Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=tjom20
Journal of Maps
ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tjom20
Biophysical regions of the Southern Highlands,
Tanzania: regionalization in a data scarce
environment with open geospatial data and
statistical methods
Joni Koskikala , Danielson Kisanga & Niina Käyhkö
To cite this article: Joni Koskikala , Danielson Kisanga & Niina Käyhkö (2020) Biophysical
regions of the Southern Highlands, Tanzania: regionalization in a data scarce environment
with open geospatial data and statistical methods, Journal of Maps, 16:2, 376-387, DOI:
10.1080/17445647.2020.1761061
To link to this article:  https://doi.org/10.1080/17445647.2020.1761061
© 2020 The Author(s). Published by Informa
UK Limited, trading as Taylor & Francis
Group on behalf of Journal of Maps
View supplementary material 
Published online: 12 May 2020. Submit your article to this journal 
Article views: 995 View related articles 
View Crossmark data
Science
Biophysical regions of the Southern Highlands, Tanzania: regionalization in a
data scarce environment with open geospatial data and statistical methods
Joni Koskikala a, Danielson Kisangab and Niina Käyhköa
aDepartment of Geography and Geology, University of Turku, Turku, Finland; bDepartment of Geography, University of Dar es Salaam, Dar es
Salaam, Tanzania
ABSTRACT
Spatially explicit, evidence-based and regionally contextualized data on biophysical landscape
characteristics is an essential basis for regionally sustainable landscape management schemes.
In many regions of the Global South, the availability of such information is poor, especially at the
subnational level, and the spatial management is often based on generic and outdated
information, leading to severe threats for land sustainability. We have developed a
biophysical regionalization of the Southern Highlands area of Tanzania. The map is based on
open-source global datasets depicting climate, soil, topography and vegetation. Through
replicable statistical and geospatial analyses, we have identified 7 regions and 18 subsections
with biophysically similar and spatially distinctive environmental conditions. The regions
provide spatially contextualized support for understanding and managing the landscapes of
the Southern Highlands. The applications for such data sets are numerous, from screening
suitability areas for e.g. afforestation schemes to evaluating the distinctiveness and
vulnerability of landscapes to degradation.
KEYWORDS
Data scarcity; open data;
global South; Africa;
regionalization; K-means
1. Introduction
Biophysical regionalization methods target the identifi-
cation of homogeneous biophysical land characteristics
for different land assessment needs (Bailey, 2004; Blasi
et al., 2014; Leathwick et al., 2003; Omernik & Griffith,
2014; Smiraglia et al., 2013). The regionalization meth-
odology relies on a spatial hierarchy of biophysical fac-
tors where macroclimate determines the broad
ecological units, and then landforms, soil and veg-
etation steer the division of the units to more detailed
level regions of particular biophysical character (Bailey,
2004; Klijn & Udo de Haes, 1994). The value of the
regionalizations lies in their ability to contextualize bio-
physical variability over geographical space. Thus,
regionalizations bring an additional layer of geospatial
data into a complex decision-making process of esti-
mating land capacity, resilience, suitability or vulner-
ability (Gallant et al., 2004; Loveland et al., 2002;
Nowak & Schneider, 2017; Sleeter et al., 2013). With
the help of semantically and spatially accurate regiona-
lizations, it is possible, for example, to estimate how
intensive or extensive are the consequences of land
use changes in space and time, or what type of services
and benefits landscapes provide to humans (Burkhard
et al., 2009; Fang et al., 2015; Willemen et al., 2008;
Zomer et al., 2013). When linked into the overall pro-
cesses of land assessment, regionalizations increase the
reliability and practical value of assessments and help
establish meaningful region-specific decisions.
During the past decade, open-access geospatial data-
sets of high spatial accuracy and with global coverage
have become widely available, allowing mapping and
modeling of environmental factors at rather detailed
scales in most parts of the world (Metzger et al.,
2013b; Mücher et al., 2016; Sayre et al., 2014). Further-
more, environmental satellite image repositories with
abundant and up-to-date image data and access to
increased computing power have created new opportu-
nities for the creation of biophysical mapping based on
quantitative and repeatable methods (Coops et al.,
2009; Song et al., 2017; Xiong et al., 2017). This has
changed the ways in which data-scarce regions in the
Global South can be studied in terms of their biophysi-
cal environment (Egoh et al., 2012; Vrebos et al., 2015).
However, the applicability of global data sets directly at
the regional level is not straightforward, since the strata
of the data sets has been designed for global coverage
and may miss essential regional variation. To overcome
this type of semantic mismatch, the data sets need
rescaling for regional needs.
In Sub-Saharan Africa, sustainable land use plan-
ning and management of land resources is challenged
due to severe degradation of the environment, rapid
population growth and urbanization, combined with
© 2020 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group on behalf of Journal of Maps
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted
use, distribution, and reproduction in any medium, provided the original work is properly cited.
CONTACT Joni Koskikala jonkosk@utu.fi Geography Section, Department of Geography and Geology, University of Turku, Turku 20014, Finland
Supplemental data for this article can be accessed at https://doi.org/10.1080/17445647.2020.1761061
JOURNAL OF MAPS
2020, VOL. 16, NO. 2, 376–387
https://doi.org/10.1080/17445647.2020.1761061
high expectations on land productivity for better food
security and energy demands of the growing cities
(Fisher, 2010; Nkonya et al., 2016; Parnell &Walawege,
2011; Salami et al., 2010). The area of the Southern
Highlands in Tanzania is a good example of Eastern
African rapidly developing landscape regions, where
people live on the land in diverse and fragmented set-
tings of montane forests, grasslands, woodlands, bush-
lands and agricultural land. The area provides
Tanzania with some of its most critical elements of
food security and especially timber (NBS, 2015,
2016). However, the landscapes of the region are threa-
tened by deforestation of montane and miombo wood-
lands, land degradation, soil erosion and seasonal water
scarcity (Kangalawe & Lyimo, 2010). At the same time,
region-wide land improvement schemes both in the
sector of agriculture and forestry have been promoted
and expectations are set high in terms of their future
development opportunities for productivity and well-
being (Milder et al., 2013). However, the region lacks
data of the variation of biophysical environment and
thus any strategic planning, which could address the
overall land capacity for ecosystem service provisioning
and steer and set limitations and possibilities for land
use improvement schemes, is missing.
