planning and implementation of the measures, by considering the spatial variability of the erosion risk over multiple spatial scales, and by implementation of location specific erosion reduction measures.


Introduction
Soil erosion has a central role in the sustainability of agricultural sector, as it has significant negative impacts on soil productivity, surface water quality and aquatic ecosystems (Wuepper et al., 2020;Borrelli et al., 2017;Montgomery, 2007;20 Pimentel et al., 1995). It contributes to eutrophication, and to increased turbidity and siltation of surface waters (Ulén et al., 2012). Erosion causes harmful structural changes in the soil surface, and in the long-term, it can reduce soil fertility through loss of the most fertile top soil (Pimentel et al., 1995). Soil erosion is also linked to climate regulation through transport and storage of carbon (Lugato et al., 2018) and altogether, it is a cross cutting issue in the land use sector (Montanarella and Panagos, 2021). 25 The key process causing soil erosion is hydrological, and soil particles are detached from the surface by the kinetic energy of rain drops and surface runoff causing slaking, swelling and dispersion (Bissonnais, 2016;Ulén et al., 2012;Jarvis et al., 1999;Wicks and Bathurst, 1996). This process is affected by multiple connected factors, including hydrometeorological conditions, varying particle detachment mechanisms, farming practices, soil physical characteristics and chemical conditions https://doi.org/10.5194/hess-2021-457 Preprint. Discussion started: 20 September 2021 c Author(s) 2021. CC BY 4.0 License. (Turunen et al., 2017;Bechmann, 2012;Ulén et al., 2012;Turtola et al., 2007;Øygarden et al., 1997) leading to high spatial 30 variability in erosion and further in the transport of suspended sediment loads from different sites and catchments (Röman et al., 2018;Ulén et al., 2012).
In Finland, the total area of agricultural land is 2.3 million ha (7.6 % of total land area) and the main crops are cereals (45% of total area) and grass type crops (31%) (data of Finnish Food Authority for 2019). The erosion process is affected by short growing period (140-180 days) and long winter, with highest erosion during rainy autumn months and spring snowmelt 35 (Puustinen et al., 2007). Experimental studies have estimated the average erosion from fields to vary from 55 to 2100 kg ha-1 yr-1 (Lilja et al., 2017a;Puustinen et al., 2010) and earlier modelling approaches have estimated the average erosion of all agricultural lands to be 418-485 kg ha-1 yr-1 (Lilja et al., 2017b;Puustinen et al., 2010). These Fig.s are relatively low in global and European scales (Borrelli et al., 2017;Panagos et al., 2015c), however, in respect to the ecological state of water bodies in northern latitudes, they result in significant negative impacts in surface waters and in the Baltic Sea (Ulén et al., 40 2012), particularly through transport of phosphorus along with the eroded soil particles (Röman et al., 2018).
The management of environmental impacts of agriculture in Finland is guided by the EU's Common Agricultural Policy (CAP) (European Commission, 2021) and the Water Framework Directive (European Commission, 2020) and is implemented through national programmes, such as the Rural Development Programme (Ministry of Agriculture and Forestry, 2014). The management focuses largely on the main agricultural areas in southern and western Finland (Ministry of Agriculture and 45 Forestry, 2014), where major river basins drain to the Baltic Sea, and includes targeting of erosion mitigation measures, such winter-time vegetation cover, reduced soil tillage, vegetated buffer zones along streams and rivers, and perennial grass type vegetation covers. The targeting of these measures is implemented through natural constraint and environment payments to the farmers.
Despite the considerable management efforts, the agricultural loading to surface waters has not reduced substantially (Räike 50 et al., 2020;Tattari et al., 2017). In the case of erosion, a major limitation has been the lack of spatial data on distribution of erosion risk, which have led to formulation of agricultural policies and programmes with limited knowledge on spatial variability of erosion risk. For example, the current targeting of mitigation measures is based on broad regions of eight river basin districts (Alahuhta et al., 2010) with less consideration of local conditions. The modest achievements in the erosion control are also partially influenced by changes in climate and weather (Räike et al., 2020), and it is likely that the erosion 55 rates will be further influenced by the climate change . These highlight the importance of improving the erosion management, and a country-wide understanding of spatial distribution of erosion risk through high-resolution data is paramount in such efforts.
Local erosion risk can be reliably estimated with various methods but generating reliable and spatially extensive erosion data is still a challenge. Direct empirical measurement campaigns provide the most accurate information on erosion and 60 sediment loads but are costly and infeasible for production of large-scale data. Process-based computational models have shown reasonable capability to describe the erosion and sediment transport process dynamics at monitored sites (e.g. Borrelli et al., 2021;Turunen et al., 2017;Warsta et al., 2013;Jarvis et al., 1999;Wicks and Bathurst, 1996), but they are also infeasible https://doi.org/10.5194/hess-2021-457 Preprint. Discussion started: 20 September 2021 c Author(s) 2021. CC BY 4.0 License.
for large scale data production due to high computational requirements. In contrast, simplified models, which aim to estimate erosion based on a few dominating factors, provide efficient means to estimate large scale spatial distribution of erosion. These 65 include models such as the empirical Universal Soil Loss Equation (USLE) (Wischmeier and Smith, 1978) and the revised USLE (RUSLE) (Renard et al., 1997), which have been widely applied in different regions and have shown capability to reproduce annual loads in different land-use, topographic and hydrometeorological conditions (Batista et al., 2019;Estrada-Carmona et al., 2017), including the high-latitude boreal conditions (Lilja et al., 2017a).
Based on the above premises, the goal of this study was to produce a publicly available high-resolution erosion risk data 70 for agricultural lands of Finland, and thereby to demonstrate the importance of considering the variability of the erosion risk for achieving effective erosion management outcomes. The goal was achieved by 1) estimating erosion risk at two-meter resolution using RUSLE, 2) analysing spatial variability of erosion risk and its management over different spatial scales, and finally 3) providing recommendations for policy development and future research. This work was well in line with targets of the national programme on enhancing the effectiveness of water protection (Ministry of the Environment, 2021), and with the 75 targets of EU's Common Agricultural Policy (European Commission, 2021), Water Framework Directive (European Commission, 2020), and the European Green Deal (Montanarella and Panagos, 2021).

