Quantifying albedo impact and radiative forcing of management practices in European wheat cropping systems

Management practices that increase the surface albedo of cultivated land could mitigate climate change, with similar effectiveness to practices that reduce greenhouse gas emissions or favor natural CO2 sequestration. Yet, the efficiency of such practices is barely quantified. In this study, we quantified the impacts of seven different management practices on the surface albedo of winter wheat fields (nitrogen fertilizer, herbicide, fungicide, sowing, harvest, tillage, and crop residues) by analyzing observed daily albedo dynamics from eight European flux-tower sites with interpretable machine learning. We found that management practices have significant influences on surface albedo dynamics compared with climate and soil conditions. The nitrogen fertilizer application has the largest effect among the seven practices as it increases surface albedo by 0.015 ± 0.004 during the first two months after application, corresponding to a radiative forcing of −4.39 ± 1.22 W m−2. Herbicide induces a modest albedo decrease of 0.005 ± 0.002 over 150 d after application by killing weeds in the fallow period only, resulting in a magnitude of radiative forcing of 1.33 ± 1.06 W m−2 which is higher than radiative forcing of other practices in the same period. The substantial temporal evolution of the albedo impacts of management practices increases uncertainties in the estimated albedo-mediated climate impacts of management practices. Although these albedo effects are smaller than published estimates of the greenhouse gas-mediated biogeochemical practices, they are nevertheless significant and should thus be accounted for in climate impact assessments.


Introduction
The biophysical effect of land management practices on albedo (the fraction of light reflected by the land surface) may cause local climate warming of the same order of magnitude as greenhouse gas (GHG) emissions on agricultural lands (Lobell et al 2006, Devaraju et al 2015, Lugato et al 2020, Liu et al 2022).However, it has not been included in countries' climate change mitigation commitments and received little attention for improving the climate resilience of agriculture due to uncertainties of albedo-mediated mitigation potential, missing representation of albedo in farm models, difficulties in reporting and verification, and dependence on the climate effects on models and assumptions (Davin and de Noblet-ducoudré 2010, Thomas et al 2016, Seneviratne et al 2018, Chen 2021).
Changes in surface albedo induced by management practices impact regional (global) climate either by directly affecting the radiative forcing at the top atmosphere, or the atmosphere hydrothermal dynamics influenced by the albedo-mediated surface energy partitioning (Bright 2015, Huang et al 2020, Li et al 2023).The variation of air temperature by the two processes mitigates or intensifies the regional (global) warming induced by GHG emissions.Managing surface albedo is therefore an option to reduce the impact of agriculture on climate and may help to deal with other issues like erosion, low water infiltration, and highly degraded soils.This can be achieved by covering the surface in fallow periods with crop residues or cover crops which reflect more radiation than dark topsoil, or by reducing tillage which decreases the surface roughness of fields, thereby increasing the albedo during the fallow period (Davin et al 2014, Seneviratne et al 2018, Sieber et al 2022).As an example, a widespread introduction of cover crops in fallow periods could increase surface albedo in Europe by about 4%, corresponding to a mitigation potential of 159.00 kg CO 2 e ha −1 yr −1 (Carrer et al 2018).Other land management practices, such as biochar application, may affect albedo more indirectly and in a non-intended way (Genesio et al 2012).Fertilizers enhance crop growth and herbicides suppress weeds, thereby affecting surface albedo by influencing the fraction of radiation absorbed by leaves.But the climate mitigation potential of such practices for albedo management remains uncertain (Hirsch et al 2018).
In this study, we used the random forest model and visualization techniques based on partial dependence plots to address two research questions: (1) To what extent (and how) do management practices influence the temporal albedo dynamics in winter wheat crops and their subsequent fallow periods in Europe?
(2) How do albedo-mediated biophysical impacts compare with estimates resulting from GHG budgets?
We addressed these two questions using machine learning and daily surface albedo dynamics derived from incoming and outgoing shortwave radiation measurements at eight European cropland sites with information on management practices.