In this study, we have developed a biophysical regio-
nalization methodology for the area of the Southern
Highlands in Tanzania, based on open-source geospa-
tial data sets of global and regional coverage. The
method is data-driven and repeatable. The outcome
of the methodology is a map of biophysical regions,
which visualizes geographical differences in biophysical
conditions of climate, topography, soil and vegetation
in the Southern Highlands. The map can be used to
steer land suitability, risk and vulnerability assessments
based on spatially explicit cell-based models. The
suggested methodology can be used outside our study
area, but since the methodology is based on data-dri-
ven statistics, its transfer to other areas needs empirical
adjustment in terms of selected variables.
2. Material and methods
2.1. Study area
The Southern Highlands is a socioeconomically valu-
able area in the regions of Mbeya, Iringa and Njombe
with 4.3 million inhabitants (NBS, 2013, Figure 1).
Despite the region being among the richest in Tanza-
nia, over 60% of the population is facing poverty
(UNDP, 2015). Small-scale, low efficiency agriculture
is the main economic activity, with minor cash crop-
ping, livestock and beekeeping and tree planting sup-
plementing local economies. Some 90% of the rural
Figure 1. Southern Highlands is a geographically diverse landscape area lying in between 7.3°S and 10°S and 33.3°E and 36.3°E. It is
accessible from East through the main roads from Morogoro and Songea and from the West through Mbeya. Largest urban centers
in addition to main cities of Iringa and Mbeya are Makambako, Njombe and Mafinga. The land cover is dominated by open and
closed woodlands, bushland and cultivated woodland (Naforma land cover/land use map, MNRT, 2013).
JOURNAL OF MAPS 377
population depends on wood fuel for daily energy
supply and collects wood extensively also for con-
struction and charcoal production from natural and
planted woodlands and forests (Kangalawe & Lyimo,
2010). Population growth accompanied with exten-
sive, low capacity agriculture and high dependency
on wood resources increase the demand and pressure
for natural resource extraction (Kangalawe, 2012;
Kangalawe & Lyimo, 2010). These pose a challenge
for sustainable land and natural resource manage-
ment, especially since the region is already suffering
from deforestation, land degradation, soil erosion
and seasonal water scarcity (Green et al., 2013;
Kajembe et al., 2003; Malley et al., 2009; Sawe et al.,
2014; Schaafsma et al., 2012).
Today, the Southern Highlands is the most impor-
tant production area of grains and potatoes, and its
southern and eastern parts are the main source of tim-
ber in Tanzania (Kangalawe, 2012). Tanzania has
ambitious development plans for the area as a future
breadbasket and key timber production region for
the country (Milder et al., 2013; PFP, 2016). The gov-
ernment of Tanzania is currently promoting sustain-
able private forestry to facilitate development and
alleviate poverty (PFP, 2016). However, plantation
establishment, for example, is not leaning on spatially
explicit knowledge of biophysical land conditions
since the available information is either completely
lacking or is too generic or outdated (De Pauw,
1984; MNRT, 2013). The Southern Highlands needs
strategic, science-based, and landscape-level planning
practices, as seen in many other resource-rich, exten-
sively used and rapidly developing landscapes in the
world.
2.2. Statistical approach for regionalization
Three important issues to consider in developing
robust statistical regionalization processes are: (1)
the study area extent, (2) the selection of variables
and (3) the selection of spatial resolution of the regio-
nalization, also known as the scale issue in Modifiable
Areal Unit Problem (MAUP) (Bakkestuen et al., 2008,
Dark & Bram, 2007; Jelinski & Wu 1996; Metzger
et al., 2005; Wu et al., 2006). In our study, we designed
the regionalization only to the Southern Highlands of
Tanzania and the demarcations are thus relevant for
the Southern Highlands and may change if the extent
is changed. For variable selection, we reviewed pre-
vious regional and global scale biophysical regionali-
zation studies (Bakkestuen et al., 2008; Chuman &
Romportl, 2010; Fairbanks & Benn 2000; Metzger
et al., 2013b; Serra et al., 2011) and selected openly
available global and regional scale spatial data sets,
depicting climate, topography, soil and vegetation
conditions (Table 1). To evaluate the effect of
MAUP and to develop a robust spatial representation
of the biophysical variation within the study area, we
developed the regionalization with 1, 5 and 10 km
grid cells.
Table 1. Biophysical variables consist of climate (temperature and precipitation related variables), topography (altitude, slope,
terrain ruggedness), soil (texture, structure, density), and vegetation (intensity and seasonality), obtained from open access
global data sets.
Dataset Derived variable
Climate, Temperature (CT)
Climate, Precipitation (CP)
Source: Worldclim2
Spatial resolution 1km
CT1) Annual Mean Temp, CT2) Mean Diurnal Range,CT3) Isothermality,
CT4) Temp seasonality, CT5) Max Temp of Warmest Month,
CT6) Min Temp of Coldest Month, CT7) Temp Annual Range,
CT8) Mean Temp of Wettest Quarter, CT9) Mean Temp of Driest Quarter,
CT10) Mean Temp of Warmest Quarter,
CT11) Mean Temp of Coldest Quarter, CP1) Annual Prec,
CP2) Prec of Wettest Month, CP3) Prec of Driest Month,
CP4) Prec Seasonality, CP5) Prec of Wettest Quarter,
CP6) Prec of Driest Quarter, CP7) Prec of Warmest Quarter,
CP8) Prec of Coldest Quarter, CP9) Global aridity index,
CT12) Global reference Evapotranspiration, CP10) Growth days
Topography (T)
Source: NASA/SRTMGL1v003
Spatial resolution 30 m
T1) Mean altitude, T2) STD altitude, T3) Mean slope, T4) STD slope,
T5) Terrain ruggedness index
Soils (S)
AfSoilGrid (ISRIC)
Spatial resolution 250 m
S1) Organic carbon, S2) pH, S3) Sand fraction, S4) Silt fraction,
S5) Clay fraction, S6) Coarse fragments, S7) Bulk density,
S8) Cation exchange capacity, S9) Total N, S10) Exchangeable acidity,
S11) Al content, S12) Exchangeable Ca, S13) Exchangeable K,
S14) Exchangeable Mg, S15) Exchangeable Na
Vegetation intensity (VI)
Vegetation seasonality (VS)