Methodology
The estimation of agricultural erosion was based on the Revised Universal Soil Loss Equation (RUSLE) (Renard et al., 1997;Wischmeier and Smith, 1978), and three different types of average erosion estimates were calculated for the Finnish 80 agricultural lands, based on average weather during 2007-2013 ( Fig. 1): • Erosion susceptibility (kg ha -1 yr -1 ) of all land areas, which describes the erosion risk according to rainfall erosivity, topography and soil erodibility, excluding the effects of vegetation cover and soil management • Potential erosion risk (kg ha -1 yr -1 ) of agricultural lands, which describes the highest potential erosion corresponding to bare fallow land without sub-surface drainage. 85 • Actual erosion risk (kg ha -1 yr -1 ) of agricultural lands, which describes the erosion under agricultural practices of the year 2019, including the prevailing sub-surface drainage These estimates were derived through a modelling framework consisting of calculation of soil erosion susceptibility in twometer resolution for all land areas (A, Fig. 1), calibration of RUSLE at seven experimental fields (B), testing of calibrated RUSLE at five small catchments and at fourteen large river basin (C), estimation of potential erosion risk for all arable lands 90 (D), and estimation of the actual erosion risk of all arable lands (E). In addition, an Erosion Management Index (EMI) (-) was developed and used to estimate the effectiveness of erosion mitigation measures.

95
The resulting data were then analysed spatially. The potential erosion risk was analysed at sub-basin scale, to identify differences in erosion risks within the landscape. The actual erosion risk and the EMI were analysed at municipal level to provide administratively relevant information for managing erosion.
The following sections provide general introduction to RUSLE and a detailed explanation of the modelling framework and the used data. Additional information is presented in supplementary material. 100

Revised Universal Soil Loss Equation (RUSLE)
The RUSLE (Eq. 1) is an empirical model for estimating soil loss due to water erosion (Renard et al., 1997;Wischmeier and Smith, 1978). The RUSLE equation is (Eq. 1) where E is the annual average erosion (t ha -1 yr -1 ), R is the rainfall erosivity factor [MJ mm ha -1 h -1 yr -1 ], K is the soil erodibility 105 factor (t ha h ha-1 MJ-1 mm-1), L is slope length factor (dimensionless) and S is the slope steepness factor (dimensionless), C is the cover-management factor, and P is the support practices factor which accounts for erosion control practices, such as buffer zones, contour tillage and sub-surface drainage. The dimensionless C and P factors vary from near 0 to 1.
While the other factors are described with single factor values, the cover-management factor (C) consists of sub-factors (Eq. 2), 110 where Ccrop accounts for the influence of crops on erosion, and Cmanagement accounts for the influence of management practices on erosion (Panagos et al., 2015b). The Ccrop and Cmanagement sub-factors are dimensionless and vary from near 0 to 1.
where Ctillage, Cresidues and Ccover quantify the effects of tillage, plant residues and cover crops on erosion, respectively (Panagos et al., 2015b). In this study, the Cresidues was not considered due to lack of data.