Datasets 2.1.1. Flux tower data
We obtained ecosystem-scale carbon, water, and energy fluxes, as well as meteorological data from flux sites under cropland use through the Integrated Carbon Observation System (ICOS) Data Portal (Bernhofer et al 2023, Brümmer et al 2023, Brut et al 2023, Buysse et al 2023, Dumont et al 2023, Schmidt et al 2023, Tallec et al 2023), and European Fluxes Database Cluster (EFDC, www.europe-fluxdata.eu/).We only considered periods during which winter wheat was cultivated as the main crop, i.e. crop growth periods from sowing to harvest and the subsequent fallow periods until the next main crop period other than winter wheat.Out of the 32 European fluxnet sites in the data, only eight sites (BE-Lon, FR-Gri, FR-Lam, FR-Aur, DE-RuS, DE-Geb, DE-Kli, CZ-KrP) had sufficient information available (supplementary figure 1).The data spans the period 2000-2020 with a total of 126 site years.The data includes measurements of incoming and outgoing shortwave radiation (SW_IN, SW_OUT), incoming shortwave radiation at the top of the atmosphere (SW_IN_POT), air temperature, soil temperature, vapor pressure deficit, wind speed, precipitation, air pressure, superficial soil water content, sensible heat flux, and latent heat flux.Except for SW_IN and SW_OUT, all fluxes were provided as daily values.Data were qualitycontrolled and gap-filled using uniform methods (Pastorello et al 2020).Two climatic variables that potentially affect surface albedo dynamics were further derived, i.e. the ratio between diffuse and total solar radiation (Kd), which reflects atmosphere conditions such as cloud cover (Boland et al 2008), and the Bowen ratio which shows the surface hydrothermal transport (supplementary text 1.1).We applied a coarse-resolution cloud cover filtering based on daily moderate-resolution imaging spectroradiometer (MODIS) MO(Y)D04_L2 product (https:// ladsweb.modaps.eosdis.nasa.gov/) to exclude days with very low incoming shortwave radiation at the surface which prevents us from estimating albedo adequately.
The daily shortwave mid-day albedo α was averaged from SW_IN and SW_OUT from 11:00 to 13:00 Central European Summer Time at each site (Lin et al 2022): (1)

Crop management data
We assessed the impact of management practices on albedo dynamics using a crop management database provided by site managers (supplementary table 1).
The legacy effect of selected management practices on albedo dynamics is represented as the number of days since the implementation of given practices with an upper bound after which the practices were assumed to have no effect anymore (see below).The practices include physical management and inputrelated ones.The former encompasses tillage, sowing and harvest events of winter wheat, the presence of crop residues on the soil surface (including stalks left standing) (coincides with harvest date), and type of cover crops.The latter involves the applications of mineral fertilizer, manure, fungicide, herbicide, insecticide, growth regulator, and stalk stabilizer (table 1).In cases where multiple events occurred for given practices, timers were reset to one at each event.
For cases where the time since the last event of a practice is unavailable (i.e. the first occurrence of a practice in the management records) we set the timer to the duration of the assumed maximum legacy effect of management practices (supplementary figure 2).We further categorized mineral fertilizer by calcium, nitrogen, sulfur, and magnesium, and we accounted for the depth of tillage and the type of cover crop.