MODIS/MOD13Q NDVI, EVI, NDWI 16-day composite and
MOD11A2 LST 8-day composite Spatial resolution 250 m–1 km
V1) Mean annual NDVI, V2) STD of annual NDVI, V3) Mean annual
EVI, V4) STD of annual EVI, V5) Mean annual NDWI,
V6) STD of annual NDWI, V7) Mean annual LST, V8) STD of annual LST
378 J. KOSKIKALA ET AL.
2.3. Creating geodatabase with biophysical
variables
We acquired bioclimatic variables related to tempera-
ture and precipitation from theWorldClim 2 global cli-
mate dataset (Fick & Hijmans, 2017). WorldClim
climate variables are especially useful when local
weather station data is not available (Li et al., 2017).
In addition to the 19 bioclimatic variables available
from WordClim2 database, we downloaded Global
Potential Evapotranspiration (Global-PET) and Global
Aridity Index (Global-Aridity) layers from CGIAR
Consortium for Spatial Information (CGIAR-CSI)
website (https://cgiarcsi.community/data/). Both Glo-
bal-PET and Global-Aridity datasets are modeled
based on WorldClim2 data (Trabucco & Zomer,
2019) and have been shown previously to be important
variables in biophysical regionalizations (Metzger et al.,
2013b). Furthermore, we calculated ‘number of annual
growth days’, as this has been shown to be an impor-
tant variable in South Africa, depicting the soil water
balance (see Fairbanks & Benn, 2000).
For topographical variables, we used National
Aeronautics and Space Administration’s (NASA)
Shuttle Radar Topography Mission digital elevation
model (SRTM DEM) dataset at 30 m resolution,
accessed through Google Earth Engine (GEE) (Google
Earth Engine Team, 2015; Jarvis et al., 2008). We cal-
culated mean altitude, STD altitude, mean slope, STD
slope and terrain ruggedness index, referring to vari-
ables found suitable to depict topographical differ-
ences in previous regionalization studies
(Bakkestuen et al., 2008; Chuman & Romportl,
2010; Serra et al., 2011).
We obtained soil variables from International Soil
Reference and Information Centre (ISRIC) AfSoil-
Grids250 m database (see details in Hengl et al.,
2015). Soil surface (depth 5–15 cm and 0–20 cm) char-
acter grids of texture, structure and density were
extracted from the database.
We extracted Normalized Difference Vegetation
Index (NDVI), Enhanced Vegetation index (EVI),
Normalized Difference Water Index (NDWI) and
Land Surface Temperature (LST) from GEE MODIS
data collection (Google Earth Engine Team, 2015).
These indices are frequently used in landscape classifi-
cation studies as they depict vegetation condition,
water content and seasonal changes (Gao, 1996; Jack-
son et al., 2004). These indices provide composites
with 8-day and 16-day intervals adding up to 22–43
composites annually with 250–1000 m spatial resol-
ution. We also calculated annual (2014) mean and
annual standard deviations of these composites.
Finally, we resampled all variables and summarized
them as means into 1, 5 and 10 km grid cells superim-
posed over the study area and stored into a
geodatabase.
2.4. Establishing biophysical gradients through
PCA
We conducted Principal Components Analysis (PCA)
in R statistics (package psych (Revelle, 2018)) based
on the original variables in order to reduce information
redundancy between the variables. Following the Kai-
ser-Guttman method (Jackson, 1993), we extracted
the principal components with greater than average
eigenvalue (eigenvalue greater than one). According
to the correlations of the 50 variables (see supplemen-
tary material for further details), we then identified the
components’ representative biophysical gradients. We
found there to be seven (7) relevant biophysical gradi-
ents representing precipitation, temperature, topogra-
phical variation, soil fertility, vegetation intensity,
vegetation seasonality and temperature variation
(Figure 2). These explain 85.6% to 88.4% of the total
variance of the data depending on the grid cell size
used (Table 2).
2.5. K-means++ clustering and identification of
biophysical regions
Through statistical clustering of the extracted com-
ponents, we formed areas with similar biophysical
characteristics. We used k-means clustering method
for its suitability for post-PCA classification (Ding &
He, 2004) and its wide application in regionalization
studies (Guitet et al., 2013; Hargrove & Hoffman,
2005; Soto & Pintó, 2010; White et al., 2005; Zhang
et al., 2012). K-means algorithm divides the data into
a set of homogenous clusters by analyzing cell value
distances in each individual cell to cluster value centers
in Euclidean space and assigning the pixels to the clo-
sest cluster. New centroids are calculated for each clus-
ter after each iteration, and pixels are assigned again to
the closest cluster. The iterative algorithm continues
until thresholds of iteration are met. We applied k-
means++ algorithm (Arthur & Vassilvitskii, 2007), an
augmentation of the k-means algorithm where the ran-
dom seeding process of ordinary k-means is replaced
by careful seeding. K-means++ has been shown to pro-
duce more robust clusters compared to k-means
(Arthur & Vassilvitskii, 2007, Zhang et al. 2012). The
clustering was run with 100 initial seeding points and
k-values ranging from 2 to 30. We used package
LICORS (Goerg, 2013) in R statistics.
