Post on 21-Dec-2015
description
transcript
Soil
Qua
lity
& F
erti
lity
Agronomy Journa l • Volume 100 , I s sue 5 • 2008 1511
Published in Agron. J. 100:1511–1526 (2008).doi:10.2134/agronj2007.0355
Copyright © 2008 by the American Society of Agronomy, 677 South Segoe Road, Madison, WI 53711. All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher.
Although western Kenya is regarded to be a region
of high potential for crop production, current yields
of the major crops in smallholder farms are much less than
yields achieved under controlled experimental conditions
on research stations. These yield gaps are largely the result
of nutrient limitations, weed infestation, pests, diseases,
and poor agronomic management that together reduce the
efficiency of use of available nutrients and water (Tittonell
et al., 2007a,b). Given the small farm sizes, the problems
of poor soil fertility, and the scarcity of labor and nutri-
ent resources in this densely populated region, mineral
fertilizers are one option to increase both land and labor
productivity. However, the use of mineral fertilizers within
smallholder systems should be designed judiciously to
ensure their effectiveness and to avoid negative environ-
mental externalities. Far from being a solution per se to
poor land and labor productivity, mineral fertilizers are a
useful and necessary means to improve productivity when
strategically allocated to specific niches within complex and
dynamic farming systems. The design of such strategies
should not overlook the effects of farm heterogeneity and
long-term sustainability of farming practices.
Use of mineral fertilizers faces high transaction costs in rural
markets. Th ey are retailed at higher prices than in urban wholesale
markets and oft en not labeled, so that farmers are unable to verify
their composition. Moreover, decisions on purchasing fertilizers
are made before planting, at a time of high demand for other
important household expenditures (e.g., paying school fees), or
when farmers have already sold their harvest from the previous
season. As a result, the amounts of fertilizers that farmers can
access are small, and therefore it is crucial that these are targeted to
fi elds within their farm that allow the highest marginal returns to
investments (van Keulen and Breman, 1990). Within smallholder
farms, fi elds can be identifi ed that exhibit diff erent patterns of
responsiveness to applied nutrients: poorly responsive fertile fi elds,
poorly responsive infertile fi elds, and responsive medium-to-infer-
tile fi elds (Tittonell et al., 2007b; Zingore et al., 2007).
Strategically targeted fertilizer use together with organic nutri-
ent resources to ensure fertilizer use effi ciency and crop productiv-
ity at farm scale are basic principles of ISFM (Vanlauwe and Giller,
2006). In particular, poorly responsive infertile fi elds require long-
term rehabilitation to build up soil fertility before crops respond to
ensure effi cient use of applied nutrients. In mixed crop–livestock
systems, the combined application of animal manure and mineral
fertilizers is one option to achieve this. Synergies and/or additive
eff ects have been observed in fi eld experiments testing diff erent
combinations of manure and mineral fertilizers (e.g., Vanlauwe et
al., 2001; Bationo et al., 2006).
Most of these results, however, have to be interpreted cautiously
since the application rates and the quality of the manure used in
most experiments are superior to those that farmers can aff ord
in practice. Under smallholder farmers’ conditions, it may be
expected that even manures with poor nutrient concentrations can
be useful to build up soil C and supply micronutrients to crops,
when applied over successive years. Such long-term strategies to
build up soil fertility are especially necessary on poorly responsive
ABSTRACTIntegrated soil fertility management (ISFM) technologies for African smallholders should consider (i) within-farm soil hetero-geneity; (ii) long-term dynamics and variability; (iii) manure quality and availability; (iv) access to fertilizers; and (v) competing uses for crop residues. We used the model FIELD (Field-scale resource Interactions, use Effi ciencies and Long term soil fertility Development) to explore allocation strategies of manure and fertilizers. Maize response to N fertilizer from 0 to 180 kg N ha–1 (±30 kg P ha–1) distinguished poorly responsive fertile (e.g., grain yields of 4.1–5.3 t ha–1 without P and of 7.5–7.5 t ha–1 with P) from responsive (1.0–4.3 t ha–1 and 2.2–6.6 t ha–1) and poorly responsive infertile fi elds (0.2–1.0 t ha–1 and 0.5–3.1 t ha–1). Soils receiving manure plus fertilizers for 12 yr retained 1.1 to 1.5 t C ha–1 yr–1 when 70% of the crop residue was left in the fi eld, and 0.4 to 0.7 t C ha–1 yr–1 with 10% left . Degraded fi elds were not rehabilitated with manures of local quality (e.g., 23–35% C, 0.5–1.2% N, 0.1–0.3% P) applied at realistic rates (3.6 t dm ha–1 yr–1) for 12 yr without fertilizers. Mineral fertilizers are neces-sary to kick-start soil rehabilitation through hysteretic restoration of biomass productivity and C inputs to the soil.
P. Tittonell, Plant Production Systems (PPS), Dep. of Plant Sci., Wageningen Univ., P.O. Box 430, 6700 AK Wageningen, Th e Netherlands and Tropical Soil Biology and Fertility Inst. of the Int. Centre for Tropical Agric. (TSBF-CIAT), United Nations Ave., P.O. Box 30677, Nairobi, Kenya; M. Corbeels, TSBF-CIAT and Centre de Coopération Int. en Recherche Agron. pour le Dév. (CIRAD), SupAgro, Bâtiment 27, 2 place Viala, 34060 Montpellier Cedex 2, France; M.T. van Wijk, PPS; B. Vanlauwe, TSBF-CIAT; K.E. Giller, PPS. Received 26 Oct. 2007. *Corresponding author (pablo.tittonell@cirad.fr).
Abbreviations: DAP, diammonium phosphate; HC, humifi cation coeffi cient; ISFM, integrated soil fertility management; PAR, photosynthetically active radiation.
Combining Organic and Mineral Fertilizers for Integrated Soil Fertility Management in Smallholder Farming Systems of Kenya: Explorations Using the Crop-Soil Model FIELD
P. Tittonell,* M. Corbeels, M. T. van Wijk, B. Vanlauwe, and K. E. Giller
1512 Agronomy Journa l • Volume 100, Issue 5 • 2008
infertile fi elds to achieve signifi cant crop responses to applied
nutrients. Within this context, of limited access to fertilizers, poor
soil fertility, and poor quality and availability of manure, options
for soil fertility management within heterogeneous farms should
be explored.
Simulation modeling can help in identifying options, and in
understanding the trade-off s between short- and long-term ben-
efi ts of ISFM. A simple, dynamic crop–soil simulation model,
FIELD (Tittonell et al., 2007c), was developed to explore crop
and soil management strategies within the existing heterogeneous
conditions of smallholder farms and to assess a range of indicators
of resource-use effi ciency. FIELD is the crop–soil module of a
farm-scale model (NUANCES-FARMSIM), in which it operates
linked to livestock, manure management, and household deci-
sions modules to analyze resource and labor allocation strategies in
African farming systems. A relatively simple modeling tool is nec-
essary to perform such analyses, given (i) the scarcity of biophysi-
cal data (of the type needed to parameterize most crop growth
simulation models) for most African cropping systems; and (ii) the
multiple interactions between crop management factors operating
at farm scale (e.g., labor allocation to weeding), which may have
a larger impact on crop productivity than the typical crop–soil
processes that are being simulated using the detailed crop growth
models.
FIELD is built around the concept of resource-use effi ciencies
(i.e., radiation, water and nutrient-use effi ciencies) for the assess-
ment of crop production. Th e model conserves the key attributes
of the approach taken in QUEFTS (Janssen et al., 1990) to
account for nutrient interactions, but incorporates long-term
plant–soil feedbacks and the interactions with other relevant driv-
ers of farm heterogeneity (i.e., management decisions). FIELD has
proven to simulate maize (Zea mays L.) and soybean (Glicine max
L.) responses to N, P, and manure applications reasonably well on
clayey and sandy soils in Zimbabwe (Tittonell et al., 2007c).
Th e objective of this study was to analyze options for ISFM
within heterogeneous smallholder farms, combining the use of
organic and mineral fertilizers, while considering application rates
that are aff ordable for the farmers. We fi rst calibrated and tested
the model FIELD for maize against a number of experimental
datasets and then used it to analyze (i) the eff ect of current soil fer-
tility status on crop responsiveness and the effi ciency of mineral
fertilizer use; (ii) the potential of diff erent ISFM strategies to
maintain or build up soil fertility in the long term; and (iii) the
capacity of diff erent categories of fi elds to support responses in
crop productivity when soil restorative measures are put in place.
In the search for options for targeting ISFM technologies and
to address the above objectives, the following research questions
were formulated: (i) how does maize – the major food and cash
crop in the region – respond to increasing rates of applied N and
P (little response to K has been observed in trials, see Tittonell et
al., 2007b) within spatially heterogeneous farms? (ii) how does
maize respond to realistic, minimum rates of mineral fertilizers
in the presence of diff erent types of manure within spatially
heterogeneous farms? (ii) if part of the crop harvest residues are
retained on the fi eld, is it possible to maintain adequate organic
carbon in the soil through increased mineral fertilizer applica-
tions (with and without manure application)? (iv) if an increase
in soil organic matter leads to improved resource use effi ciency,
better use of applied mineral fertilizers, and crop productivity,
what is the capacity of diff erent management interventions to
restore soil productivity through soil organic matter buildup for
fi elds that underwent diff erent intensities of soil degradation?
Th is capacity of soil restoration, or capacity of the system to react
to management practices aimed at rehabilitating soil productivity
is referred to as hysteresis of soil restoration, in analogy to the path-
dependent process of hysteresis occurring in natural systems (e.g.,
in drying-and-rewetting soils, Scanlon et al., 2002).1 Although not
strictly similar, we believe that the behavior of soils that undergo
degradation and rehabilitation resembles the phenomenon of hys-
teresis (see also: Lal, 1997), on the basis that when soil C or crop
yields are followed in time for a soil undergoing degradation they
tend to follow a concave decline; when measures are put in place
to restore productivity, these indicators tend to follow an upward,
convex trajectory.
