From shifting cultivation to teak plantation: effect on overland flow and sediment yield in a montane tropical catchment

Soil erosion supplies large quantities of sediments to rivers of Southeastern Asia. It reduces soil fertility of agro-ecosystems located on hillslopes, and it degrades, downstream, water resource quality and leads to the siltation of reservoirs. An increase in the surface area covered with commercial perennial monocultures such as teak plantations is currently observed at the expanse of traditional slash-and-burn cultivation systems in steep montane environments of these regions. The impacts of land-use change on the hydrological response and sediment yields have been investigated in a representative catchment of Laos monitored for 13 years. After the gradual conversion of rice-based shifting cultivation to teak plantation-based systems, overland flow contribution to stream flow increased from 16 to 31% and sediment yield raised from 98 to 609 Mg km−2. This result is explained by the higher kinetic energy of raindrops falling from the canopy, the virtual absence of understorey vegetation cover to dissipate drop energy and the formation of an impermeable surface crust accelerating the formation and concentration of overland flow. The 25-to-50% lower 137Cs activities measured in soils collected under mature teak plantations compared to soils under other land uses illustrate the severity of soil erosion processes occurring in teak plantations.

main supply of sediment export at the outlet needs to be more thoroughly investigated, as this phenomenon is in competition with other processes (e.g. reinfiltration and sedimentation, subsurface flow contribution). Although the direct measurement of overland flow is possible at the plot and hillslope scales, it is difficult to carry out similar measurements at the catchment scale. At this larger scale, indirect tracer-based approaches provide an effective method to calculate the respective contributions of subsurface and surface flow. Typically, water mixing models using geochemical properties (e.g. electric conductivity, δ 18 O) are applied to quantify these contributions 15,16 .
In forested catchments, subsurface contributions were shown to provide the dominant contribution to streamflow 17,18 , whereas surface overland flow was identified as the main supply in cultivated catchments where an almost impermeable crust may be formed at the soil surface 15,19,20 . Most land-use change studies investigated the conversion of forests into cropland, showing the associated increase in soil erosion 21,22 . To our knowledge, only few studies have been conducted on the conversion of annual crops into commercial tree monoculture plantations on the steep hillslopes of humid tropics although this rapid land-use change is affecting extensive areas across this region 23 . The main objective associated with the reforestation of degraded land is often to restore the soil infiltration capacity and to limit soil erosion. However, it was shown that this reforestation may also increase plant water uptake and consequently reduce river baseflow 24,25 .
To investigate the impact of land-use change on sediment yield, the monitoring of paired catchments is often conducted in nearby sites where contrasted land uses prevail while biophysical characteristics are as similar as possible. This synchronic approach suffers from several limitations because of the difficulty to select two perfectly identical catchments 26 . In contrast, this approach is easier to conduct in similar zero-order headwater catchments (i.e. hollow valleys referred to as S7 and S8) with similar soil characteristics although different land uses. In addition to this synchronic approach, we applied a diachronic investigation in the larger catchment affected by land use change and draining these hollow valleys. The 0.6-km² Houay Pano catchment (Fig. 1), located in Northern Laos, is representative of areas that experienced large-scale afforestation with commercial teak tree (Tectona grandis L.f.) plantations at the expense of traditional slash-and-burn (i.e. shifting) cultivation. Discharge and sediment yield were monitored for 13 years in this experimental catchment.
The contributions of surface (i.e. overland flow) and sub-surface flow (i.e. groundwater) to river water at the catchment outlet were assessed by measuring the river water electrical conductivity. This technique is easier to implement than the more conventional method based on δ 18 O, and both techniques were found to provide similar results 15 . The contributions of surface and sub-surface soil sources to stream sediment exports at the catchment outlet were discriminated by measuring the activity of Caesium-137 ( 137 Cs) emitted by thermonuclear bomb tests in the 1960s in the sediment transiting the river 27 . This radioisotope characterised by a half-life of 30 years shows highly contrasting activities between the surface of cultivated soils exposed to atmospheric fallout and gully/ channel bank material sheltered from this fallout [28][29][30] .