After clustering, we analyzed the explained variance
of each cluster representation in order to evaluate the
optimal number of clusters to depict the biophysical
regions. Compared to the approaches using a fixed
number of clusters (e.g. Metzger et al., 2013b) or spatial
aggregation of data sets (e.g. Sayre et al., 2014), the
data-driven approach enables statistical evaluation of
the significance of the clusters and their explained var-
iances and allows for contextually sensitive cluster
JOURNAL OF MAPS 379
representation. We plotted the total sum of squares
between clusters for each of the 29 regionalization
approaches and the optimal number of clusters was
chosen visually (see Supplementary). Furthermore,
we tested spatial robustness of the clustering approach
by calculating spatial goodness of fit (GOF) between
the consequent cluster maps using the Mapcurves
function (Hargrove et al., 2006) (package sabre (Now-
osad & Stepinski, 2018)). The Mapcurves method is not
sensitive for the varying amount of categories between
the compared maps. We calculated the GOF statistics
for map comparisons following the equation given by
(Hargrove et al., 2006):
GOF =
∑ C
B+ C ×
C
A+ C
( )
,
Where GOF is the goodness-of-fit of a cluster, A is the
map to be compared, B is the reference map and C is
the overlap between A and B.
We analyzed hierarchical relationships between
clusters with Ward’s method (Ward, 1963) (package
stats (R Core Team, 2019)). The hierarchy of clusters
was based on the squared Euclidean distance of each
cluster centroid in relation to the biophysical com-
ponents. We organized cluster distances as a
dissimilarity matrix and drew in a dendrogram that
shows the scaled distance of each cluster and their hier-
archy (Supplementary material). For characterizing the
formed homogeneous clusters into biophysical regions,
we gathered descriptive information on the clusters on
the basis of variables that had high loadings on the
principal components.
3. Results
3.1. Biophysical regions in the Southern
Highlands
The Southern Highlands comprises seven (7) main bio-
physical regions (A–G) and 18 subsections (Main
map). Each region encompasses distinct biophysical
character, as shown in Figure 3. Region A is character-
ized by high vegetation intensity and low seasonality in
vegetation cover, depicting the evergreen montane for-
ests of Eastern Arcs and Southern Rift montane forest-
grassland mosaic in the North East and South West,
respectively. Region B depicts the fertile flatlands of
the Usangu plains with high seasonality in vegetation
cover. Region C encapsulates the rugged, topographi-
cally variating landscapes, while region D is
Figure 2. Seven biophysical gradients derived from principal components of the 50 original variables at 1 km resolution.
380 J. KOSKIKALA ET AL.
characterized by poor soils and high vegetation season-
ality typical in wooded savannas developed on poor
soils of the central African plateau. Region E encapsu-
lates the moist and hot flatlands of the Kilombero val-
ley with high temperature variation caused by the
adjacent Udzungwa scarp. Region F encapsulates the
very moist flatland of the Kyela valley with a precipi-
tation pattern governed by Lake Nyasa. The region is
characterized by high values in precipitation and temp-
erature gradients and low values in temperature vari-
ation gradient i.e. stable warm and moist climate.
Region G depicts the highland range with low annual
average temperatures and high temperature variation.
The subsections are specifications of each main bio-
physical region and they provide more spatially
detailed stratification of the biophysical environment
Table 2. Principal components 1–7, their eigenvalues, cumulative variance, strongest and weakest correlative variables (high
loadings, low loadings) and gradient name based on correlative variables. For all variable loadings, see Supplementary material.
PC Eig. Tot var (%) High loadings Low loadings Gradient Name
1 km Resolution
1 13 26 CP1, CP9, CP10 CP4 Precipitation
2 10.8 47.6 CT1, CT10, CT9 T1 Temperature
3 5.3 58.2 T2, T3, T5 Topography
4 4.9 67.9 S13, S12, S14 Soil Fertility
5 4.1 76.1 VI1, VI2, VI3 CT7 Vegetation Intensity
6 3 82.1 VS3, VS2, VS1 Vegetation Seasonality
7 1.7 85.6 CT3, CT2 Temperature variation
5 km Resolution
1 14.8 29.6 CP1, CP9, CP10 CP4 Precipitation
2 11.1 51.7 CT10, CT1, CT9 T1 Temperature
3 5.2 62.1 T2, T4, T5 Topography
4 5 72.1 S13, S12, S8 Soil Fertility
5 3.2 78.4 VI1, VI2 CT7, CT2 Vegetation Intensity
6 2.9 84.2 VS3, VS2, VS1 Vegetation Seasonality
7 1.8 87.7 CT3 Temperature variation
10 km Resolution
1 14.3 28.6 CP9, CP5, CP2 S3, CP4 Precipitation
2 11 50.6 CT10, CT1, CT8 T1 Temperature
3 5.6 61.8 T4, T2, T5 Topography
4 5 71.9 S13, S12, S8 Soil Fertility
5 4.2 80.3 VI1, VI2 VS1, VS2 Vegetation Intensity
6 2.3 84.8 CT2, CT7 Vegetation Seasonality
7 1.8 88.4 CT3 Temperature variation
Figure 3. Biophysical regions presented according to the median values of selected biophysical variables. Prec = Yearly precipi-
tation (mm), Temp = Average temperature (°C), Std. Altitude = Standard deviation of altitude (m/km2), CEC = Cation Exchange
Capacity (cmolc/kg), NDVI = yearly average NDVI, Std. NDVI = Standard deviation of NDVI throughout a year, Temp Variation =
Mean Diurnal Range of temperature (Mean of monthly (max temp – min temp)).
JOURNAL OF MAPS 381
(Figure 4). For example, the region E consists of three
subsections (E1–E3) (Figure 4, graphs 4 and 5). The
subsections differ mostly on their topography and veg-
etation characteristics (NDVI and Std. of NDVI) with
E1 having flat terrain and moderate vegetation inten-
sity and seasonality while E2 has high topographic vari-
ation and closed and more evergreen vegetation
structure.
The created biophysical regionalization map of the
Southern Highlands is freely available at PANGAE
(https://doi.pangaea.de/10.1594/PANGAEA.909589).
4. Discussion
Our mapping of the biophysical regions shows the
practical value of open access global and continental
Figure 4. Biophysical subsections presented according to the median values of selected biophysical variables within a subsection. P
= Yearly Precipitation (mm), T = Average temperature (°C), Std. Alt = Standard deviation of altitude (m/km2), CEC = Cation
Exchange Capacity (cmolc/kg), NDVI = yearly average NDVI, Std. NDVI = Standard deviation of NDVI throughout a year, TV =
Mean Diurnal Range of temperature (Mean of monthly (max temp – min temp)).