MATERIALS AND METHODSSystem Characterization and Background
Th e study sites in western Kenya comprise highland and mid-
land agroecological zones that receive 1300 to 2100 mm of annual
rainfall in a bimodal pattern. In normal years, 60 to 70% of the
rainfall occurs during the long rains season, between February and
June, while the rest falls during the short rains between August
and November. Farms sizes are small (0.5 to 2 ha), and although
soil types vary within the landscape, soils are in general inherently
fertile (70% of the area is considered to be of high agricultural
potential). Diff erential long-term management of the fi elds within
the farm has led to strong heterogeneity in soil productivity within
individual farms (Tittonell et al., 2007b). In general, current soil
fertility is poor as a result of continuous cultivation with little
nutrient input through organic and/or mineral fertilizers, which is
oft en aggravated by soil water erosion. Cultivation without inputs
is the result of poor availability of, or limited access to, nutrient
resources (Table 1). For farmers who own cattle, manure applica-
tion rates vary (on average, between 0.9 to 4 t fresh mass ha–1)
1In a deterministic system with no hysteresis and no dynamics, it is possible to predict the output of the system at a given moment in time, knowing only the input to the system at that moment. If the system has hysteresis, in order to predict the output it is necessary to consider also the path that the response follows before reaching its current value.
Table 1. Nitrogen use in farms from different wealth classes in west-ern Kenya as derived from analysis of resource fl ow maps (adapted from Tittonell, 2003). Area cropped, livestock owned and potential availability of manure and C, N and P for application to crops.
Village†Resource
endowmentLand
cropped Livestock
heads
Potential manure
availability
Potential application
rates‡C N P
ha no. farm–1 t year–1 kg ha–1
Ebusiloli higher 2.1 4.0 8.4 960 38 6.1medium 1.1 2.2 3.6 785 31 5.0poorer 0.5 0.8 1.1 528 21 3.3
Among’ura higher 2.3 2.3 3.5 212 8 1.3medium 2.2 2.0 2.9 218 9 1.4poorer 1.0 1.7 2.0 408 16 2.6
† Ebusiloli (Vihiga district) is located in a highly populated area (ca. 1000 inhabitants km–2), closer to urban centers with easier access to markets; intensive (zero grazing, Friesian) livestock production systems predominate. Among’ura (Teso district) area is less populated (200–300 inhabitants km–2), land is available for fallow, markets are far, and the local (zebu) livestock graze in communal land.
‡ Calculated over the total area of cropped land, assuming optimum manure handling and an average dry matter content of 80%, C content 30%, N content 1.2%, and P content 0.19%.
Agronomy Journa l • Volume 100, Issue 5 • 2008 1513
across farms of diff erent resource endowment and
across localities where diff erent livestock management
systems prevail (e.g., free grazing vs. stall feeding).
Despite the scarcity of animal manure, only a rela-
tively small number of farmers use small quantities
of mineral fertilizers. For example, in the case of N
fertilizers, the wealthiest farmers in the region may
apply up to 60 to 80 kg N ha–1 on small portions (10
to 40%) of their cropped land (Tittonell et al., 2005).
Among the poorest farmers, those who use fertilizers
apply them on less than 10% of their land area with N
application rates below 20 kg ha–1. Crop productivity
in the region is mostly limited by N and P; localized K
defi ciencies were also reported (Shepherd et al., 1996).
Overview of the FIELD ModelFIELD is the crop-soil module of the bio-eco-
nomic model NUANCES-FARMSIM (FArm-scale
Resource Management SIMulator; www.africanu-
ances.nl), which simulates household objectives and
constraints, resource allocation patterns, labor and
economic balances, and nutrient fl ows at farm level
(Fig. 1A). FARMSIM is designed for analyzing trade-
off s between farming systems and the environment,
focusing on strategic decision-making and embracing
the spatial and temporal variability of smallholder sys-
tems. FARMSIM consists of a crop–soil (FIELD), a
livestock (LIVSIM), and a manure (HEAPSIM) mod-
ule that are integrated functionally to allow capturing
feedbacks between these identities at farm scale, as
aff ected by farmers’ management decisions. FIELD
simulates long-term changes in soil fertility (C, N, P,
and K), interactions between nutrients that determine
crop production, and crop responses to management
interventions such as mineral fertilizer and/or manure
applications. Diff erent fi elds within a farm represent
combinations of crop types and sets of soil proper-
ties, which are simulated as diff erent instances of the
FIELD module. Simulation of livestock productivity,
growth, and herd dynamics is done with LIVSIM,
while nutrient cycling through manure is simulated
using HEAPSIM (Rufi no et al., 2007).
Here, we used a stand-alone version of the FIELD
model. Th e simulation of soil processes in FIELD was
described by Tittonell et al. (2007a). Th e approach
for the simulation of crop production is illustrated in Fig. 1B.
Total dry matter and grain yields are calculated on the basis of
seasonal resource (light, water, and nutrients) availabilities and
use effi ciencies, according to the generic equation:
Crop production = Resource availability × Resource
capture effi ciency × Resource conversion effi ciency
From the total amount of incident photosynthetically active
radiation (PAR) during the growing season, only a fraction is
intercepted by the crop (FRINT), and this is converted into crop
biomass using a light conversion effi ciency coeffi cient (Fig. 1B).
It calculates the light-determined yield that is aff ected by man-
agement factors such as cultivar choice, planting date, or stand
density (and thus it cannot be considered to be the potential yield in a strict sense). Water-limited crop production is cal-
culated on the basis of seasonal rainfall and a site- and crop-
specifi c rainfall use effi ciency coeffi cient. Th e way in which this
coeffi cient is estimated for a given case study depends on avail-
ability of data. When suffi cient data are available, detailed crop
growth models simulating the soil water balance can be used to
generate functional relationships (e.g., fraction of rainfall infi l-
trated vs. runoff as a function of soil texture and slope) that are
then built into FIELD (explained below). When no data are
available, rainfall use effi ciency coeffi cients (i.e., yield per mm
of rain) derived from literature and/or experiments are used
to estimate water-limited crop yields. For example, crop yields
measured on experimental plots receiving full-nutrient treat-
Fig. 1. (A) Schematic representation of the relationships between differ-ent modules of the dynamic farm system model FARMSIM (FArm-scale Resource Management SIMulator). The various modules simulating soil-crop (FIELD), livestock (LIVSIM), and manure storage (HEAPSIM) dynamics are functionally integrated through C and nutrient flows (full and dotted black arrows). The household module dictates management and allocation of the various farm resources, represented by land, labor, water, nutrients, and financial resources (gray arrows). Different instances of FIELD and LIVSIM represent the various crop and livestock activities on the farm, while different instances of FARMSIM represent different farm types in a community. (B) Schematic representation of how crop production is simulated in FIELD (Field-scale resource Interactions, use Efficiencies, and Long-term soil fertility Development). See text for fur-ther explanation.
1514 Agronomy Journa l • Volume 100, Issue 5 • 2008
ments under controlled conditions may be considered to be
close to the water-limited yields for a given site, and can thus be
used to calibrate the rainfall capture and conversion (or tran-
spiration) effi ciency coeffi cients for the given crop at that site.
Nutrient availabilities and use effi ciencies determine
nutrient-limited crop production. Nutrient capture effi ciency
results from the partitioning of available nutrients between
crop uptake and other processes that act as nutrient sinks (e.g.,
leaching and gaseous losses of N, immobilization into soil
organic matter). Nutrient conversion effi ciency is the inverse
of the weighted average nutrient concentration in the crop and
range between a crop-specifi c minimum and maximum value
(Nijhof, 1987). Resource-limited crop production in FIELD
is then calculated as the minimum of water-limited produc-
tion and the production determined by the availability and
use effi ciency of N, P, and K and their interactions following
Liebscher’s Law of the Optimum (van Keulen, 1995). Actual
crop production is then calculated by applying a reduction
factor for weed competition. Actual grain yield is fi nally deter-
mined by multiplying actual biomass production with a harvest
index coeffi cient. More details on FIELD can be found in
Tittonell et al. (2007a).
Model Set-up, Calibration, and TestingIn this section we describe how FIELD was calibrated and
tested for the conditions of the study area, using four indepen-
dent datasets. We fi rst used a dynamic crop growth simulation
model running on a daily time step, already tested for maize in
western Kenya, to derive functional relationships that describe
light and water use effi ciencies. Th en, we calibrated FIELD
against long-term datasets on changes in soil C with and
without manure application, and fi nally tested the model to
simulate crop responses to applied manure and mineral fertil-
izers. Once the model was calibrated and tested, we used the
model to respond to our research questions by running a set of
scenario simulations of ISFM strategies. Th ese are described in
a later section.
Data SourcesTh e following four datasets were used in the various steps
of model calibration and testing: (1) data on soil organic C
dynamics, from a chronosequence of agricultural fi elds of dif-
ferent age following forest clearance (up to 100 yr of continu-
ous crop cultivation) around the Kakamega National Forest
Reserve in western Kenya (Solomon et al., 2007); (2) soil
organic C and crop biomass data from a long-term experiment
(1989–2003) on eff ects of manure application (5 and 10 t ha–1
yr–1) in maize-based cropping systems at Machang’a, Kenya
(Micheni et al., 2004); (3) data on maize responses to increas-
ing rates of manure application (0, 1.2, and 4 t C ha–1) with
and without mineral N applied at a rate of 120 kg ha–1 (in the
presence of P and K fertilizers) from an experiment that was
conducted during two consecutive growing seasons (long and
short rains of 2005) at two localities in Aludeka and Nyabeda
(c. 20 km from Emuhaya) (unpublished); (4) crop biomass
and soil fertility data from an on-farm N-P-K (100:100:100
kg ha–1) nutrient-omission trial with maize conducted on
18 farms in three localities in western Kenya during the
short rains season of 2002: Aludeka, Emuhaya, and Shinyalu
(Vanlauwe et al., 2006).