Statistical analysis 2.2.1. Isolation of the management practice impacts on surface albedo dynamics
We trained random forest models to disentangle the effects of management practices on daily surface albedo dynamics from other influences (see below) (figure 1).The random forest model effectively handles nonlinear relationships and can be used to disentangle the effects of different predictors on the target variables (surface albedo) (Friedman 2001, Svetnik et al 2003).We selected 28 potential features for which site information is available (table 1), including climatic and ecological features (n = 9), soil properties (n = 3), site identity (n = 1), days since management practices (n = 13, i.e. one for each management practice considered), types of cover crops (n = 1) and depths of tillage (n = 1).The selected features do not include crop growth or phenology which drive albedo dynamics during the cropping period, but instead included the climatic and edaphic drivers of crop growth.The nine daily climatic and ecological features were derived from the flux tower dataset (see section 2.1.1).The three time-invariant soil properties were obtained from the Harmonized World Soil Database (HWSD) based on the site locations (Wieder et al 2014).They were employed to characterize soil properties across all data points for each site.Site identity reflects site heterogeneity due to factors not accounted for.Each of the eight sites is identified by a categorical label ranging from one to eight.All features are continuous except for site identity, tillage depth, and cover crop type.At first, all features were included in the random forest models and recursive feature elimination was then implemented to select relevant features for a given period (supplementary text 1.2).Two random forest models were trained using the combined data from all eight sites, respectively: (1) for both growing and fallow periods spanning winter wheat sowing to the next main crop sowing (2645 data points); (2) for fallow periods after winter wheat harvest to the next main crop sowing (798 data points).The latter was done to identify factors influencing the albedo during the fallow period specifically (3-7 months).Harvest events always coincide with the occurrence of crop residues, and it is impossible to distinguish their effects (not shown).Thus, for interpretation of results we refer to the albedo dynamics on the days after harvesting events as the impact of harvest for growing and fallow periods and as the impact of crop residues for fallow periods (figures 3(a) and (c)).
The dataset was split randomly into 80% for training and tuning and 20% for testing.We applied 10-fold cross-validation to the training dataset for hyperparameters tuning (i.e.number of decision trees (ntrees) from 300 to 1000, number of variables sampled (mtry) from 2 to 7 at each split).The model with the optimal set of parameters was selected, trained on the whole training dataset, and evaluated using the testing dataset.We used the coefficient of determination (R 2 ) and root mean square error (RMSE) as quality indicators during tuning, training, and testing (supplementary table 6).Recursive feature elimination was used during the training step to reduce the number of predictors (Guyon et al 2002).We tested for overfitting by comparing the final R 2 and the R 2 during the cross-validation process.For more information see supplementary text 1.2 and supplementary figure 2. The duration of the legacy effects of management practices on daily surface albedo differs among practices.This analysis focuses on the short-term impacts of management practices, as the observation period is too short to disentangle effects which operate on a time scale of years and longer.Therefore, we artificially limited the maximum duration before the random forest training to 210 d for growing and fallow periods, and 150 d for fallow periods.These two durations were selected by testing a range of 30-300 d when the improvements of R 2 and RMSE with increasing duration in random forest models were levelling off (supplementary figure 4).A risk of confused feature contribution might occur if the duration is not limited, especially in fallow periods (not shown).
Table 1.Features used in the analysis, including climatic and ecological features, soil properties, site identity, and features related to management practices.Features of management practices contain timing and characteristics.

Features Labels Explanations
Air temperature ( Partial dependence plots were utilized to quantify changes in daily albedo since the application of each management practice (Friedman 2001).For each practice i, albedo change is defined as the difference between albedo estimated at a given day d (noted α pdp,m ) and the baseline albedo (α av,m ) for the wheat growing and fallow periods together (m = w) and fallow period alone (m = f ), respectively: ( Baseline albedo is estimated from the averaged albedo for time points after the maximum duration of a management effect till the same event occurs again (supplementary table 2).This estimate was used as a proxy for bare soil albedo, as no bare soil control was available at the sites included in this analysis.Partial dependence plots are not suitable to quantify interactions among features here, because of the small sample sizes available for each pair of features.

Climate impact resulting from management practices on surface albedo
The radiative forcing (W m −2 ) caused by surface albedo variation is defined as the change in energy flux at the top of the atmosphere (Jones et al 2015, Bright and Lund 2021).To quantify the impact of management practices on the climate via the induced changes in surface albedo, the radiative forcing was derived as follows (Carrer et al 2018): where RF ∆α(d, i) is the average of radiative forcing of albedo change in N days caused by implementing management practices; T au (d) represents upward atmospheric transmittance (Unitless), which can be retrieved from SW_IN/SW_IN_POT.∆α (d, i) has the same meaning as in equation ( 2).We computed the radiative forcing for (1) N days = 210 for the combined growing and fallow periods and (2) N days = 150 for fallow periods.We further converted RF ∆α(d, i) to equivalent CO 2 emission or removal using global warming potential (GWP) to compare with estimates from the literature of albedo (GHG)-induced radiative forcing (supplementary table 3).The calculation of GWP can be seen in supplementary text 1.3.

Comparison of the importance of features for daily albedo predictions
The trained random forest model, with optimized hyperparameters (ntrees = 600 and mtry = 3), showed a good performance in predicting the daily albedo over both the growing and fallow period with R 2 and RMSE of 0.64 and 0.02, respectively.When considering fallow periods only, the R 2 and RMSE were 0.58 and 0.02, respectively (figures 2(a) and (c)).The small difference between the R 2 derived from the test dataset and the R 2 obtained during the crossvalidation process (4%) revealed very low overfitting.
The ten most important features selected to predict the daily albedo depended on the periods of simulation.For growing and fallow periods, the most influential features were related to management practices as well as climatic and ecological features (in descending order of importance): Nitrogen fertilizer Herbicide, tillage, and harvest (crop residues) were identified as important drivers for both periods, while only fungicides and sowing were selected for growing and fallow periods as they were applied during the growing period of winter wheat only.With respect to climatic and ecological features, Kd affects surface albedo most significantly, followed by soil temperature and soil water content in both periods.Tests of feature importance in different durations of maximum legacy effects additionally support the dominant effect of nitrogen fertilizer and Kd on surface albedo dynamics (supplementary figure 5).