Results
Land-use change. The proportion of secondary forests in the catchment decreased from 16% (2002) to 8% (2014) (Fig. 2b). During this period, the variations in the respective proportions of fallow (29-69%) and annual crops (4-42%) were negatively correlated and they showed large inter-annual variations, reflecting the management strategies of replacing fallows with annual crops through slash-and-burn, followed by the regrowth of vegetation in the temporarily abandoned cultivated fields. This entire cycle may take 2 to 6 years (Fig. 2b). Until 2007, the percentage of catchment area covered with teak plantations (Teak) remained low and did not exceed 4%. From 2008, it continuously increased, almost linearly, to reach a maximum of 36% of the catchment area in 2014 (Fig. 2b), illustrating the transition from an annual crop-based system towards an organisation dominated by tree plantations. Accordingly, the monitoring period was divided into a first sub-period (2002)(2003)(2004)(2005)(2006)(2007) characterized by the alternation between upland rice (Oryza sativa L.), Job's tear (Coix lacryma-jobi L.) annual crops and bush fallow, in addition to sparse teak plantations, and a second sub-period (2008-2014) characterised by the concomitant expansion of teak plantations and the reduction of annual crops. During the last five years of this second sub-period, most of the teak plantations were more than 3 years old and characterised by limited understorey vegetation cover, which may enhance soil degradation. In 2014, land use greatly differed between the two hollow valleys investigated in details (Fig. 1b), with a dominant contribution of typical shifting cultivation landscape mosaics at S7 compared to a major contribution of teak plantations at S8.
A correlation analysis was performed to investigate the relationships between the catchment hydrological behaviour, land uses and sediment yields. The main objective of this analysis was to evaluate how the transition from an annual crop-based system towards a tree plantation-based system affected the hydro-sedimentary response of the catchment. We evaluated the significance of the correlation between several variables reported in Table 1. Among these parameters, the areal percentage of more than 3-year-old teak tree monocultures (Teak >3 ) was considered. Teak >3 was used in order to exclude the youngest plantations which were not yet submitted to burning (Fig. 4a), and in which soils were not yet degraded.
Contrary to the total annual overland flow depth (OLF), which is correlated to Rain (p-value < 0.05;), F OLF is not. In contrast, F OLF is positively correlated to Teak >3 (p-value < 0.05; Table 1) and negatively correlated to the sum of fallow and forest areas (Fw&Fo) (p-value < 0.05), and it provides therefore a better proxy of land use change. This is also confirmed by the regression analysis with a positive correlation with annual crops cultivated alone or in combination with one-or-two-years-old teak (AC) (Fig. 4a). InfRate is very significantly and negatively correlated to Teak and Teak >3 (p-value < 0.01) ( Table 1).
Sediment yield was strongly correlated to the annual suspended load (SL) (p-value < 0.001), which is the main export route for soil particles (SL account for 80 and 78% of SY, on average, during the first and the second sub-periods, respectively). Annual bed load (BL) is also well correlated to SL (p-value < 0.01). Interestingly, SY (i.e. BL + SL) is correlated to Rain_F (p-value < 0.001, < 0.01 and < 0.001 respectively; Table 1) and F OLF (p-value < 0.05, 0.01 and 0.05 respectively), which agrees with the regression equation (Fig. 4b). A negative correlation was observed between the areal percentage of Fw&Fo and soil losses. However, it was significant for BL only (p-value < 0.05) ( Table 1).