Deriving Functional Relationships Using Dynamic Crop Growth Models
Many of the current parameters and functions describing
resource use effi ciency within FIELD are directly derived from
experimental observations. To make the model more generic
and yet maintain a low degree of complexity in its formulation
and parameterization, functional relationships for key pro-
cesses in FIELD were developed using more detailed, dynamic
crop growth simulation models that have a shorter time step
of integration. For example, in an earlier study (Chikowo et
al., 2008), we used the crop growth model APSIM (Keating
et al., 2003) to generate relationships such as rainfall capture
effi ciency as a function of total seasonal rainfall and soil type.
In the present study, we used the crop growth model DYNBAL
(DYnamic Nutrient BALances) (Tittonell et al., 2006) to
generate functional relationships for FIELD, since it has been
calibrated and tested for maize under the conditions of western
Kenya. We parameterized DYNBAL using the soil data from
the nutrient-omission experiments (Dataset 4) and ran it with
daily radiation and rainfall data to simulate light-determined
and water-limited maize yields (i.e., with the N module of
DYNBAL switched off ). Daily rainfall was recorded during
the short rains of 2002 at each experimental location totaling
641 mm in Aludeka, 654 mm in Emuhaya, and 716 mm in
Shinyalu. Daily global radiation was measured at the Maseno
Experimental Station (western Kenya) and used to calculate
the total amount of PAR reaching the crop throughout the
short rains 2002 growing season (on average, 1200 MJ PAR
m-2 season–1). Examples of parameter values for FIELD that
were derived using DYNBAL are presented in Table 2. Figure 2
illustrates how the FIELD model parameters, seasonal fraction
of intercepted radiation (intercepted over incident PAR), and
effi ciency of rainfall capture (transpiration over rainfall) were
derived from daily-step simulations with DYNBAL.
Calibration of FIELD Th e soil organic matter module of FIELD was calibrated
against data from the chronosequence around Kakamega
Forest (Dataset 1), simulating changes in soil C under continu-
ous crop cultivation following forest clearance. Measurements
of soil bulk density made in the forest and in farmers’ fi elds
were used to adjust soil bulk density values with decreasing soil
organic C (bulk density = 1719.2 – 33.1 × SOC, r2 = 0.61).
Table 2. Examples of parameters used in FIELD that were derived running the dynamic model DYNBAL. Average values presented for Emuhaya.
Parameter UnitAverage value
(Emuhaya)Light-determined yield Incoming PAR† (season) MJ m–2 1208 Fraction of PAR intercepted – 0.58 PAR conversion effi ciency g MJ–1 3.43
Water-limited yield Cumulative rainfall (season) Mm 616 Rainfall capture effi ciency – 0.23 Rainfall conversion effi ciency kg ha–1 mm–1 134† PAR, photosynthetically active radiation.
Agronomy Journa l • Volume 100, Issue 5 • 2008 1515
Values for soil input parameters were as follows: 46% clay,
19% sand, 11.8 mg kg–1extractable (Olsen) P, 0.4 cmol(+) kg–1
exchangeable K; relative C losses by soil erosion were set at
0.01 yr–1. Th e model, run with average rainfall from historical
30-yr weather data (1635±218 mm yr–1; National Agricultural
Research Laboratory, 1994), simulated an exponential decrease
in soil C in the upper 20 cm from 140 to 27 t ha–1 over 100
yr, with an average net loss rate of 1.13 t C ha–1 yr–1 (0.8% per
year in relative terms) (Fig. 3A). Th e comparison of observed vs.
simulated soil organic C (0–20cm) produced a RMSE of 13.3
t ha–1, with r2 = 0.94 (P < 0.01). Simulated maize grain yields
decreased from 6.7 t ha–1 at the beginning of cultivation period
(1 yr aft er forest clearance) to 3.4 t ha–1 aft er 20 yr of cultiva-
tion, 2.4 t ha–1 aft er 40 yr and 1.4 t ha–1 aft er 100 yr. With the
same FIELD model parameters and input variables as above,
but assuming an annual manure application rate of 5 t dry mat-
ter ha–1 (Maseno FTC manure; Table 3), equilibrium soil C
was achieved aft er 60 yr of cultivation with a C content of 71
t ha–1 in the upper 20 cm (c. 30 g C kg–1 soil). Th is is slightly
greater than the soil C contents that are found in similar soils
of continuously manured home gardens in the region (Tittonell
et al., 2005).
Using the 1989–2003 seasonal rainfall records, we calibrated
FIELD against the long-term dataset on maize yields (two
crops per year, respectively, during the long and short rains)
and changes in soil C contents at Machang’a (Dataset 2), simu-
lating eff ects of annual application rates of 0, 5, and 10 t dry
matter manure ha–1. Initial soil (chromic Cambisol) properties
were set as follows: 31% clay, 13% silt, and 56% sand; 5.9 g kg–1
soil organic C (C to N ratio 12.7); and 0.6 mg kg–1 extractable
P. Mimicking the experiment, manure was applied at the start
of the long rains season before planting of the maize crop (i.e.,
only once a year despite the two crops per year). Th e quality
parameters of the applied manure are shown in Table 3. When
compared with observed values, FIELD satisfactorily predicted
crop aboveground biomass over the 26 growing seasons (overall
RMSE = 1.7 t ha–1; r2 = 0.51). By adjusting the annual humi-
fi cation coeffi cient for manure (HC = 0.27 yr–1), we were able
to fi t the model to the observed soil C values (Fig. 3B) with
RMSE of 0.8, 2.1, and 3.8 t C ha–1, respectively, for the treat-
ments receiving 0, 5, and 10 t manure ha–1 (overall r2 = 0.66).
Testing FIELD to Simulate Effects of Manure Application
We then tested the model against the data on maize
responses to manure application (0, 1.2, and 4 t C ha–1, cor-
responding to 3.4 and 11.4 t dry matter manure ha–1 yr) in
Aludeka and Nyabeda during the long and short rains of
2005 (data set 3). Soil properties at both sites are presented in
Table 3. All treatments received 60 kg P ha–1 and 60 kg K ha–1,
while only the +N treatments received 120 kg N ha–1. Average
nutrient contents of the manure used in this experiment are
Fig. 2. Simulations using the dynamic crop model DYNBAL. (A) Incident and intercepted radiation by maize; (B) cumulative crop transpiration vs. cumulative rainfall during the growing season at three locations in western Kenya.
Table 3. Parameters used in the model simulations. Soil properties at the experimental sites of the manure application experi-ment. Dry matter (DM), C, and nutrient content of manures from different sources.
Locality Clay Sand
Soil organic
CTotal soil N
Extractable P
Exchangeable bases pH
(water 1:2.5) Manure origin DM C N P KK Ca Mg
% g kg–1 mg kg–1 cmol(+) kg–1 %Aludeka 8 85 8.3 0.8 6.0 0.39 5.0 0.6 5.5 Machang’a experiment 80 26 2.0 0.48 na‡Nyabeda 58 29 15.4 1.4 2.4 1.01 4.9 1.8 4.9 Maseno FTC† 80 35 1.4 0.18 1.8
Experimental Dairy Farm 82 39 2.1 0.22 4.0Farm A 56 30 1.2 0.32 2.0Farm B 59 29 1.0 0.30 1.6Farm C 77 25 1.0 0.10 0.6Farm D 43 35 1.5 0.12 3.3Farm E 41 23 0.5 0.10 0.6
† Manure from the farm at Maseno Farmer Training Centre, Maseno, western Kenya.‡ na, not available.
1516 Agronomy Journa l • Volume 100, Issue 5 • 2008
presented in Table 3 (Maseno FTC), together with those of
manures sampled from the experimental dairy farm of Maseno
University and from fi ve farms in western Kenya (Castellanos-
Navarrete, 2007). By adjusting the value of the HC of manure,
we fi tted the model to the observed crop responses at both sites,
minimizing the RMSE (resulting in HC = 0.53 season–1, or
0.22 yr–1). Th is value was used as default for the other types of
manure in Table 3. Although we acknowledge that manures
Fig. 3. (A) Calibration of the model FIELD against soil C across a chronosequence of 100 yr of cultivation around Kakamega Forest Reserve, western Kenya; (B) Simulated and measured soil C increase after 13 yr (26 seasons) under 0, 5, and 10 t ha–1 manure ap-plications in a Cambisol at Machang’a, central Kenya; (C) Observed (x axis) and simulated aboveground biomass of maize in the long (LR) and short rains (SR) of 2005 with different rates of manure and mineral N in Aludeka (Alu) and Nyabeda (Nya), western Kenya; (D) Aboveground biomass production of maize with application of manure (0, 1.2 and 4 t C ha–1), with and without appli-cation of mineral N (120 kg ha–1), during the long and the short rains of 2005 at Aludeka – bars: measured values (plus standard deviation), asterisks: FIELD simulations; (E) Observed (x axis) and simulated aboveground biomass of maize in the case study fields (Table 4) with all combinations of N, P, and K in the nutrient-omission trial; (F) Measured biomass yield of all NPK treatments and simulated water-limited yields as a function of soil organic C.
