Composition of weed communities in seasonally flooded rice environments in East Africa is determined by altitude

Weeds are major biotic constraints to rice production worldwide. Compared to other sub-regions, weed communities of rice are not well described for East Africa and there is limited information on environmental factors affecting the distribution of species. This study aimed to address these knowledge gaps. Seasonally flooded rice production fields of 31 sites in Rwanda, Tanzania, Kenya and Uganda, across three altitude classes (Low: <200 m; Medium 200-1,000 m; High: >1,000 m), were surveyed for weed species using quadrats. Data analyses involved multivariate approaches, non-parametric Kruskal–Wallis tests and logistic regressions, followed by calculation of ranked species abundance and Shannon Weiner Index diversity analyses. A total of 286 weed species, belonging to 59 families, were recorded with 42 species not previously reported as lowland rice weed in the sub-region. Twenty-four species were identified as abundant across altitudes. Weed species diversity was higher at medium altitudes compared to high and low altitudes. Significant patterns of floristic distinction between altitudinal classes were observed, with 80% of dissimilarity. The high altitude was dominated by Echinochloa colona , Leptochloa squarrosa and Sphaeranthus suaveolens, the medium altitude was dominated by Crassula granvikii , Pycreus lanceolatus and Ageratum conyzoides while the low altitude was dominated by E. colona, Cyperus difformis and Cyperus esculentus . The weed species composition of seasonally flooded rice fields in East Africa is diverse. Identification of a limited group of (24) commonly abundant weed species as well as the articulation of altitude-specific weed species groups will facilitate the development of better tailored weed control programmes.


Introduction
Rice is an increasingly important staple food in Africa (Seck et al., 2012).In 2013, the total area under rice in Africa was estimated at nearly 10 million ha, of which 61% in West and 26% in East Africa (Diagne et al., 2013).It is primarily grown in three types of environment, conventionally classified as irrigated lowland, rainfed lowland and rainfed upland.The classification in lowland and upland is not referring to the altitude of the production site, but rather to their soil and water characteristics.Upland rice fields have freely draining soils that are rarely flooded while lowland fields have waterlogged soils that are subject to controlled (irrigated lowlands) or uncontrolled (rainfed lowlands) seasonal flooding.In East Africa 55% of the area under rice is characterised as rainfed lowland, and 27% as irrigated lowland (Nhamo et al., 2014).Compared to West Africa a relatively large share of these production areas is located on higher altitudes, between 700 and 1,600 m a.s.l.(Balasubramanian et al., 2007).
Among the biological constraints to rice production, competition from weeds is the most important (Seck et al., 2012).Across rice systems in Africa weeds have been conservatively estimated to cause US $1.4 billion worth of produce losses annually (Rodenburg and Johnson, 2009), of which parasitic weeds may be responsible for US $200 million (Rodenburg et al., 2016a).At the field level, weed-inflicted yield losses despite control efforts have been estimated at 15% in irrigated and 23% in rainfed lowlands (Becker et al., 2003;Becker andJohnson, 2001a, 2001b) and weed-inflicted yield loss abatements by frequent weeding interventions were estimated at 1.5 t ha -1 in rainfed lowland and 2 t ha -1 in irrigated lowland rice (Rodenburg et al., 2019).African smallholder rice farmers are however often limited by poor access to the necessary resources, technologies, markets and information that would enable them to implement effective control to prevent such weedinflicted yield losses (Rodenburg et al., 2019).One of the requirements for effective weed control strategies is knowledge about the species composition of the targeted weed vegetation.Compared to other important rice production areas worldwide, the available information about weed species in rice in East Africa is fragmentary at best and therefore a thorough inventory is required.A survey in this region, where rice is grown at a wide range of altitudes, also enables the analysis of an elevation effect on weed species distribution.
There are two, non-exclusive, overarching factors influencing the distribution of weed species in arable agriculture.Climatic and edaphic properties determine the occurrence and distribution of weed species (Pinke et al., 2012(Pinke et al., , 2010)).Topographical variables such as slope, altitude, longitude and latitude further influence the plant diversity and community composition (Hanzlik and Gerowitt, 2011;Lososová et al., 2004;Nowak et al., 2016).
Human interventions-through different agronomic practices, including the use of herbicides, tillage, crop rotations and organic or inorganic fertilisers-affect presence and abundance of species in the weed community (Fried et al., 2008;José-María et al., 2011;Lim et al., 2015;Pinke et al., 2014).For rice in Africa previous studies assessed effects of flooding depth and duration (Kent and Johnson, 2001) or water management and crop diversification (Kent and Johnson, 2001) on weed species composition.Information concerning the distribution of rice weed species along an altitudinal gradient is lacking.If weed species groups could be established according to relatively simple environmental criteria, tailored and more effective weed management strategies could be developed for smallholder farmers.
The objectives of the current study are therefore to (1) assess the current composition of weed species in lowland rice fields in East Africa, and (2) determine whether species composition and abundance in East Africa is influenced by altitude, allowing weed species grouping to facilitate the development of better tailored weed control programmes.