382 J. KOSKIKALA ET AL.
data sets for regional scale mapping efforts. The 7 bio-
physical regions and 18 subsections of the Southern
Highlands were able to depict regionally important
landscape areas such as the Usangu plains and Kilo-
mbero valley, and identify essential geographical differ-
ences in biophysical conditions over a relatively large
geographical area with a data-driven statistical method.
The map of biophysical regions will enhance possibili-
ties for regional land use planning and strategic level
discussions, since it differentiates the region into
unique biophysical areas, which previously have not
existed in a map form, and as a digital database (Gal-
lant et al., 2004; Loveland & Merchant, 2004; Metzger
et al., 2013a). The map has many other application pos-
sibilities. It can be used, for example when planning
suitable target areas for agricultural and forestry
schemes, which will have a major impact on the success
of the region’s food security and sustainable forest
management (Mücher et al., 2016; Milder et al., 2013;
Nijbroek & Andelman 2015; Williams et al., 2008).
The biophysical map is a major improvement, consid-
ering that at present, the incentive schemes rely on lim-
ited knowledge, like rainfall (PFP, 2016) or general
environmental classifications (De Pauw, 1984). The
map can also have an important role in targeting future
discussions of the risks posed by climate change and
land degradation to different regions in the Southern
Highlands (Zomer et al., 2013).
The additional value of the biophysical map is that
as a digital database, it can be analyzed in relation to
other spatial data of the region for improved under-
standing. For example, semantic accuracy and rel-
evance of land use/land cover (LULC) data sets can
be enriched with the help of biophysical information
(Auch et al., 2011; Bryce et al., 1999; Sleeter et al.,
2013). Assessment of LULC patterns in relation to
underlying biophysical context, can help to assess
which features of the current land cover (e.g forests)
are located in a unique biophysical set-up and how
rare or abundant are certain critical land assets in
each region (Martínez-Harms et al., 2016; Maselli
et al., 2009). Thus, the biophysical regionalization can
be used to enrich the available geospatial data sets.
This is especially vital in Southern Highlands and Tan-
zania in general, where the generic LULC map demar-
cations are not capable of depicting the multifunctional
land use patterns. The enriching process is inherently
spatial since the biophysical conditions are homo-
geneous and temporally relatively constant within the
designated regions.
Despite the advantages of the statistical approach in
creating objectively homogeneous biophysical regions,
there are still uncertainties and subjectivity in the
choice of the input variables. We chose the input vari-
ables based on their relevance in depicting biophysical
characters (climate, topography, soil and vegetation)
according to previous studies (Bailey, 2004; Bakkestuen
et al., 2008). Most of the previous regionalizations have
used land use data and socio-economic data as input
variables (Bernert et al., 1997; Chuman & Romportl,
2010; Coops et al., 2009) but we decided to leave
LULC data out from the regionalization since the avail-
able data from Tanzania is spatially generic and could
steer the regionalization. Furthermore, the spatial accu-
racy of the global and continental input variables at
regional scale may vary (Li et al., 2017) but consider-
ations of the effect of the data quality on the regionali-
zation was not included in our study since they are the
only available data sets from the area.
The number of meaningful homogeneous regions
and their aggregation remains the most uncertain
issue in the regionalization process (Metzger et al.,
2005; Metzger et al., 2013b; Serra et al., 2011; Zhang
et al., 2012). For a quantitative and statistical approach,
this is more important than for qualitative approaches,
since the approach is data-driven and thus dependent
on the quality of the input variables and the geographi-
cal extent of the study. In our case, the biophysical
regions and subsections are relative to the area of the
Southern Highlands and consequently, the variances
of different input variables and the suitable number
of clusters are regionally restricted. However, the clus-
tering method provided spatially robust delineations of
homogeneous biophysical areas and repetition of the
study would result in similar demarcations.
Another crucial step to consider when the regiona-
lizations are based on GIS and data sets derived from
remote sensing is the scale issue in MAUP (Dark &
Bram 2007). The optimal solution of scale is dependent
on the application of the regionalization product. In
our case, the smallest common scale was 1 km, gov-
erned by the climate data. Nevertheless, the order
and loadings of biophysical gradients, derived from
the principal components of the original variables
were consistent in 1, 5 and 10 km resolutions. This
indicates that with the range of 1 km to 10 km the
regionalization is not scale-dependent and the aggrega-
tion MAUP scale problem is not present. On the other
hand, the DEM is available at higher resolution com-
pared to the other biophysical variables and requires
significant upscaling from 30 m to 1 km. Thus, some
of the topographical variance may have already been
averaged out on 1 km resolution and the MAUP aggre-
gation problem is not present when upscaling from 1
km resolution. Applications targeted for local scale
should focus on the spatial hierarchy of different bio-
physical factors (Bailey, 2004; Klijn & Udo de Haes,
1994) and seek ways to combine various datasets
with respect to their scale of spatial variance within
the landscape.
Tanzania, like many other rapidly growing regions
in the world, faces concrete impacts of land degra-
dation due to unsustainable use of lands, increasing
demand for natural resources of the growing
JOURNAL OF MAPS 383
population and changes in climate patterns (Fisher,
2010; Lambin & Meyfroidt, 2011; Nkonya et al.,
2016; Parnell & Walawege, 2011). These challenges
require practical solutions to improve planning and
management of natural resources. Spatially explicit,
real data based assessments are needed at relevant con-
text and resolution to support sustainable regional
strategies and planning processes. We have shown
through our case study that biophysical regionalization
can be statistically established and rescaled to regional
context using open access continental and global cover-
age data sets. Such information reflects the land poten-
tial and capacities and offers valuable information to
assess and steer human actions, particularly in the
rapidly developing areas of the Global South, where
global open access data sets may provide an essential
source of spatial information to create regional scale
landscape assessments (Lu et al., 2015; Sarvajayakesa-
valu, 2015). Furthermore, such information supports
evidence-based decision-making, which is one of the
most critical bottlenecks for sustainable land manage-
ment in rapidly developing countries (Sarvajayakesa-
valu, 2015).
Software
All the geospatial processing steps and map viaualisa-
tions were performed with QGIS 3.4. All the statistical
analysis were performed with R Statistics version 3.6.1.
