URN:NBN:fi:tsv-oa6813 The future of environmental sustainability in the Taita Hills, Kenya: assessing potential impacts of agricultural expansion and climate change EDUARDO EIJI MAEDA Maeda, Eduardo Eiji (2012). The future of environmental sustainability in the Taita Hills, Kenya: assessing potential impacts of agricultural expansion and cli- mate change. Fennia 190: 1, pp. 41–59. ISSN 1798-5617. The indigenous cloud forests in the Taita Hills have suffered substantial degrada- tion for several centuries due to agricultural expansion. Currently, only 1% of the original forested area remains preserved. Furthermore, climate change im- poses an imminent threat for local economy and environmental sustainability. In such circumstances, elaborating tools to conciliate socioeconomic growth and natural resources conservation is an enormous challenge. This article tackles essential aspects for understanding the ongoing agricultural activities in the Taita Hills and their potential environmental consequences in the future. Initially, an alternative method is proposed to reduce uncertainties and costs for estimating agricultural water demand. The main characteristic of the approach proposed in this study is the use of satellite data to overcome data availability limitations. Furthermore, a modelling framework was designed to delineate agricultural ex- pansion projections and evaluate the future impacts of agriculture on soil ero- sion and irrigation water demand. The results indicate that if current trends per- sist, agricultural areas will occupy roughly 60% of the study area by 2030. Rain- fall erosivity is likely to increase during April and November due to climate change and slight decrease during March and May. Although the simulations indicate that climate change will likely increase total annual rainfall volumes during the following decades, irrigation requirements will continue to increase due to agricultural expansion. By 2030, new cropland areas may cause an in- crease of approximately 40% in the annual volume of water necessary for irriga- tion. Keywords: Land changes, climate change, simulation models, water resources, soil erosion Eduardo Eiji Maeda, Department of Geosciences and Geography, University of Helsinki, Gustaf Hällströmin katu 2, 00014, Helsinki, Finland. E-mail address: eduardo.maeda@helsinki.fi. Introduction The world population has grown from 2.5 billion people in the 1950s to approximately 6.9 billion people in 2010 (UN 2010). The ability of mankind to cultivate crops and raise livestock, together with recent advances in agricultural techniques, is per- haps the main factor that allowed this fast popula- tion increase. Nevertheless, agriculture has changed the face of the planet’s surface and con- tinues to expand at alarming rates. Currently, al- most one-third of the world’s land surface is under agricultural use and millions of hectares of natural ecosystems are converted to croplands or pastures every year (Foley et al. 2005). If current trends per- sist, it is expected that by 2050 around 10 billion hectares of natural ecosystems will be converted to agriculture (Tilman et al. 2001). The development of the agricultural sector is es- sential to provide food for the population and combat food insecurity in poor countries. How- ever, the expansion of croplands without logistical and technological planning is a severe threat to the environment. Hence, the dilemma of integrat- 42 FENNIA 190: 1 (2012)Eduardo Eiji Maeda ing economic and population growth with envi- ronmental sustainability is an undeniable issue that needs to be addressed. Fresh water is perhaps the natural resource mostly affected by agricultural activities. Current- ly, roughly 70% of freshwater withdraws are used for agriculture (FAO 2005). Although global with- drawals of water resources are still below the criti- cal limit, more than two billion people live in highly water-stressed areas due to the uneven dis- tribution of this resource in time and space (Oki & Kanae 2006). In Kenya, currently over 55% of the rural population do not have access to quality drinkable water (FAO 2005). Another major environmental problem associ- ated with the expansion of agriculture is soil ero- sion. Although soil erosion is a natural process, changes in the landscape structure caused by the replacement of natural vegetation are likely to re- sult in accelerated rates of soil loss. Increased ero- sion rates are directly associated with nutrient loss, which may reduce agricultural productivity (Bakker et al. 2007) and cause water bodies’ eu- trophication (Istvánovics 2009). In some cases, ad- vanced stages of soil erosion, such as rill and gully erosions, can devastate entire areas, turning them unsuitable for agricultural purposes (Kirkby & Bracken 2009). Furthermore, variations in precipitation and temperature patterns associated with climate change also have important impacts on the sus- tainability of agricultural systems. For instance, changes in precipitation volume and intensity may increase the energy available in rainfall for detach- ing and carrying sediments, accelerating soil ero- sion. According to Yang et al. (2003), the global average soil erosion is projected to increase ap- proximately 9% by 2090 due to climate change. The climate also exerts great influence on water needs for agriculture. Projections indicate that, without proper investments in water management, climate change may increase global irrigation wa- ter needs by roughly 20% by 2080 (Fischer et al. 2007). Currently, science is facing new challenges to advance in the direction of environmental sustain- ability. One major challenge lies in the need for understanding the interactions and feedbacks be- tween human activities and the environment (Fig. 1). Hence, interdisciplinary studies are essential to improve our knowledge on the relationships be- tween different components of environmental sys- tems. Another important challenge is the acquisition of reliable and appropriate data for environmental modelling. For instance, solar radiation, relative Fig. 1. Flow chart showing a simplified illustration of interactions between agricultural expansion, climate and environment addressed in this article. FENNIA 190: 1 (2012) 43The future of environmental sustainability in the Taita Hills, … humidity and wind speed are some of the varia- bles usually necessary to estimate evapotranspira- tion (ET). However, assembling and maintaining meteorological stations capable of measuring such variables is, in general, expensive. In many poor regions, meteorological stations are insufficient to acquire the information necessary to represent the spatial-temporal variation of ET. As a result, the ir- rigation management in such areas is usually inap- propriate, increasing the risks of water scarcity and water conflicts. Therefore, in order to conciliate agricultural sys- tems productivity and environmental sustainabili- ty it is imperative to create appropriate tools for monitoring current activities and delineating ap- propriate strategies for coping with expected changes in the future. This study addresses impor- tant elements of this challenge, focusing on envi- ronmental issues and methodological drawbacks currently faced in the Taita Hills region, Kenya. The Taita Hills is home for an outstanding diversity of flora and fauna, with a high level of endemism (Burgess et al. 2007). Despite the huge importance of this region from environmental and biological conservation perspectives, the Taita Hills have suf- fered substantial degradation for several centuries due to agricultural expansion. Hence, the area is considered to have high scientific interest, and there is an urgent need for tools and information that are able to assist the sustainable management of agricultural systems and natural resources. This article presents a series of interdisciplinary studies, which integrate different technologies and model- ling techniques aiming to understand specific en- vironmental aspects and delineate future environ- mental scenarios for the Taita Hills. The specific research problems and objectives are delineated below. Research problems and objectives I. The availability of ground meteorological data is extremely limited in the Taita Hills. This limi- tation is a serious