Contributions of lateral flow and groundwater to the spatio

advertisement
Contributions of lateral flow and groundwater to the spatio-temporal variation of
irrigated rice yields and water productivity in a West-African inland valley
Petra Schmitter1,2,3, Sander J. Zwart1, Alexandre Danvi,,5, Félix Gbaguidi5
1
Africa Rice Center, Cotonou, Benin
Department of Civil & Environmental Engineering, National University of Singapore, Singapore
3
Present address: International Water Management Institute, East Africa & Nile Basin Office, Addis, Ethiopia
4
Department of Geography, University of Bonn, Bonn, Germany
5
Inland Valley Unit, Directorate of Rural Engineering, Ministry of Agriculture, Livestock and Fisheries, Porto Novo,
Benin
2
Abstract
Water management techniques to elevate rice yields and productive use of water resources in Africa
frequently lack a substantial spatial assessment as they are often based on plot level measurements without
taking into account toposequential effects present in the landscape. These effects have been shown to
significantly affect spatio-temporal variations in water availability and rice productivity in Asia. Therefore,
this study addresses the spatio-temporal variations of the various water components within irrigated
toposequences in an African inland valley and assesses its effect on water productivity and respective rice
yields for two irrigation practices: i) continuous flooding (CF), a well-known water management practice
in rice cultivation used worldwide and ii) a reduced irrigation scheme (RI) where irrigation is applied every
5 days resulting in a 1-2 cm water layer after irrigation.
The lateral flow observed in the inland valley had a strong two dimensional character, contributing to water
gains between fields, located at the same toposequential level as well as along toposequences. The
toposequential effect on sub-surface hydrological processes masked the overall effect of water management
treatment on rice production. Additionally, the associated water productivity (WP) was not found to differ
significantly between the treatments when standard calculations (i.e. net irrigation and evapotranspiration)
were used but a clear toposequential effect was found for the fertilized lower lying fields when the net
irrigation was corrected by the lateral flow component. Results of the established mixed regression model
indicated that based on the groundwater table, rainfall and standard soil physico-chemical characteristics
rice yields can be predicted in these African inland valleys under continuous flooding and reduced irrigation
practices. Validation of the established regression function of inland valleys, representing various
groundwater tables in the region, could lead to improved regression functions suitable to estimate spatial
1
variation in rice production and water consumption across scales as affected by water management,
fertilizer application and groundwater tables.
Keywords: Groundwater, Irrigation, Rice, Water management, Water productivity, West-Africa
2
1
1
Introduction
2
African countries depend on 40% of international food markets and rice imports, mostly from Southeast
3
Asia, to fulfil their demands (Tsujimoto et al., 2013). In 2011, 11 million tons of rice was produced in West
4
Africa and an additional 6.3 million was imported to meet its growing rice demand (FAO Stat, 2011). This
5
dependency on food markets makes West-African countries vulnerable to price shocks such as those
6
witnessed in 2008 (Seck et al., 2013), which can lead to political instability and social unrest. Most West-
7
African countries have developed National Rice Sector Development Strategies outlining the pathways for
8
self-sufficiency by the year 2018. The development of new irrigated rice systems in lowland environments
9
and intensification of lowland rice productivity through rehabilitation of irrigation infrastructure are often
10
prioritized by national governments to increase national rice production. Since the food crisis in 2008, a
11
strong growth in rice production has been witnessed in Sub-Saharan African, which has been attributed to
12
a 70% increase in productivity and 30% area expansion (Saito et al., 2013). However, achievements to
13
increase national production and become self-sufficient are still outpaced by a strong growth in consumer
14
demands.
15
16
Over the past decade, rice consumption has increased exponentially and become one of the main food items
17
in household diets (Becker and Johnson, 2001). With exception of Egypt and Senegal, the productivity in
18
African rice systems is still low compared to Asia. Average paddy rice yields range from a low 0.85 ton ha-
19
1
20
2.6 ton ha-1. Reasons for this low productivity can be attributed to bio-physical and socio-economic
21
constraints found at many levels (Djagba et al., 2014) and include sub-optimal functioning markets for
22
acquiring fertilizers and commercialization of rice products, lack of financial services to make necessary
23
investments for intensification, poor management and maintenance of irrigation infrastructures, insufficient
24
national policies, amongst others (Saito et al., 2014).
(Gambia), a moderate 3.3 ton ha-1 (Benin) to 5.3 ton ha-1 (Mauretania), resulting in a regional average of
25
26
In West Africa the inland valleys are a common landscape and believed to have great potential for rice
27
development (Rodenburg et al., 2014). This counts especially for those located in the agro-ecological
28
Guinea savannah zone due to the favourable rainfall patterns and the shallow groundwater to flooding water
29
tables in the rainy season (Andriesse and Fresco, 1991). However, high inter-annual variability of
30
precipitation as well as the onset of the rainy season poses large uncertainties for rainfed agriculture (Gruber
31
et al., 2009). As such, these inland valleys, which are characterized by coarse to medium textured soils with
32
low water retention potential (Andriesse and Fresco, 1991), remain, to date, largely underutilized and if
33
cultivated, restricted to one crop in the rainy season due to the absence of adequate irrigation systems
3
34
(Djagba et al., 2014; Saïdou and Kossou, 2009). Irrigated rice production in these valleys remains low and
35
only reaches 2% of its total potential according to Gruber et al. (2009). Therefore, as many of the inland
36
valley rice systems are rainfed, field based water management strategies such as bunds, levelling or drainage
37
systems are frequently absent (Worou et al., 2013). In those regions where irrigation systems are present,
38
poor infrastructure and maintenance lead to sub-optimal water availability and hence low rice production
39
(Djagba et al., 2014; Totin et al., 2012).
40
41
To improve rice yields and water productivity, research experiments in the region have been carried out to
42
study the effect of bunding and levelling, which in Asia are inherent practices in the so called Sawah system,
43
on improving water productivity and consecutively rice production (Becker and Johnson, 2001; Rodenburg
44
et al., 2014; Worou et al., 2012, 2013). Subsequently, water saving techniques such as alternate wetting
45
drying (AWD) and the system of rice intensification (SRI) have been tested to elevate rice yields in the
46
region (de Vries et al., 2010; Krupnik et al., 2012). These studies are based on plot level measurements in
47
controlled environments that try to eliminate the interaction between the plots. However, due to field
48
conditions in the inland valleys, toposequential effects along the slopes are present in the landscape and
49
will influence water distribution among fields and subsequently rice productivity.
50
51
Lateral flow through bunds as well as sub-surface drainage in coarse to medium textured soil is substantial
52
in Asian rice paddy systems (Janssen and Lennartz, 2009; Tsubo et al., 2005). Due to the spatial variability
53
of the soils physical and chemical properties within toposequences, large variations can be expected of
54
water balance components as well as rice performance (Tsubo et al., 2007). Tsubo et al. (2005) observed
55
water gains up to 9.8 mm day-1 due to lateral flow in lower lying rainfed rice fields along toposequences in
56
Thailand. In a similar study in Laos, Tsubo et al. (2006) positively linked the water gain in these lower
57
lying positions with increasing grain yields. As irrigation water alters soil physico-chemical characteristics
58
due to sediment deposits along cascades with higher sandy – low organic matter content in the upper lying
59
paddies (Schmitter et al., 2010), negatively affecting rice productivity (Schmitter et al., 2011), a similar
60
variability of lateral flow components can be expected within irrigated toposequences.