The spider diagrams were generated with Python 3.6
and modified with Inkscape version 0.92.
Acknowledgements
The authors wish to thank the University of Turku, Depart-
ment of Geography and Geology and the UTU Geospatial
Labs (geospatial.utu.fi), and the University of Dar Es Salaam,
Department of Geography for facilitating this research. The
research was funded by the Academy of Finland (project:
Sustainability, scale relations and structure–function-
benefit chains in the landscape systems of the Tanzanian
Southern Highlands (SUSLAND), 276126).
Disclosure statement
No potential conflict of interest was reported by the author(s).
ORCID
Joni Koskikala http://orcid.org/0000-0001-8655-8424
References
Arthur, D., & Vassilvitskii, S. (2007, January 7–9). k-means+
+: the advantages of careful seeding. Proceedings of the
eighteenth annual ACM-SIAM symposium on discrete
algorithms (pp. 1027–1035). New Orleans, LA: Society
for Industrial and Applied Mathematics.
Auch, R. F., Sayler, K., Napton, D. E., Taylor, J. L., & Brooks,
M. S. (2011). Ecoregional differences in late 20th century
land use and land-cover change in the U.S Northern great
plains. Great Plains Research: A Journal of Natural and
Social Sciences, 1190. http://digitalcommons.unl.edu/
greatplainsresearch/1190
Bailey, R. (2004). Identifying ecoregion boundaries.
Environmental Management, 34(S1), S14–S26. https://
doi.org/10.1007/s00267-003-0163-6
Bakkestuen, V., Erikstad, L., & Halvorsen, R. (2008). Step-
less models for regional environmental variation in
Norway. Journal of Biogeography, 35(10), 1906–1922.
https://doi.org/10.1111/j.1365-2699.2008.01941.x
Bernert, J. A., Eilers, J. M., Sullivan, T. J., Freemark, K. E.,
& Ribic, C. (1997). A quantitative method for delineat-
ing regions: An example for the western Corn Belt
plains ecoregion of the USA. Environmental
Management, 21(3), 405–420. https://doi.org/10.1007/
s002679900038
Blasi, C., Capotorti, G., Copiz, R., Guida, D., Mollo, B.,
Smiraglia, D., & Zavattero, L. (2014). Classification and
mapping of the ecoregions of Italy. Plant Biosystems -
An International Journal Dealing with all Aspects of
Plant Biology, 148(6), 1255–1345. https://doi.org/10.
1080/11263504.2014.985756
Bryce, S. A., Omernik, J. M., & Larsen, D. P. (1999).
Ecoregions: A geographic framework to guide risk charac-
terization and ecosystem management. Environmental
Practice, 1(3), 141–155. https://doi.org/10.1017/S1466046
600000582
Burkhard, B., Kroll, F., Muller, F., & Winfhorst, W. (2009).
Landscapes’ capacities to provide ecosystem services – a
concept for land-cover based assessments. Landscape
Online, 15, 1–22. https://doi.org/10.3097/LO.200915
Chuman, T., & Romportl, D. (2010). Multivariate classifi-
cation analysis of cultural landscapes: An example from
the Czech Republic. Landscape and Urban Planning, 98
(3–4), 200–209. https://doi.org/10.1016/j.landurbplan.
2010.08.003
Coops, N. C., Wulder, M. A., & Iwanicka, D. (2009). An
environmental domain classification of Canada using
earth observation data for biodiversity assessment.
Ecological Informatics, 4(1), 8–22. https://doi.org/10.
1016/j.ecoinf.2008.09.005
Dark, S. J., & Bram, D. (2007). The modifiable areal unit pro-
blem (MAUP) in physical geography. Progress in Physical
Geography: Earth and Environment, 31(5), 471–479.
https://doi.org/10.1177/0309133307083294
De Pauw, E. (1984). Consultant’s final report on the soils,
physiography and agroecological zones of Tanzania. FAO.
Ding, C., & He, X. (2004, July 4–8). K -means clustering via
principal component analysis. ICML ‘04 Proceedings of
the twenty-first international conference on Machine
learning, Banff, Canada, (p. 29).
Egoh, B., Drakou, E. G., Dunbar, M. B., Maes, J., &
Willemen, L. (2012). Indicators for mapping ecosystem ser-
vices: A review (Report EUR25456EN). Publications Office
of the European Union.
Fairbanks, D. H. K., & Benn, G. A. (2000). Identifying
regional landscapes for conservation planning: A case
study from KwaZulu-Natal, South Africa. Landscape
and Urban Planning, 50(4), 237–257. https://doi.org/10.
1016/S0169-2046(00)00068-2
Fang, X., Zhao, W., Fu, B., & Ding, J. (2015). Landscape ser-
vice capability, landscape service flow and landscape ser-
vice demand: A new framework for landscape services
and its use for landscape sustainability assessment.
384 J. KOSKIKALA ET AL.
Progress in Physical Geography: Earth and Environment,
39(6), 817–836. https://doi.org/10.1177/030913331561
3019
Fick, S. E., & Hijmans, R. J. (2017). Worldclim 2: New 1-km
spatial resolution climate surfacesfor global land areas.
International Journal of Climatology, 37(12), 4302–4315.
https://doi.org/10.1002/joc.5086
Fisher, B. (2010). African exception to drivers of deforesta-
tion. Nature Geoscience, 3(6), 375–376. https://doi.org/
10.1038/ngeo873
Gallant, A., Loveland, T., Sohl, T., & Napton, D. (2004).
Using an ecoregion framework to analyze land-cover
and land-use dynamics. Environmental Management, 34
(S1), S89–S110. https://doi.org/10.1007/s00267-003-
0145-3
Gao, B. (1996). NDWI-a normalized difference water index
for remote sensing of vegetation liquid water from
space. Remote Sensing of Environment, 58(3), 257–266.
https://doi.org/10.1016/S0034-4257(96)00067-3
Goerg, G. M. (2013). LICORS: Light cone reconstruction of
states – predictive state estimation from spatio-temporal
data. R package version 0.2.0. https://CRAN.R-project.
org/package=LICORS
Google Earth Engine Team. (2015). Google earth engine: A
planetary-scale geospatial analysis platform. https://
earthengine.google.com
Green, J. M. H., Larrosa, C., Burgess, N. D., Balmford, A.,
Johnston, A., Mbilinyi, B. P., Platts, P. J., & Coad, L.