Time-variant impact of management practices on surface albedo
The top 10 ranked management practices lead to albedo changes ranging from −0.006 ± 0.001 (harvest) to 0.004 ± 0.007 (nitrogen fertilizer) (mean ± standard deviation) when averaged over 210 d during growing and fallow periods, and from −0.005 ± 0.002 (herbicide) to 0.003 ± 0.001 (crop residues) when averaged over 150 d during fallow periods (table 2).The largest change in albedo during growing and fallow periods was due to nitrogen fertilizer application, while during fallow periods, herbicides had the largest influence (figures 3(a) and (c)).
The albedo impact of the majority of practices varies strongly within the periods considered (figure 3).In growing and fallow periods, nitrogen fertilizer, fungicide, and herbicide increase surface albedo within the first 30-60 d after applications, followed by neutral or negative impacts (figure 3(a)).The impact of fungicide diminishes more rapidly over time compared with nitrogen fertilizer.Different from the input-related practices, the physical management practices show fewer variations in their influence on surface albedo over 210 d (figure 3(c)).Only the impact of sowing changes from negative (−0.005 ± 0.003) to positive (0.002 ± 0.001) after 120 d, attributed to the growing crop with increased albedo covering the soil surface that was destroyed during seedbed preparation.In fallow periods, the negative contribution of herbicide to the albedo change within 150 d of application is most significant, while the effect of tillage remains minor, especially after 20 d since application (figure 3(c)).

Climate impacts resulting from the effects of management practices on surface albedo
The transient albedo-mediated radiative forcing induced by management practices is small and ranges from −1.25 ± 2.15 W m −2 (nitrogen fertilizer) to 1.21 ± 0.93 W m −2 (tillage) in average over 210 d after application during growing and fallow periods.During fallow periods the range is from −0.66 ± 0.53 W m −2 (nitrogen fertilizer) to 1.33 ± 1.06 W m −2 (herbicide) in average over 150 d (table 2).Climatic and ecological features have comparable magnitudes of radiative forcing between −0.27 ± 0.62 and −1.31 ± 2.27 W m −2 averaged in the same two periods, respectively (supplementary table 4).Note that large temporal variations in all cases weaken their overall climate impacts.In growing and fallow periods, the nitrogen fertilizer and fungicide applications result in negative albedo-mediated radiative forcing within the first 60 and 30 d, with average values of −4.39 ± 1.22 and −1.47 ± 1.80 W m −2 , respectively.Subsequently, their radiative forcing becomes positive with average values of 0.03 ± 0.45 and 0.42 ± 0.24 W m −2 (figure 3(b)).A positive albedo-mediated radiative forcing of herbicide occurs after 30 d, with an average value of 0.92 ± 0.55 W m −2 .The tillage and harvest exhibit averaged positive albedo-mediated radiative forcing in 210 d, while the sowing has a minor negative albedo impact on climate (table 2).
In fallow periods, nitrogen fertilizer and crop residues lead to negative albedo-mediated radiative forcing, while herbicide and tillage induce negative albedo effects due to the decreased surface albedo with exposed soil (table 2).