Agronomy Journa l • Volume 100, Issue 5 • 2008 1517
of varying chemical composition will have diff erent HCs,
we lacked experimental data to derive a generic relationship
between the HC and manure quality. We thus assume that
diff erences in simulated crop responses for the various types of
manure are directly due to diff erences in their C and nutrient
contents. Maize responses to manure application were satis-
factorily simulated by FIELD, although with a slight tendency
to underestimate aboveground biomass yields without N and
overestimate the response when N was added (Fig. 3C). In
the long rains and at both locations, maize responded almost
linearly to manure applications without N, while responses to
N with and without manure were only observed in the sandier
soils of Aludeka, in both rainy seasons (Fig. 3D).
Testing FIELD to Simulate Crop Responses to Fertilizers on Heterogeneous Farms
Finally, we tested FIELD for simulating maize responses
to mineral fertilizers using the soil and yield data from the
on-farm fertilizer trials at Aludeka, Emuhaya, and Shinyalu
(data set 4). Th e model was parameterized for a combination of
three localities × six farms per locality × three positions within
the farm (home-, mid- and outfi elds) totaling 54 independent
observations. Th e soil C module of FIELD was initialized by
running 100-yr simulations (approximately the period since
land cultivation started on the oldest fi elds in the region) under
diff erent scenarios of manure inputs to represent the historical
management that led to current fertile and poor fi elds match-
ing their observed soil C contents (Tittonell et al., 2007c). Th e
model was then run to simulate the experimental treatments:
control without fertilizer, full N-P-K fertilization (100 kg N
ha–1, 100 kg P ha–1, 100 kg K ha–1) and three treatments with
one of the nutrients (N, P, or K) missing. Other crop and man-
agement parameters for the model (e.g., plant density, planting
dates, length of growing period, harvest index) were defi ned as
in the experiments.
Given the large variability in the data from the on-farm
experiment, the performance of FIELD to simulate maize
production was satisfactory (overall RMSE 2.8 t ha–1), as
illustrated in Fig. 3E for total aboveground biomass under all
fertilizer treatments in the case-study fi elds of Table 4. Th e
water-limited yield calculated by FIELD using the summary
functions derived with DYNBAL increased as a function of
increasing soil C, as did the maize yields measured in the plots
with full-NPK fertilization (Fig. 3F). However, a large number
of fi elds receiving full-NPK and having between 10 and 20 g
kg–1 soil organic C produced yields that were smaller (up to
40% less) than the simulated water-limited yield. Yields under
full-NPK are assumed to be close to water-limited yield levels,
unless other factors that limit or reduce crop growth are pres-
ent (e.g., micronutrient defi ciencies or Striga spp. infestations).
Th is gap between simulated water-limited and measured full-
NPK yields may further suggest that DYNBAL overestimated
water availability and therefore water-limited yields in soils
with greater C content, or that the application rates of N, P
and/or K in the experiment were suboptimal.
Scenario Analysis Once FIELD was parameterized and tested for the condi-
tions of the study area in Western Kenya, we used the model to
address the four research questions around ISFM posed earlier.
Th ree farms from three localities in western Kenya that were
included in Dataset 4 were used as case studies for scenario
analysis: Aludeka division in Teso district (0°35́ N, 34°19´ E),
Emuhaia division in Vihiga district (0°4́ N, 34°38´ E), and
Shinyalu division in Kakamega district (0° 12´ N, 34° 48´ E).
Th ese farms had been characterized earlier and visited on sev-
eral occasions, and exhibited marked variability in soil quality,
maize productivity, and responses to mineral fertilizers on their
home- to their outfi elds (Tittonell et al., 2005). Soil properties
Tab
le 4
. Sce
nari
o an
alys
is. U
pper
por
tion
incl
udes
soi
l pro
pert
ies,
cro
p m
anag
emen
t pa
ram
eter
s, a
nd m
aize
abo
vegr
ound
bio
mas
s yi
eld
unde
r fa
rmer
man
agem
ent
(fi r
st s
easo
n, lo
ng r
ains
) an
d in
the
exp
erim
ent
(sec
ond
seas
on, s
hort
rai
ns)
mea
sure
d in
thr
ee fa
rms
acro
ss s
ites
. Low
er p
orti
on
show
s m
ulti
plie
rs u
sed
to s
imul
ate
rain
fall
vari
abili
ty (
deri
ved
from
mea
sure
d da
ta)
thro
ugho
ut t
he 1
2 yr
of t
he s
imul
atio
n pe
riod
.
Farm
er (
site
) an
d po
siti
on w
ithi
n fa
rm
Soil
prop
erti
esR
esul
ts u
nder
farm
er m
anag
emen
t†M
aize
bio
mas
s in
ex
peri
men
tC
lay
+ Si
ltO
rgan
ic
CE
xtra
ctab
le P
Exc
hang
. KB
ulk
dens
ity
Fiel
d sl
ope
Bio
mas
s yi
eld
Har
vest
in
dex
Res
idue
re
mov
edR
esid
ue
biom
ass
Use
of
com
post
Con
trol
yi
eld
NP
yi
eld
NP
K
yiel
d%
g kg
–1m
g kg
–1cm
ol(+
) kg–
1kg
m–3
%t
ha–1
%–t
ha–
1 ––t
ha–
1 –J.
Obo
njo
(Alu
deka
)H
omefi
eld
4412
.213
.40.
3913
402.
06.
10.
3720
3.1
014
.710
.713
.9M
idfi e
ld44
7.6
2.9
0.47
1470
2.5
4.0
0.20
03.
20
4.8
11.7
11.9
Outfi e
ld47
7.0
1.8
0.30
1440
1.5
2.5
0.28
02.
50
3.7
11.4
10.5
D. N
akay
a (E
muh
aya)
Hom
efi e
ld50
20.8
15.3
1.96
1150
7.5
8.9
0.43
05.
12.
317
.815
.616
.4M
idfi e
ld54
14.4
3.6
0.63
1340
9.0
8.3
0.43
601.
90.
57.
316
.616
.3O
utfi e
ld56
12.6
1.9
0.28
1400
3.0
6.9
0.43
100
0.0
0.0
9.0
10.6
14.8
S. Sh
ivon
je (
Shin
yalu
)H
omefi
eld
7624
.013
.60.
4911
909.
55.
50.
4020
2.6
1.8
12.2
17.4
13.2
Midfi e
ld72
17.3
2.5
0.08
1200
29.5
4.2
0.32
800.
81.
08.
213
.510
.9O
utfi e
ld72
16.1
2.1
0.10
1070
11.0
3.0
0.26
201.
80.
03.
711
.610
.6
Year
12
34
56
78
910
1112
Rai
nfal
l var
iabi
lity
0.66
1.20
0.76
1.23
1.02
1.04
0.61
1.25
1.31
0.83
0.85
1.23
† Re
sults
der
ived
from
par
ticip
ator
y re
sour
ce fl
ow m
appi
ng a
nd o
n-fa
rm y
ield
mea
sure
men
ts; t
he a
mou
nts
of c
rop
resi
due
and
com
post
inco
rpor
ated
wer
e ca
lcul
ated
usi
ng fa
rmer
s’ ow
n es
timat
ions
.
1518 Agronomy Journa l • Volume 100, Issue 5 • 2008
and maize yields under farmer management and under con-
trolled experimental conditions are presented in Table 4. In total,
we simulated maize responses on nine fi elds, representing three
case-study farms with three fi eld types each. However, for clarity
we oft en plotted in graphs subsets of fi elds, those that showed
typical patterns of responsiveness to management interventions.
In the model simulations, we used manure of diff erent qual-
ity, from high-quality manure such as that from the experi-
mental dairy farm of Maseno University to low-quality manure
such as that on farm E (Table 3). For simplicity, and to repre-
sent common practices in the area, we assumed that through
proper manure management 1.8 t dry matter manure was avail-
able for application on 1 ha of cropland per season (Table 1).
Since concentrating the available manure on small portions
of land is also a common practice in the area (Tittonell et al.,
2005), application rates of 5 t dry matter ha–1 to restore soil
productivity were also simulated. Th e minimum mineral fertil-
izer application rates were set based on the assumption that a
farmer was able to buy a 50-kg bag of diammonium phosphate
(DAP) (18:46:0) and a 50-kg bag of urea (46:0:0) to apply on
1 ha of maize (equivalent to 32 kg N ha–1 and 23 kg P ha–1).
An application of the (recommended) 60 kg N ha–1 and 30
kg P ha–1 was defi ned as basal fertilizer. Application of 140
kg N ha–1 and 40 kg P ha–1 was defi ned as replacement fertil-izer, as this provides roughly the same amounts of N and P as
a combined application of basal fertilizer + 5 t ha–1 of manure
of average quality. Th e model was run for 12 yr (or 24 seasons)
using variable seasonal rainfall. To allow for comparisons
across localities, we assumed a
similar rainfall variability pat-
tern over the seasons, which was
calculated from historical rainfall
records in the region. Coeffi cients
of variability were calculated and
multiplied by the average rainfall
at each locality to generate 12 yr of
variable rainfall but with a similar
pattern across localities (Table 4).
Th e following sets of treatments
were applied in the scenario simu-
lations with FIELD: (i) applica-
tion of, respectively, 0, 30, 60, 90,
120, 150, and 180 kg N ha–1 with
and without 30 kg P ha–1 for a
single season to all the fi elds in
Table 4; (ii) application of basal
and replacement fertilizer rates,
good-quality manure (5 t dm
ha–1) and combined basal fertil-
izer + manure, for 12 consecutive
years to all fi elds in Table 4, with
diff erent proportions of crop har-
vest residues retained on the fi eld;
(iii) application of manure (1.8
t dm ha–1) of diff erent qualities
(Table 3) with and without appli-
cation of a minimum fertilizer
rate (32 kg N ha–1 + 23 kg P ha–1)
for 12 consecutive years to all
fi elds in Table 4; (iv) no nutrient
inputs during 12 consecutive years, followed by a 12-yr rehabil-
itation treatment applying manure (1.8 t dm ha–1) of diff erent
qualities (Table 3) with and without application of a minimum
fertilizer rate (32 kg N ha–1 + 23 kg P ha–1).