(2013). Deforestation in an African biodiversity hotspot:
Extent, variation and the effectiveness of protected areas.
Biological Conservation, 164, 62–72. https://doi.org/10.
1016/j.biocon.2013.04.016
Guitet, S., Cornu, J., Brunaux, O., Betbeder, J., Carozza, J., &
Richard-Hansen, C. (2013). Landform and landscape
mapping, French Guiana (South America). Journal of
Maps, 9(3), 325–335. https://doi.org/10.1080/17445647.
2013.785371
Hargrove, W. W., & Hoffman, F. M. (2005). Potential of
multivariate quantitative methods for delineation and
visualization of ecoregions. Environmental Management,
34(S1), S39–S60. https://doi.org/10.1007/s00267-003-
1084-0
Hargrove, W. W., Hoffman, F. M., & Hessburg, P. F. (2006).
Mapcurves: A quantitative method for comparing categ-
orical maps. Journal of Geographical Systems, 8(2), 187–
208. https://doi.org/10.1007/s10109-006-0025-x
Hengl, T., Heuvelink, G. B. M., Kempen, B., Leenaars,
J. G. B., Walsh, M. G., Shepherd, K. D., Sila, A., et al.
(2015). Mapping soil properties of Africa at 250 m resol-
ution: Random forests significantly improve current pre-
dictions. PloS one, 10(6), e0125814. https://doi.org/10.
1371/journal.pone.0125814
Jackson, D. A. (1993). Stopping rules in principal com-
ponents analysis: A comparison of heuristical and statisti-
cal approaches. Ecology, 74(8), 2204–2214. https://doi.org/
10.2307/1939574
Jackson, T., Chen, D., Cosh, M., Li, F., Anderson, M.,
Walthall, C., Doriaswamy, P., & Hunt, E. (2004).
Vegetation water content mapping using Landsat data
derived normalized difference water index for corn and
soybeans. Remote Sensing of Environment, 92(4), 475–
482. https://doi.org/10.1016/j.rse.2003.10.021
Jarvis, A., Reuter, H. I., Nelson, A., & Guevara, E. (2008).
Hole-filled SRTM for the globe Version 4. Retrieved
May 11, 2019, from https://research.utwente.nl/en/
publications/hole-filled-srtm-for-the-globe-version-4-
data-grid
Jelinski, D. E., & Wu, J. (1996). The modifiable areal unit
problem and implications for landscape ecology.
Landscape Ecology, 11, 129–140. https://doi.org/10.1007/
BF02447512
Kajembe, G. C., Mbwilo, A. J., Kidunda, R. S., &
Nduwamungu, J. (2003). Resource use conflicts in
Usangu plains, Mbarali District, Tanzania. International
Journal of Sustainable Development & Ecology, 10(4),
333–343. https://doi.org/10.1080/13504500309470109
Kangalawe, R. Y. M. (2012). Food security and health in the
southern highlands of Tanzania: A multidisciplinary
approach to evaluate the impact of climate change and
other stress factors. African Journal of Environmental
Science and Technology, 6(1), 50–66. https://doi.org/10.
5897/AJEST11.003
Kangalawe, R. Y. M., & Lyimo, J. G. (2010). Population
dynamics, rural livelihoods and environmental degra-
dation: Some experiences from Tanzania. Environment,
Development and Sustainability, 12(6), 985–997. https://
doi.org/10.1007/s10668-010-9235-y
Klijn, F., & Udo de Haes, H. (1994). A hierarchical approach
to ecosystems and its implications for ecological land
classification. Landscape Ecology, 9(2), 89–104. https://
doi.org/10.1007/BF00124376
Lambin, E. F., & Meyfroidt, P. (2011). Global land use
change, economic globalization, and the looming land
scarcity. Proceedings of the National Academy of
Sciences, 108(9), 3465–3472. https://doi.org/10.1073/
pnas.1100480108
Leathwick, J. R., Overton, J. M., & McLeod, M. (2003). An
environmental domain analysis of New Zealand, and its
application to biodiversity conservation. Conservation
Biology, 17(6), 1612–1623. https://doi.org/10.1111/j.
1523-1739.2003.00469.x
Li, G., Messina, J., Peter, B. G., & Snapp, S. (2017). Mapping
land suitability for agriculture in Malawi: Agricultural
land suitability mapping. Land Degradation &
Development, 28(7), 2001–2016. https://doi.org/10.1002/
ldr.2723
Loveland, T., & Merchant, J. (2004). Ecoregions and ecore-
gionalization: Geographical and ecological perspectives.
Environmental Management, 34(S1), S1–S13. https://doi.
org/10.1007/s00267-003-5181-x
Loveland, T. R., Sohl, T. L., Stehman, S. V., Gallant, A. L.,
Sayler, K. L., & Napton, D. E. (2002). A strategy for esti-
mating the rates of recent United States land-cover
changes. Photogrammetric Engineering & Remote
Sensing, 68(10), 1091–1099.
Lu, Y.,Nakicenovic,N., Visbeck,M.,& Stevance, A. (2015). Five
priorities for theUN sustainable developmentGoals.Nature,
520(7548), 432–433. https://doi.org/10.1038/520432a
Malley, Z. J. U., Taeb, M., & Matsumoto, T. (2009).
Agricultural productivity and environmental insecurity
in the Usangu plain, Tanzania: Policy implications for sus-
tainability of agriculture. Environment, Development and
Sustainability, 11(1), 175–195. https://doi.org/10.1007/
s10668-007-9103-6
Martínez-Harms, M. J., Quijas, S., Merenlender, A. M., &
Balvanera, P. (2016). Enhancing ecosystem services
maps combining field and environmental data.
Ecosystem Services, 22(Part A), 32–40. https://doi.org/10.