Management practices and climate influence the seasonal albedo dynamics
Management practices and climate jointly modulate surface albedo dynamics during growing and fallow periods, while the climate effect dominates during fallow periods (figure 2).We discuss the correlation between the surface albedo and the top five features in these two periods.D N , D Fug , and D res relate to management practices, while Kd, soil temperature, soil water content, and wind speed refer to the climate in the following text.
Nitrogen fertilizers, fungicides, herbicides and other chemical applications impact surface albedo through sustained chemical effects on plant growth and short-lived physical alteration of the soil surface during application.The chemical effects of fertilizers and fungicides lead to an increase in albedo during the cropping period (figure 3(a)) which is in line with theory.Changes in leaf traits (e.g.pigmentation, cell structure), canopy structure, and changed soil exposure as leaf area index varies, can lead to a positive relationship between albedo and nitrogen and fungicides which declines over time.Canopy nitrogen, which increases after fertilizer addition, was shown to be positively correlated with albedo in a wide range of ecosystems, including croplands (Tranavic ˇienė et al 2007, Hollinger et al 2010).Fungicides can increase surface albedo as infected leaves are less pigmented (e.g. for tan spot or septoria leaf blotch infection Table 2.The averaged changes in surface albedo and the corresponding albedo-mediated radiative forcing due to the selected management practices.The albedo change is the difference between the albedo on the first day since application and the baseline albedo over growing and fallow periods (fallow periods).Shown includes the comparison of radiative forcing of different management practices (kg CO2e ha −1 yr −1 ) related to their induced albedo changes and previous estimates on the albedo-mediated biophysical impacts and the greenhouse gas (GHG)-mediated biogeochemical impacts (kg CO2e ha −1 yr −1 ).All radiative forcing in previous research were converted to the unit of kg CO2e ha −1 yr −1 for comparison.Details of previous estimates of radiative forcing can be found in supplementary table 3.During the application of chemicals, the surface is altered depending on the machinery used (e.g.type of sprayer) and the conditions during the time of application.Solid fertilizer in the form of white or lightcolored granules which make up 70% of fertilizers might have a direct effect on surface albedo till they completely dissolve (Ni and Pacholski 2022).This could explain the initial increase in albedo during the first 5 d after application (figure 3(a)).Spraying of liquids leaves reflective water droplets on leaf surfaces which can stay up to a few days (Rasool et al 2022, Vázquez et al 2022) which could also explain the initial increase in surface albedo within a few days of fertilizer and fungicide application (figure 3(a)).In addition, the vehicles used for application leave tire tracks which might decrease surface albedo because the vehicle tracks can exhibit wetter soil for a couple of hours.Less area is affected for a shorter time by physical effects than by the chemical ones the effects should be smaller.This can explain why the importance of fertilizer and herbicide application was low during fallow periods where chemical effects are marginal due to the absence of crops (figure 2(d)).
Covering crop residues on the soil surface tends to increase albedo due to light color (figure 3(c)) (Sieber et al 2022).This effect varies with factors such as orientation (standing or on the ground), surface coverage, decomposition stage, and water content (dry or wet) of residues, as well as tillage practices (Daughtry et al 2010).Crop residues can also reduce soil albedo by influencing the surface energy balance (Horton et al 1996, Du et al 2022) and soil properties over the course of years (Jin et al 2020).Both were not investigated in this study.
Distinguishing the effect of crop residues from harvesting is challenging as they often co-occur at the same time (supplementary table 5).The opposite albedo changes in growing and fallow, and fallow periods are mainly caused by different baseline albedo (see below).Overlapping between management practices also occurs for tillage and sowing (54%), and nitrogen fertilizer, fungicide and herbicide (30%), thus, the independent impacts of these practices on albedo dynamics are inevitably influenced by their concurrent implementation.
Unlike management practices, climatic factors are equally important for the cropping and fallow period.Kd modulates surface albedo by affecting the quality and quantity of the surface solar radiance reaching the surface.During the cropping period it also affects the fraction intercepted by crop canopy which affects photosynthesis and subsequent crop growth, canopy structure and leaf traits (Pinty et al 2005, Yang et al 2019, Liu et al 2022).
Surface albedo, soil temperature, and soil water content are tightly interconnected via surface energy balance (e.g.latent and sensible heat fluxes) which is illustrated by the high importance of soil temperature and soil water content for both periods (figures 2(b) and (d)).In addition, soil water content negatively modulates surface albedo by controlling soil color: the higher the soil water content (darker soil), the lower the surface albedo (supplementary figures 6 and 7) (Ni et al 2019, Yang et al 2020).Soil water content and temperature are interconnected and are influenced by management practices which were not assessed in this study.The variation in wind speed can modulate indirectly surface albedo during fallow periods by influencing the soil evaporation rates and thereby the soil water content (soil color) (Luu et al 2023).It is also considered as a feedback mechanism in albedo-induced land-atmosphere interaction (hydrothermal transport) related to the surface roughness (e.g.near-surface turbulence intensity) (Bhimireddy et al 2022).The indirect impact of wind speed on surface albedo explains the lower importance ranking of wind speed, compared with Kd, soil temperature, and soil water content which are directly related to the surface energy balance (figure 2(d)).
Management practice-induced changes in surface albedo could affect the two-way albedo-climate interaction by relating to variations of Kd, soil temperature, soil water content, and wind speed.This includes the radiative process (e.g.solar radiance reflectance) and non-radiative process (e.g. the land-atmosphere hydrothermal transport) (Chen et al 2020).The former drives the albedo-mediated radiative forcing at the top of the atmosphere (figures 3(b) and (d)), while the latter modulates the surface and atmosphere dynamics (supplementary figures 6 and 7).For example, the modified surface albedo by practices changes Kd by affecting the regional cloud and aerosol concentration via a non-radiative process, depending on aerodynamic conductance and temperature-humidity gradients (Zeng et al 2017, Chen et al 2020).
Management practices have varied legacy effects (supplementary figures 6 and 7).The nitrogen fertilizer and fungicide have cooling potential lasting 150 and 90 d after application and herbicide impacts after 60 d in growing and fallow periods (supplementary figure 8).While influences of all management practices remain relatively stable within 150 d during fallow periods (supplementary figure 9).The climate impact of management practices also depends on the incoming solar radiance and atmospheric transmittance (Hansen et al 1997), but this is not accounted for in our analysis.For example, applications of crop residues and cover crops usually occur in summer and fall, respectively.The seasonal variations of surface solar radiance and atmospheric transmittance that encompass the cloud cover effect and the effect of aerosols in the atmosphere might drive different albedo-mediated radiative forcing induced by these two practices even though they have similar albedo changes.
The albedo-induced climate impact of management practices depends on the considered periods and the baseline albedo which is used to derive the climate impact from partial dependence plots (see equation (2)).For example, herbicide application during the growing season improves the growth of the main crop and increases surface albedo, but its application during fallow periods exposes bare soil and decreases surface albedo (supplementary figures 8(d) and 9(d)).Differences in the baseline albedo between the two periods, explain the difference in the albedo-mediated radiative forcing for tillage between periods (table 2).The higher baseline albedo for growing and fallow periods which includes periods of high crop coverage, unlike during fallow periods, explains the higher albedo impact of tillage which exposes the bare soil.Hence, assessing and comparing the climate impacts of management practices require harmonization of baseline albedo and periods under consideration.
Our approach has shortcomings.The amount of fertilizer is omitted due to insufficient data, but instead, we assumed an optimal amount was applied at each site which might not always had been the case.Cover crops and biochar are also not evaluated here but were shown to have been addressed as having significant impacts on surface albedo through alterations in crop structure and leaf traits, and soil pigmentation, respectively (Genesio et al 2012, Lugato et al 2020).In our few sites, there was no biochar application, and only 106 data of cover crops (13% of total 798 data in fallow periods) with 11 types.Therefore, cover crop types were not retained after recursive feature elimination.