Multiple log-linear and linear regressions were conducted to identify explanatory variables that best predicted the inter-annual variations of F OLF and SY. All the variables listed in Table 1 were considered in this analysis. A best sub-set of two explanatory variables was selected in each regression: Teak >3 and the areal percentage of annual crops cultivated alone or in combination with one-or-two-years-old teak (AC) for F OLF (Fig. 3a); Rain_F and F OLF for SY (Fig. 4b). While the regression model predicting SY was linear, a stronger log-linear relationship

Discussion
Impact of land-use change on infiltration rates and overland flow. In contrast to teak plantations 9 , fallows are characterised by higher infiltration rates associated with a higher biological activity maintaining greater porosity and lower surface crusting 31 . Several vegetation strata, including understorey, are observed in fallows. By dissipating the kinetic energy of raindrops, they reduce the soil surface crusting. In contrast, under teak canopies where large leaves concentrate rainfall, heavier rain drops fall from higher interception heights, impacting the soil with a greater kinetic energy 30 , clogging the soil surface porosity and disturbing the soil surface structure, both processes contributing to crusting and the associated reduction in infiltration rates. Consequently, the production of overland flow increases and detaches more soil particles by sheet flow. The contribution of surface overland flow to the total stream flow increased during the second sub-period. This evolution reflects a change in the hydrological behaviour of the catchment, which coincides with the increase   of the teak cover at the expense of fallow and annual crops. Until it is two-year old, teak is generally interspersed with annual crops (Fig. 2b). From the third year, if the natural understorey vegetation is maintained, it will protect the soil from the raindrop impact and the subsequent surface crusting 9 . However, in the study area, understorey vegetation is systematically removed and/or burnt by the farmers, and soil is left bare and exposed to raindrop and overland flow (Fig. 6a,b and c). Consistently, F OLF increases in correlation with the increase in Teak >3 (Table 1). Nevertheless, during the four last years of the second sub-period, overland flow slightly decreased with the increase of Fw&Fo (Fig. 2b). These observations are consistent with plot-scale measurements conducted in the same catchment, showing that infiltration rates under fallow are higher than those found under teak plantations 9 .  Fig. 2e). These high values cannot be related to the areal percentage of annual crops, which decreased from 40 to 4% during the second sub-period. Rainfall has long been known as a major control factor for soil erosion at the plot scale 32 , possibly explaining the high SY values observed after 2011. Indeed, high annual rainfall depths were measured in 2011 (1884 mm) and 2013 (1524 mm). However, the SY value measured in 2004 (232 Mg km −2 ) is lower than that of 2012 (609 Mg km −2 ), despite a 23% higher rainfall depth (978 mm and 1305 mm, respectively) and a similar areal percentage of annual crops (29% and 24% respectively). However, the areal percentage of Teak >3 increased from 2 to 18% between 2004 and 2012. Therefore, the most likely factor that may explain this difference is the increase in the areal percentage of teak plantations. This is demonstrated by the different behaviour observed in contrasted sub-catchments (S7 Since sediment yield is best explained and positively correlated to Rain_F and F OLF (Fig. 4b) and F OLF is best explained and positively correlated to Teak >3 and AC (Fig. 4a), it is consistent that the higher the surface covered with fallow and secondary forest, the lower is the sediment yield. In contrast, the higher the surface covered with teak plantations and annual crops, the higher is the soil loss.

Impact of land-use change on sediment yields.
Whatever the period considered, subsurface flow was the major contributor to stream discharge during floods. This process can lead to the more frequent collapse of channel banks and the remobilisation of material deposited in the channel 33 . Groundwater outflow can also decrease the concentration of suspended sediments in the river 16,29 .
Measurements of 137 Cs activities showed that soil erosion was so intense under teak plantations that the content of their soil surface in 137 Cs decreased to reach values measured in subsurface material. This trend is reflected by the decrease in 137 Cs activities with the increasing age of the teak plantations, revealing the major role of erosion in removing surface soil layers through the increased generation of rills and gullies 12 . This result is further supported by the fact that teaks were preferentially planted in the most accessible zones from the village, i.e. in areas with the deepest and most fertile soils.

Synthesis and recommendations for future land-use management in Southeastern Asia. In
Southeastern Asia, incentives to reduce shifting cultivation have induced rapid land-use changes, leading to a dramatic reduction in the cropland area and to a shortening of fallow periods 34 , which counter-productively resulted in an increase in land degradation and soil erosion as well as in a decline of crop yields 35,36 . The lack of sustainability of these transitional farming systems 37 has prompted the need for developing alternative production systems. These include perennial and monoculture tree plantations such as rubber (Hevea brasiliensis, Muell. Arg.) and teak (Tectona grandis, L.), which are rapidly expanding throughout montane environments 23 , as observed in the Houay Pano catchment (Fig. 7).
Most studies that investigated the impact of afforestation in montane cultivated catchments have demonstrated a reduction of soil losses at the hillslope level 38 . This trend is attributed to rainfall interception by the canopy, and to a reduction of the throughfall kinetic energy and the resulting splash erosion as the soil is better protected by various strata of vegetation cover 39,40 . Furthermore, under these conditions, the soil surface hydrodynamic properties (i.e. hydraulic conductivity) recovered, as a result of the intensification of biological activity 41 , improving infiltration and the vertical drainage of rainfall and thereby reducing soil losses due to overland flow.