Study sites
A survey was conducted between October 2010 and November 2013 in 31 lowland rice production sites in Rwanda, Tanzania, Kenya and Uganda (Table 1; Figure 1).Along an altitudinal gradient ranging from 4 to 1,596 m above sea level, from sites located between -10.92 and 2.20 latitude and 29.74 and 40.19 longitude, weed vegetation was sampled (Table 1).The schemes were grouped in three altitudinal classes: Low (<200 m a.s.l.), Medium (200 -1,000 m a.s.l.) and High (>1,000 m a.s.l.).All surveyed fields in our study area were categorised as flooded rice as they were subject to season-long flooding, with water management ranging from fully controlled (irrigated) to uncontrolled (rainfed).
In the highlands (>1,000 m a.s.l.) the annual temperature ranged between 10° C and 30° C.
Except for rice production sites located in coastal regions in Kenya (e.g.Garsen-Tana Delta Irrigation Project) and Tanzania (e.g.Pwani-Bagamoyo and Pwani-Chauru/Ruvu), flooded rice is commonly produced on foot slopes and flood plains on hydromorphic soils (Moormann and van Breemen, 1978;Nawaz, 2010).The cultivated rice species in the study sites was Oryza sativa L., commonly known as Asian rice.

Sampling methods and species identification
Rice production sites were visited between 60 and 120 days after planting.In each site, one field was randomly selected.Quadrats of 1 m 2 were thrown randomly in each rice field for weed sampling.The number of quadrats varied between three and ten per rice field, depending on the field size.A total of 201 quadrats were sampled.To enable a fair comparison across altitudes, we only used a random sub-sample of three quadrats from each rice scheme for statistical analyses, hence 93 quadrats in total.All 201 quadrats were however considered to construct the list of weed species presented in the Supplementary Information section (S1).
Identification and recording of weed species were done alongside an estimation of their soil cover inside each quadrat.Coverage is the percentage of the surface area of the sample plot covered by a given species, and not normally limited by the size or distribution of individuals (Floyd and Anderson, 1987)).It is scored based on a cover-abundance-dominance scale estimated as the vertical projection of vegetation on the ground (Daubenmire, 1968).
Plant soil coverage of each species was visually scored using a scale from one to five: 1 = 0-10% coverage, 2 = 10-30%, 3 = 30-60%, 4 = 60-80%, 5 = 80-100% (Savary and Castilla, 2009).In the current study, the species abundance refers to the percentage of coverage of each species calculated from the cover-abundance-dominance scale of species to each altitude class.Species were grouped into one of the three taxonomic categories (i.e.monocotyledons, dicotyledons and ferns), into one of the five life-cycle groups (i.e.annuals, short-lived perennials/bi-annuals, perennials, parasitic and unknown and into three photosynthesis pathways (C3, C4 and CAM); Supplementary Information S1).Identification of species was done with the help of field guides (Agyakwa and Akobundu, 1987;Ivens, 1967;Johnson, 1997), regional flora (Beentje, 2000;Gillett, 2017;Gillett et al., 1971;Haines and Lye, 1983;Troupin, 1982Troupin, -1988;;Whitehouse and Gardens, 2001) and the AFROweeds identification tool (Rodenburg et al., 2016b).All weed species that could not be identified immediately in the field were collected for later determination at the National Herbarium of Rwanda (NHR), University of Dar es Salaam (UDSM) and Eastern African Herbarium (EA) of the National Museum of Kenya.Taxonomy and nomenclature of the plant names followed that of the Plant List Data Base (2010).Registration of altitude (m), longitude and latitude (expressed in the Universal Transverse Mercator coordinate system; UTM) were recorded using a GPS device (eTrex Legend ®; GARMIN International Inc., USA).