Albedo-mediated biophysical impacts can cancel or amplify the impacts of GHG emissions
The albedo-mediated impact of a management practice can be in the same direction or opposite to the biogeochemical-mediated one (table 2).In the following, GHG-mediated impacts include CO 2 and N 2 O emissions unless other specified uncertainty ranges are given if available.The albedo-mediated biophysical impact of nitrogen fertilizer is much smaller than the 270 day GHG emission-based biogeochemical impacts (798.35-929.34kg CO 2 e ha −1 yr −1 ) estimated for seven European winter wheat sites (Ceschia et al 2010).Because of the two opposite impacts, the overall climate impact of nitrogen fertilization is likely slightly smaller than previously estimated from biogeochemical impacts only.We found albedo-mediated radiative forcing of herbicide and fungicide which are in the same order of magnitude or larger than the ones from GHG emission of pesticide (30.07 ± 20.17 kg CO 2 e ha −1 yr −1 ) (Ceschia et al 2010).However, as the signs of biophysical impacts differ among the types of pesticides and the dependence of results on the period under consideration (section 4.1), a comparison is not straightforward.
The biophysical radiative forcing from sowing has the same sign but is much smaller than the impact related to the carbon contained in seeds of −227.87 kg CO 2 e ha −1 yr −1 in Ceschia et al (2010).The positive albedo-mediated impact of harvest estimated in our analysis impacts climate negatively but is smaller than the carbon removal of 6545 kg CO 2 e ha −1 yr −1 from harvesting winter wheat biomass (Ceschia et al 2010).Note that the harvested organic matter, if used in buildings, could lead to small sinks but not accounted for here (Ahmadi et al 2020).Crop residues that remain on the surface after harvest can lead to N 2 O emissions equivalent to 108.95 kg CO 2 e ha −1 yr −1 which is comparable in magnitude to the albedo-mediated biophysical ones either in this analysis or −79.30kg CO 2 e ha −1 yr −1 in previous research (Sieber et al 2022).The interaction of the biophysical and biogeochemical impacts of residues depends on the area covered and the lifetime of residues (Flower et al 2022).Previous analysis has shown the biogeochemical impact of 312 kg CO 2 e ha −1 yr −1 from tillage via decreased soil carbon storage (Liu et al 2022).A decreased N 2 O emission due to tillage was not found for winter wheat but for other crop types (Niu et al 2019).The positive albedo-mediated biophysical impacts from tillage were 34.37-255.55kg CO 2 e ha −1 yr −1 in this analysis, and 288.00-234.00kg CO 2 e ha −1 yr −1 in 52 d for 15 plots in Sweden (Sieber et al 2022).Both the biogeochemical and albedo-mediated biophysical impacts indicate an overall favorable impact of reduced use of tillage.