Th e results of the simulated treatments under (iv) were used
to calculate the hysteresis of soil restoration in total above-
ground biomass yield units and the number of years necessary
for restoring the initial crop productivity of a certain fi eld (i.e.,
the productivity at t1 = 0 is the beginning of the 12-yr simula-
tion without inputs).
RESULTSMaize Response to Mineral Fertilizers
Simulations using FIELD indicated diff erent responses of
maize grain yield to increasing application rates of N fertil-
izers across the three case-study farms, and even wider diff er-
ences across the various fi elds of each individual farm (Fig. 4).
Considering the treatments that received only N (Fig. 4A, C, E),
three patterns of responsiveness can be observed: poorly
responsive infertile fi elds (e.g., outfi elds at Aludeka), responsive
fi elds (e.g., midfi elds at Shinyalu, homefi elds at Aludeka), and
poorly responsive fertile fi elds (e.g., homefi elds at Emuhaya).
Crops in most fi elds responded to application of 30 kg P
ha–1 alone or in combination with N (Fig. 4B, D, F). In most
cases and particularly in the homefi elds at the three loca-
tions, the sole addition of P led to a doubling of maize grain
yields. Adding P to the homefi elds caused a saturation of the
Fig. 4. Simulated maize grain yields with increasing application of N (0 to 180 kg ha–1), with and without application of P (–P = 0 and +P = 30 kg P ha–1), as mineral fertilizers in home-fields, midfields, and outfields of three case-study farms (Table 4) in Aludeka (A, B), Emuhaya (C, D), and Shinyalu (E, F), western Kenya.
Agronomy Journa l • Volume 100, Issue 5 • 2008 1519
simulated response curve with N application rates of 60 kg N
ha–1 at Aludeka, 0 kg N ha–1 at Emuhaya, and 120 kg N ha–1
at Shinyalu. Yields attained with N + P in the homefi elds of
Emuhaya are close to the potential yields as observed under
on-station experimental conditions in the area (Tittonell et al.,
2007b). Th e addition of P induced almost linear yield responses
to N from 0 to 180 kg ha–1 in the outfi elds at the three loca-
tions. Th e recommended fertilizer rate of 60 kg N ha–1 and
30 kg P ha–1 led to widely varying results across locations and
fi elds, ranging between 1.1 and 5.9, 2.6 and 7.6, and 2.7 and
4.5 t grain yield ha–1 in Aludeka, Emuhaya, and Shinyalu,
respectively. Th e simulated response to N applied at rates >100
kg ha–1 in the presence of P indicates that, indeed, the N fertil-
izer rate applied in the on-farm experiment was suboptimal.
Combined Application of Manure and Mineral Fertilizers
Application of 5 t dry matter ha–1 of good-quality manure
(Experimental Dairy Farm, Table 3) led to substantially
increased maize productivity in the mid to long term in four
fi elds with diff erent initial patterns of responsiveness to fertil-
izers (Fig. 5A, D, J, M). Simulated crop productivity was larger
during the fi rst three to four seasons with application of mineral
fertilizer at the basal rate (60 kg N ha–1 and 30 kg P ha–1) than
with application of 5 t dry matter ha–1 of manure (of the best
quality found in the region). In subsequent seasons, maize yields
were greater with manure applications in the fi elds that were
initially poorer (Fig. 5D, J, M), and did not diff er from yields
obtained with basal fertilizer in the homefi eld of Emuhaya
(a poorly responsive, fertile fi eld, Fig. 5A). Positive interac-
tions between combined basal fertilizer and manure were only
observed during the fi rst season in the responsive fi elds (Shinyalu
midfi eld and Aludeka homefi eld), while virtually the same
performance as basal fertilizer was observed in the nonrespon-
sive fi elds. However, the combination of mineral fertilizer and
manure led to the highest long-term crop productivity in the
degraded outfi elds of Aludeka, three times larger than crop pro-
ductivity with basal fertilizer alone. Replacement fertilizer, i.e.,
application of the same amounts of N and P as in manure + basal
fertilizer, led to similar productivity levels in the responsive and
fertile fi elds in wetter seasons, but less in drier seasons or in all of
the seasons in the poorly responsive outfi eld in Aludeka.
Fig. 5. Simulation of maize production under various nutrient management strategies during 12 yr (24 seasons) in fields with dif-ferent patterns of responsiveness. A, B, C: Emuhaya homefield (nonresponsive fertile field); D, E, F: Shinyalu midfield (responsive field); J, K, L: Aludeka homefield (responsive field); M, N, O: Aludeka outfield (nonresponsive poor field). Left panes: aboveground biomass against time; central panes: cumulative aboveground biomass against cumulative rainfall; right panes: soil organic carbon against cumulative crop C inputs to the soil (roots + stover).
1520 Agronomy Journa l • Volume 100, Issue 5 • 2008
Larger long-term maize productivity as a consequence of
improved nutrient management is refl ected in higher rain-
fall productivities (Fig. 5B, E, K, N), which in the case of
the homefi eld in Emuhaya (Fig. 5B) reached values of about
15 kg aboveground biomass ha–1 mm–1 of seasonal rainfall.
Calculations for western Kenya using the dynamic crop growth
model DYNBAL indicated maximum attainable water-limited
yields in the order of 20 kg aboveground biomass ha–1 mm–1
of seasonal rainfall (Tittonell et al., 2006). Simulated rainfall
productivity attained under the control treatment without
inputs (as under farmers’ management) ranged between 1 and
5 kg biomass ha–1 mm–1. Th us, a maize aboveground biomass
production of about 15 t dm ha–1, as simulated for the wetter
seasons in Fig. 5A, D and J, represents a ceiling productivity
level that is, however, hardly achieved in reality (e.g., Kipsat et
al., 2004). Th e various simulated treatments varied in the rate
at which they build up soil organic C, assuming that all crop
residues were retained on the fi elds, basically due to their large
diff erences in crop productivity and associated C inputs to the
soil (Fig. 5C, F, L, O). In the nonresponsive fertile homefi eld at
Emuhaya, both rates of mineral fertilizer without manure con-
tributed almost the same amounts of crop residue C (Fig. 5C).
In the other fi elds, application of fertilizer at replacement rates
led to more C input to the soil compared with the basal fertil-
izer treatment, and in the Aludeka outfi eld even to more C
input than in the manure application treatment (Fig. 5O).
Since farmers have many diff erent uses for crop residues,
including livestock feeding and bedding, fencing, or using
them as fuel, they normally remove a large part of the residues
from the fi elds aft er harvest (also to facilitate tillage activi-
ties in these double-cropping systems). Our simulation results
indicate that the initial soil C contents can practically be
maintained on the fertile fi elds with basal fertilizer rates if 70%
of the crop residue is retained in the fi eld (assuming alterna-
tive uses for the remaining 30%), except on the poor fi elds
(Fig. 6A). In the latter, replacement fertilizer rates increased
soil C by 2.3 t ha–1 aft er 12 yr (24 growing seasons) with
respect to the initial value at t1 = 0. It must, however, be noted
that maintaining the initial soil C contents of poor fi elds is
insuffi cient. Soil C needs to be increased in such fi elds and
this was only achieved with manure application every season
(twice a year) in our simulations. Th e largest diff erences in
soil C buildup amongst the various simulated treatments were
observed in the Shinyalu midfi eld, which is characterized by
the steepest slopes and clayey soils (Table 4). If farmers remove
most (90%) of the crop residues, as they commonly do, soil C
is only built up by manure and root-C inputs (Fig. 6B). In such
case, the use of fertilizers is insuffi cient to build soil organic
matter, and the contribution of roots is minimal since a slower
soil organic matter buildup also leads to less crop productivity.
The Attractiveness of Soil-Improving TechnologiesOft en the implementation of ISFM technologies represents
a trade-off between the immediate concern of increasing yields
and the long-term sustainability of the system. Th e combined
application of manure and fertilizers may be attractive for
responsive fi elds, as this may induce positive interactions in
responsive fi elds (Fig. 5D, J) in the short term and maintain
soil C in the long term. However, that may not be the case
for the poor outfi elds during the fi rst seasons (Fig. 5M), and
especially not when more realistic manure application rates
and average manure qualities are considered. For example, in
the outfi eld at Aludeka, where soil C buildup is deemed neces-
sary, seasonal application of 1.8 t dry matter ha–1 of manure
of the various qualities sampled in western Kenya (Table 3) led
to diverse simulated long-term outcomes in terms of restoring
productivity and soil organic C (Fig. 7A, B). With the sort of
manure qualities as sampled from case-study farms in western
Kenya, soil C can only be maintained, at most, with this rate of
manure application.
Farmers’ decisions on technology adoption are oft en
conditioned by attractive short-term crop yield responses.
Zooming-in on the fi rst 4 yr, Fig. 7C, D shows simulated maize
grain yields on the outfi eld at Aludeka with repeated manure
applications, with and without application of a minimum fer-
tilizer rate (32 kg N ha–1 and 23 kg P ha–1). Th e crop residue
was retained in the fi eld. Beyond the variability induced by sea-
sonal rainfall, yields in the second year (and increasingly there-
aft er) were substantially larger with all manure types when
mineral fertilizer was applied, achieving larger grain yields
than aft er 4 yr without fertilizers. However, the response to
fertilizer without manure (control) was poor in the fi rst seasons
(Fig. 5M). Th ese simulation results suggest that without small
amounts of mineral fertilizers to boost crop productivity in the
second year, soil C contents could not be improved with any of
the manure qualities sampled in western Kenya farms (Farms A
to E) applied at the (quite realistic) rate of 1.8 t dm ha–1.