Data analysis
The relationship between weed vegetation and altitude were analysed using multivariate statistical ordination and classification techniques previously used for similar purposes (Hanzlik and Gerowitt, 2016).For data analyses and presentation purposes, scientific names of weed species were replaced by their respective five-character EPPO codes (European and Mediterranean Plant Protection Organization; Supplementary Information S1).Prior to data analysis, species scored on the abundance-dominance scale (220 weed species from 93 quadrats) in the main matrix were not transformed for preservation of original information, whereas topographic descriptors in the second matrix were relativized by standard deviate (column).This transformation represents each transformed value as number of standard deviations that it differs from the mean and considered very useful for environmental variables, putting them on the same footing.
First a Nonmetric Multidimensional Scaling (NMS; Kruskal, 1964) was done of the 93 quadrats by general patterns of all weed species using 'Medium' autopilot mode as setting, Kruskal's first approach for non-penalization for ties in the distance matrix and using the Euclidean distance method (McCune et al., 2002).NMS was selected because it does not require normality or linearity assumptions.We then correlated the topographic descriptors (altitude, longitude and latitude) with the NMS axes in the three dimensions that yielded a stable solution to our data set.Finally, the first axis of the NMS reflecting the principal dimension of the variation (showed a moderate positive correlation to altitude, r = 0.55) was examined as a proxy for altitude influencing weed vegetation.We tested whether the scores of the first axis of the NMS differed between altitudinal classes-hence have consistent weed vegetation according to altitude.Prior to the analysis of variance (ANOVA), Levene's test was used to assess whether the homogeneity of variance (homoskedasticity) assumption was met.As this assumption was not satisfied (with P<0.05, indicating variances are not equal and further parametric tests such as ANOVA are not suited), the non-parametric Kruskal-Wallis test was employed.
Second, to identify the indicators species at each altitudinal class, an Indicator Species Analysis (ISA; Dufrêne and Legendre, 1997)  Third, to test the null hypothesis of no difference in community species composition between elevational classes, we performed the analysis of similarities (ANOSIM) through a one-way ANOSIM test (9999 permutations) of weed vegetation based on Euclidean distance where an R-value close to 1 indicates strongly dissimilar vegetation groups, and an R-value close to zero indicates that vegetation groups are barely distinguishable (Clarke, 1993).Next, a Similarity Percentages (SIMPER; Clarke, 1993) analysis concentrated on Euclidean distance (9999 permutations) was done to identify those species that contributed most to the observed altitude class distinction in community species composition.A threshold of 50% for cumulative percentage was chosen to identify abundant species with the highest dissimilarity contribution as good discriminators for comparison (Clarke, 1993) between altitudes.
Finally, after identifying abundant species that contributed most consistently to differences between altitudes, calculation of percentage abundance of each of those 24 species follows as mean coverage.A cut-off (superior to 4%) was chosen as threshold for dominant (or most abundant) species to each altitude class.In addition, five weed species with total high abundance values per country and rice production site between altitude classes were calculated as: (2) Diversity of weed species was evaluated using Shannon Weiner Index (H') (Shannon and Weaver, 1949): (3 where Pi is the proportion of all observations in the i th species, Ln = logbasen and S is the total number of species within each altitude class or species richness.The minimum value of H' is zero -a value representing a community with a single species-and this increases as species richness and evenness increases. In order to determine if species abundance correlated with altitude, a simple Mantel test (Mantel, 1967) using asymptotic approximation method was employed.Euclidean distance matrix was calculated both for species abundance data, and for altitude.
In order to investigate whether the occurrence of weed species could be explained by altitudinal gradient or photosynthesis pathway, a logistic regression model was fitted to our data, with occurrence (absent or present) as response variable and photosynthesis pathway and altitude variables as predictors.This relationship was examined at P<0.05 significance level.