In the region where the current research was conducted, teak tree afforestation induced an opposite hydro-sedimentary response (Fig. 7). Most farmers intentionally kept the soil bare under mature teak trees through the frequent burning of the understorey. This practice resulted from a mix of local beliefs and practical considerations including the supposed need for a reduced competition for soil water and nutrients, and an improved access to and a better circulation within plantations. The soil surface in these managed teak tree plantations is exposed to raindrops having a kinetic energy exceeding that recorded in non-intercepted rainfall; the reason for this high kinetic energy is that the large leaves of teak intercept and concentrate rain drops which subsequently fall from an effective height exceeding the threshold (8 m) above which velocity no longer increases with height due to air resistance 42 . Besides detachment, exposition of soil to the splash process generates the formation of crusts (Fig. 6c) and the progressive clogging of soil porosity. Because of these two processes, surface soil structure is destroyed and soil surface becomes impermeable, which limits subsurface flow and intensifies overland flow, increasing the generation of rill and gully erosion occurring on these steep slopes. Moreover, reforestation with teak may also decrease baseflow due to the increased root water uptake and evapotranspiration 43 .
Based on the log-linear and linear relationships found in this current research (Fig. 4a,b), overland flow and total sediment yield by 2020 were estimated under the following scenario: (i) an expansion rate of teak tree plantations similar to that observed during the second sub-period (2008-2014) with teak plantations covering 75% of the catchment area by 2020; (ii) an unchanged fractional cover of annual crops and (iii) the stability of triggering-flood annual rainfall depth at a value corresponding to the average measured during the entire study period. Under these conditions, the statistical relationships predicted an increase of overland flow contribution during floods, from 28% in 2014 to approximately 40% in 2020, with a concomitant tremendous increase in annual sediment yield, from 429 Mg km −2 year −1 in 2014 up to approximately 600 Mg km −2 year −1 in 2020, which exceeds by far the soil loss rate of 250 Mg km −2 year −1 considered by most soil scientists as the highest tolerable at the plot scale 44 , and which generally exceeds by far soil loss rates observed at the catchment scale.
As a recommendation, removal of ground vegetation regrowth and burning of the understorey vegetation under teak tree plantations should be strongly discouraged to avoid the degradation of the soil surface (soil surface crusting) and the subsequent increased generation of overland flow and soil loss. Alternatively, planting a shade-tolerant crop as an understorey potentially providing some economic return could be tested; candidate species include ginger (Zingiber officinale), cardamom (Elettaria cardamomum), sweet potato (Ipomoea batatas) or broom grass (Thysanolaena maxima). However, improper harvesting practices of these tuber or rhizome crops may also generate severe soil erosion. One promising avenue to improve the system is to train the farmers to forestry with lower planting densities without necessarily reducing the timber yield per surface unit.

Methods
Study site and land uses. The Houay Pano catchment (60 ha) is located 10 km south of Luang Prabang in northern Laos, in the Mekong River basin (Fig. 1). Mean annual temperature is 25.3 °C 10 .The tropical monsoon climate of the region is characterized by the succession of dry and wet seasons, with 80% of annual rainfall occurring from May to October 45 . Water and sediment fluxes were monitored at the outlet of the 60-ha catchment (Fig. 1a) 1 since 2001 (at the so-called S4 station). In addition, during the rainy season in 2014, fluxes were monitored at the outlet of two zero-order headwater nested catchments (i.e. hollow valleys, so-called S7 and S8 stations). Soils consist of deep ( > 2 m) and moderately deep ( > 0.5 m) Alfisols, except along crests and ridges where Inceptisols are found 46 . The main land uses in the area are annual crops (upland rice and Job's tear), fallow, teak plantations and secondary forests 8,47 . Their respective dominance in the catchment changed during the last decade: the slash-and-burn cultivation system that dominated between 2002 and 2007 has gradually been replaced with teak tree plantations 48 (Fig. 2b). Detailed land-use surveys and mapping were conducted each year. The areal percentages of each land use (±10%) in the catchment were calculated using QGis (http://www.qgis. org/). S7 and S8 sub-catchments have similar soil characteristics (Fig. 1a), surface areas (0.60 and 0.57 ha) and mean slope gradients (62 and 54%) although, during the monitoring year 2014, they greatly varied in land use. In S7, the main land uses were teak monoculture with an abundant understorey preserved from burning (22% of surface area), and typical shifting cultivation mosaics (78% of the surface including upland rice 23%, Job's tears 14%, fallow 34%, secondary forest 7%). In contrast, the S8 sub-catchment was covered with teak monoculture with a degraded soil surface in lower parts (42% of the total surface), and secondary forest (58%) in upper parts.