Erosion susceptibility
The erosion susceptibility is described with the R, K, L, S factors, and their values were set for all land areas of Finland at twometer scale using spatial data as described below. The calculated erosion susceptibility was used in calibration at experimental 120 fields, testing at catchments and river basins, and in estimation of the potential and actual erosion data (Fig. 1).
The R factor was taken from a 1 km resolution gridded European scale dataset that is based on observational data (Panagos et al., 2015a). For Finland, R is calculated from hourly precipitation data measured at 64 stations covering a period of 2007-2013. The average R-value for Finland is 273 MJ mm ha -1 t -1 yr -1 with annual average precipitation of 660 mm, while the European average is 722 MJ mm ha -1 t -1 yr -1 . 125 The K factor was taken from Finnish Soil Database (Lilja et al., 2017c;Lilja and Nevalainen, 2006) supplemented with soil specific K values (Lilja et al., 2017a(Lilja et al., , 2017b. The soil database is a vector data with scale of 1:200 000 and the smallest feature in the data is 6,25 ha. The K values are based on calibration of ICECREAM model for Finnish soils (Rekolainen and Posch, 1993), except for clay soils. For the clay soils, the K value is derived from field studies in Poland (Lilja et al., 2017a;Święchowicz, 2012). The soils of Finnish Soil database and the soil specific K values are shown in Tab. S1. 130 The L and S factors were calculated in this study from a two-meter resolution LiDAR-based digital elevation model (DEM) of Finland (National Land Survey of Finland, 2020). A combined LS-factor was calculated with SAGA-GIS Module LS Factor (Conrad, 2003) using the method of Desmet and Govers (1996) with default settings. The DEM was used as such and it was not treated for sinks, as it would have introduced more errors. For example, filling of sinks would fill fields up to the level of nearby roads, and breaching would create artificial erosion areas in the fields. For the LS calculations Finland was divided in 135 301 units and these were based on river basins, sub-basin groups (Finnish Environment Institute, 2010), and in offshore areas on groups of multiple islands (Fig. S1).
The R factor was resampled, and the K factor was rasterised to the same two-meter resolution as the LS factor data using bilinear (R) and nearest neighbour (K) interpolation methods. The rasterised K-factor data was also extrapolated (nearest neighbour method) to account for finer details of shorelines of water bodies, as the scale of the Finnish Soil Database does not 140 describe the shorelines in detail.
The erosion susceptibility for all land areas of Finland was then calculated by multiplying the R, K and LS factor data. The erosion susceptibility data for agricultural areas was thereafter extracted using the field parcel data from Finnish Food Authority. The calculations were done in high-performance computing environment (CSC -IT Center for Science, Finland) using RSAGA (Brenning et al., 2018) and terra (Hijmans et al., 2021)

Data Description and source
Rainfall erosivity (R) European rainfall erosivity data with 1 km resolution (Panagos et al., 2015a). Data for Finland is calculated using 60 min precipitation data from 64 stations over the period 2007-2013.

RUSLE calibration and testing
The RUSLE was calibrated at seven monitored field sites with year-round soil loss measurements (Tab. 2, Fig. S2). Aurajoki, 150 Liperi, Kotkanoja, Nummela and Toholampi sites are experimental fields with multiple plots and practices, while Gårdskulla and Hovi sites are single field areas in normal agricultural use. The fields have varying soil and topographical conditions, and all, except Aurajoki site, were subsurface drained during the measurement campaigns (Tab. 2).
The fields were under different crops and management practices during the measurements, including spring cereals (wheat, oat, barley) with conventional autumn ploughing, shallow autumn stubble tillage, autumn cultivator tillage, no autumn till 155 (winter-time stubble) and direct sowing (winter-time stubble); winter cereals (wheat, rye); perennial grass; and perennial pasture. From the data, each crop and management practice with a minimum of four years of measurements was included in the calibration. This provided 20 crop and management cases that were divided to six treatment groups for the calibration: cereals with autumn ploughing, cereals with reduced autumn tillage, cereals with winter-time stubble, winter cereals, perennial grass and perennial pasture (Tab. 4). 160 The model was calibrated against the average annual soil loss of the measurement periods. The sum of soil loss via surface and sub-surface drainage was considered, as a large share of the eroded material can be transported from the soil surface via https://doi.org/10.5194/hess-2021-457 Preprint. Discussion started: 20 September 2021 c Author(s) 2021. CC BY 4.0 License.
The C factor was chosen as the calibration parameter as the sensitivity analyses show that the C factor is the largest source 165 of uncertainty (Estrada-Carmona et al., 2017) and it can vary by location depending on cultivation practices (Hudson, 1993). Therefore, in the calibration the difference between the RUSLE erosion estimates and the measured soil losses was minimized by adjusting the C value of each treatment group with least squares method.
The sub-surface drainage was considered in the P factor. The research on the effect of sub-surface drainage on erosion is limited, but studies in the North-Western US found a reduction effect of 28-51% (Formanek et al., 1987;Istok et al., 1985). A 170 study in Finland in turn found that substituting of old drainage pipes with new ones reduced erosion up to 15% on a clay soil (Turtola and Paajanen, 1995). In this study, however, a reduction effect of 40% (P=0.6) was used, following Lilja et al. (2017a).