Results
In the 93 quadrats used for the statistical analyses, 220 different weed species were observed.
The NMS ordination shows a three-dimensional (3 axes) representation as best solution for our dataset (93 quadrats by 220 species) with a final stress value of 17.74.A final stress value between 10 and 20 is generally considered to be good and reliable when the NMS technique is applied to ecological community datasets (McCune et al., 2002).Proportions of variance represented by the three dimensions were 32.5%, 30.1% and 20.5% respectively (cumulative r = 83.2%).Only the graph presenting principal dimensions (NMS 1 and NMS 2 axes) with a high proportion variation of the altitude variable as joint plot is shown (Figure 2).The Monte Carlo test suggested a significant separation of weed vegetation across altitude (P=0.0196).
The Kruskal-Wallis test, performed on the first NMS axis reflecting the principal ordination, showed a significant pattern of weed vegetation between altitude classes (H = 26.34,Hc = 26.34,nHigh = 33, nMedium = 30, nLow = 30, P<0.0001) suggesting an important role of altitude in the variation of weed vegetation.
In the complete set of quadrats ( 201), a total of 286 weedy species were recorded, belonging to 156 genera and 59 families, with 127 monocotyledonous, 151 dicotyledonous and eight fern species (Supplementary Information S1).The Cyperaceae, Poaceae and Asteraceae were the most important families in terms of weed species numbers across altitudes.The Cyperaceae was the prevalent plant family at high and medium altitudes ( 23and 26 species, respectively) whereas at low altitude the Poaceae family was most common (19 species).The Fabaceae, Commelinaceae and Onagraceae were also well represented across elevations (Table 2).The Shannon-Weiner index diversity (H') of weed species suggested that the medium altitudes had a higher diversity than the high and low altitudes.
The high altitude, in turn, recorded greater diversity than the low altitude (Table 2).
The analysis of similarity (ANOSIM) in community species composition performed between altitude classes revealed significant differences (Global RANOSIM=0.Vahl and Melochia melissifolia Benth.(Table 3).Annual species were dominating this group (Supplementary Information S2).

Weed species diversity in rice in East Africa
This study reveals a rich species diversity of the weed flora in seasonally flooded rice fields in East Africa.Overall, the proportion of dicotyledonous contributed more to the total reported flora than monocotyledonous (53% vs 45%).Species from two monocotyledons families, the Cyperaceae and the Poaceae, were the most frequently encountered.This confirms earlier reports on weeds of rice in sub-Saharan Africa, reviewed by Rodenburg and Johnson (2009).They calculated that 43% of the observed weed species were Poaceae, and 37% were Cyperaceae.Our findings are also in agreement with reports from rice fields in Brazil (Linke et al., 2014;Mesquita et al., 2013) and rice agroecosystems in Vietnam and the Philippines (Fried et al., 2017).Species of the Poaceae (grasses) and Cyperaceae (sedges) families are more important than others in rice because they are most adapted to the wet growing conditions and because grasses and sedges are most difficult to control, due to their resemblance (both physiologically and morphologically) to rice, in particular at the early phenological stages.
Weed species diversity at medium altitudes is relatively high in comparison to high and low altitudes.High diversity at medium altitudes, could be a result from a relatively high variability in landscape, climate and soil at this elevation, which in turn cause differences in water and nutrient availability.Gabriel et al. (2005) demonstrated that environmental heterogeneity promotes the diversity of plant species.Of the 286 observed species, 42 did not feature in any previous weed inventory from this region, including some recent reports (e.g.(Makokha et al., 2017).The majority of these species even represent new rice weed observations for the African continent.