bottleneck for the manage- ment of water resources used for irrigation, given that it prevents an accurate assessment of evapotranspiration (ET). To overcome this drawback, the combination of ET models with remote sensing data is considered a promising alternative to obtaining temporally and spatial- ly continuous variables necessary for ET calcu- lation. This study evaluates three empirical ET models using as input land surface temperature data acquired by the MODIS/Terra sensor, aim- ing to delineate an alternative approach for es- timating ET in the Taita Hills. II. Despite the large importance of agricultural ac- tivities for the economy and food security in the Taita Hills, the expansion of croplands imposes serious threats for the environment. Under- standing the driving forces, tendencies and pat- terns of land changes is an essential step for elaborating policies that can conciliate land use allocation and natural resources conserva- tion. This article investigates the role of land- scape attributes and infrastructure components as driving forces of agricultural expansion in the Taita Hills and simulates future landscape scenarios up to the year 2030. III. Land use and soil erosion are closely linked with each other and with local climate. The ex- pansion of agricultural areas in the Taita Hills and changes in precipitation patterns associat- ed with climate change are imminent threats for soil conservation. In this context mathemat- ical models and scenario exercises are useful tools to assist stakeholders in delineating soil conservation practices that are consistent with plausible environmental changes in the future. One of the objectives of this study is to investi- gate the potential impacts of future agricultural expansion and climate change on soil erosion in the Taita Hills. IV. In Africa, as well as in most parts of the world, the agricultural sector is the main consumer of water resources. As agricultural areas increase in the Taita Hills, there is an escalating concern regarding the sustainable use of water resourc- es. Furthermore, future changes in temperature and rainfall patterns may affect the water re- quirements for agricultural activities. Under- standing the relations between these compo- nents is crucial to identify potential risks of water resources depletion and delineate appro- priate public policies to deal with the problem. This study evaluates prospective changes on ir- rigation water requirements caused by future agricultural expansion and climate change. Study area The Taita Hills are the northernmost part of the Eastern Arc Mountains of Kenya and Tanzania, situated in the middle of the Tsavo plains in the Coast Province, Kenya (Fig. 2). The Eastern Arc 44 FENNIA 190: 1 (2012)Eduardo Eiji Maeda Mountains sustain some of the richest concentra- tions of endemic animals and plants on Earth, and thus it is considered one of the world’s 25 biodi- versity hotspots (Myers et al. 2000). The Taita Hills cover an area of approximately 850 km2. The population of the whole Taita-Taveta county has grown from 90,146 persons in 1962 to approximately 280,000 in the year 2009 (KNBS 2010). According to Clark (2010), population growth has been a central driving factor behind rising environmental pressure. The indigenous cloud forests have suffered substantial loss and degradation for several centuries as abundant rain- fall and rich soils have created good conditions for agriculture. Between 1955 and 2004, approxi- mately half of the cloud forests in the hills have been cleared for agricultural lands (Pellikka et al. 2009). Population growth and increasing areas un- der cultivation for subsistence farming have caused scarcity of available land in the hills and contributed to the clearance of new agricultural land in the lowlands (Clark 2010). Currently, it is estimated that only 1% of the original forested area remains preserved (Pellikka et al. 2009). The agriculture in the hills is characterised by intensive small-scale subsistence farming. In the lower highland zone and in upper midland zone, the typical crops are maize, beans, peas, potatoes, cabbages, tomatoes, cassava and banana (Jaetzold & Schmidt 1983; Soini 2005). In the slopes and lower parts of the hills with average annual rainfall between 600 and 900 mm, early maturing maize, sorghum and millet species are cultivated. In the lower midland zones with average rainfall be- tween 500 and 700 mm, dryland maize varieties and onions are cultivated, among others. Supplementary irrigation practice is common, especially in the highlands, and profitable produc- tion is highly dependent on the availability of wa- ter resources (Jaetzold & Schmidt 1983). Despite the small average farm size, the income of many families in the Taita Hills relies solely on agricul- tural production. Although the technological level of farmers is not high, many carry out basic soil conservation practices, such as terraces. Methods Alternative methods for estimating reference evapotranspiration In order to identify feasible approaches for esti- mating reference ET (ETo) in the Taita Hills, three Fig. 2. Geographical location of the study area shown in a TM-Landsat image from April 3, 2001. FENNIA 190: 1 (2012) 45The future of environmental sustainability in the Taita Hills, … empirical ETo models that require only air tem- perature data were evaluated, namely the Har- greaves (Hargreaves & Samani 1985), the Thorn- thwaite (Thornthwaite 1948) and the Blaney-Crid- dle (Blaney & Criddle 1962) methods. To overcome the low data availability from ground meteorological stations, this study made use of land surface temperature (LST) data ob- tained from the MODIS sensor, on board of Terra and Aqua satellites (Wang et al. 2005). In order to clearly distinguish this approach, when LST data is used in replacement of air temperature data from ground stations, the Hargreaves, the Thornthwaite and the Blaney-Criddle models will be hereafter denominated as Hargreaves-LST, Thornthwaite- LST, and Blaney-Criddle-LST, respectively. The empirical equations were calibrated using as a reference the FAO Penman–Monteith (FAO- PM) method. The FAO-PM method is recommend- ed as the standard ETo method and has been ac- cepted by the scientific community as the most precise, this is because of its good results when compared with other equations in different regions worldwide (Cai et al. 2007; Jabloun & Sahli 2008). Although the FAO-PM method also carries intrin- sic uncertainties and errors, it has behaved well under a variety of climatic conditions, and for this reason the use of such methods to calibrate or validate empirical equations has been widely rec- ommended (Allen et al. 1998; Gavilán et al. 2006). The meteorological data necessary for the FAO- PM equations were obtained from a synoptic sta- tion placed at Voi town and operated by the Kenya meteorological department. ETo values were also calculated for this exact point using the empirical models and MODIS LST data. The calibration pa- rameters were defined using the following equa- tion (Allen et al. 1998): ETo cal = a + b · ETo LST (1) where ETo cal represents the calibrated ETo values, in which the calibration parameters a and b are determined by regression analysis using as a refer- ence the FAO-PM method; ETo lst is the ETo values estimated using the empirical models and MODIS LST as input. The estimates obtained by each mod- el were compared using standard statistics and lin- ear regression analysis (Douglas et al. 2009). Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) were calculated using the equations described below: RMSE = (1n n ∑ 1 (ETo cal – ETo R )2)0.5 (2) MAE = 1 n n ∑ 1 |ETo cal – ETo R | (3) Agricultural expansion modelling in the Taita Hills This study integrated remote sensing, GIS tech- niques and a spatially explicit simulation model of landscape dynamics, DINAMICA-EGO (Soares- Filho et al. 2007), to assess the agricultural expan- sion driving forces in the study area and simulate future scenarios of land use. A general description of the applied method is illustrated in Figure 3. The Land Use-Cover Change (LUCC) model re- ceives as inputs land use transition rates, land- scape variables and landscape parameters. The landscape parameters are intrinsic spatially dis- tributed features, such as soil type and slope, which are kept constant during the simulation process. The landscape variables are spatial-tem- poral dynamic features that are subjected to changes by decision makers, for instance roads and protected areas. Ten landscape attributes (var- iables/parameters) were used as inputs for the model. Land use global transition rates refer to the total amount of changes for each type of land use/land cover transition given in the simulation period, without taking into account the spatial distribution of such changes. The transition rates were calcu- lated by cross-tabulation, which produced as out- put a transition matrix between the land cover maps from 1987 and 2003. The dates of the land cover maps were chosen based on two criteria. The first criterion was that the landscape changes between the initial and final landscape should ac- curately represent the ongoing land change activi- ties in the study area. That is to say, the agricul- tural expansion rates between 1987 and 2003 were assumed to retrieve a consistent figure of the current trends. The second criterion relied on the availability of cloud free satellite images to assem- ble the land cover maps. According to a study car- ried out by Clark (2010), between 1987 and 2003 cropland has expanded by 10,478 ha, reflecting an expansion rate of approximately 650 ha year–1. The local transition probabilities, different from the global transition rates, are calculated for each grid cell considering the natural and anthropogen- ic characteristics of the site. The transition proba- 46 FENNIA 190: 1 (2012)Eduardo Eiji Maeda bility of each cell was calculated in DINAMICA- EGO using the weights of evidence (WoE) method (Soares-Filho et al. 2002; Almeida et al. 2008). The WoE is a Bayesian method in which the effect of each landscape variable on a transition is calcu- lated independently of a combined solution (Soares-Filho et al. 2002). The spatial probability of a transition is given by the following equation (Bonham-Carter 1994): O{T } × e n i= ∑ 1 W + x,y P x,y {T/V 1 ∩ V 2 ∩ … ∩ V n } = ,(4) 1 + O{T } × t j= ∑ 1 e n i= ∑ 1 W + x,y where P x,y is the probability of transition in a cell with coordinates x,y; T represents the land use/ land cover transition; V n accounts for all possible landscape variables i selected to explain transition T; O{T} is the odd of a transition, represented by the ratio between a determined transition proba- bility and the complementary probability of non- occurrence, described by equation 5: O{T} = P P { { T T̄ } } , (5) where P{T} is the probability of occurrence of tran- sition T, given by the number of cells where the concerned land use/land cover transition occurred divided by the total number of cells in the study area; P{T̄} is the probability of non-occurrence of transition T, given by the number of cells where the concerned land use/land cover transition is ab- sent divided by the total number of cells in the study area, and W+ x,y is the weight of evidence for a determined landscape variable range, defined by the following equation: W+ = log e P P { { V V i i / / T T̄ } } , (6) where P{V i /T} is the probability of occurring vari- able V i in face of the previous presence of transi- tion T, given by the number of cells where both V i and T are found divided by the total number of cells where T is found and P{V i / T̄} accounts for the probability of occurring variable V i in face of the previous absence of transition T, given by the number of cells where both V i and are found di- vided by the total number of cells where T is not found. The W+ values represent the attraction between a determined landscape transition and a certain variable. The higher the W+ value is, the greater is the probability of a certain transition to take place. On the other hand, negative W+ values indicate Fig. 3. General description of the modeling framework, in which landscape attributes obtained using remote sensing and GIS techniques are used as inputs for a LUCC model. The model evaluates the role of each attribute in the land changes and simulates future landscape scenarios (adapted from Maeda et al. 2011a). FENNIA 190: 1 (2012) 47The future of environmental sustainability in the Taita Hills, … lower probability of a determined transition occur- ring in the presence of the respective variable range. Based on the W+ values of each range for every considered variable, DINAMICA-EGO gen- erates a spatially explicit probability map, in which the cell receives as attributes the probability for a determined transition. After defining the weight of each landscape var- iable, transition probability maps are created for every simulated year. Based on spatial probabili- ties, new agricultural patches are stochastically al- located using two algorithms: ‘expander’ and ‘patcher’. The expander function performs the ex- pansion of previously existing patches of a certain class. The patcher function, in turn, is designed to generate new patches through a seed formation mechanism (Soares-Filho et al. 2002). Assessment of potential impacts on soil erosion Future agricultural expansion and climate change scenarios were used to evaluate their potential im- pacts on soil erosion in the Taita Hills. To achieve this objective a modelling framework was assem- bled by coupling a landscape dynamic simulation model, an erosion model and synthetic precipita- tion datasets generated through a stochastic weather generator. This approach aimed to evalu- ate how agricultural expansion, together with cli- mate change, can modify the variables of a widely used soil erosion model, allowing a qualitative as- sessment of the impacts of these changes for soil conservation. The soil erosion model used was the Universal Soil Loss Equation (USLE) (Wischmeier & Smith 1978). Remote sensing and GIS techniques were combined to provide the necessary inputs for the modelling framework. A flow chart illustrating the components of the modelling framework is pre- sented in Figure 4. The USLE and its revised version, RUSLE (Re- nard et al. 1997), have been extensively used worldwide during the last decades (Kinnell 2010). Even though these models are known for their sim- plicity, their effectiveness has been demonstrated in many recent studies (e.g. Beskow et al. 2009; Nigel & Rughooputh 2010), including several studies in Kenya (e.g. Angima et al. 2003; Mutua et al. 2006). The USLE is given as: A = R × K × LS × C × P (7) where A is the annual average soil loss [t ha–1 ano–1], R is the rainfall erosivity factor [MJ mm ha–1 Fig. 4. Flow chart illustrating the integrated modelling framework concept used to estimate the impacts of agricultural ex- pansion and climate change on soil erosion (adapted from Maeda et al. 2010b). 48 FENNIA 190: 1 (2012)Eduardo Eiji Maeda h–1], K is soil erodibility [t ha h MJ–1 mm–1], L and S are the topographical factor [–], C is the vegetation cover factor [–], and P represents erosion control practices [–]. Provided the fact that the K and LS factors are intrinsic characteristics of the landscape, they can be kept constant in all simulated scenarios. On the other hand, land changes directly affect the C fac- tor. These changes were analysed by evaluating the average C factor value in the study area during 1987, 2003 and in the simulated scenario for 2030. The potential impacts of agricultural expan- sion for soil conservation were also assessed by analysing the spatial distribution of croplands in relation to the K and LS factors. Possible changes on the P factor were not addressed in the present study. The rainfall erosivity factor (R) is a numerical index that expresses the capacity of the rain to erode a soil (Wischmeier & Smith 1978). Hence, the R factor is directly affected by changes in pre- cipitation pattern. These changes were evaluated at monthly and yearly time steps. Additionally, the soil erosion potential was calculated by excluding the anthropogenic variables from the USLE equa- tion (C and P). This approach is needed to clearly understand the role of external factors in the sys- tem without the influence of the changes in the landscape cause by human activities. Although the USLE provides a simple and useful tool for soil conservation, studies commonly ne- glect the calibration and validation of this model. Given