61
62
Lateral flow losses respectively gains, are often not accounted for when comparing water productivity of
63
new irrigation practices as water productivity is either calculated based on evapotranspiration or net
64
irrigation (i.e. rainfall and irrigation) (Zwart, 2013). Moreover, shallow groundwater tables can significantly
65
contribute to water productivity of rice systems. Shen et al. (2014) used stable isotope techniques and
66
showed that 80% of the water uptake under a continuous flooding scheme in Jiangxi Province, China was
67
attributed to soil water uptake within the first 10 cm of the soil profile due to the shallow water table.
4
68
69
In the African inland valleys however, little is known about spatial variation of water losses, its significance
70
when assessing water saving techniques and its impact on rice productivity. A greater understanding of the
71
variation along a toposequence is believed to improve irrigation system design and management, which
72
may lead to higher performance of rice-based production systems. The purpose of this study is twofold.
73
Firstly, we aimed to assess the spatio-temporal variations of the hydrological balance within irrigated
74
toposequences in an inland valley located in Benin and assessed its impact on water productivity and rice
75
yields. Two irrigation practices along toposequences were evaluated: i) continuous flooding (CF), a well-
76
known water management practice in rice cultivation used worldwide, and ii) a reduced irrigation scheme
77
(RI) where irrigation is applied every 5 days resulting in a 1-2 cm water layer after irrigation. Additionally,
78
fertilizer experiments were performed to assess the effect of water management on nutrient availability and
79
therefore rice production. Each water management treatment had two fertilizer treatments: i) zero fertilizer
80
and ii) fertilizer application according to the local recommendations. The second aim of this study was to
81
predict the spatial variation of rice productivity in the inland valley toposequences. A mixed regression
82
model was developed to predict the spatio-temporal variation of rice yields along toposequences as affected
83
by the water availability, soil nutrient status and groundwater table. Such a model could help irrigation
84
managers and farmers to recommend and apply site-specific management practices to improve rice yield
85
and use water resources efficiently.
86
2
87
2.1
88
The experiment was conducted during four seasons (2011-2012) in the inland valley of Bamé, Zagnanado
89
commune, 100 km north of Cotonou (7°12’N, 2°24’E, elevation of 36 m a.s.l.). The inland valley is
90
delineated by two perennial streams both draining to the Ouémé river downstream of the experimental site.
91
The upland area, draining towards the inland valley, is mainly under shifting cultivation with the main crops
92
being maize (Zea mays L.), cassava (Manihot esculenta Crantz), groundnut (Arachis hypogaea), cowpea
93
(Vigna spp.) and sweet potato (Ipomoea batatas) and has an average slope of 1 to 2%. The inland valleys
94
are used mainly for rice (Oryza sativa L.) production, sugar cane (Saccharum officinarum), taro (Colocasia
95
esculenta) and vegetables. The soils are hydromorphic alluvial sandy-loam soils with a shallow
96
groundwater to standing water tables all year round (Table 1).
Methodology
Experimental site
97
5
98
Precipitation follows a bimodal pattern (Figure 1). The longer rainy season runs from March to July and
99
shows higher variability and more erratic rainfall events. The shorter season from September to October is
100
characterized by lower intensities and less variability.
101
2.2
102
The experiment was conducted for four consecutive seasons over two years in 2011 and 2012 (Table 1).
103
The first and third season were conducted from February/March till May/June, covering the drier season
104
with lower groundwater tables and few erratic rainfalls. The second and fourth season were conducted
105
during the wetter season extending from July/August till November/December. This period is characterized
106
by a higher groundwater table and higher occurrence of less intensive rainfall. An automatic weather station
107
(DeltaT, WS-GP1) with a tipping bucket rain gauge (0.2 mm accuracy) was installed measuring minimum
108
and maximum air temperature (°C), relative humidity (%), solar radiation (kW.m-2) and wind speed (m s-1)
109
(Table 1). In total 1096 mm of rainfall was recorded in 2011 while 1132 mm was measured in 2012.
Experimental design
110
111
The experimental layout followed a split-split plot toposequential design resulting in 8 toposequences,
112
consisting out of 6 successive fields each with an area of 100 m2 covering a total of 48 fields (0.48 ha)
113
(Figure 2). Difference in elevation between successive fields within a toposequence was determined by
114
measuring the difference in bund height. The upper fields of the toposequence received water from the main
115
irrigation channel. Irrigation and drainage channels were constructed parallel to the toposequences to divert
116
water to and from each field separately (Figure 2). All fields were bunded but not shielded to allow water
117
losses due to sub-surface lateral flow. As rice cultivation is a relatively young agricultural practice in the
118
area no hard pan is present at 30 cm below the soil surface. Within each repetition, water treatment was the
119
main factor and fertilizer application the sub-factor. Firstly, the water treatment was randomized followed
120
by the fertilizer treatment within the water management treatment. In order to avoid fertilizer contamination
121
each field was irrigated separately and the same fertilizer treatment was applied for all 6 fields in one
122
toposequence (Figure 2). The two water management treatments were: i) continuous flooding (CF)
123
(ponding water table) and ii) reduced irrigation (RI) (no ponding water table). The ponding water layer in
124
the fields under CF were measured on a daily basis and received, when necessary, irrigation to maintain a
125
ponding water layer around 5 cm during each cropping season till maximum tillering and 10 cm till two
126
weeks after flowering. Fields under the reduced irrigation treatment (RI) were irrigated till the ponding
127
table reached 1-2 cm every five days. When necessary, fields were drained after rainfall and after irrigation
128
to remove excess water from the fields in order to maintain the necessary water level or soil saturation,
129
respectively.
130
6
131
The two fertilizer sub-treatments were: (i) no fertilizer application (-F) and ii) fertilizer (+F) according to
132
Africa Rice recommendations for the chosen variety, via split application which resulted in a total
133
application of 105 kg N ha-1, 84 kg P ha-1 and 42 kg K ha-1. The first dressing applied at transplanting
134
entailed 40% N, 100% P and 100% K of the total amount in the form of NPK. No irrigation was performed
135
until four days after fertilizer application. The second and third dressings, each contained 30% N of the
136
total amount of fertilizer in the form of Urea and were applied after weeding at maximum tillering and at
137
flowering. The fertilizer application (+F) treatment assisted in assessing rice and water productivity in
138
absence of nutrient availability for the various water management strategies while the unfertilized fields (-
139
F) revealed the rice and water productivity due to inherent soil fertility. The latter one is of importance in
140
Africa as fertilizers are not always available on the market; farmers cannot afford them or choose not to
141
apply fertilizers.
142
2.3
143
For all four seasons, the same lowland Nerica variety Nerica-L-19 was used. In order to assess crop
144
performance in each field, nine sub-plots of 1 m2 were marked within each 100 m2 plot. The distance of
145
each sub-plot was measured according to the x and y direction of the experimental layout where the upper
146
left corner of the RI–F treatment (Repetition 2) was taken as origin (Figure 2). Land preparation entailed
147
puddling and levelling in all treatments. For each season, transplanting was done fifteen days after nursery,
148
with 25 cm spacing for all treatments. All fields within one repetition were planted within one day, resulting
149
in two successive days for the entire experiment. Manual weeding was carried out twice within each
150
cropping season.
