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.