175
The validation of erosion models, such as the RUSLE, is typically difficult and rarely done mainly due to limitations in data availability (Batista et al., 2019). To get an indication of the performance of the RUSLE on larger spatial scales, the current model was tested at river basin and small catchment scales against total suspended solid measurements (TSS) from streams and rivers. The test was done by analysing the statistical relationship of estimated potential erosion risk (t yr -1 ) of agricultural lands by RUSLE and measured average TSS (t yr -1 ). However, this test is considered only indicative of RUSLE's performance 180 due to three reasons. First, the potential erosion risk describes only erosion from agricultural lands, whereas TSS measurements account for erosion from all land uses. Second, potential erosion risk emphasizes the source of erosion rather than later phases of the erosion-transport-sedimentation process that affect actual TSS quantities in rivers. Third, the agricultural practices have varied in the catchments and basins over the TSS measurement periods that could not be accounted for, and therefore, the potential erosion risk was used. Despite these limitations, the test provides useful information for understanding the 185 performance of the RUSLE beyond the calibration conditions and in a larger scale, but it is noteworthy that it does not equal model validation.
The test catchments and basins were selected so that the share of agricultural land was higher than 10% of total and large lakes and major dams with reservoirs were absent, since these surface water features reduce the transport of sediments and would reduce the commensurability between the measurements and the model outputs. Data was available for five small 190 catchments with sizes varying from 5.3 to 15.2 km 2 and share of agricultural land varying from 17 to 63% (Tab. S2 and

Potential erosion risk
The potential erosion risk (kg ha -1 yr -1 ) describes the maximum potential erosion of agricultural land, which was defined here as bare fallow land without sub-surface drainage. Bare fallow is as land that is not under crop rotation and has no planted vegetation cover. The potential erosion risk was calculated from the erosion susceptibility data by multiplying it with a C factor value of 0.5 suggested for bare fallow land by the literature (Panagos et al., 2015b). The resulting two-meter resolution potential 200 erosion risk data allows a spatially consistent analysis of erosion risk without the effects of crops and management. In this study, the data was analysed at 2 m resolution at two case study areas, and at sub-basin scale for the whole Finland.
In addition, the potential erosion risk was analysed in the proximity (< 50 m) of main water bodies that were defined according to the stream network, river area, and lake area data (Finnish Environment Institute, 2010).

Actual erosion risk 205
The actual erosion risk (kg ha -1 yr -1 ) was calculated using the erosion susceptibility data and by considering the agricultural practices of 2019 in the C factor and the sub-surface drainage in the P factor. The calculation of the actual erosion risk was done by using average value of the erosion susceptibility for each field parcel and by multiplying this with field parcel specific C and P values. The resulting data is a vector data with actual erosion risk estimate for each field parcel.
The agricultural practices of 2019 were taken from the field parcel data of Finnish Food Authority, which contains field 210 parcel specific information on cultivated crops and erosion reduction measures, including reduced autumn tillage, winter-time vegetation cover and buffer zones. The data are collected annually from farmers through government controlled self-reporting process, and it is also the basis for payment of agricultural and environmental subsidies. According to this data, Finland had 2,34 million hectares of agricultural land with 1,09 million field parcels with 212 different crop and vegetation cover types.
The crops in the field parcel data were parametrised in the Ccrop, and the calibration provided values for 89% of the 215 agricultural area (cereals and grasses). The literature (Panagos et al., 2015b) provided further Ccrop values for many crops, but not for all. Remaining crops were divided to groups according to their similarities and a Ccrop value of most similar crop in the RUSLE calibration or literature were assigned for those. For example, all large root vegeTab.s were placed in the same group and they were given the Ccrop value of potato and sugar beet in the literature (Panagos et al., 2015b). There were still few annual crops that could not be given Ccrop values according to calibration and literature, and for these, calibrated Ccrop value of cereals 220 was used. The parameterisation is summarized in Tab. 3.
The erosion reduction measure of reduced autumn tillage was parametrised in the Ctillage. The Ctillage of normal, conventional autumn ploughing was assumed to have a value of 1, similarly to Panagos et al. (Panagos et al., 2015b), and the Ctillage for reduced autumn tillage was defined as the ratio of calibrated C value of cereals with reduced autumn tillage (cultivator, shallow stubble tillage) and C value of autumn ploughing. 225 The winter-time vegetation cover was parametrised in the Ccover and it was defined as the ratio of calibrated C value of cereals with winter-time stubble (no autumn till, direct sowing) and C value of normal autumn ploughing. Thus, in the calculation of actual erosion risk the winter-time vegetation corresponds to winter-time stubble. However, in the field parcel data the winter-time vegetation cover can consist of different types, including grasses, stubble, vegetation covered fallow, over-wintering vegetation and perennial plants, but these were not distinguished in the data. 230 The subsurface drainage data was from the Finnish Field Drainage Association and it contained information on drainage status of field parcels. It is based on regional reporting that has been arranged into a database. The data is the best available on field parcel level with adequately comprehensive coverage, but it may lack information on drainage status of some individual fields. The P factor value of 0.6 was used for the sub-surface drainage, similarly to Lilja et al. (2017a) and RUSLE calibration. The retention effect of buffer zones is typically considered in the P factor, but due to limitations in the data this could not 235 to be considered in the calculation of actual erosion risk. The retention effect refers to retention of eroded soil and solids that are transported by overland flow from the uphill field area over the buffer zone. The field parcel data identifies buffer zones as individual field parcels, but it does not identify the field parcel from which the buffer zone is intended to capture the eroded and transported soil and solid material. In addition, entire fields have been classified as buffer zones in the field parcel data due to regulatory issues. Therefore, an analytical and systematic approach for quantifying retention effect was not possible. 240 The vegetation cover of buffer zone areas themselves was, however, considered in the Ccrop, and calibrated Ccrop value of grass was used.
After the calculation of actual erosion risk data, it was analysed on municipal level, which is a suitable administrative level from policy perspective. The vector data was first rasterized to 10 m resolution before calculating zonal statistics for municipal areas. 245