the absence of reliable data for calibration, the presented study did not attempt to provide soil loss estimation figures. Instead, the evaluation of the soil erosion potential among the different sce- narios was based mainly on a comparative analy- sis of changes, following the procedure proposed by Kepner et al. (2004) and Miller et al. (2002). Such procedure assumes that, using percent change observations, the parameters incorporated in an eventual calibration would be partially or to- tally cancelled, providing more realistic figures than absolute values of soil loss. The absolute changes in soil erosion potential were analysed only qualitatively, taking into account the spatio- temporal distribution of changes. The R factor was calculated using the method proposed by Renard and Freimund (1994), and re- cently applied by Beskow et al. (2009). The meth- od is based on an empirical relationship between rainfall erosivity and the Fournier Index (FI). The FI indicates climatic aggressiveness, which has a high correlation with the amount of sediment washed into the stream by surface runoff. Future climatic conditions were simulated using a stochastic weather generator. Three greenhouse- gas emission scenarios were considered (SRES, Special Report on Emissions Scenarios): SRA1B, SRA2 and SRB1 (Nakicenovic et al. 2000). For each scenario, a respective synthetic precipitation datasets was created: SyA1B, SyA2 and SyB1. The data necessary for this procedure were obtained from the IPCC data distribution centre (http:// www.ipcc-data.org). A detailed description of the procedures used to generate the synthetic climate datasets are described in Maeda et al. (2010b). The K factor was calculated using the method proposed by Williams & Renard (1983). This ap- proach was chosen for being broadly used in re- cent studies (e.g. Xiaodan et al. 2004; Rahman et al. 2009) and for requiring input variables that are commonly available worldwide. The data neces- sary for calculating the K factor were obtained from the Soil and Terrain Database for Kenya (KENSOTER), which provides a harmonized set of soil parameter estimates for Kenya (Batjes & Gicheru 2004). The topographical factor (LS) was calculated in the software USLE2D (Van Oost et al. 2000), using the algorithm proposed by Wischmeier and Smith (1978). This calculation was performed based on a 20 m spatial resolution DEM. Assessment of potential impacts on irrigation water requirement In this last assessment, agricultural expansion and climate change scenarios were used to evaluate their potential impacts on Irrigation Water Re- quirements (IWR). Remote sensing and GIS tech- niques were combined to provide the necessary inputs for the modelling framework described in Figure 5. Crop water requirement (CWR) is defined as the amount of water required to compensate the ET loss from a cropped field (Allen et al. 1998). In cases where all the water needed for optimal growth of the crop is provided by rainfall, irriga- tion is not required and the IWR is zero. In cases where all water has to be supplied by irrigation the IWR is equal to the crop ET (ETc). However, when part of the CWR is supplied by rainfall and the re- maining part by irrigation, the IWR is equal to the difference between the ETc and the Effective Pre- FENNIA 190: 1 (2012) 49The future of environmental sustainability in the Taita Hills, … cipitation (Peff). In such cases, the IWR was com- puted using the following equation (FAO 1997): IWR m = (Kc m × ETo m × 30) – Peff m (8) where: IWR m = monthly average crop water re- quirement in month m, [mm]; Kc m = crop coeffi- cient in month m, [–]; ETo m = mean daily reference evapotranspiration in month m, [mm day–1]; Peff m = average effective precipitation in month m, [mm]. The Peff is defined as the fraction of rainfall re- tained in the root zone, which can be effectively used by the plants. That is, the portion of precipita- tion that is not lost by runoff, evaporation or deep percolation. The monthly total rainfall was con- verted to Peff using a simplified method proposed by Brouwer and Heibloem (1986), which is based on empirical observations and requires only the total monthly volume of precipitation. The Hargreaves model was chosen to estimate the ETo in the study area. The Kc values were ob- tained from tables recommended by FAO (Allen et al. 1998). Nevertheless, to assign the appropriate Kc values it is essential to identify the agriculture calendar in the study area, that is, the period of the year when crops are planted, grown and harvest- ed. For this, monthly Normalized Difference Veg- etation Index (NDVI) obtained from MODIS satel- lite imagery were used to identify the phenological stages of croplands during the year. A stochastic weather generator was used to esti- mate future precipitation and temperature condi- tions. Three SRES greenhouse-gas emission sce- narios were considered: SRA1B, SRA2 and SRB1 (Nakicenovic et al. 2000). For each scenario, a re- spective synthetic precipitation and temperature datasets were created: SyA1B, SyA2 and SyB1. For a detailed description of this method, please refer to Maeda et al. (2011b). Results Remote sensing based methods for estimating evapotranspiration The results obtained in the evaluation of the ETo models are summarized in Table 1. The global av- erage RMSE and MAE are fairly homogeneous for each of the evaluated models. The RMSE ranged from 0.47 mm day–1, with the Hargreaves-LST Fig. 5. Flow chart illustrating the integrated modelling framework concept used in assessing the potential impacts of agricul- tural expansion and climate change on irrigation water requirements (adapted from Maeda et al. 2011b). 50 FENNIA 190: 1 (2012)Eduardo Eiji Maeda model, to 0.53 mm day–1, with the Blaney-Criddle- LST model. The MAE achieved similar figures, ranging from 0.39 mm day–1, with the Hargreaves- LST model, to 0.46 mm day–1, with the Blaney- Criddle-LST model. The monthly errors obtained by the tested mod- els, in comparison with the reference method, are presented in Figure 6. Although the monthly per- formance of the models is uniform between March and July, important differences are observed in January, February and between August and De- cember (Fig. 6). In particular for the Blaney-Crid- dle-LST model, it is possible to notice a clear in- crease of the RMSE and MAE in January, Novem- ber and December. The Blaney-Criddle-LST model performed better in months when air temperature was closely related to LST, but had its performance reduced in months when air temperature and LST are less correlated. However, the Hargreaves-LST model was more efficient in minimizing the effects of the differences observed between air tempera- ture and LST during November, December and January. The model performed well during these months, retrieving RMSEs of 0.51, 0.61 and 0.51 mm day–1, respectively. Considering the results achieved in this study and comparisons with previous research, it is con- cluded that the Blaney-Criddle-LST model is not appropriated for this region when using the pro- posed methodology. However, the Hargreaves- LST model achieved the best results in the linear regression and in the analysis of errors. The results obtained using the Hargreaves-LST are compatible with the errors observed by Gavilán et al. (2006), which evaluated the Hargreaves equation under semiarid conditions in southern Spain, finding RMSE ranging from 0.46 to 1.65 mm d–1. Further- more, the correlation coefficients obtained by the Hargreaves-LST and Thornthwaite-LST models are consistent with the results reported by Narongrit Table 1. Summary of the results obtained from the models’ error analysis and linear regression analysis Hargreaves-LST Thornthwaite-LST Blaney-Criddle-LST Correlation coefficient (R) 0.67 0.66 0.55 RMSE (mm day–1) 0.47 0.49 0.53 MAE (mm day–1) 0.39 0.42 0.46 Calibration parameter (a) 3.221 3.507 –1.980 Calibration parameter (b) 0.497 0.543 1.379 Fig. 6. Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). The errors are used to quantify the differences between the ETo estimated using the reference method (FAO-PM) and the estimates obtained using the empirical models parameterized using MODIS LST data (adapted from Maeda et al. 2011a). FENNIA 190: 1 (2012) 51The future of environmental sustainability in the Taita Hills, … and Yasuoka (2003), which achieved R2 of 0.57 and 0.60 when comparing these respective mod- els with the FAO-PM method. The driving forces of agricultural expansion and scenarios for 2030 The highest conversion rates were observed in the transition from woodlands to agriculture. Howev- er, considering absolute numbers, shrubland areas are the most affected, given that currently they represent the predominant vegetation type in the region. The small regions covered with broad- leaved forests were nearly untouched, presenting low conversion rates, the total area decreased from 7.7 to 6.9 km2 during the observed period. The most relevant W+ values obtained during the model calibration are shown in Fig. 7. This in- formation represents the attraction between a de- termined landscape transition and a certain land- Fig. 7. W+ values at- tributed for each range of six landscape attributes most related to the ‘shrublands to croplands’ transition (adapted from Maeda et al. 2010a). 52 FENNIA 190: 1 (2012)Eduardo Eiji Maeda scape attribute. Distance to rivers, insolation, dis- tance to croplands, DEM, distance to roads and distant to markets were particularly associated with the land-use transitions. The distance to cro- plands is an important driving factor for all transi- tions indicating that the proximity to previously established croplands is a key factor for agricul- tural expansion in this region. Although areas close to rivers did not retrieve high positive W+ values, the importance of water bodies for croplands is clearly reflected in regions distant from rivers, where high negative W+ values are observed. Hence, the results indicate that patches further than 1 km from water bodies have lower probability of being converted to cropland. Distance to roads also presents a clear pattern in influencing the transition from shrublands to crop- lands. Nevertheless, this attribute did not retrieve very high W+ values, possibly due to the fact that the Taita Hills comprise a relatively dense road network, diminishing the contrast between areas nearby and away from roads. The distance to mar- kets, here represented by the Euclidean distance to the main villages, was the most representative driving force for the agricultural expansion. After the model is calibrated and the role of each landscape variable is defined, transition probability maps are created for each simulated year. The spatial probabilities are used to guide the distribution of new simulated agricultural patches, which are stochastically allocated by the ‘expand- er’ and ‘patcher’ algorithms. In Figure 8, the land use maps for 1987 and 2003 are displayed (upper left and upper right) together with the land use maps for 2030 resulted from the simulation. It is observed that, in 1987, croplands were already clearly established along highlands (central area in the maps). This is explained by the favourable cli- matic and edaphic conditions for agricultural ac- tivities (e.g. high precipitation rates), which result- ed in the clearance of large areas of forest during the last century. Cropland areas expanded to Fig. 8. Land use maps for 1987 and 2003 (upper left and upper right) and simulated scenario for 2030 (lower left) (adapted from Maeda et al. 2010b). FENNIA 190: 1 (2012) 53The future of environmental sustainability in the Taita Hills, … around 515 km2 in 2030, corresponding to about 60% of the study area. This represents an increase of 40% in comparison to the year 2003, when cro- plands occupied around 365 km2. Potential impacts on soil erosion by 2030 The results show that agricultural patches estab- lished during the last decades were carefully set- tled in areas with favourable topography from the soil conservation perspective, i.e. lower LS factor. This pattern was reinforced after 1987, when the availability of space in the hills was scarce and the agriculture started expanding to flat areas along the foothills. In the agricultural expansion simu- lated for 2030, a slight increase is noted in crop- land patches settled in areas with LS factor be- tween 1 and 10, whereas the most significant in- crease was again in areas with a topographic fac- tor between 0 and 1. The soil erodibility factor in the study area var- ied from 0.0139 to 0.0307, allowing the distinc- tion of eight different erodibility classes. Because soil erodibility figures may vary according to the method used for calculation, the results were ana- lysed only in a comparative base, taking into ac- count the soil erodibility range found in the study area. In 1987 and 2003 agricultural areas were es- tablished mainly in soils with medium erodibility (0.0205−0.0255). The simulated agricultural ex- pansion for 2030 was also higher in these soils. The low occurrence of agricultural activities in soils with higher erosivity is explained firstly by the low area occupied by these soils in the study area and secondly by the fact that such soils, to- gether with climatic variables, create unfavoura- ble conditions for agricultural practices. Therefore, the results indicate that agricultural activities are unlikely to expand into areas with higher soil erodibility. The average R factor for the study area was ap- proximately 3040 MJ mm ha–1 h–1 year–1 when considering current climate conditions. This result is consistent with figures obtained in other semi- arid regions. Da Silva (2004) found that erosivity varied from 2000 to 4000 MJ mm ha–1 h–1 year–1 in semi-arid regions in the north-east of Brazil. How- ever, in regions with high topographic heterogene- ity, such as the Taita Hills, it is crucial to consider local variations at detailed spatial scales. The erosivity values obtained in the SyA2 sce- nario resulted in the most evident differences in comparison with current conditions. In January, March, May and December the changes in pre- cipitation resulted in a clear but slight decrease in rainfall erosivity. The erosivity reduction during these months varied from 4 to 120 MJ mm ha–1 h–1 month–1. However, still for the SyA2 scenario, a large increase was observed in April (280 MJ mm ha–1 h–1 month–1) and November (260 MJ mm ha–1 h–1 month–1). For the SyB1 scenario, the increases in rainfall erosivity during April and November were lower, approximately 217 and 40 MJ mm ha–1 h–1 month–1, respectively. A slight decrease was also observed during March, May and December, but in contrast with the SyA2 scenario, the erosivity during Janu- ary was almost constant, with a minor increase of 27 MJ mm ha–1 h–1 month–1. The SyA1B was the most conservative scenario, although clear chang- es are still present. Namely, it showed the highest erosivity increases during January and December, while it confirmed the tendency of a decrease in erosivity during March and May. In general, it is plausible to assert that the cli- mate changes simulated for the study area are likely to decrease rainfall erosivity during March and May due to a slight reduction in precipitation rates in these months. However, the model indi- cates the possibility of an increase, of much higher magnitudes during April and November. The disa- greements between the simulated scenarios in January and December indicate higher uncertain- ties during these months. For June, July, August and September rainfall erosivity values are likely to continue to be very low. For all scenarios it is noted that low or no change occurs in the lowlands (roughly 500 m above sea level). The changes, however, start to be more evident along areas higher than 1000 m a.s.l., where average precipitation rates are higher. In particular for the SyA2 scenario, changes are very high above 1500 m, reaching absolute differ- ences up to 1500 MJ mm ha–1 h–1 year–1 when compared with current conditions. Potential impacts on irrigation water requirement by 2030 Spatial and temporal variations on ETo are strong both in the lowlands and in the hills. In general ETo is higher in the months of September and Oc- tober, and reaches the lowest values between April and June. The correlation between ETo and alti- tude varies according to season and altitude range. ETo follows more closely the changes in the alti- 54 FENNIA 190: 1 (2012)Eduardo Eiji Maeda tude in October than in May. In May, the variation ranges from 4.5 to 5.4 mm day–1 while