Crop performance as influenced by water and fertilizer management
151
152
Plant height and the relative leaf chlorophyll content, using a Soil Plant Analysis Development (SPAD)
153
chlorophyll meter, were measured at sub-plot level during three cropping stages: i) two weeks after tillering,
154
ii) maximum tillering and iii) flowering. Data was used to calculate changes in crop evapotranspiration due
155
to the various treatments. At harvest, grains and total above ground biomass were determined in each sub-
156
plot and grain moisture content upon harvest was measured. Additionally, all fields were harvested and
157
grains were weighed. Grain yields at sub-plot and plot level were converted to 14% moisture content and
158
the harvest index at sub-plot level was calculated.
159
2.4
160
Soil samples were taken of various sub-plots to investigate inherent soil spatial variability prior to the field
161
experiment. Topsoil samples (0-5 cm) within the sub-plots were taken after field preparation to ensure a
162
homogenous soil profile after ploughing and puddling. The samples taken in the middle position of the field
163
were analysed for particle size (pipet method) and soil organic carbon (SOC) (Walkley-Black) using
Soil spatial variability within the experiment
7
164
standardized laboratory analysis (Houba et al., 1995). Furthermore, five samples per field (Figure 2) were
165
selected and measured on pH (H2O), total nitrogen (Ntot) (Kjeldahl), exchangeable iron (Fe2+), potassium
166
(K+), magnesium (Mg2+) (Mehlich) and phosphorus (PO43-) (Bray) (International Institute or Tropical
167
Agriculture, 1979). Exchangeable iron was measured to test whether iron toxicity was prevailing and
168
compared to the content taken up by the plants after the third season.
169
2.5
170
Seasonal water balances for each water management and fertilizer treatment were calculated at field scale
171
within each toposequence. Rainfall (R) and groundwater table (G) were measured at daily time steps
172
whereas percolation (P) was determined the day after irrigation, irrigation (I) and drainage (D) were
173
measured when applicable. The adjusted crop evapotranspiration (ETc,adj ) was calculated on a daily time
174
step for each field within each season according to the FAO Penman-Monteith equation (Allen et al., 1998)
175
while seasonal lateral flow (L) was calculated using the water balance formula.
Spatio-temporal variation in water availability and water productivity
176
177
Plastic PVC pipes with a diameter of 50 mm were installed in each field for irrigation and drainage purposes
178
and closed when no irrigation was needed. Irrigation and drainage duration was recorded using a stopwatch
179
and the corresponding water depth in the pipe was measured at regular intervals. Irrigation continued till
180
the required ponding water depths in the field for the various treatments were obtained (see section 2.2).
181
Irrigation and drainage volumes were computed using the Manning formula for more or less than half full
182
pipes (i.e. unpressurized flow) depending on the measured water depth in the pipe. Manning value of 0.009
183
for PVC pipes was used and adjusted depending on the water depth/pipe diameter ratio. Both, calculated
184
irrigation and drainage volume where converted to corresponding water depth (mm).
185
186
Percolation was measured using cylindrical tubes of 200 mm diameter and 400mm length allowing solely
187
for a downward water movement in every second field along the toposequence (Tsubo et al., 2005). Tubes
188
were covered with a lid to prevent evapotranspiration losses and rainfall input. Water level in the tube was
189
brought to the water depth observed in the field after irrigation or drainage and recorded the day after.
190
191
Tubes for daily groundwater measurements were installed upstream of the first field in each toposequence
192
and within the last field. PVC pipes were implemented with a total length of 1000 mm and 50 mm diameter,
193
perforated 500 mm from the top of the pipe and installed 800 mm in the soil profile. For the fields at
194
toposequence level, two to five (i.e. P2 to P5, Figure 2), daily groundwater values were linearly interpolated
195
between the top and the bottom field, for each field within a toposequence.
196
8
197
At field level, daily crop requirement under soil water stress conditions were calculated using the FAO-
198
dual crop coefficient approach (Allen et al., 1998) and if necessary corrected for soil water limiting
199
conditions.
200
๐พ๐‘๐‘ ,the basal crop coefficient for rice as function a of the initial, mid and end development stages
201
(i.e.๐พ๐‘๐‘๐‘–๐‘›๐‘– , ๐พ๐‘๐‘๐‘š๐‘–๐‘‘ and ๐พ๐‘๐‘๐‘’๐‘›๐‘‘ , respectively)
202
development stage, measured plant height at the three stages and occurring climatic conditions. The soil
203
evaporation coefficient (๐พ๐‘’ ) was estimated based on the daily water balance calculations under local soil
204
physical properties. The water stress coefficient (๐พ๐‘  ) was calculated at daily time steps based on the
205
estimated readily available water content as a function of measured soil texture and calculated root zone
206
depletion. However, due to the high groundwater table, the value of ๐พ๐‘  remained 1 throughout the
207
experiment for both water management treatments.
was adjusted depending on the length for each plant
208
209
The total seasonal net lateral water flow (L) for all four seasons was calculated at field level and represented
210
percolation under the bunds as well as seepage through the bunds. Negative values showed a clear loss
211
while positive values indicate a net water gain within the field due to lateral water movement. As
212
groundwater was observed within 50 cm of the sub-surface, the seasonal change in groundwater table was
213
used to calculate the lateral flow. The assumption is made that within one season there is a steady
214
groundwater movement within the sub-surface layer. As such the depletion or recharge of groundwater
215
within one season must be equal to all incoming and outgoing water components:
216
217
218
๐‘ณ = ๐‘น + ๐‘ฐ − ๐‘ซ − ๐‘ท − ๐‘ฌ๐‘ป๐’„,๐’‚๐’…๐’‹ − ๐œŸ๐‘ฎ
Eq. 1
219
where ๐ฟ is the lateral flow (mm), ๐›ฅ๐บ seasonal change in groundwater table within the field (mm), ๐‘… total
220
rainfall (mm), ๐ผ total irrigation amount (mm), ๐ท total drainage amount (mm), ๐‘ƒ total estimated percolation
221
(mm) and ๐ธ๐‘‡๐‘,๐‘Ž๐‘‘๐‘— the calculated seasonal evapotranspiration (mm).
222
223
To calculate the net seasonal lateral flow during each rice season, total percolation needed to be estimated
224
at field level throughout the period. Using PROC MIXED (SAS v9.2), a mixed effect model was built to
225
investigate the effect of fertilizer, water management and toposequence position on percolation losses at
226
field level. No significant changes between fertilizer, water management or toposequence position were
227
found and overall average percolation was estimated to be 3 mm day-1. In case the monitored groundwater
228
level was found to be above soil surface, percolation losses were assumed to be zero. Soil water fluctuation,
229
in absence of a standing water layer, was based on sub-surface groundwater fluctuation (G < 0) and
230
estimated using soil porosity as a function of bulk density. Sand was the main soil fraction and soil bulk
9
231
density was estimated to be 1.6 g cm-3 (Pedosphere, 2013) resulting in a porosity of 0.4. When groundwater
232
level was above soil surface no adjustments were made. As such, ๐›ฅ๐บ represented the overall seasonal
233
depletion (<0) or recharge (>0) within the field.