Erosion Management Index
A quantitative Erosion Management Index (EMI) was developed to estimate the level of erosion management over specific 250 areas. The index is dimensionless, and it varies from 0 to 1. Higher values indicate that the area is closer to minimum potential erosion and thus the erosion management efforts are more effective. The EMI can be calculated as (Eq. 4) where EMIi is the index value for an area i, Emax,i is the maximum and Emin,i is the minimum potential erosion (kg/ha/yr), and Ei is the crop and management specific erosion (kg/ha/yr). The Emax, Emin and Ei can be defined case specifically. The strength 255 of the index is that it can be used for spatially and temporally consistent evaluation of erosion management.
In this study, the Emax was defined as the calculated potential erosion risk, corresponding to field conditions with bare fallow land and with no sub-surface drainage. Emin was defined as erosion under field conditions with perennial grass cover and with sub-surface drainage. The Ei was defined as the calculated actual erosion risk, which meant that the erosion management measures considered in the EMI were crop and vegetation cover type, winter-time vegetation cover, reduced 260 tillage, and sub-surface drainage. The buffer zones were considered only partially, as in the actual erosion risk.
The calculated EMI was then analysed on municipal level together with the agricultural area data from field parcel data (Finnish Food Authority) and with the calculated actual erosion risk data. The used methods included Pearson's linear correlation (Pearson, 1920), Kendal's rank correlation (Kendall, 1975) and Welch's t-test (Welch, 1951).

RUSLE performance
The overall RUSLE performance was reasonable, although with some limitations. The calibrated RUSLE estimated erosion relatively accurately at five experimental fields -Aurajoki, Gårdskulla, Hovi, Liperi and Toholampiand underestimated it at two clayey experimental fields -Kotkanoja and Nummelaas shown in Tab. 4. The mean error at the five accurately estimated fields was -2% and varied from -43 to +22%, and at the two underestimated fields the errors were -90 and -49%. 270 The R 2 for all seven fields was 0.75 (p-value < 0.000) (Fig. S3A), and for the five relatively accurately estimated fields 0.98 (p-value < 0.000). The average ratio of all estimated to measured erosion rates of the seven fields was 0.83. The calibrated C values are shown in Tab. 5, and they provide estimates for the effect of crops and management on erosion.
According to the C factor values for cereals, winter-time stubble reduces erosion by 66%, reduced autumn tillage by 23%, winter cereals by 29% compared to autumn ploughing. The perennial grass and pasture have 69% and 54% lower erosion than the cereals with normal ploughing, respectively. The Ccover for winter-time vegetation and Ctillage for reduced autumn tillage that were calculated from calibrated C factor values, are 0.341 and 0.768, respectively (Tab. 5). 280 The testing of RUSLE against TSS measurements at the five small catchments and fourteen river basins (Tab. S2) indicated 285 good performance at large spatial scales. At the small catchments the R 2 for estimated potential erosion risk and TSS measurements was 0.49 (p-value = 0.1896), but the tau of Kendall's rank correlation was 1.00 (p-value=0.0167). This indicates that while the estimated potential erosion risk deviated from the measurements, RUSLE was able to rank the magnitude of erosion correctly between the catchments. At the river basins the R 2 was 0.90 (p-value < 0.000) and the Kendall's tau 0.78 (pvalue < 0.000) (Fig. S3B).

Potential erosion risk
Samples of the estimated two-meter resolution potential erosion risk data from two neighbouring basins -Karjaanjoki and Paimionjokiin the Southern coast of Finland are shown in Fig. 2, to exemplify how the potential erosion risk varies in the landscape and between basins. At the Karjaanjoki basin, the average potential erosion risk was estimated to be 4530 kg ha -1 yr -1 , the average field slope varies between 1.5-5.0° by sub-basin, and high erosion areas are scattered in the landscape. At the 295 Paimionjoki basin, in turn, the average potential erosion risk was estimated to be lower, 2020 kg ha -1 yr -1 , with lower average slope of 0.4-2.1°, and the areas of highest erosion are concentrated near the river and stream channels.