Fig. 6. Simulated changes in soil organic C after 12 yr of maize cultivation under different management strategies with retention of 70% (A) or 10% (B) of crop residues in the field after harvest, in fields with different responsiveness: nonresponsive fertile field (Emuhaya homefield), responsive fields (Shinyalu midfield and Aludeka homefield), and nonresponsive infertile field (Aludeka outfield).
Agronomy Journa l • Volume 100, Issue 5 • 2008 1521
Hysteresis of Productivity Restoration
By analogy to the phenomenon of hysteresis in dynamic sys-
tems, we defi ned the hysteresis of restoration as the capacity of
the system to react to ISFM interventions aimed at rehabilitating
soils, restoring their productivity. Figure 8 shows FIELD simula-
tions of crop productivity during 24 yr: 12 initial years without
inputs and 12 subsequent years with application of manure, min-
eral fertilizer, or manure + mineral fertilizer (at rates of 32 kg N
ha–1 and 23 kg P ha–1 and 1.8 t dm ha–1 of good quality manure)
for a nonresponsive fertile fi eld (Emuhaya homefi eld), a respon-
sive fi eld (Aludeka homefi eld), and a nonresponsive infertile fi eld
(Aludeka outfi eld). For simplicity, average constant instead of
variable rainfall was used in these simulations. In Fig. 8A, C, E,
the rehabilitation phase (r) has been plotted reversing the time x axis, to illustrate the magnitude of the hysteresis (h). Figures 8B,
D, F show the number of years (t) necessary to achieve the initial
crop production levels with the respective interventions and the
net productivity gains (g) that may be achieved. Th e rate of resto-
ration was faster with mineral fertilizers (Fig. 8C, D) than with
manure (Fig. 8A, B), at the simulated application rates, and much
faster with combined manure and fertilizers (Fig. 8E, F) (note
the diff erences in the scale of the y axes). Taking the initial crop
productivity as the threshold, however, is not always appropriate.
In the case of the poor outfi eld of Aludeka, the low initial pro-
ductivity is achieved aft er 3 yr of manure application or aft er 1 yr
of fertilizer application. On the contrary, the initial high produc-
tivity of the fertile homefi eld in Emuhaya is not achieved aft er
12 yr of manure application. Th erefore, a desirable or achievable
threshold yield (Tittonell et al., 2007c) should be defi ned and
used in the calculations.
In general, the hysteresis of restoration will depend on the type
of technology implemented to restore soil productivity (mineral
and/or organic fertilizers, rotations with legume crops, soil erosion
control measures, improved crop germplasm, etc.), on the inherent
properties and initial conditions of the soil, and on complementary
management measures such as retaining crop residues in the fi eld
or water harvesting measures in drier areas etc. Table 5 presents
the results of the calculations of the hysteresis of restoration for
the three fi elds (Fig. 8) with diff erent responsiveness aft er 12 yr
of cropping without inputs, using the various manure qualities
in Table 3 applied at 1.8 t dm ha–1, with and without minimum
fertilizer rates (32 kg N ha–1 and 23 kg P ha–1), and retaining crop
residues in the fi eld. Th e degree of hysteresis measured in crop
biomass units varied strongly for the various types of manure, with
little reaction of the three systems to the application of poor qual-
ity manures without fertilizer, and greater reactions to mineral
fertilizers than to any type of manure.
Th e simulated eff ects of soil properties on the hysteresis of
restoration are illustrated in Fig. 9A–9C, depicting the results
of FIELD simulations of 12 yr of degradation followed by 12 yr
Fig. 7. Rehabilitation of nonresponsive fields (outfield at Aludeka) with application of 1.8 t dm ha–1 manure of different qualities (Table 3). Simulated aboveground maize biomass (A) and soil organic carbon (B) during a 12-yr period. Zooming-in on the first 4 yr of the simulation, grain yield increase with application of different manure types (C) and with manure plus a minimum fertilizer rate (32 kg N ha–1 and 23 kg P ha–1) (D).
1522 Agronomy Journa l • Volume 100, Issue 5 • 2008
of rehabilitation for all the fi elds in the on-farm experiment
(Dataset 4; n = 54) with application of high-quality manure
(1.8 t dm ha–1). For a wide range of initial soil C contents, the
hysteresis of the system remained below 2 t ha–1 (Fig. 9A);
the few cases above that threshold correspond to fi elds where
available (Olsen) P was larger. Th e topographic slope of the
fi elds aff ects water and nutrient capture effi ciencies, and soils
in the study area are generally of less quality than on fl at
landscape positions (Tittonell et al., 2005). While fi elds with
slopes between 0 and 10% could experience either low or high
hysteresis, fi elds on abrupt slopes showed consistently poor
capacity of reaction to rehabilitation with manure applications.
Combination of manure and minimum fertilizer rates led to
positive interactions in most fi elds (particularly in those with
less soil C), as illustrated for Aludeka in Fig. 9D: the simulated
hysteresis of rehabilitation of fi elds with soil C < 10 g kg–1 with
manure + fertilizer combined was larger than the sum of the
hysteresis with sole manure and sole fertilizer.
Fig. 8. Hysteresis of soil restoration. (A, C, E) Simulated biomass yields during the degradation (d) and rehabilitation (r) phases; (B, D, F) Absolute difference with respect to the initial yield (at t1) over the years of rehabilitation (t2), indicating the time needed to achieve initial yield levels (t) and the net productivity gain (g), for three fields in western Kenya (HF: homefield, OF: outfield). In A, C, E, the rehabilitation phase was plotted inverting the direction of the time axis to indicate the magnitude of the hysteresis (h). Rehabilitation treatments included application of manure (A, B), N-P mineral fertilizers (C, D), and combined manure + fertilizers (E, F).
Agronomy Journa l • Volume 100, Issue 5 • 2008 1523
DISCUSSIONTh e application of a given rate of mineral fertilizer produced
widely variable yield responses of maize across the various fi elds
of individual farms (Fig. 4), confi rming experimental results
of other studies across sub-Saharan Africa (e.g., Carter and
Murwira, 1995; Vanlauwe et al., 2006; Wopereis et al., 2006).
Manure application in farmers’ fi elds results oft en in poorer
crop responses than those measured in controlled experiments
(e.g., Misiko, 2007), basically due to the wide diff erences in
manure qualities across diff erent farms (Table 3, Fig. 7). Th e
combination of small amounts of mineral fertilizer and realis-
tic application rates of animal manure of farmers’ average qual-
ity looks most promising as an ISFM strategy, as indicated by
the simulations with FIELD (Fig. 7, Table 5). Th e simulation
results suggest that when manures are poor in nutrients, the
presence of fertilizer is essential to increase soil organic matter.
Even with the poorest-quality manure (Farm E in Table 3) in
combination with minimum amounts of mineral fertilizers,
Table 5. Hysteresis of rehabilitation brought about by appli-cation of 1.8 t ha–1 of manure of different qualities with and without addition of mineral fertilizer to fi elds of different ini-tial fertility (responsiveness).
Field†No
manure
Manure quality typeExp. dairy farm
Farm A
Farm B
Farm C
Farm D
Farm E
No fertilizer t dm ha–1
Emuhaya HF – 2.46 1.20 1.13 0.72 0.51 0.17 Aludeka HF – 1.98 0.68 0.63 0.44 0.33 0.06 Aludeka OF – 0.94 0.32 0.29 0.15 0.12 0.0232 kg N ha–1 + 23 kg P ha–1
Emuhaya HF 3.73 12.26 7.51 7.18 5.97 5.79 4.52 Aludeka HF 2.14 8.89 4.35 4.18 3.65 3.70 2.75 Aludeka OF 1.15 4.62 2.47 2.38 2.04 2.11 1.50† HF, homefi eld; OF, outfi eld.
Fig. 9. Hysteresis of soil restoration (i.e., the value of h, Fig. 8) with repeated applications of animal manure calculated for all the fields in the nutrient-omission experiment and plotted against (A) their initial soil organic carbon, (B) extractable P contents, and (C) their topographic slope. (D) Hysteresis of soil restoration with application of animal manure, mineral fertilizer, and manure + mineral fertilizer shown only for the fields at Aludeka.
1524 Agronomy Journa l • Volume 100, Issue 5 • 2008
some responses in crop yield are expected that could make
the technology attractive to farmers. Th e limited amounts of
manure available to farmers should be better targeted so that
nonresponsive infertile fi elds are rehabilitated into responsive
fi elds in the mid to long term. Nonresponsive fertile fi elds (e.g.,
homefi eld in Emuhaya; Fig. 4) may be managed with mini-
mum, maintenance fertilization rates (mainly with mineral P)
to sustain their current productivity.
Within the boundaries of its agroecological requirements,
maize is a well-suited crop to build up soil organic matter
through crop residue inputs due to its large potential for biomass
production and responsiveness to applied nutrients (Fig. 5).
However, the competing uses that farmers have for maize sto-
ver in such integrated crop–livestock systems oft en prevent its
use as soil amendment (Waithaka et al., 2006). If crop residues
are removed from the fi elds aft er harvest, their C and nutrient
inputs must be replaced with other organic amendments, such
as animal manure and green manures or through transfer of
plant biomass from outside the fi eld. In contrast to mineral
fertilizer use, continuous application of organic manures, even
if in small amounts, would in principle: (i) allow building up
more balanced soil nutrient stocks and a larger capacity of the
soil to retain nutrients (and water) by increasing soil organic
matter in the long term (Woomer et al., 1994); (ii) help to miti-
gate other potential soil fertility problems, such as micronutri-
ents defi ciencies, soil acidity, or soil physical impediments (e.g.,
Zingore et al., 2008). Fortifying mineral fertilizers by the addi-
tion of more nutrients in their composition can partly solve this
problem, although C inputs to the soil are not guaranteed.