in October the variation is from 5.3 to 6.9 mm day–1. From 1987 to 2003, a large number of cropland patches were created in areas with ETo between 5.7 and 5.9 mm day–1, while few new patches were imple- mented in areas with lower ETo (<5 mm day–1). This tendency was strongly sustained during the scenario simulated for 2030. Hence, it is feasible to assert that by 2030 new croplands are likely to take place in areas where ETo values are histori- cally higher. Considering an invariable climate condition, the agricultural expansion observed between 1987 and 2003 resulted in an IWR increase of 42%. An integrated analysis of this result with the historical precipitation volumes in the study area indicates that the agricultural expansion has likely reached unsustainable levels from the water resources point of view. That is to say, by 2003 the annual average precipitation volume in the entire study area (~390 million m3 year–1) was already insuffi- cient to meet the water resource requirements necessary to achieve optimal crop production in every agricultural property. Among the scenarios simulated for 2030, the scenario considering current climate conditions resulted in the highest annual IWR volume, reach- ing approximately 610 million m3 year–1. All the climate change scenarios (SyA1B, SyA2 and SyB1) caused a slight reduction in the IWR when com- pared with current conditions. Therefore, the re- sults indicate that the climate change tendencies up to 2030 are likely to decrease the total annual volume of IWR in this study area. A comparison between the annual IWR maps for 1987, 2003 and 2030, considering they Sy scenario, is presented in Figure 9. Figure 10 shows the monthly IWR values for all simulated scenarios. The increase in IWR caused by the agricultural expansion component is clearly identified in an offset of the curves among different years. Considering only the curves for the year 2030, the climate changes simulated in the SyA1B, SyB1 and SyA2 datasets indicate a tendency of in- crease in the IWR during March and May when compared with a scenario with unchanged cli- mate conditions (Sy2030). This tendency is invert- ed in April and November, when a slight decrease is observed in the IWR. During the other months, the IWR is kept relatively constant among the dif- ferent climate scenarios. In these cases, eventual increases in the temperature were likely to be compensated by increases in rainfall volume, keeping the IWR constant. From the practical point of view, the results indicate a higher water demand during the seeding season, in February/ March, and during the period of maximum devel- opment of the crops, in May. Climate changes are likely to decrease the water demand for irrigation during both crop growing seasons in April and No- vember. Fig. 9. Annual irrigation water requirement maps for 1987, 2003 and 2030. The black areas in the map represent regions where no agricultural activities are taking place (adapted from Maeda et al. 2011b). FENNIA 190: 1 (2012) 55The future of environmental sustainability in the Taita Hills, … Discussion The Taita Hills are among the most degraded areas in the Eastern Arc Mountains, having lost approxi- mately 99% of their original forest during the last few centuries (Pellikka et al. 2009). Although this number may sound discouraging, it also makes the Taita Hills a unique learning environment for the protection of more preserved areas in the Eastern Arc, such as Udzungwas, East Usambaras and Ul- ugurus. Therefore, the causes behind this massive forest loss must not be ignored. In particular, the fact that these highlands have significant impor- tance for providing food and income to local pop- ulation should be acknowledged. Previous studies have shown that agriculture was the main cause of forest conversion in the Taita Hills (e.g. Clark 2010). Consequently, it is al- ready known that agricultural activities are active- ly present in the Taita Hills, and continues to ex- pand. Nevertheless, the answers for other impor- tant questions remain unclear or unsolved. To bet- ter understand the degradation process in the East- ern Arc it is essential to determine what type of agricultural activities are taking place, to where it may expand in the future and what impacts it will cause. The combined results of this study provide initial arguments to deal with these important questions. Public policies need to take into consideration the areas where agriculture is more likely to ex- pand in the future. The answer for this question can only be achieved through a detailed assess- ment of the forces driving the land changes. In this context, this study presents a pioneer assessment of the driving forces of agricultural expansion in the Taita Hills. The results obtained in this analysis closely agree with previous studies carried out in other locations in Kenya. For instance, studying the Narok District in Kenya, Serneels and Lambin (2001) found that the expansion of small-holder agriculture is mainly controlled by proximity to permanent water, land suitability and vicinity to villages. In addition to delineating agricultural expansion scenarios for the year 2030, this article also ad- dressed potential environmental impacts caused by land changes. The highlands of Kenya and Tan- zania are considered natural water towers, given that the higher precipitation rates in these areas historically provide water resources for large re- gions along the lowlands throughout the year (Ae- schbacher et al. 2005; Viviroli & Weingartner 2008). However, the results presented in this study, Fig. 10. Monthly irrigation water re- quirements volume for the entire study area during the years 1987, 2003 and 2030. For the year 2030, four climate scenarios are considered (adapted from Maeda et al. 2011b). 56 FENNIA 190: 1 (2012)Eduardo Eiji Maeda and confirmed by field work observations, indi- cate that the average precipitation rates in the Taita Hills are no longer able to provide water resources for the lowlands during the entire year. It is plausi- ble to consider that the changes observed in water resources availability in the Taita Hills are likely to be caused by two main factors. Firstly, the increas- ing use of water resources for irrigation in the hills might be causing a decrease in rivers’ flow, dimin- ishing or in some case depleting the volume of wa- ter in the downstream portion of the rivers. The second, but not less important, factor contributing to this issue is likely to be related to changes in the hydrological response of the rivers’ basins caused by the replacement of natural vegetation in favour of croplands. Previous studies have shown that land changes may increase surface runoff (e.g. Germer et al. 2010) and have direct impacts on water balance (e.g. Li et al. 2009). These factors can potentially reduce water retention in the wa- tershed, decreasing the flow during the dry sea- sons. The replacement of natural vegetation cover will also contribute for accelerated soil erosion. Therefore, the undertaken soil conservations prac- tices must be set as a priority action in the Taita Hills. Although changes in erosion control prac- tices (P factor) were not considered in this study, previous investigations have shown that appropri- ate land management can significantly decrease soil erosion. For instance, studying soil erosion risk scenarios in Calabria, southern Italy, Terrano- va et al. (2009) show that erosion control practices can cause a significant reduction of the erosive rate, decreasing from roughly 30 to 12.3 Mg ha–1 year–1. Feng et al. (2010) demonstrated that soil conservation measures taken by the Chinese gov- ernment (Grain-for-Green project) significantly de- creased soil erosion in the Loess Plateau between the years 1990 and 2005. Finally, it is important to mention that the issues involving water resources and soil conservation can and should be tackled in an integrated man- ner. For instance, the adoption of practices to re- duce surface runoff, such as terraces, is an impor- tant step to avoid soil erosion, and at the same time it enhances water infiltration into the soil, al- lowing a better recharge of groundwater reser- voirs. Furthermore, opting for cultivars capable of maintaining some vegetation protection over the soil for longer