300
On the country scale, the average potential erosion risk of agricultural lands was estimated to be 2,010 kg ha -1 yr -1 , and it varied between 110 and 14,030 kg ha -1 yr -1 by sub-basin, as shown in Fig. 3  The topography of the fields was the most influential factor in the estimation of potential erosion risk. The linear correlation between slope length and steepness factor (LS) and the potential erosion risk at sub-basin level was 0.67 (p-value < 0.000) whereas it was it was 0.51 (p-value < 0.000) for soil erodibility (K) and 0.39 (p-value < 0.000) for rainfall erosivity (R). 310 The LS factor was also a major contributing factor in the two regions of high erosion risk identified in Fig. 3. The LS factor had heightened values in those same regions as shown in Fig. 4. Similarly, the lower LS factor values in the western coast were contributing to the lower erosion risk in those areas. According to the K factor, large areas of erosive soils were found in the Southwest, and highly erosive soils were found in the western coast, and particularly in the river valleys (Fig. 5b). The areas with highest rainfall erosivity were found in the western coast of Southern Finland (Fig. 5a).  The potential erosion risk within 50 m distance from main water bodies was estimated to be on average 3,140 kg ha -1 yr -1 , 320 which is 1.6 times the average of all agricultural lands. In 10% of the sub-basins this ratio was higher than 2.3, as shown in Fig. 5. The agricultural areas within 50 m distance from main water bodies account for 6% of arable land, but their total erosion risk (t/yr) was 9% of all agricultural lands, which demonstrates their importance as sources of erosion.
Two regions with considerably higher potential erosion risk near the water bodies were identified, and both are situated by the coast of the Baltic sea (Fig. 5). The largest one is in the Southwest Finland and the smaller in South Finland (Fig. 5). Also, 325 the western coast seems to have several sub-basins with higher potential erosion risk near the water bodies. Southeast Finland in turn has a large region where the erosion is more uniform in all agricultural areas and where lakes form a large proportion of the area. In the Northern Finland, with low proportion of agricultural land, the situation is mixed.

Actual erosion risk
RUSLE estimate for the actual erosion risk with management practices of 2019 was on average 426 kg ha -1 yr -1 and it varied 335 by municipality from 102 to 1288 kg ha -1 yr -1 as shown in Fig. 6A (Fig. S4, Tab. S4). The spatial distribution of the actual erosion risk resembled the erosion risk in Fig. 3. The two areas with the highest erosion were similarly detected in central and coastal area in Southern Finland, and a large area with low erosion was also identified in the coastal area in central Western

Finland.
The estimate for the actual total erosion risk of agricultural land was 985,942 t yr -1 , and it varied by municipality from 13 340 to 33,088 t yr -1 (Tab. S4). Majority of the total erosion occurs in Southwest Finland, as shown in Fig. 6B. Agriculture is most intensive in Southern and Western Finland, where over 50% of the land area of some municipalities can be agricultural land. Individual municipalities with high actual total erosion are also situated in the inland parts of Central Finland. Altogether, 37% of actual total erosion occurred in 10% of the municipalities (n=31) with highest total erosion risk.

Erosion Management Index 350
The average EMI of municipalities was 0.84, and the variability between the municipalities was high, ranging from 0.60 to 0.99, as shown in Fig. 6D and Fig. 7 (Tab. S4). The visual examination of EMI values, contrasted with ranked municipalities in Fig. 7 does not reveal strong patterns that would indicate better erosion management in areas with high field area (%), high actual erosion risk (kg ha -1 yr -1 ) or high actual total erosion risk (t yr -1 ). However, in areas with very low field area (<10%) and actual total erosion risk (<160,000 t yr -1 ), the EMI values were lower. Most interestingly, the EMI values tend to be lower in 355 municipalities with high actual erosion risk (Fig. 7B), indicating that the erosion management measures have not been targeted according to the erosion risk. Many municipalities with high field area, high actual erosion risk and high actual total erosion risk have also below average EMI values.  The statistical analyses support the visual interpretation of EMI, but they also reveal some statistically significant patterns. 365 Pearson's and Kendal's correlation analyses show statistically significant but weak relationships between EMI (-) and field area (%), and between EMI and actual total erosion risk (t yr -1 ), but there was no statistically significant relationship between https://doi.org/10.5194/hess-2021-457 Preprint. Discussion started: 20 September 2021 c Author(s) 2021. CC BY 4.0 License.
targeted slightly more to intensive agricultural areas, but not specifically to areas with high erosion risk.