Conclusion
Comparing the albedo dynamics and related climate impacts by seven management practices, this analysis shows that albedo impacts are usually small but of similar levels as biogeochemical effects.It can provide valuable information for dedicated field trials and monitoring systems able to consider the substantial temporal variation we found here.Moreover, our study is the first to reveal an effect of nitrogen fertilizer on albedo, which suggests that changes in leaf color should be further investigated.Our findings also highlight that the effects of practices during growing and fallow periods such as nitrogen fertilizer and fungicides are stronger than the effects of practices during the fallow periods, like crop residues and tillage.This indicates that the crop period is more suited for climate mitigation than the fallow period.Uncertainties in our results are primarily due to limited data availability, such as few data assessing the effect of cover crops and difficulties in establishing the baseline albedo.Accurate bare soil albedo estimates from earth observation, albedo monitoring of fields, and dedicated field experiments are needed.The inclusion of other biophysical climate impacts (e.g.evapotranspiration) is needed to identify climate solutions by accounting for the interactions between agriculture and climate change.The feasibility of management practices should also consider the socio-technical-economic constraints, the interests of farmers, and the need for climate resilience.

Figure 1 .
Figure 1.The flow chart of the data processing procedure.
D N , Kd, soil temperature, soil water content, D Fug , D Har , D Her , air pressure, D sow , and D Til .For fallow periods, climatic and environmental characteristics were more influential than management-related features: Kd, soil temperature, soil water content, wind speed, D res , sand and clay contents, D N , D Til , and D Her (figures 2(b) and (d)).

Figure 2 .
Figure 2. The performance of the random forest model for predicting daily surface albedo for (a) and (b) growing and fallow periods, and (c) and (d) fallow periods.(a)-(c) are the scattering plots between the predicted surface albedo from the random forest model and the observed surface albedo for the test dataset.(b)-(d) show the importance ranking of selected features using the Gini index.Kd, Temps, SWC, ws, PA, SiteIden, DN, DFug, DHar, DHer, Dsow, Dss, D Til , and Dres represent the ratio of diffuse solar radiance, soil temperature, soil water content, wind speed, air pressure, site identity, days after nitrogen fertilizer, fungicide, harvest, herbicide, winter wheat sowing, stalk stabilizer, tillage, and crop residues.Orange, blue and green boxes are features related to climatic and ecological features, days since the applications of management practices and site identity, respectively.

Figure 3 .
Figure 3. Changes in surface albedo and corresponding albedo-mediated radiative forcing due to the application of management practices as a function of time since onset of practice.Shown are daily albedo differences averaged over eight sites estimated from partial dependence and baseline albedo (see equation 2).(a) and (b) results for growing and fallow periods (c) and (d) for fallow periods.DN, DFug, DHar, DHer, Dsow, D Til , and Dres represent days since the applications of nitrogen fertilizer, fungicide, harvest, herbicide, winter wheat sowing, tillage and crop residues, which are in the top 10 ranked important features selected by the random forest model.