Hydro-sedimentary monitoring. Rainfall was measured every 6-min using 1 automatic tipping-bucket gauge and 6 manual gauges across the catchment (Fig. 1). The maximal spatial variability of annual rainfall is estimated at 20%. Stream discharge was calculated from the continuous monitoring of the water level since 2002. Water level was recorded with a OTT Thalimedes device after every 0.5 cm variation. Uncertainties on river discharge are estimated to be below 10%. Suspended sediment was collected using an automatic sampler (Fig. 6e) during each flood event during the 2002-2014 period (samples were collected after every 2-cm water level variations during the rising stage and after every 5-cm water level variations during the falling stage). Bedload sediment was estimated each month using a sediment trap (Fig. 6e). Suspended and bed load sediments were estimated following the approach described by Valentin et al. 1 . An average of 20 river water (600 mL) samples was collected ~10 cm below the river surface during each flood event. Overland flow was estimated based on electrical conductivity 6 measured in these samples. Infiltration rates were estimated by calculating the difference between rainfall and overland flow divided by the duration of the rainfall event. Total sediment fluxes correspond to the export of both bedload and suspended sediment, and these values are provided with 10% uncertainties 1 . All hydro-sedimentary data are available on-line (http://msec.obs-mip.fr/). Soil and sediment sampling. Soil surface samples (top 0-5 cm) were collected with 100-cm 3 soil density cylinders along toposequences 8 between 2000 and 2005 (n = 55) and in 2012 (n = 32) to characterize potential sources of sediment (Fig. 1) 29 . In addition, gully and channel bank samples (n = 23) were collected in 2012 by scraping the material over the entire exposed surface of channel banks (up to 50 cm) and gullies (1-5 cm) to characterize subsurface sources (Fig. 1) 27 . In teak plantations, soil surface samples (top 0-5 cm, n = 35) were collected along toposequences by scraping soil, 20 cm upstream and downstream of trees, as well as between trees, under teak plantations of different ages (5, 19, 25 years). Samples were then regrouped for each type of plantation. Finally, a composite sediment sample representative of the total annual export at the catchment outlet (S4 station) was prepared for 2002 (dominance of slash-and-burn cultivation) and 2012 (period of expansion of teak plantations).
Fallout radionuclide activities of the soil and sediment samples. Caesium-137 ( 137 Cs) was measured at the Laboratoire des Sciences du Climat et de l'Environnement (Gif-sur-Yvette, France) following the methods detailed by Gourdin et al. 29 . All results are expressed in Bq kg −1 with all activities decay-corrected to 2012. Uncertainties on radionuclide activities reached up to 30%, depending on the counting time and the quantity of material available for analysis.

Statistical analyses.
Multiple regressions were computed to determine linear or log-linear relationships 49,50 between the independent variables calculated for each year, namely contribution of overland flow during floods (F OLF , annual median percentage of total stormflow) and total annual soil loss (SY), and several candidate explanatory variables calculated for each year (i.e. Rain, Rain_F, OLF, F OLF , InfRate, SL, BL, SY, Fw&Fo, AC, Teak, Teak >3 ). To predict each independent variable, the selection of the best set of explanatory variables was guided by 'best subsets' and 'step-wise' regressions, two selection algorithms -available in the Minitab 17 free trial statistical packages (https://www.minitab.com/en-us/products/minitab/) -that intend to maximize the prediction R-squared (R 2 pred ) calculated by leave-one-out cross-validations. R 2 pred reflects the ability of the model to predict observations which were not used in the model calibration. An explanatory variable was considered to be statistically significantly different from zero when its p-value, derived from the Student's t-test, was lower than 0.05. The required homoscedasticity of the model residuals was verified by visual inspection. Possible multi-collinearity among the explanatory variables was controlled with the variance inflation factor 51 .
The Tukey-Kramer Honestly Significant Difference (HSV) test was applied to compare 137 Cs activities measured in various soil and sediment sample groups, after rejection of the ANOVA null hypothesis of equal means ( Table 2).