Pearsons' r (p-value) Kendal's tau (p-value)
Field area (%) 0.36 (< 0.000) 0.3 (< 0.000) Actual erosion risk (kg ha -1 yr -1 ) -0.05 (0.341) 0.06 (0.09) Actual total erosion risk (t yr -1 ) 0.24 (< 0.000) 0.2 (< 0.000) Also, Welch's t-test suggests that the 10% of the municipalities with highest field area (%) and actual total erosion risk (t yr -1 ) have slightly higher (+2.5% and +4.5%, respectively) EMI values than the rest of the municipalities, but in the case of actual 375 erosion risk (kg ha -1 yr -1 ) the EMI values did not differ between the ranked top 10% and rest of the municipalities, as shown in Tab. 7. This also supports the interpretation that erosion management efforts have been targeted slightly more to intensive agricultural areas, but not to areas with highest erosion risk. Tab. 7. Comparison of average erosion management index (EMI) of top 10% of municipalities having highest field area, highest actual erosion risk, and highest actual total erosion risk with the average EMI of the rest of the municipalities using Welch's t-test.  (Panagos et al., 2015c). National scale studies, based on RUSLE2015 and VIHMA models and crop and management data for 2010, estimated the average erosion to be on average 418 and 485 kg ha -1 yr -1 , respectively (Lilja et al., 2017b;Puustinen et al., 2010). In our study, the average erosion was estimated to be 425 kg ha -1 yr -1 , and thus the modelling approaches agree on the average magnitude of erosion in agricultural lands of Finland. https://doi.org/10.5194/hess-2021-457 Preprint. Discussion started: 20 September 2021 c Author(s) 2021. CC BY 4.0 License.

Ranking of municipalities
The performance of RUSLE in this study was also similar to previous evaluation of RUSLE that was based on calibration 390 with partially same experimental fields (Lilja et al., 2017a). There the model passed the ± 50% error criteria in 72% of the 19 cover and management cases, while in this study, the model passed the same criteria in 76% of the 20 cases. The evaluation of (Lilja et al., 2017a) focused only on erosion via surface runoff and sub-surface drainage was excluded, and therefore, the C factor values were also lower than in our study. The C factor values in the European scale study (Panagos et al., 2015b) were similar, although slightly lower, than in this study, as shown in Tab. 5. 395

Uncertainties
The strength of the RUSLE is in its capability to estimate spatial distribution of erosion magnitude within landscapes, and it is shown in this study and in earlier research (Renard et al., 1997;Wischmeier and Smith, 1978) that the emergent long-term bulk erosion magnitude can be a result of the a few dominant controlling factors. However, it is also recognized that the model includes uncertainties rising from a range of generalizations and simplifications in its description of the erosion process, and 400 even validated models are subject to uncertainties when model predictions are conducted (e.g., Højberg and Refsgaard, 2005;Refsgaard et al., 2006). The quantification of the total uncertainties remains as a challenge regarding the large spatial estimates, as well as for model applications at large (Batista et al., 2019;Beven, 2016).
According to sensitivity analyses, the LS and C factors are the largest sources of uncertainty in RUSLE (Estrada-Carmona et al., 2017). The resolution of DEM for computing the LS factor is known to affect the estimation of slope in gently sloping 405 areas and the estimation of up-slope contributing area in steep areas, and high-resolution DEM's reduce the resulting uncertainty (Gertner et al., 2002). The choice of computation method of LS factor is also found to affect the magnitude of the erosion estimates (Hrabalíková and Janeček, 2017) and the emphasis between rill and inter-rill erosion (Erskine et al., 2006).
In this study, the LS factor was computed using a high-resolution DEM (National Land Survey of Finland, 2020) and with a well-established method (Desmet and Govers, 1996), and is therefore expected to provide an adequate description of the LS-410 factor and a reasonable estimate of the location of high erosion areas within the fields for erosion management purposes.
The C factor values for crops and management vary by location depending on local conditions and practices (Panagos et al., 2015b) and they may contain uncertainties. However, 89% of the crop area was parameterised according calibration at Finnish experimental fields and, therefore, the parameterisation is largely expected to be representative of Finnish conditions and practices. The parametrisation of winter-time vegetation was, however, based on winter-time stubble (no autumn till), 415 whereas the crop and management data (Finnish Food Authority) considers various cover types as winter-time vegetation cover, but does not specify them.
The used K factor was based on national 1:200,000 scale soil data (Lilja et al., 2017c;Lilja and Nevalainen, 2006) and it does not capture all local heterogeneities affecting the erosion process. The estimation of K factor values of different soils is also limited in Finland, and the underestimation of erosion observed at Kotkanoja and Nummela fields suggest further research 420 is needed, particularly in the case of heavy clay soils. Likely the soil structure dynamics of clay soils (e.g., Bissonnais, 2016;Turunen et al. 2017) is one source of uncertainties in the estimates regarding the cohesive soils. The sub-surface drainage was parameterised in the P factor with the support of literature, but the empirical research on the effect of sub-surface drainage on erosion is scarce. The effect of sub-surface drainage is likely to vary depending on the soil and catchment properties as well as the type, condition, and design parameters of the drainage system. 425 Furthermore, the exclusion of the retention effect of buffer zones in the P factor may have led to overestimation of erosion risk, although the overestimation in this study is expected to be small. A study in Finland, on the effect of 10-m-wide buffer zones at the lower end of a subsurface drained field on clay soil, found that the buffer zones reduced erosion loading via overland flow by 11-58% depending on crop and management types (Uusi-Kämppä and Jauhiainen, 2010). However, according to the experimental field data (clayey soils of Gårdskulla, Kotkanoja, Nummela and sandy soil of Toholampi) used 430 in this study 50-92% of erosion matter is transported via sub-surface drain flow, which corresponds findings in earlier research (Turunen et al., 2017;Warsta et al., 2013;Uusitalo et al., 2001;Turtola and Paajanen, 1995;Øygarden et al., 1997). Thus, at sub-surface drained fields the retention effect is likely smaller than in non-subsurface drained fields, but the effect may vary considerably between fields. Buffer zones can, however, have a significant role in preventing erosion from their own area, as they are often located in sloping lands near water bodies. 435 Despite these limitations, the calibration of RUSLE at field parcel scale and testing at catchment and basin scales indicate that the results and the developed data are of reasonable quality, and they markedly improve the understanding of distribution of erosion and the possibilities of erosion management in Finland. One of the most promising features of the model application was that the calibration of the RUSLE differentiated well between crops and management types, which provides a good basis development and assessment of different crop and management scenarios. However, certain level of care is needed in the 440 interpretation of data, particularly at the field parcel scale, as the RUSLE underestimated erosion at two of the seven fields.