Weed abundance along the altitude gradient in flooded rice in East Africa
The current study showed that altitude was correlated with species' abundance.Along the altitude gradient the abundance of certain species (e.g.E. colona, L. adcendens and S.
africana) increased with increasing elevation, whereas the abundance of other species (e.g. C. difformis, F. littoralis and A. uliginosa) showed an opposite trend.In this study, E. colona was the most abundant species in high and low altitudes.It has been considered as the most serious grass weed of rice with a wide distribution in the subtropics and tropics (Holm et al., 1977) and a wide altitude range, as it was recorded from 0 to above 2,000 m a.s.l.(Lazarides, 1980).It is also the most cited weed species of lowland rice in sub-Saharan Africa (Rodenburg and Johnson, 2009).The two broad-leaved weed species S. suaveolens and C.
hepperi were observed as most abundant at high altitudes.These species have previously been described as widespread in East Africa over a range of altitudes (Ivens, 1967).The grass L. squarrosa reported for the first time by Makokha et al. (2017)was determined as a major weed of rice at high altitude ranges (between 1093 and 1391 m a.s.l.) in the current study, while knowledge about the species' ecology and management is largely lacking.The broadleaved weed C. granvikii and the sedge weed P. lanceolata were among the most abundant at medium altitudes (between 721 -1,596 m a.s.l) in this study.Crassula granvikii grows along permanent streams up to 3,200 m a.s.l.from Central, East and Southern Africa (Exell and Fernandes, 1960;Troupin, 1983) while P. lanceolata was previously reported to thrive between 750 and 1,160 m a.s.l. in East Africa (Haines and Lye, 1983).In West Africa, P.
lanceolata was reported as a common weed of rice on hydromorphic soils (Agyakwa and Akobundu, 1987;Johnson, 1997).In this study, we also observed the sedges C. difformis (4 -1,596 m a.s.l),K. polyphylla (0-40 m a.s.l.) and C. esculentus (0-798 m a.s.l.) among the most abundant species at low altitudes.The species K. polyphylla and C. esculentus were reported to occur along streams in the forest zone of East Africa up to 1,200 m a.s.l. and 2,000 m a.s.l., respectively (Haines and Lye, 1983) while C. difformis has been widely reported as a common weed of rice over a range of altitudes (e.g.Ivens, 1967).Other grasses such as L. hexandra and S. africana, considered as important perennial grasses in lowland rice production systems in Africa (Rodenburg and Johnson, 2009), were reported in the current study between 5 and 1,596 m a.s.l and between 721 and 1,596 m a.s.l, respectively.In Asia, L. hexandra has been reported as moderately to highly competitive in lowland rice (up to 2,200 m a.s.l.; Caton, 2010).