234
235
Water productivity at field scale was calculated for each season using the following formulas:
236
237
๐‘Š๐‘ƒ๐ผ๐‘›๐‘’๐‘ก = ๐‘Œ⁄(๐‘… + ๐ผ − ๐ท)
Eq. 2
238
๐‘Š๐‘ƒ๐ผ๐‘›๐‘’๐‘ก +๐ฟ = ๐‘Œ⁄(๐‘… + ๐ผ − ๐ท + ๐ฟ)
Eq. 3
239
๐‘Š๐‘ƒ๐ธ๐‘‡ = ๐‘Œ⁄(๐ธ๐‘‡
๐‘,๐‘Ž๐‘‘๐‘— )
Eq. 4
240
241
Where ๐‘Š๐‘ƒ๐ผ๐‘›๐‘’๐‘ก , ๐‘Š๐‘ƒ๐ผ๐‘›๐‘’๐‘ก +๐ฟ and ๐‘Š๐‘ƒ๐ธ๐‘‡ are field based seasonal water productivity based on (i) net irrigation,
242
(ii) net irrigation including lateral sub-surface gains or losses and (iii) crop evapotranspiration, respectively.
243
๐‘Œ represented the harvested rice at field scale (kg m-2), ๐ผ total irrigation amount (10 -3 m), ๐ท total drainage
244
amount (10 -3 m), ๐ฟ the lateral flow (10 -3 m) and ๐ธ๐‘‡๐‘,๐‘Ž๐‘‘๐‘— the calculated seasonal evapotranspiration (10 -3
245
m).
246
2.6
247
Descriptive statistics were obtained using the univariate procedure in SAS v9.2. A first indication regarding
248
soil, rice, water productivity and water balance parameter variability were obtained through the coefficient
249
of variation (CV) with values < 10% and > 90% indicating low and high variability, respectively (Wei et
250
al., 2008).
Statistical analysis on spatio-temporal variability of soil properties, water balance and rice
251
252
To evaluate further whether fertilizer, water management and toposequence had significant impact on rice
253
and water productivity, firstly the spatial variability of the initial soil parameters prior to the field
254
experiments was investigated. Two mixed effects models: (i) soil texture and soil organic carbon, and (ii)
255
soil nutrients (Ntot, PO43- Bray, K+, Fe2+, Mg2+) and pH, were built due to differences in available samples and
256
the associated covariance structure. For both models the fixed effects that were tested were: repetition
257
(repetition), water management (WM: CF or RI), fertilizer (F: -F and +F), toposequence position (position:
258
1-6), their two way interaction as well as the overall interaction between water management (WM), fertilizer
259
and position. The random effects included the interaction between repetition and water management
260
(repetition*WM) as well as the interaction between repetitions, water management and fertilizer
261
(repetition*WM*F). For the nutrient models, sub-plot (i.e. location: 1 to 9) was added to the random
262
statement, a repeated statement was added in which the field was taken as subject and an exponential spatial
10
263
covariance structure was added using the distance in x- and y-direction (m), according to the overall defined
264
origin of the experiment, for each sample location within the field (Figure 2). The dependent variables were,
265
where necessary, log transformed in order to obtain homogeneity of variance and normality of the residuals.
266
267
Similarly, the spatio-temporal variation for plant height, SPAD, rice yield and biomass were analysed using
268
the same fixed effects as described for the soil spatial analysis. Seasonal variation was accounted for by
269
grouping the monitored seasons into two groups: i) dry season: lower groundwater table and higher rainfall
270
variability, more erratic rainfall (i.e. season one and three) and ii) wet season: higher groundwater table and
271
less rainfall variability (i.e. season two and four). Grouping of the monitored seasons was valid as no
272
significant difference was found between the same measurements taken in different cropping seasons under
273
the dry (i.e. first and third season) respectively wet season (i.e. second and fourth season). Within this
274
model, the same random effects (i.e. repetition*WM, repetition*WM*F and sub-plot) were added as for the
275
nutrient models described above. As seasonality was shown to have significant impact on rice yields,
276
additional models for the dry and wet season were tested separately.
277
278
2.7
Predicting rice productivity in inland valleys
279
The prediction of rice yield at field scale as a function of soil fertility and water availability was obtained
280
through a multiple mixed regression. As for all four seasons each field was kept under the same management
281
practices it can be expected that observations have dependent errors throughout the entire experiment.
282
Therefore, all observations at the same location were considered as repeated measures with field taken as
283
subject. The residual variance-covariance structure matrix was then taken as a spatial power function with
284
the harvesting date (in Julian days) as the coordinate. As such, the correlation between yields of the various
285
seasons at the same location decays exponentially in time. The transplanting date of the first season was
286
taken as origin, bringing the harvest date of the first season on 100 Julian days, the second season on 217,
287
etc. Prior to the mixed regression all variables were standardized and a stepwise variable selection was
288
performed using the water balance components, spatial and soil fertility variables. In order to obtain one
289
model which is able to predict rice yields as a function of water and fertilizer management the model was
290
allowed to have management specific coefficients for all selected variables if appropriate. This is translated
291
into a maximum of four management specific coefficients per variable (i.e. two water and two fertilizer
292
management). The model was validated using a five-fold cross validation resulting in a random allocation
293
of the data into five folds from which four were used as calibration and one was used as validation. This
294
procedure was performed five times. The average Pearson’s correlation (r) between the observed and the
295
predicted values, resulting from the cross-validation, was calculated.
11
296
3
Results
297
3.1
298
Textural analysis of the soil samples showed that the rice fields are sand dominated ranging between 70.4
299
and 92.4%, followed by silt (8-21%) and clay (1.5-15%) (Table 2). The fields were classified as sandy
300
loam, sand and loamy sand corresponding to the USDA classification (Pedosphere, 2013). Soil organic
301
carbon ranged from 0.3 to 1.6 % with an average 0.8 ± 0.3%. Overall, average values of 1.5 ± 0.3 g kg-1
302
and 27.2 ± 13.2 mg kg-1 were found for Ntot and PO43- Bray while 0.2 ± 0.1 cmol kg-1 and 0.3 ± 0.1 cmol kg-1
303
was obtained for K+ and Mg2+, respectively. Values between 140 and 275 ppm were obtained for Fe2+. The
304
coefficient of variation showed moderate variability, ranging between 13-57%, for all parameters except
305
pH and the sand fraction. Particle size, soil organic carbon and total nitrogen analysis for the samples taken
306
prior to the field experiments revealed no significant difference between the toposequences (i.e. between
307
the toposequences allocated for a specific water and fertilizer treatment) or within one toposequence (i.e.
308
position of the field within a toposequence) (Table 2). With regard to the pH a significant difference
309
(p<0.0001) was found between the positions. The other nutrients and Fe2+ showed significant variability
310
between fields allocated to a specific management or at a specific location. However, no significant pattern
311
was detected.