Policy and management implications
The results showed considerable spatial variability in erosion risk over multiple spatial scales and this heterogeneity includes large potential to be accounted for in the Finnish agricultural policy and environmental programmes. The results also suggest that the erosion reduction measures have not been targeted efficiently to high erosion risk areas, which is also largely due to 445 lack data on erosion risk. These limitations call for improvements in policy and management, and this study formulates how improvements can be made through an erosion management approach, where 1) the erosion management is guided by spatially explicit erosion risk data, 2) the spatial distribution and magnitude of erosion risk is considered in addition to location and total area of the fields, 3) the high erosion risk areas and their sizes and locations are identified with systematic data analysis over multiple 450 spatial scales and erosion management measures are targeted accordingly, and 4) the erosion management measures are chosen and implemented according to the local erosion conditions.
In addition to these, the study provides a generalisation of the effect of different management practices on erosion, that is based on RUSLE calibration with measurement data from seven Finnish experimental fields. In cereal cultivation, the most effective erosion reduction measure was found to be winter-time stubble (-66%), whereas winter cereals (-29%) and reduced autumn 455 https://doi.org/10.5194/hess-2021-457 Preprint. Discussion started: 20 September 2021 c Author(s) 2021. CC BY 4.0 License. tillage (-23%) were found to be less effective. Perennial grass type vegetation cover, in turn, was found to reduce erosion even more (-69%) than cereals with winter-time stubble. Therefore, in the targeting of environmental measures, winter-time stubble could be preferred over winter cereals as a winter-time vegetation cover. Perennial grass type vegetation cover could be emphasized at field parcels with high erosion risk and potentially highest off-site impacts, such as those with steep slopes near water bodies. 460

Future research directions
The uncertainties in the RUSLE erosion estimates can be reduced with further empirical data and consequently by improving the parameterization and model testing approaches. In the Finnish case, the improvement of spatial accuracy of soil data and its parameterisation in the K factor would yield more accurate erosion estimates. Improvements in the parameterisation of erosion mitigation measures in the C and P factor would improve the estimation of the effect of erosion reduction measures. 465 The RUSLE does not consider sediment transport and approaches, such as the sediment connectivity (Najafi et al., 2021;Heckmann et al., 2018;Bracken et al., 2015) would complement the RUSLE erosion risk estimates and improve the targeting of erosion reduction measures. Sediments are likely to be transported differently from each field parcel to water bodies and to the outlet of the basin, and this may have influence on how the erosion management efforts should be targeted.
The developed data enables improved location specific erosion management strategies, and this potential should be further 470 investigated. The current study revealed that the targeting of erosion management measures can be improved, but the erosion reduction potential of such improvements is yet to be quantitatively evaluated.
In addition, the developed RUSLE data provides a basis for estimation of losses of soil bound phosphorous and carbon, and research in these directions would further improve understanding of agricultural loading to water bodies and carbon balances in agricultural soils. 475

Conclusions
A major impediment for efficient agricultural erosion management in Finland has been the lack of comprehensive spatial data on erosion risk, which has affected the formulation of policy and targeting of the erosion reduction measures. This limitation was addressed in this study by developing a two-meter resolution erosion risk data for Finland using RUSLE, and by analysing the spatial distribution of erosion risk and its management in agricultural lands. 480 The developed data considerably improves the understanding on erosion risk in Finland, and it was found that erosion risk varies substantially in the landscape over multiple spatial scales. The average erosion of agricultural lands was estimated to be 430 kg ha -1 yr -1 with agricultural practices of 2019, and it varied from 100 to 1290 kg ha -1 yr -1 by municipality. On more local scales, the erosion risk had even greater variability. The findings also suggest that erosion management has not been well-