However, the use of animal manure as soil amendment in
western Kenya (and much of sub-Saharan Africa) is strongly
conditioned by the lack of suffi cient quantities at farm scale
(Table 1). Castellanos-Navarrete (2007) measured effi cien-
cies of N cycling in crop–livestock systems of western Kenya of
around 30% on average. Th is implies that for each 100 kg N fed
to livestock (e.g., in 10 t of maize stover) only 30 kg N is avail-
able for application to crops (e.g., in 2.5 t of manure), of which
probably half becomes available to the crop in the fi rst season.
Considerable N (and C) inputs to the soil could still be achieved
if the available amount of manure is concentrated on a small
portion of the farm (e.g., 0.25 ha). However, given the current
crop productivity of western Kenya of 1 t grain ha–1 on average,
an equivalent of about 10 ha would be necessary to produce the
10 t of maize stover needed to feed 100 kg N to livestock. Th is
implies that nutrients must be brought into the farming system,
either as mineral fertilizers or as feedstuff for livestock.
Although some of the ISFM options explored with FIELD
showed a strong hysteretic restoration of soil productivity
(Fig. 8), the buildup of soil fertility (e.g., organic matter stocks)
may be much slower. Th e average soil C accumulation simulated
by FIELD across fi elds and for all fertilizer, manure, and crop
residue retention treatments was 0.37 t C ha–1 yr–1 (Fig. 6). Th e
maximum C capture effi ciency in the soil for the 12-yr period
simulated was 0.18 (i.e., increase in soil C/total C input),
whereas average C losses attributable to heterotrophic respira-
tion and soil erosion were around 4.6 t C ha–1 yr–1. On a Nitisol
in central Kenya, Kapkiyai et al. (1999) measured diff erences
in total soil organic C in the order of 6 t C ha–1 in the upper 15
cm of the soil aft er 18 yr between control plots without fertilizer
or manure and plots that each year had received fertilizers (120
kg N ha–1, 52 kg P ha–1) and manure (10 t dry matter ha–1 yr,
20.5% C). In that experiment, which was conducted under con-
trolled, on-station conditions, average maize grain yields were
1.5 and 5 t ha–1 yr–1 for the control and fertilizer + manure
treatments, respectively, and crop residues were removed from
the control plots, while not from the fertilized plots.
In rehabilitating nonresponsive infertile fi elds it may be of
practical use to identify threshold values for soil organic matter
that indicate a positive shift into responsive fi elds. For example,
studies with organic matter amendments on sandy soils in
Zimbabwe (Mtambanengwe and Mapfumo, 2005) pointed
toward the existence of a minimum soil C threshold of around
5 g C kg–1 soil for substantial responses to mineral fertilizers by
maize. In our study, however, soil C explains only part of the crop
response to fertilizers; soil P availability seems even to play a more
important role (see also Tittonell et al., 2007b). Soil P availabil-
ity determines not only the short-term crop response to applied
nutrients but also the capacity of the system to react to restora-
tion measures in the longer term (i.e., available P had a tighter
relationship with the hysteresis of restoration than soil C; Fig. 9).
Most soils under cultivation in western Kenya are extremely defi -
cient in available P (<2 mg kg–1) (Tittonell et al., 2007a).
Th e concept of hysteresis of soil restoration provides an inte-
grative measure of the capacity of reaction and response of the
system to restorative ISFM interventions in the long term–as
much as the response of crops to applied nutrients does in the
short term–refl ecting both the eff ect of system properties (e.g.,
soil condition, rainfall variability, type of crops) and the per-
formance of diff erent rehabilitation technologies. In our case,
the simulated reaction of degraded soils to the application of
mineral fertilizers (Fig. 8C, D) indicated almost immediate
responses in the fi rst year. Th is might, however, overestimate
the actual capacity of reaction of the system. In reality, it may
take longer to restore soil productivity when degraded soils
exhibit other limitations such as physical degradation or acid-
ity that were not simulated by FIELD. Th e calculated values of
hysteresis are only relevant within the system (or set of systems)
under study, and extrapolations outside these boundaries are
of little value. Here, the hysteresis of restoration was measured
in crop productivity units, but it could also be expressed in soil
C units, annual crop C inputs to the soil, value of production
(at constant prices), etc. If calculated with comparable meth-
ods and with standard assumptions, the concept of hysteresis
of restoration could be used in scenario analysis across farm-
ing systems within diff erent biophysical and socioeconomic
environments; for example, comparing the impact of certain
interventions across regions diff ering in agroecology or under
varying market situations.
A disadvantage in the implementation of this concept is
the need for long-term data, either to calculate the hysteresis
of restoration directly from measured changes in the relevant
indicators, or to calibrate simulation models to calculate
changes in the long term. Availability (and accessibility) of data
from long-term experiments to calibrate models constitutes a
bottleneck for studies in sub-Saharan Africa using modeling
scenario analysis. In the present study, for example, reliable
data were lacking for calibrating the model to simulate the dif-
ferences in nutrient release and soil organic C buildup between
Agronomy Journa l • Volume 100, Issue 5 • 2008 1525
animal manures of variable quality. In this study, we simply
used the same HC for all manures. However, manure composi-
tion, as well as soil properties can have a signifi cant eff ect on
the HC and, thus, on the rate of soil C buildup. In light of such
shortcomings, we and others (e.g., Smaling et al., 1997; Andrén
et al., 2004) argue that simulation models for scenario explo-
ration in data-scarce environments should be kept simple. By
taking a seasonal time step as in FIELD, processes can be sum-
marized into functional relationships that capture key aspects
of the dynamics of cropping and farming systems relevant to
the research questions raised.
SUMMARY AND CONCLUSIONS Th e exploration of ISFM options based on combined organic
and mineral fertilizer applications across heterogeneous farms
of western Kenya using the crop-soil model FIELD highlighted
the following facts: (i) mineral N and P fertilizers induce
widely variable yield responses of maize across the various fi elds
of individual farms, questioning the validity of the current
blanket fertilizer recommendations for maize (i.e., based on
agroecological zones or coarse soil maps); (ii) in most of the
fi elds evaluated, P limitation of maize yields was more critical
than N; (iii) locally available animal manure applied at aff ord-
able rates for smallholder farmers in the study area have a weak
eff ect on restoring the productivity of degraded soils, which
may discourage farmers from investing eff orts in soil rehabilita-
tion; (iv) application of poor-quality manure in combination
with small amounts of mineral fertilizer may generate more
attractive responses in the short term and a more balanced
buildup of soil C and nutrient stocks in the long term; (v) soils
that underwent severe degradation, or soils that are inherently
infertile, exhibit low hysteresis of restoration, and require major
long-term investments to restore their productivity.
In rehabilitating degraded fi elds, small amounts of mineral
fertilizers can be used to kick-start soil restoration, to jump to a
higher crop productivity that will generate favorable feedbacks
within the crop-soil system. In this sense, mineral fertilizers
are a clear option for soil fertility management by smallholder
farmers in areas of high population densities such as western
Kenya, characterized by small farm sizes that prevent the
practice of fallow or growing green manures, lack of nutrient
infl ows from communal grazing lands via animal manure, and
generalized soil degradation. Th e Africa Fertilizer Summit of
2006 in Abuja, Nigeria, set the goal of raising the average fertil-
izer use in sub-Saharan Africa from its current 10 kg ha–1 yr–1
to 50 kg ha–1 yr–1. Measures to promote fertilizer use among
farmers (e.g., reducing transaction costs and improving their
accessibility in rural areas) should go hand-in-hand with strate-
gies to improve the effi ciency of use of the applied nutrients,
taking into account the impact that farm heterogeneity may
have on crop response to fertilizers.
However, greater crop productivity induced by the use of
mineral fertilizers does not translate into better soil fertility
in the long term when large amounts of C and nutrients are
removed every season from the fi elds with the crop harvest
residue. In this sense, and under current circumstances, the
speculation on the capacity of smallholder farmers in Africa
to commercialize their crop residue as raw materials for bio-
fuels would have serious consequences for the sustainability
of these systems (see: www.africa-ata.org/aatf for the call by
the Director General of the UN Industrial Development
Organization to make Africa a world leader in biofuel produc-
tion). Although animal manure remains an option to manage
soil fertility in mixed smallholder crop–livestock systems,
its availability and quality are oft en poor compared with the
application rates and nutrient concentration of manures (nor-
mally obtained from commercial farms) used to evaluate ISFM
options in most fi eld experiments.
Research on and design of truly integrated soil fertility
management strategies in the context of African smallholder
farming systems should embrace these key features: strong
management-induced soil heterogeneity, limited availability of
poor quality manure, competing uses for crop residues within
the farm, lack of labor, and limited access to mineral fertilizers.
While strategies such as point-placing and/or microdosing
of mineral fertilizers, maintenance fertilization of the fertile
home gardens, or concentration of the available manure on
degraded fi elds may serve to increase nutrient use effi ciency
at plot scale, research is also needed to help redesign current
farming systems (e.g., growing alternative sources of fodder to
reduce the need of using crop residues), aiming at their sustain-
able intensifi cation.
ACKNOWLEDGMENTSWe thank the European Union for funding this research through the
AfricaNUANCES Project (Contract No. INCO-CT-2004-003729),
the Rockefeller Foundation for providing financial support in the
framework of the project on Valuing Within-Farm Soil Fertility
Gradients to Enhance Agricultural Production and Environmental
Service Functions in Smallholder Farms in East Africa (2003 FS036),
and Alain Albrecht and Regis Chikowo for their critical discussion of
initial results.