312
3.2
313
For all four successive seasons, the groundwater tables in the upper lying fields were lower compared to
314
the lower lying fields. Lower lying fields had groundwater tables above surface level throughout the
315
majority of the rice seasons (Figure 1). Groundwater table measurements showed distinct differences in
316
both the toposequential direction (i.e. within one treatment) as between treatments (i.e. in the x direction,
317
Figure 3). The relatively high sand content of the fields resulted in high irrigation requirements to maintain
318
the standing water layer in the continuous flooding treatment. As such, a large variation in applied irrigation
319
depth between the fields within the toposequence as well as between the two toposequences within the same
320
treatment existed. In the dry seasons, total irrigated depths varied between 131 and 2772 mm resulting in
321
an overall average application of 1011 mm whereas in the wet seasons it varied between 0 and 1780 mm
322
resulting in an overall average application of 584 mm. For the reduced irrigation trials the applied depth
323
ranged between 19 and 657 mm (average: 251 mm) and between 0 and 102 mm (average: 28 mm) for the
324
dry and wet seasons, respectively (Figure 3). Within the same management toposequence, irrigation was
325
highest in the upper (position 1) and decreased substantially towards the lowest lying fields (position 6)
326
(Figure 3). The seasonal ETc,adj showed a slight spatial but no significant variation between fertilizer and
327
water management treatments (CV = 3-7 %). Average seasonal crop evapotranspiration for all fields was
328
489 mm and 437 mm during the dry and wet season, respectively. Seasonal percolation losses showed a
Soil spatial variability along toposequences
Spatio-temporal variation in water flow components along toposequences
12
329
slightly higher variability and ranged between 49 and 315 mm with an average around 176 mm and a CV
330
of 52 %.
331
332
The sandy soils and the shallow water table within the experimental layout as well as the high irrigation
333
amounts for the fields under the CF treatment resulted in a large two dimensional spatial variation of sub-
334
surface drainage and therefore water losses due to lateral flow (CV = -700 to 201 %) (Figure 3). Taking
335
into account the length of each cropping season, average lateral flow losses/ gains for all four seasons were
336
-19.9 to 3.1 mm day-1, respectively. The ratio of seasonal lateral flow vs. total water input (i.e. rainfall and
337
irrigation) showed on average water losses up to 71 % in the upper lying fields and water gains up to 117%
338
in the lower positions. Furthermore, losses attributed to sub-surface drainage resulted in larger water gains
339
in the fields under reduced irrigation management explaining the large variation of irrigation and drainage
340
amounts found between fields within the same toposequence and treatment (Figure 3). This points towards
341
a strong gradient in both the x as well as y direction of the experimental setup (Figures 2 and 3).
342
3.3
343
Detailed crop performance analysis of all the sub-plots at the various crop development stages showed a
344
significant seasonal effect with regards to plant height for all stages (p<0.0001) while only a significant
345
effect for SPAD values was found at the flowering stage (p<0.0001). With the exception of the first crop
346
development stage (i.e. two weeks after tillering) in the dry season, a significant fertilizer effect on plant
347
height was observed in all stages for both seasons pointing towards a clear fertilizer uptake (p<0.05).
348
Additionally, in the both seasons the field position within the toposequence showed to have an additional
349
effect on plant height in the flowering stage (-F: p<0.0001; +F: p = 0.006) but not in the tillering and
350
maximum tillering stage. No significant effect of water management during all stages and in both seasons
351
was observed. The SPAD values were only significantly affected by fertilizer treatment in the flowering
352
stage during the dry season (p<0.0001) and from the maximum tillering stage (p= 0.023) onwards in the
353
wet season. No significant effect of water treatment or position was found for the obtained SPAD values at
354
sub-plot level for any of the three stages.
Effect of water management on rice and water productivity
355
356
In both seasons, rice yields and biomass at sub-plot level showed a high variability for both fertilizer
357
treatments as well as the various toposequence positions (Figure 4). In the dry season the measured biomass
358
and yield median was 2.9 and 2.7 t ha-1 (CF treatment, CV= 39 and 33%), 2.4 and 2.4 t ha-1 (RI treatment,
359
CV= 41 and 26%) for the unfertilized while for the fertilized fields these increased to 4.8 and 3.9 t ha-1 (CF
360
treatment, CV= 31 and 26%) and 4.4 and 4.8 t ha-1 (RI treatment, CV= 85 and 30%), respectively. In the
361
wet season, slightly lower biomass and rice yields were observed for the unfertilized fields resulting in 2.5
13
362
and 2.6 t ha-1 (CF treatment, CV= 44 and 29%) and 2.1 and 2.2 t ha-1 (RI treatment, CV= 39 and 37%) while
363
fertilized fields produced 3.9 and 3.9 t ha-1 (CF treatment, CV= 59 and 21%) and 3.1 and 3.2 t ha-1 (RI
364
treatment, CV= 43 and 29%), respectively. Biomass and yield both showed a significant seasonal
365
(p<0.0001) as well as a seasonal dependent fertilizer effect (p=0.015 to p<0.0001) while no significant
366
impact of water management was found (Table 3). Seasonal analysis showed that for the wet season a
367
significant toposequential effect and its respective interaction with water management and fertilizer was
368
found for the observed rice yields. For biomass, on the other hand, the toposequential effect was significant
369
for both seasons in absence of the interactions with water management or fertilizer.
370
371
The high inter-field variability in rice yields resulted in a large variability in water productivity (Table 4).
372
Depending on its definition (Equations 1-4), water productivity at field level for the continuous flooding
373
treatment, varied between 0.06 and 2.14 kg m-3, 0.28 and 0.93 kg m-3, 0.22 and 0.88 kg m-3 for WPInet, WPET
374
and WPInet+L respectively. For the reduced irrigation trials values ranged between 0.15 and 2.35 kg m-3, 0.13
375
and 0.88 kg m-3, 0.13 and 0.74 kg m-3, respectively (Table 4). The water productivity based on net irrigation
376
(i.e. WPInet) showed a significant fertilizer effect for the dry season (p=0.038) while WPET and WPInet+L
377
revealed a significant fertilizer effect in both the dry and wet season (p<0.0001). However, the effect of
378
water management on water productivity was only found to be significant in the wet season for the
379
unfertilized trials based on the ET calculations (p=0.047). While water management did not significantly
380
affect water productivity in the dry season when the lateral flow was taken into account (i.e. WP Inet+L), the
381
toposequential (i.e. field position) did seem to be positively affecting water productivity in both seasons for
382
the lower lying fertilized fields (dry season: p= 0.025; wet season: p<0.0001).
383
3.4
384
The following variables were selected in the mixed regression model; (i) water balance components: total
385
seasonal rainfall (mm), seasonal groundwater table depth at field level within one toposequential position
386
(mm); (ii) spatial variable: distance of the field in the x direction according to the origin of the experiment
387
(i.e. upper left field) (m); and iii) field scale soil fertility variables: clay fraction (%), ferric content (Fe2+,
388
ppm) and total nitrogen (Ntot, g kg-1). The five-fold cross validation for all water management treatments
389
and their respective fertilizer application resulted in an overall regression model with a Pearson’s
390
correlation coefficient of 0.88 when prediction values were compared with the observed yields at field level
391
(Figure 5). The root mean square error (RMSE) of the model resulting from the five-fold cross validation
392
resulted in 0.31 ton ha-1 with a corresponding R2 of 0.79. It clearly showed its capability to predict the
393
spatial variation observed within toposequences as function of both water management practices as well as
394
the two fertilizer treatments.