Altitude effects on weed species composition in flooded rice in East Africa
The present study identified the weed species of flooded rice fields in East Africa and revealed that species composition of the weed vegetation in these environments is influenced by altitude.Despite a clear distinction in terms of weed species between altitude classes, the proportion of variance between the principal dimension (first NMS) was not very high in comparison to other two dimensions, indicating other ecological or management descriptors associated with altitude are important to explain the variation of weed species.These other descriptors, such as water depth and flooding period, weed management, crop rotation, climate, soil parameters and even topographic descriptors (latitude and longitude) were not considered in this context.The results presented in this study encompass however a very useful first step allowing the development of more tailored weed management strategies along elevation gradients.Follow-up research could refine the resolution obtained in this study.For instance, within each altitude class a further weed group differentiation could be made based on weed management practices, soil properties and climate as shown before (Pinke et al., 2012).Lososová et al. (2004) and Fried et al. (2008) have shown that significant changes in weed species composition are associated with a complex gradient of altitude, precipitation, clay content, texture and pH.
This study showed that some weed species were clearly restricted to a certain altitude.
Floristic composition and distribution of weeds often serve as indicators of field conditions and environments (Moody, 1989) regime may exert an important effect on the structure of weed communities even within an altitude gradient.Several studies suggested the C3 pathway to provide a higher degree of physiological flood-tolerance than their C4 counterparts (Carmo-Silva et al., 2008;Ghannoum, 2009;Kalapos et al., 1996).In a situation of seasonally flooded rice fields, as in our study, species with the C3 pathway might have a competitive advantage over those with the C4 pathway.For example, a study by Kent and Johnson (2001) indicated that continuous flooding conditions to 2 -4 cm in 2 -4 days every week increased density of the annual C3 S. zeylanica and decreased that of annual C4 E. colona and Echinochloa crus-pavonis (Kunth) Schultes.According to Caudle and Maricle (2012) and Maricle and Lee (2007), tolerance of plants under flooded conditions is explained from the ability to maintain photosynthetic activity, high stomatal conductance and oxygen transport to submerged tissues in order to avoid shortage of oxygen to roots.The current study also revealed that species with the C4 pathway have a relative higher representation at low altitudes, perhaps because of associated higher temperatures, than high altitudes.(Ghannoum, 2009).In rice systems, C4 weeds have been previously suggested to favour rainfed upland conditions over lowland conditions (Rodenburg et al., 2011).
The current study represents the first large-scale survey of species of weed vegetation of flooded rice in East Africa.The 24 most abundant species are known to exert a high level of competition to flooded rice across altitudes and should be prioritised in weed management strategies.The identified altitude-specific weed grouping will also enable the development of more tailored weed management strategies.The data was reorganized to come up with three major variables with their corresponding categories; PP (CAM-0, C3-1 and C4-2), Altitude (Low-0, Medium-1 and High-2) and Occurrence (Absent-0 and Present-1).The occurrence variable was an indicator variable of the PP to show if the categories CAM, C3 and C4 were present or absent in the corresponding altitudes.Thus, the variable occurrence, becomes the response (dependent) variable to test for the independence of the presence of the PP categories across the three-altitude classes.

Percentage of Species
was performed.A Monte Carlo test with 1,000 randomization runs evaluated statistical significance of Indicator Values (IV) at a P<0.05 cutoff threshold.Indicator values (IV) are measures of faithfulness (closeness) of occurrence of a species in a particular group and ranges from zero (no indication) to 100 (perfect indication).The objective was to determine which species were characteristic of each altitude class.Names of three weed species with the highest IV were used to name vegetation type of each altitude class.

Figure 2 .
Figure 2. Nonmetric Multidimensional Scaling (NMS) ordination of abundance data from 220 species by 93 quadrats studied in East Africa showing the altitudinal gradient along the axis 1.Only graph for dimensionality (NMS1 and NMS2) with high proportion variation of altitude variable as joint plot is shown.Graphs for dimensionalities NMS1 & NMS3 and NMS2 & NMS3 are not shown.

Figure 3 .
Figure 3. Frequency distribution of C3, C4 and CAM photosynthetic pathway (PP) of species along altitudinal gradients based on occurrence of 220 weed species, with known photosynthetic pathways, in seasonally flooded rice systems in East Africa.The data was reorganized to come up with three major variables with their corresponding categories; PP (CAM-0, C3-1 and C4-2), Altitude (Low-0, Medium-1 and High-2) and Occurrence (Absent-0 and Present-1).The occurrence variable was an indicator variable of the PP to show if the categories CAM, C3 and C4 were present or absent in the corresponding altitudes.Thus, the variable occurrence, becomes the response (dependent) variable to test for the independence of the presence of the PP categories across the three-altitude classes.

Table 1 .
Description of study sites in East Africa: name of country, district/region and site, altitude range (ALT), altitude classes (AC: HA=High; MA=Medium; LA=low), latitude (LAT), longitude (LON).

Table 2 .
Taxonomic classes (Class), proportion of important families (only the ones with more than one species are shown) and weed species diversity indices (H'= diversity index of Shannon-Weiner per altitude (HA: High Altitude, MA: Medium Altitude and LA: Low Altitude) of lowland rice fields in East Africa.

Table 3 .
Indicator species of weed species for each altitude range (HA: High altitude, MA: Medium altitude and LA: Low altitude).Indicator value (IV = 0 no indication to IV = 100 perfect indication), level of significance of the IV value is indicates with probability (p-value).Names of the three species with high indicator values in each group were used for the group name.Only indicator species with P<0.05 are shown in the table with their standard deviation.