periods can contribute to avoiding the direct impact of rainfall in the soil and improve soil structure. These factors not only reduce soil erosion, but also contribute to lower soil evapora- tion rates and a better water infiltration. In this context, the consolidation of agroforestry systems using native plant species may be highly benefi- cial to the local agriculture, and an excellent alter- native for replacing the eucalyptus plantation for- ests in the Taita Hills. Conclusions and further studies Two general contributions can be highlighted from the results obtained in this study. The first contri- bution lies in the development and assessment of alternative approaches to improve the acquisition of data related to key aspects of agricultural activi- ties in the Taita Hills. The second relates to the establishment of novel knowledge by delineating unprecedented insights on future environmental scenarios for the region. Considering the above contributions, the results of this study have a large potential to attain researchers, policy makers and local population. An alternative method for estimating ETo was evaluated by integrating remote sensing data and empirical models. The combined use of the Har- greaves ETo model and MODIS LST data retrieved an average RMSE close to 0.5 mm d–1. This out- come is consistent with results obtained by previ- ous studies reported in the literature using weather data collected by ground stations. Moreover, the errors and uncertainties identified in the use of re- mote sensing LST can be tolerated considering the reduced weather data collection network in this region. Further studies are necessary to expand this method for other regions in East-Africa. In par- ticular, the spatial variability of the calibration pa- rameters for different climate conditions over East- Africa needs to be identified. Moreover, the meth- od can be significantly improved by using low cost direct methods (e.g. lysimeters) to calibrate the empirical equations. In relation to the agricultural expansion model- ling in the Taita Hills, a connected relation be- tween villages and roads is evident in the defini- tion of new cropland patches. The proximity to already established crop fields is also one of the key factors driving the agricultural expansion. If current trends persist, it is expected that agricul- tural areas will occupy 60% of the study area by 2030. LUCC simulations indicate that agricultural expansion is likely to take place predominantly in lowlands and foothills throughout the next 20 FENNIA 190: 1 (2012) 57The future of environmental sustainability in the Taita Hills, … years. Current trends indicate that the small resid- ual areas of tropical cloud forest, home to a great part of the biodiversity in the Taita Hills, is likely to remain intact throughout the coming years. Never- theless, the impact of the increasing habitat frag- mentation in such biodiversity is a relevant issue that must be addressed in further studies. The replacement of shrublands and woodlands in favour of croplands expected for the next dec- ades is very likely to reduce the vegetation cover protecting the soil against the direct impact of rainfall, resulting in accelerated soil erosion. By the year 2030, rainfall erosivity is likely to increase during April and November. All scenarios con- verge to a slight erosivity decrease tendency dur- ing March and May. The highest uncertainties were observed in January and December, when some scenarios indicate a small reduction in ero- sivity while some indicate an increase. Account- ing for land changes and climate changes in an integrated manner, it is plausible to conclude that the highlands of the Taita Hills must be prioritized for soil conservation policies during the next 20 years. Although new croplands are likely to be set- tled in lowlands over the next decades, increases in precipitation volumes are expected to be higher in highlands. Moreover, it was demonstrated that in areas with elevated LS factor, typically in the highlands, increasing rainfall will have significant- ly higher impacts on soil erosion potential. Due to the limited availability of non-agricultur- al land in the highlands, new cropland areas are being settled in areas with low precipitation and higher temperatures. The continuity of this trend is likely to drive agricultural lands to areas with a higher IWR, increasing the spatial dependence on distance to rivers and other water bodies. Although the simulated scenarios indicate that climate change will likely increase annual volumes of rainfall during the following decades, IWR will continue to increase due to agricultural expan- sion. By 2030, new cropland areas may cause an increase of approximately 40% in the annual vol- ume of water necessary for irrigation. NOTES This article is a reduced version of the synopsis of the academic dissertation “Agricultural expansion and climate change in the Taita Hills, Kenya: an assess- ment of potential environmental impacts”, University of Helsinki, 2011. ACKNOWLEDGEMENTS The author would like to thank Prof Petri Pellikka for providing scientific supervision during this study. Part of the dataset used in the land-change assessments was provided by Dr Barnaby Clark and Dr Mika Sil- jander. REFERENCES Aeschbacher J, Liniger H & Weingartner R 2005. Riv- er water shortage in a highland–lowland system. Mountain Research and Development 25: 2, 155– 162. Allen RG, Pereira LS, Raes D & Smith M 1998. Crop evapotranspiration − Guidelines for computing crop water requirements. FAO Irrigation and drainage paper 56. FAO, Rome. Almeida CM, Gleriani JM, Castejon EF & Soares-Filho BS 2008. Using neural networks and cellular au-- tomata for modeling intra-urban land use dynam- ics. International Journal of Geographical Infor- mation Science 22: 9, 943−963. Angima SD, Scott DE, O’Neill MK, Ong CK & Weesi- es GK 2003. Soil erosion using RUSLE for central Kenyan highlands conditions. Agricultural Eco- systems and Environment 97: 1−3, 295–308. Bakker MM, Govers G, Jones RA & Rounsevell MDA 2007. The effect of soil erosion on Europe’s crop yields. Ecosystems 10: 7, 1209−1219. Batjes NH & Gicheru P 2004. Soil data derived from SOTER for studies of carbon stocks and change in Kenya (ver. 1.0; GEFSOC Project). Report 2004/01. ISRIC – World Soil Information, Wage- ningen. Beskow S, Mello CR, Norton LD, Curi N, Viola MR & Avanzi JC 2009. Soil erosion prediction in the Grande River Basin, Brazil using distributed mod- elling. Catena 79: 1, 49−59. Blaney HF & Criddle WD 1962. Determining con- sumptive use and irrigation water requirements. USDA Technical Bulletin 1275. US Department of Agriculture, Beltsville. Bonham-Carter G 1994. Geographic information sys- tems for geoscientists: Modelling with GIS. Else- vier Science, Pergamon, Kidlington. Brouwer C & Heibloem M 1986. Irrigation water management. Training Manual 3. FAO, Rome. Burgess ND, Butynski TM, Cordeiro NJ, Doggart NH, Fjeldså J, Howell KM, Kilahama FB, Loader SP, Lovett JC, Mbilinyi B, Menegon M, Moyer DC, Nashanda E, Perkin A, Rovero F, Stanley WT & Stuart SN 2007. The biological importance of the Eastern Arc Mountains of Tanzania and Kenya. Biological Conservation 134: 2, 209−231. Cai J, Liu Y, Lei T & Pereira LS 2007. Estimating refer-- ence evapotranspiration with the FAO Penman- Monteith equation using daily weather forecast 58 FENNIA 190: 1 (2012)Eduardo Eiji Maeda messages. Agricultural and Forest Meteorology 145: 1−2, 22−35. Clark B 2010. Enhanced processing of SPOT multi- spectral satellite imagery for environmental moni- toring and modelling. Doctoral dissertation. Uni- versity of Helsinki, Faculty of Science. Da Silva AM 2004. Rainfall erosivity map for Brazil. Catena 57, 251–259. Douglas EM, Jacobs JM, Sumner DM & Ray RL 2009. A comparison of models for estimating potential evapotranspiration for Florida land cover types. Journal of Hydrology 373: 3−4, 366−376. Frenken K 1997. Irrigation potential in Africa: A ba- sin approach. (FAO Land and Water Bulletin 4) Land and Water Development Division of the Food and Agriculture Organization of the United Nations, Rome. Italy. FAO 2005. AQUASTAT information system on water and agriculture. 