Predicting spatio-temporal variation of rice yields in an inland valley
14
395
396
The intercept of the regression model varied between 3.09 and 4.30 and represented the average expected
397
grain yield for a specific water management as well as the fertilizer treatment within the water management
398
application with the highest potential obtained for the CF+F treatment followed by the RI+F (Table 5). The
399
interaction between groundwater table and the x-gradient of the field trial showed the highest positive
400
contribution (0.13) when predicting rice yields. Additionally, the clay fraction positively affected rice yields
401
whereas the interaction between clay and the iron content as well as the rainfall to the third power seemed
402
to decrease rice yields, with the latter having the strongest influence (-1.11). The seasonal groundwater
403
table fluctuation to the power two at field level seemed to only significantly affect the unfertilized
404
treatments while the squared rainfall only significantly influenced the fertilized treatments. However, the
405
latter revealed a different coefficient as a function of the water management treatment. In terms of soil
406
fertility, the contribution of total nitrogen content at field level in predicting rice yields along toposequences
407
showed a significant difference between water management treatments resulting in a negative effect for the
408
continuous flooding and a positive effect for the reduced irrigation experiments.
409
4
410
4.1
411
The lateral flow observed in Bamé had a strong 2-dimensional character; contributing to water gains
412
between fields, located at the same toposequential level as well as along a toposequence. The average lateral
413
flow losses/gains of -19.9 and 3.1 mm day-1, respectively are significantly higher than the ones reported by
414
Tsubo et al. (2005). This can be explained by the relatively higher sand content (84 %) in this study
415
compared to the study conducted by Tsubo et al. (2005) in Thailand (i.e. 5 – 20%). The contribution of
416
lateral flow to water gains in these fields influenced the required irrigation depths in both the RI as well as
417
the CF experiments. Especially for the fields under the CF treatment, significantly larger irrigation amounts
418
were applied in the upper lying compared to the lower lying fields within a toposequence. The 2-
419
dimensional sub-surface flow might contribute to the long term spatial variability of nutrients observed
420
within the field trial prior to the start of the experiment in 2011 (Table 2). As indicated by studies in Asia,
421
variation in soil chemical properties along toposequences is highly influenced by sediment deposition as
422
well as runoff processes and consequently alterations in soil physical properties (Haefele et al., 2010;
423
Schmitter et al., 2010). Sandy soils have lower water holding as well as nutrient binding capacity and
424
infiltration is known to increase with decreasing clay content. Therefore, a flushing effect can be expected
425
in those fields where lateral flow is substantially high, removing the easy soluble nutrients.
Discussion
Contribution of lateral flow to the spatio-temporal variation of rice and water productivity
426
15
427
The spatio-temporal variation of crop performance indicators at the various stages as well as the obtained
428
total biomass and rice yields seemed to be positively affected by toposequential position rather than the
429
water management practice. The spatial pattern was found to be complex and crop performance seemed
430
higher at the middle fields within a toposequence compared to the lower lying fields. This is mainly due to
431
the extreme flooding conditions resulting in a high standing water table (> 10 cm) in some of these fields
432
during the rainy season, negatively affecting grain development during the flowering stage. Although the
433
CF treatment seems on average to produce higher yields, no significant differences in crop performance or
434
rice production was found between the CF and the RI treatments for the same fertilizer treatment. However,
435
for both seasons, plant height at the flowering stage showed a clear toposequential influence within one
436
fertilizer treatment but the toposequential expression was only significantly reflected in rice yields for the
437
wet season.
438
439
The toposequential influence on rice performance was further confirmed when evaluating water
440
productivity. The calculated mean water productivity based on net irrigation as well as the
441
evapotranspiration, for both water management strategies and fertilizer treatments, fall within the ranges of
442
other field studies synthesized by Zwart (2013). To date, no comparative study was found that takes into
443
account the lateral flow losses, respectively gains when calculating water productivity. However, the results
444
in this study delineate the importance of including lateral flow when calculating water productivity. While
445
the fertilizer effect was found to significantly increase water productivity for WPInet, WPET and WPInet+L the
446
effect of water management practices was overshadowed by the strong lateral flow component. As such,
447
the upper lying fields within the toposequences under CF treatment tend to have significantly lower water
448
productivity (i.e. WPInet) due to the high amounts of irrigation water applied compensating the water losses.
449
As net irrigation does not take water losses/gains due to sub-surface drainage into account, the
450
toposequential effect on water productivity was masked and only found to be significant when the total
451
water availability was corrected for lateral flow at field level in both seasons.
452
4.2
453
The mixed model was able to predict rice yields as affected by water and fertilizer treatment within the
454
inland valley with a Pearson correlation of 0.88, resulting from a 5-fold cross validation. Although no
455
significant difference was found between water management treatments for the same fertilizer treatment
456
due to the high intra-field spatial variability, the mixed regression model did reveal, aside from a fertilizer
457
specific, a water management specific intercept indicating significant differences in average expected yields
458
for each treatment within the inland valley. The intercepts reflect that reduced irrigation could lead to a
459
slight but significant reduction of rice yields compared to the continuous flooding treatment.
Predicting spatio-temporal rice yields using groundwater tables in inland valleys
16
460
461
Aside from the sub-surface flow contribution on rice productivity, as discussed above, the role of
462
groundwater showed a similar strong two dimensional character. The groundwater table at field level
463
reflects inherently the toposequential position, as each field had a significant different groundwater table
464
with lower tables observed in the upper lying fields compared to the lower ones. The importance of shallow
465
ground tables when predicting grain yields was revealed in the established mixed regression model and
466
highly significant for the unfertilized toposequences. This can be linked to the presence of nutrients in the
467
groundwater available for plant uptake in absence of fertilizer. The significant importance of shallow
468
groundwater tables on paddy rice was indicated by Shen et al. (2014).
469
470
The inter-linkage between fields at the same toposequential position was represented by the interaction of
471
the groundwater component with the distance of the field in the x-direction, indicating the recharge of
472
groundwater tables as influenced by the lateral flow. The contribution of this variable was found to be
473
independent of the water and fertilizer treatment applied as dominated by soil physico-chemical
474
characteristics. This is supported by the presence of the clay component in the model having one coefficient
475
for all treatments. The incorporation of the clay component in the model has a dual functionality; it can be
476
seen as a proxy for the soil nutrient status in the field as well as the porosity of the field and associated
477
water losses. Furthermore, a significant interaction between clay and the presence of Fe2+ in the topsoil was
478
found, decreasing rice yields in the inland valley. However, in this study iron uptake in plant material and
479
rice grains were examined within one season but not found to reach toxicity levels (data not shown). Its
480
presence in the model however, does indicate the sensitivity of rice yields to the iron concentrations in the
481
top soil and reflects its negative influence on yield potentials in the valley as affected by textural
482
characteristics. Due to the age of the inland valleys as well as sedimentation of iron rich sediments in the
483
lowlands, iron toxicity is one of the main constraints in West African inland valleys (Becker and Asch,
484
2005). The presence of iron in the top soil and the redox processes occurring during water logging
485
conditions reduces the uptake of other minerals such as nitrogen and phosphorus (Worou et al., 2013).