REFERENCESAndrén, O., T. Kätterer, and T. Karlsson. 2004. ICBM regional model for
estimations of dynamics of agricultural soil carbon pools. Nutr. Cycling Agroecosyst. 70:231–239.
Bationo, A., J. Kihara, B. Vanlauwe, B. Waswa, and J. Kimetu. 2006. Soil organic carbon dynamics, functions and management in West African agro-ecosystems. Agric. Syst. 97:13–25.
Carter, S., and H. Murwira. 1995. Spatial variability in soil fertility manage-ment and crop response in Mutoko Communal Area, Zimbabwe. Ambio 24:77–84.
Castellanos-Navarrete, A. 2007. Cattle feeding strategies and manure management in Western Kenya. MSc thesis. Wageningen Univ., Th e Netherlands.
Chikowo, R., M. Corbeels, P. Tittonell, B. Vanlauwe, A. Whitbread, and K.E. Giller. 2008. Generating functional relationships for inclusion in a farm-scale summary model for smallholder African cropping systems: Use of the dynamic simulation model APSIM. Agric. Syst. 97:151–166
Janssen, B.H., F.C.T. Guiking, D. van der Eijk, E.M.A. Smaling, J. Wolf, and H. Reuler. 1990. A system for quantitative evaluation of the fertility of tropical soils. Geoderma 46:299–318.
Kapkiyai, J.J., N.K. Karanja, J.N. Qureshi, P.C. Smithson, and P.L. Woomer. 1999. Soil organic matter and nutrient dynamics in a Kenyan nitisol under long-term fertilizer and organic input management. Soil Biol. Biochem. 31: 1773–1782.
Keating, B.A., P.S. Carberry, G.L. Hammer, M.E. Probert, M.J. Robertson, D. Holzworth, N.I. Huth, J.N.G. Hargreaves, H. Meinke, Z. Hochman, G. McLean, K. Verburg, V. Snow, J.P. Dimes, M. Silburn, E. Wang, S. Brown, K.L. Bristow, S. Asseng, C.S. Chapman, R.L. McCown, D.M. Freebairn, and C.J. Smith. 2003. An overview of APSIM, a model designed for farming systems simulation. Eur. J. Agron. 18:267–288.
Kipsat, M.J., H.K. Maritim, and J.R. Okalebo. 2004. Economic analysis of non-conventional fertilizers in Vihiga district, western Kenya. p. 535-
1526 Agronomy Journa l • Volume 100, Issue 5 • 2008
544. In A. Bationo et al. (ed.) Improving human welfare and environ-mental conservation by empowering farmers to combat soil fertility degradation. Proc. Int. Symp., Yaoundé, Cameroun, May 17–22. African Network for Soil Biology and Fertility, Nairobi, Kenya.
Lal, R. 1997. Degradation and resilience of soils. Philos. Trans. Royal Soc. London 1356:997–1010.
Micheni, A., F. Kihanda, and J. Irungu. 2004. Soil organic matter (SOM): Th e basis for improved crop production in arid and semi-arid climates of eastern Kenya. p. 239–248. In A. Bationo et al. (ed.) Improving human welfare and environmental conservation by empowering farmers to com-bat soil fertility degradation. Proc. Int. Symp., Yaoundé, Cameroun, May 17–22 African Network for Soil Biology and Fertility, Nairobi, Kenya.
Misiko, M. 2007. Fertile ground? Soil fertility management and the African smallholder. PhD thesis. Wageningen Univ., Th e Netherlands.
Mtambanengwe, F., and P. Mapfumo. 2005. Organic matter management as an underlying cause for soil fertility gradients on smallholder farms in Zimbabwe. Nutr. Cycling Agroecosyst. 73:227–243.
National Agricultural Research Laboratory. 1994. Final report of the fertil-izer use recommendation program (FURP), Vol. V and VII, Busia and Kakamega districts. National Agric. Res. Lab., Kenya Ministry of Agric. and Livestock, Nairobi.
Nijhof, K. 1987. Th e concentration of macro-elements in economic products and residues of (sub)tropical fi eld crops. Staff working paper SWO-87–08. Cent. for World Food Studies, Wageningen, Th e Netherlands.
Rufi no, M.C., P. Tittonell, M.T. van Wijk, A. Castellanos-Navarrete, N. de Ridder, and K.E. Giller. 2007. Manure as a key resource to sustain-ability of smallholder farming systems: Analysing farm-scale nutrient cycling effi ciencies within the NUANCES framework. Livestock Sci. 112(3):273–287.
Scanlon, B.R., B.J. Andraski, and J. Bilskie. 2002. Methods of soil analysis: Physical methods: Miscellaneous methods for measuring matric or water potential. Soil Sci. Soc. Am. J. 4:643–670.
Shepherd, K.D., E. Ohlsson, J.R. Okalebo, and J.K. Ndufa. 1996. Potential impact of agroforestry on soil nutrient balances at the farm scale in the East African Highlands. Fert. Res. 44:97–99.
Smaling, E.M.A., S.M. Nandwa, and B.H. Janssen. 1997. Soil fertility in Africa is at stake. p. 47–61. In R.J. Buresh et al. (ed.) Replenishing soil fertility in Africa. ASA, CSSA, SSSA, Madison, WI.
Solomon, D., J. Lehmann, J. Kinyangi, W. Amelung, I. Lobe, A. Pell, S. Riha, S. Ngoze, L. Verchot, D. Mbugua, J. Skjemstad, and T. Schäfer. 2007. Long-term impacts of anthropogenic perturbations on dynamics and speciation of organic carbon in tropical forest and subtropical grassland ecosystems. Glob. Change Biol. 13:511–530.
Tittonell, P. 2003. Soil fertility gradients in smallholder farm of western Kenya. Th eir origin, magnitude and importance. Quantitative Approaches in Systems Analysis No 25. Wageningen, UR, Th e Netherlands.
Tittonell, P., P.A. Leff elaar, B. Vanlauwe, M.T. van Wijk, and K.E. Giller. 2006. Exploring diversity of crop and soil management within small-holder African farms: A dynamic model for simulation of N balances and use effi ciencies at fi eld scale. Agric. Syst. 91:71–101.
Tittonell, P., K.D. Shepherd, B. Vanlauwe, and K.E. Giller. 2007a. Unravelling factors aff ecting crop productivity in smallholder agricultural systems of western Kenya using classifi cation and regression tree analysis. Agric. Ecosyst. Environ. 123:137–150.
Tittonell, P., B. Vanlauwe, M. Corbeels, and K.E. Giller. 2007b. Yield gaps, nutrient use effi ciencies and responses to fertilisers by maize across het-erogeneous smallholder farms in western Kenya. Plant and Soil, doi: 10.1007/s11104-008-9676-3.
Tittonell, P., B. Vanlauwe, P.A. Leff elaar, K.D. Shepherd, and K.E. Giller. 2005. Exploring diversity in soil fertility management of smallholder farms in western Kenya. II. Within-farm variability in resource alloca-tion, nutrient fl ows and soil fertility status. Agric. Ecosyst. Environ. 110:166–184.
Tittonell, P., S. Zingore, M.T. van Wijk, M. Corbeels, and K.E. Giller. 2007c. Nutrient use effi ciencies and crop responses to N, P and manure applica-tions in Zimbabwean soils: Exploring management strategies across soil fertility gradients. Field Crop Res. 100:348–368.
Vanlauwe, B., and K.E. Giller. 2006. Popular myths around soil fertility man-agement in sub-Saharan Africa. Agric. Ecosyst. Environ. 116:34–46.
Vanlauwe, B., P. Tittonell, and J. Mukalama. 2006. Within-farm soil fertil-ity gradients aff ect response of maize to fertilizer application in western Kenya. Nutr. Cycling Agroecosyst. 76:171–182.
Vanlauwe, B., J. Wendt, and J. Diels. 2001. Combined application of organic matter and fertilizer. p. 247–279. In Tian et al. (ed.) Sustaining soil fertil-ity in West-Africa. Special Publ. ASA, Madison, WI.
van Keulen, H. 1995. Sustainability and long-term dynamics of soil organic matter and nutrients under alternative management strategies. p. 353–375. In Bouma et al. (ed.) Ecoregional approaches for sustainable land use. Kluwer, Th e Netherlands.
van Keulen, H., and H. Breman. 1990. Agricultural development in the West African Sahelian region: A cure against land hunger? Agric. Ecosyst. Environ. 32:177–197.
Waithaka, M.M., P.K. Th ornton, M. Herrero, and K.D. Shepherd. 2006. Bio-economic evaluation of farmers’ perceptions of viable farms in western Kenya. Agric. Syst. 90:243–271.
Woomer, P.L., A. Martin, A. Albrecht, D.V.S. Resck, and H.W. Scharpenseel. 1994. Th e importance and management of soil organic matter in the tropics. p. 47–80. In P.L. Woomer, et al. (ed.) Th e biological management of tropical soil fertility. J. Wiley, Chichester, UK,
Wopereis, M.C.S., A. Tamélokpo, K. Ezui, D. Gnakpénou, B. Fofana, and H. Breman. 2006. Mineral fertilizer management of maize on farmer fi elds diff ering in organic inputs in the West African savanna. Field Crops Res. 96:355–362.
Zingore, S., R.J. Delve, J. Nyamangara, and K.E. Giller. 2008. Multiple benefi ts of manure: Th e key to maintenance of soil fertility and restoration of depleted sandy soils on African smallholder farms. Plant Soil 80:267–282.
Zingore, S., H.K. Murwira, R.J. Delve, and K.E. Giller. 2007. Soil type, his-torical management and current resource allocation: Th ree dimensions regulating variability of maize yields and nutrient use effi ciencies on African smallholder farms. Field Crops Res. 101:296–305.