10.09.2012. Feng X, Wang Y, Chen L, Fu B & Bai G 2010. Mode- ling soil erosion and its response to land-use change in hilly catchments of the Chinese Loess Plateau. Geomorphology 118: 3−4, 239−248. Fischer G, Tubiello FN, van Velthuizen H & Wiberg DA 2007. Climate change impacts on irrigation water requirements: Effects of mitigation, 1990−2080. Technological Forecasting and So- cial Change 74: 7, 1083−1107. Foley J, de Fries R, Asner GP, Barford C, Bonan G & Carpenter SR 2005. Global consequences of land use. Science 309, 570−574. Gavilán P, Lorite IJ, Tornero S & Berengena J 2006. Regional calibration of Hargreaves equation for es- timating reference ET in a semiarid environment. Agriculture Water Management 81, 257–281. Germer S, Neill C, Krusche AV & Elsenbeer H 2010. Influence of land-use change on near-surface hy- drological processes: Undisturbed forest to pas- ture. Journal of Hydrology 380: 3−4, 473−480. Hargreaves GH & Samani ZA 1985. Reference crop evapotranspiration from temperature. Applied En- gineering in Agriculture 1: 2, 96–99. Istvánovics V 2009. Eutrophication of lakes and res- ervoirs. Encyclopedia of Inland Waters, 157−165. Jabloun M & Sahli A 2008. Evaluation of FAO-56 methodology for estimating reference evapotran- spiration using limited climatic data. Application to Tunisia. Agricultural Water Management 95: 6, 707−715. Jaetzold R & Schmidt H 1983. Farm Management Handbook of Kenya, vol. II. East Kenya. Ministry of Agriculture, Nairobi, Kenya. Kepner WG, Semmens DJ, Bassett SD, Mouat DA & Goodrich DC 2004. Scenario analysis for the San Pedro River, analysing hydrological consequenc- es of a future environment. Environmental Moni- toring and Assessment 94, 115–127. Kinnell PIA 2010. Event soil loss, runoff and the uni- versal soil loss equation family of models: A re- view. Journal of Hydrology 385: 1−4, 384−397. Kirkby MJ & Bracken LJ 2009. Gully processes and gully dynamics. Earth Surface Processes and Landforms 34: 14, 1841−1851. KNBS (Kenya National Bureau of Statistics) 2010. 2009 Population and Housing Census. 10.09.2012 Li Z, Liu W, Zhang X & Zheng F 2009. Impacts of land use change and climate variability on hy- drology in an agricultural catchment on the Loess Plateau of China. Journal of Hydrology 377: 1−2, 35−42. Maeda EE, Clark BJF, Pellikka PKE & Siljander M 2010a. Modelling agricultural expansion in Ken- ya’s eastern arc mountains biodiversity hotspot. Agricultural Systems 103: 9, 609−620. Maeda EE, Pellikka P, Siljander M & Clark BJF 2010b. Potential impacts of agricultural expansion and climate change on soil erosion in the Eastern Arc Mountains of Kenya. Geomorphology 123: 3−4, 279−289. Maeda EE, Wiberg DA & Pellikka PKE 2011a. Estimat- ing reference evapotranspiration using remote sensing and empirical models in a region with limited ground data availability in Kenya. Applied Geography 31: 1, 251−258. Maeda EE, Pellikka PKE, Clark BJF & Siljander M 2011b. Prospective changes in irrigation water re- quirements caused by agricultural expansion and climate changes in the eastern arc mountains of Kenya. Journal of Environmental Management 92: 3, 982−993. Miller SN, Kepner WG, Mehaffey MH, Hernandez H, Miller RC, Goodrich DC, Devonald KK, Heggem DT & Miller WP 2002. Integrating landscape as- sessment and hydrologic modeling for land cover change analysis. Journal of the American Water Resources Association 38, 915−929. Mutua BM, Klik A & Loiskandl W 2006. Modelling soil erosion and sediment yield at a catchment scale: the case of Masinga catchment, Kenya. Land Degradation & Development 17, 557– 570. Myers N, Mittermeier RA, Mittermeier CG, Fonseca GAB & Kent J 2000. Biodiversity hotspots for con- servation priorities. Nature 403: 6772, 853−858. Nakicenovic N, Alcamo J, Davis G, de Vries B, Fen- hann J, Gaffin S, Gregory K, Grübler A, Jung TY, Kram T, Lebre la Rovere E, Michaelis L, Mori S, Morita T, Pepper W, Pitcher H, Price L, Riahi K, Roehrl A, Rogner HH, Sankovski A, Schlesinger M, Shukla P, Smith S, Swart R, van Rooijen S, Vic- tor N & Dadi Z 2000. Emissions scenarios: A spe- cial report of the intergovernmental panel on cli- mate change (IPCC). Cambridge University Press, Cambridge. Narongrit C & Yasuoka Y 2003. The use of Terra- MODIS data for estimating evapotranspiration and its change caused by global warming. Envi- ronmental Informatics Archives 1, 505−511. Nigel R & Rughooputh S 2010. Mapping of monthly soil erosion risk of mainland Mauritius and its ag- FENNIA 190: 1 (2012) 59The future of environmental sustainability in the Taita Hills, … gregation with delineated basins. Geomorpholo- gy 114: 3, 101−114. Oki T & Kanae S 2006. Global hydrological cycles and world water resources. Science 313: 1068, 1068−1072. Pellikka PKE, Lötjönen M, Siljander M & Lens L 2009. Airborne remote sensing of spatiotemporal change (1955−2004) in indigenous and exotic forest cov- er in the Taita Hills, Kenya. International Journal of Applied Earth Observations and Geoinforma- tion 11: 4, 221−232. Rahman MR, Shi ZH & Chongfa C 2009. Soil erosion hazard evaluation − An integrated use of remote sensing, GIS and statistical approaches with bio- physical parameters towards management strate- gies. Ecological Modelling 220: 13−14, 1724−1734. Renard KG & Freimund JR 1994. Using monthly pre- cipitation data to estimate the Rfactor in the re- vised USLE. Journal of Hydrology 157, 287–306. Renard KG, Foster GR, Weesies GA, McCool DK & Yoder DC 1997. Predicting Soil Erosion by Water: A Guide to Conservation Planning with the Re- vised Universal Soil Loss Equation (RUSLE). USDA Agriculture Handbook 703. USDA-ARS, Tucson, USA. Serneels S & Lambin EF 2001. Proximate causes of land-use change in Narok district, Kenya: A spa- tial statistical model. Agriculture, Ecosystems and Environment 85: 1−3, 65−81. Soares-Filho BS, Pennachin CL & Cerqueira G 2002. DINAMICA – a stochastic cellular automata mod- el designed to simulate the landscape dynamics in an Amazonian colonization frontier. Ecological Modelling 154: 3, 217−235. Soares-Filho BS, Garcia RA & Rodrigues HO 2007. Coupling socio economic and demographic dimensions to a spatial simulation model of deforestation for the Brazilian Amazon. Paper presented at the LBA- ECO 11th Science Team Meeting. Soini E 2005. Livelihood capital, strategies and out- comes in the Taita hills of Kenya. ICRAF Working Paper no. 8. World Agroforestry Centre, Nairobi. Terranova O, Antronico L, Coscarelli R & Iaquinta P 2009. Soil erosion risk scenarios in the Mediter-- ranean environment using RUSLE and GIS: An ap- plication model for Calabria (southern Italy). Geo- morphology 112: 3−4, 228−245. Thornthwaite CW 1948. An approach toward a ra- tional classification of climate. Geographical Re- view 38, 55–94. Tilman D, Fargione J, Wolff B, D’Antonio C, Dobson A, Howarth R, Schindler D, Schlesinger WH, Sim- berloff D & Swackhamer D 2001. Forecasting ag- riculturally driven global environmental change. Science 292, 281–284. UN (United Nations) 2010. World Population Pros- pects: The 2010 Revision Population Database. 10.09.2012. Van Oost K, Govers G & Desmet P 2000. Evaluating the effects of changes in landscape structure on soil erosion by water and tillage. Landscape Ecol- ogy 15: 6, 577−589. Viviroli D & Weingartner R 2008. “Water Towers” − A global view of the hydrological importance of mountains. In Wiegandt E (ed). Mountains: Sourc- es of water, Sources of Knowledge. Springer, Dor- drecht, The Netherlands. Wang K, Wan Z, Wang P, Sparrow M, Liu J, Zhou X & Haginoya S 2005. Estimation of surface long wave radiation and broadband emissivity using moder- ate resolution imaging spectroradiometer (MODIS) land surface temperature/emissivity products. Journal of Geophysical Research 110, D11109. Williams JR & Renard KG 1983. EPIC: a new method for assessing erosion’s effects on soil productivity. Journal of Soil and Water Conservation 38, 381– 383 Wischmeier WH & Smith DD 1978. Predicting rain- fall erosion losses. A guide to conservation plan- ning. Agricultural Handbook 537. United States Department of Agriculture, Washington, DC. Xiaodan W, Xianghao Z & Jianrong F 2004. Assess- ment and spatial distribution of sensitivity of soil erosion in Tibet. Journal of Geographical Sciences 14: 1, 41−46. Yang D, Kanae S, Oki T, Koike T & Musiake K 2003. Global potential soil erosion with reference to land use and climate changes. Hydrological Proc- esses 17: 14, 2913−2928.