486
Furthermore, Worou et al. (2013) observed a clear effect of toposequential position on the presence of iron
487
and subsequently showed its negative impact on nitrogen uptake by the rice plants. Although the clay
488
content reduces Fe dynamics due to complexation in the clay mineral surfaces compared to sandy soils, it
489
is highly dependent on the types of clay minerals present in the inland valley (Becker and Asch, 2005).
490
Prior to the field experiment a spatial variation of Fe2+ was observed between the fields, mainly with regards
491
to the overall toposequential position of the field as well as within a fertilizer and water management
492
toposequence (Table 2). As such it can be expected that in these inland valleys the combination of
17
493
groundwater tables, clay content and the associated availability of Fe2+ might influence the spatial
494
variability of rice yields.
495
496
Aside from the negative influence of iron on rice production, the total nitrogen content in the top soil seems
497
to negatively affect production in the fields under continuous flooding compared to the reduced irrigation
498
trials. No significant spatial pattern was found prior to the commencement of the field experiment. The
499
presence of this variable in the model is most probably linked with the increased plant available nitrogen
500
(NO3-) losses due to increased irrigation and associated sub-surface losses influenced by the high sand
501
content of the soil. Due to the interaction of the soils physical properties and the occurring hydrological
502
processes in the field no significant difference of this variable as a function of fertilizer treatment can be
503
expected which is reflected in the model by the coefficient dependency on water management rather than
504
fertilizer application.
505
5
506
This study revealed the significant contribution of shallow groundwater tables and lateral flow when
507
assessing water management strategies and their effect on the spatio-temporal variation of rice production
508
and water productivity in a sandy inland valley. The toposequence position clearly influenced water
509
productivity due to sub-surface lateral flow which was found to significantly influence water gains in the
510
lower lying fields, reducing irrigation requirements within a specific water management treatment. The
511
results indicate that based on the groundwater table, rainfall and standard soil physico-chemical
512
characteristics, rice yields can be predicted in these inland valleys for continuous flooding and reduced
513
irrigation practices. Groundwater contribution was found to play an important role in predicting rice yields,
514
especially for the unfertilized treatments which is potentially linked to the high nutrient concentrations
515
resulting from the sub-surface nutrient drainage of surrounding fields. However, as this study was restricted
516
to one inland valley comparative studies in similar regions with varying groundwater tables (i.e. shallow to
517
deep) would help to further elucidate and validate the applicability of this study. As such, rice production
518
as well as water consumption could be estimated for various inland valleys in West-Africa under similar
519
agro-climatological conditions. Optimizing the established function within a large scale assessment would
520
lead to improved regression functions suitable in the region to estimate rice production and water
521
consumption across scales.
Conclusion
18
522
6
Acknowledgements
523
This study was carried out in the framework of the project ‘Sawah, Market Access and Rice Technologies
524
for Inland Valleys’ (SMART-IV). The authors acknowledge the Ministry of Agriculture, Forestry and
525
Fisheries (MAFF) of Japan for funding the project. We would like to thank Hernaude Agossou, Maurice
526
Mondegnon, Bjorn Nikolaus and Gertrude Tongnite for their contributions in data collection and analysis.
527
19
528
References
529
530
531
Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration-Guidelines for computing
crop water requirements - FAO Irrigation and drainage paper 56. Food and Agriculture Organization of the
United Nations, Rome, 300, 6541.
532
533
Andriesse, W., Fresco, L.O., 1991. A characterization of rice-growing environments in West Africa.
Agriculture, Ecosystems & Environment 33, 377-395.
534
535
Becker, M., Asch, F., 2005. Iron toxicity in rice - Conditions and management concepts. Journal of Plant
Nutrition and Soil Science 168, 558-573.
536
537
Becker, M., Johnson, D.E., 2001. Improved water control and crop management effects on lowland rice
productivity in West Africa. Nutrient Cycling in Agroecosystems 59, 119-127.
538
539
de Vries, M.E., Rodenburg, J., Bado, B.V., Sow, A., Leffelaar, P.A., Giller, K.E., 2010. Rice production
with less irrigation water is possible in a Sahelian environment. Field Crops Research 116, 154-164.
540
541
542
Djagba, J.F., Rodenburg, J., Zwart, S.J., Houndagba, C.J., Kiepe, P., 2014. Failure and success factors of
irrigation system developments: A case study from the Ouémé and Zou valleys in Benin. Irrigation and
Drainage 63, 328-339.
543
544
FAO (Food and Agricultural Organization), 1985. Guideline for Land Evaluation for Irrigated Agriculture.
Soils Bulletin, 55. FAO, Rome. Land and Water Development Division
545
FAO Stat, 2011. Agricultural production of Benin in 2009. FAO, Rome, Itally.
546
547
Gruber, I., Kloos, J., Schopp, M., 2009. Seasonal water demand in Benin's agriculture. Journal of
Environmental Management 90, 196-205.
548
549
Haefele, S.M., Sipaseuth, N., Phengsouvanna, V., Dounphady, K., Vongsouthi, S., 2010. Agro-economic
evaluation of fertilizer recommendations for rainfed lowland rice. Field Crops Research 119, 215-224.
550
551
Houba, V.J.G., Lee, J.J.V.d., Novozamsky, I., 1995. Soil Analysis Procedures Other Procedures.
Department of Soil Science and Plant Nutrition, Wageningen, The Netherlands.
552
553
International Institute or Tropical Agriculture, 1979. Selected Methods for Soil and Plant Analysis. Oyo
Road, PMB 5320 Ibadan, Nigeria.
554
555
Janssen, M., Lennartz, B., 2009. Water losses through paddy bunds: Methods, experimental data, and
simulation studies. Journal of Hydrology 369, 142-153.
556
557
558
Krupnik, T.J., Shennan, C., Rodenburg, J., 2012. Yield, water productivity and nutrient balances under the
System of Rice Intensification and Recommended Management Practices in the Sahel. Field Crops
Research 130, 155-167.
559
560
Pedosphere,
2013.
Soil
Bulk
Density
http://www.pedosphere.ca/resources/bulkdensity/.
Calculator
(U.S.
Texture
Triangle).
20
561
562
563
Rodenburg, J., Zwart, S.J., Kiepe, P., Narteh, L.T., Dogbe, W., Wopereis, M.C.S., 2014. Sustainable rice
production in African inland valleys: Seizing regional potentials through local approaches. Agricultural
Systems 123, 1-11.
564
565
566
567
Saïdou, A., Kossou, D., 2009. Water management for enhancing land productivity in Benin: Exploring
opportunities for enhancing land productivity by smallholder farmers through water management. Towards
Enhancing Innovation System Performance in Smallholder African Agriculture, Proceedings of the First
CoS-SIS International Conference, CoS-SiS Programme, 48-52.
568
569
Saito, K., Dieng, I., Toure, A.A., Wopereis, M.C.S., 2014. Rice yield trend analysis for 24 African countries
over 1960–2012. Global Food Security (submitted).
570
571
572
573
Saito, K., Nelson, A., Zwart, S.J., Niang, A., Sow, A., Yoshida, H., Wopereis, M.C.S., 2013. Towards a
better understanding of biophysical determinants of yield gaps and potential expansion of rice area in
Africa, in: Wopereis, M.C.S., Johnson, D., Ahmadi, N., Tollens, E., Jalloh, A. (Eds.), Realizing Africa's
Rice Promise. CABI (Wallingford, United Kingdom), Africa Rice Center (Cotonou, Benin), pp. 264-274.
574
575
576
Schmitter, P., Dercon, G., Hilger, T., Hertel, M., Treffner, J., Lam, N., Duc Vien, T., Cadisch, G., 2011.
Linking spatio-temporal variation of crop response with sediment deposition along paddy rice terraces.
Agriculture, Ecosystems and Environment 140, 34-45.
577
578
579
Schmitter, P., Dercon, G., Hilger, T., Thi Le Ha, T., Huu Thanh, N., Lam, N., Duc Vien, T., Cadisch, G.,
2010. Sediment induced soil spatial variation in paddy fields of Northwest Vietnam. Geoderma 155, 298307.
580
581
582
583
Seck, P.A., Touré, A., Coulibaly, J.Y., Diagne, A., Wopereis, M.C.S., 2013. Africa’s rice economy before
and after the 2008 rice crisis, in: Wopereis, M.C.S., Johnson, D., Ahmadi, N., Tollens, E., Jalloh, A. (Eds.),
Realizing Africa's Rice Promise. CABI (Wallingford, United Kingdom), Africa Rice Center (Cotonou,
Benin), pp. 264-274.
584
585
Shen, Y.J., Zhang, Z.B., Gao, L., Peng, X., 2014. Evaluating contribution of soil water to paddy rice by
stable isotopes of hydrogen and oxygen. Paddy and Water Environment, 1-9.
586
587
588
Totin, E., Van Mierlo, B., Saïdou, A., Mongbo, R., Agbossou, E., Stroosnijder, L., Leeuwis, C., 2012.
Barriers and opportunities for innovation in rice production in the inland valleys of Benin. NJAS Wageningen Journal of Life Sciences 60-63, 57-66.
589
590
591
Tsubo, M., Basnayake, J., Fukai, S., Sihathep, V., Siyavong, P., Sipaseuth, Chanphengsay, M., 2006.
Toposequential effects on water balance and productivity in rainfed lowland rice ecosystem in Southern
Laos. Field Crops Research 97, 209-220.
592
593
594
Tsubo, M., Fukai, S., Basnayake, J., To, P.T., Bouman, B., Harnpichitvitaya, D., 2005. Estimating
percolation and lateral water flow on sloping land in rainfed lowland rice ecosystem. Plant Production
Science 8, 354-357.
595
596
597
Tsubo, M., Fukai, S., Basnayake, J., To, P.T., Bouman, B., Harnpichitvitaya, D., 2007. Effects of soil clay
content on water balance and productivity in rainfed lowland rice ecosystem in Northeast Thailand. Plant
Production Science 10, 232-241.
21
598
599
600
Tsujimoto, Y., Yamamoto, Y., Hayashi, K., Zakaria, A.I., Inusah, Y., Hatta, T., Fosu, M., Sakagami, J.I.,
2013. Topographic distribution of the soil total carbon content and sulfur deficiency for rice cultivation in
a floodplain ecosystem of the Northern region of Ghana. Field Crops Research 152, 74-82.
601
602
603
Wei, J.B., Xiao, D.N., Zeng, H., Fu, Y.K., 2008. Spatial variability of soil properties in relation to land use
and topography in a typical small watershed of the black soil region, northeastern China. Environmental
Geology 53, 1663-1672.
604
605
606
Worou, O.N., Gaiser, T., Saito, K., Goldbach, H., Ewert, F., 2012. Simulation of soil water dynamics and
rice crop growth as affected by bunding and fertilizer application in inland valley systems of West Africa.
Agriculture, Ecosystems and Environment 162, 24-35.
607
608
609
Worou, O.N., Gaiser, T., Saito, K., Goldbach, H., Ewert, F., 2013. Spatial and temporal variation in yield
of rainfed lowland rice in inland valley as affected by fertilizer application and bunding in North-West
Benin. Agricultural Water Management 126, 119-124.
610
611
612
Zwart, S.J., 2013. Assessing and Imrpoving Water Productivity of Irrigated Rice Systems in Africa, in:
Wopereis, M.C.S., Johnson, D., Ahmadi, N., Tollens, E., Jalloh, A. (Eds.), Realizing Africa's Rice Promise.
CABI (Wallingford, United Kingdom), Africa Rice Center (Cotonou, Benin), pp. 264-274.
613
614
22
615
Figures
616
617
Figure 1: Monthly rainfall distribution (mm, vertical bars) and corresponding 50% of the monthly total
618
potential evapotranspiration (0.5 ETo, mm, black line) according to FAO guideline (1985) , indicating dry
619
periods where rainfall is below the 0.5 ETo line (top). Corresponding daily average measured groundwater
620
fluctuations (mm) of the top fields (black) and bottom fields (grey) (bottom) are given below for 2011-2012
621
at the experimental site of Bamé. Dashed vertical lines indicate the four measured rice seasons and the
622
dashed zero line represents the surface level of the rice fields.
623
624
Figure 2: Schematic representation of the field experiment with RI representing the reduced irrigation and
625
CF the continuous flooding treatments. Within each treatment -F represents the unfertilized and +F the
626
fertilized toposequence. Each field within the toposequence is irrigated and drained through the
627
neighbouring irrigation and drainage channel, respectively. Within each field, nine sub-plots of 1 m2 were
628
delineated (squares). The five black coloured sub-plots represent the soil samples analysed. The blue
629
channels represent the irrigation channels while the drainage channels are in green.
630
631
Figure 3: Averaged water balance components for the two dry seasons (left) and for the two wet seasons
632
(right) for each of the water and fertilizer treatment at their respective distance of 10 (RI–F), 20.5 (RI+F), 31
633
(CF+F), 41.5 (CF-F), 52 (CF-F), 62.5 (CF+F), 73 (RI+F) and 83.5 m (RI–F) in the experimental layout (Figure
634
2). Position refers to the field position in the toposequence with position 1 being the first top field and
635
position 6 the last field.
636
637
Figure 4a: Box plots of measured rice yields in the dry season within the sub-plots as function of field
638
position along a toposequence (y-axis). Separate box plots are made for each water management treatment
639
(left to right) (i.e. RI and CF, respectively) and each fertilizer treatment top to bottom (unfertilized and
640
fertilized, respectively) within one repetition (i.e. repetition 1 and 2). Crossbars, boxes and whiskers
641
represent the median, quartile range (5th and 95th percentile) and range, respectively. The dashed line
642
represents the mean of the measured rice yields within a field.
643
644
Figure 4b: Box plots of measured rice yields in the wet season within the sub-plots as function of field
645
position along a toposequence (y-axis). Separate box plots are made for each water management treatment
646
(left to right) (i.e. RI and CF, respectively) and each fertilizer treatment top to bottom (unfertilized and
647
fertilized, respectively) within one repetition (i.e. repetition 1 and 2). Crossbars, boxes and whiskers giving
23
648
the median, quartile range (5th and 95th percentile) and range, respectively. The dashed line represents the
649
mean of the measured rice yields within a field.
650
651
Figure 5: Observed vs. predicted rice yields (t ha-1) at field level as function of water management, fertilizer
652
application and topographical location after cross validation of the established mixed effects model.
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
24
682
683
684
685
686
687
688
25
689
26
27
690
691
692
28
693
694
29
695
30
Download