1016/j.ecoser.2016.09.007
Maselli, F., Di Gregorio, A., Capecchi, V., & Breda, F. (2009).
Enrichment of land-cover polygons with eco-climatic
information derived from MODIS NDVI imagery.
Journal of Biogeography, 36(4), 639–650. https://doi.org/
10.1111/j.1365-2699.2008.01981.x
JOURNAL OF MAPS 385
Metzger, M., Bunce, R., Jongman, R., Mucher, C., &Watkins,
J. (2005). A climatic stratification of the environment of
Europe. Global Ecology and Biogeography, 14(6), 549–
563. https://doi.org/10.1111/j.1466-822X.2005.00190.x
Metzger, M. J., Brus, D. J., Bunce, R. G. H., Carey, P. D.,
Goncalves, J., Honrado, J. P., Jongman, R. H. G.,
Trabucco, A., & Zomer, R. (2013a). Environmental stra-
tifications as the basis for national, European and global
ecological monitoring. Ecological Indicators, 33, 26–35.
https://doi.org/10.1016/j.ecolind.2012.11.009
Metzger, M. J., Bunce, R. G. H., Jongman, R. H. G., Sayre, R.,
Trabucco, A., & Zomer, R. (2013b). A high-resolution bio-
climate map of the world: A unifying framework for global
biodiversity research and monitoring. Global Ecology and
Biogeography, 22(5), 630–638. https://doi.org/10.1111/
geb.12022
Milder, J. C., Buck, L. E., Hart, A. K., Scherr, S. C., & Shames,
S. A. (2013). A framework for agriculture Green growth:
Greenprint for the Southern agricultural growth Corridor
of Tanzania (SAGCOT). SAGCOT Centre. Retrieved
December 15, 2016, from http://ecoagriculture.org/
resources/publications/
MNRT. (2013). Tanzania Mainland Land use-Land Cover,
Tanzania Forest Service Agency (TFS), Dar Es Salaam,
Tanzania.
Mücher, S., De Simone, L., Kramer, H., De Wit, A., Roupioz,
L., Hazeu, G., Boogaard, H., Schuiling, R., Fritz, S.,
Latham, J., & Cormont, A. (2016). A new global Agro-
environmental stratification (GAES) (Report 2761).
Wageningen Environmental Research.
National Bureau of Statistics. (2013). 2012 population and
housing census: Population distribution by administrative
areas, National Bureau of Statistics, Ministry of Finance,
Dar es Salaam, Tanzania and Office of Chief Government
Statistician, President’s Office, Finance, Economy and
Development Planning, Zanzibar. Retrieved October 12,
2016, from http://www.nbs.go.tz/
National Bureau of Statistics. (2015). Environment statistics
in Tanzania Mainland 2014. Retrieved October 12,
2016, from http://www.nbs.go.tz/
National Bureau of Statistics. (2016). 2014/2015 annual agri-
culture sample survey report. Retrieved December 20,
2016, from http://www.nbs.go.tz/
Nijbroek, R. P., & Andelman, S. J. (2015). Regional suitability
for agricultural intensification: A spatial analysis of the
Southern agricultural growth Corridor of Tanzania.
International Journal of Agricultural Sustainability, 14
(2), 231–247. https://doi.org/10.1080/14735903.2015.
1071548
Nkonya, E., Johnson, T., Kwon, H. Y., & Kato, E. (2016).
Economics of land degradation in Sub-Saharan Africa.
In E. Nkonya, A. Mirzabaev, & J. von Braun (Eds.),
Economics of land degradation and improvement – a glo-
bal assessment for sustainable development (pp. 215–
256). Springer.
Nowak, A., & Schneider, C. (2017). Environmental charac-
teristics, agricultural land use, and vulnerability to degra-
dation in Malopolska Province (Poland). Science of the
Total Environment, 590-591, 620–632. https://doi.org/10.
1016/j.scitotenv.2017.03.006
Nowosad, J., & Stepinski, T. F. (2018). Spatial association
between regionalizations using the information-theoreti-
cal V-measure. International Journal of Geographical
Information Science, 32(12), 2386–2401. https://doi.org/
10.1080/13658816.2018.1511794
Omernik, J. M., & Griffith, G. E. (2014). Ecoregions of the
conterminous United States: Evolution of a hierarchical
spatial framework. Environmental Management, 54(6),
1249–1266. https://doi.org/10.1007/s00267-014-0364-1
Parnell, S., & Walawege, R. (2011). Sub-Saharan African
urbanisation and global environmental change. Global
Environmental Change, 21(Supplement 1), S12–S20.
https://doi.org/10.1016/j.gloenvcha.2011.09.014
PFP. (2016). Panda Miti Kibiashara – private forestry pro-
gramme- programme document. Retrieved December
20, 2016, from www.privateforestry.or.tz/en/resources/
category/reports
R Core Team. (2019). R: A language and environment for
statistical computing. R Foundation for Statistical
Computing. https://www.R-project.org/
Revelle, W. (2018). Psych: Procedures for personality and
psychological research. Northwestern University. https://
CRAN.R-project.org/package=psych Version=1.8.12
Salami, A., Kamara, A. B., & Brixiova, Z. (2010). Smallholder
agriculture in East Africa: Trends, constraints and oppor-
tunities (Working Papers Series n. 105). African
Developmet Bank.
Sarvajayakesavalu, S. (2015). Addressing challenges of devel-
oping countries in implementing five priorities for sus-
tainable development goals. Ecosystem Health and
Sustainability, 1(7), 24. https://doi.org/10.1890/EHS15-
0028.1
Sawe, T. C., Munishi, P. K. T., & Maliondo, S. M. (2014).
Woodlands degradation in the Southern Highlands,
miombo of Tanzania: Implications on conservation and
carbon stocks. International Journal of Biodiversity and
Conservation, 6(3), 230–237. https://doi.org/10.5897/
IJBC2013.0671
Sayre, R., Dangermond, J., Frye, C., Vaughan, R., Aniello, P.,
Breyer, S., Cribbs, D., Hopkins, D., Nauman, R.,
Derrenbacher, W., &Wright, D. (2014). A new map of glo-
bal ecological land units – an ecophysiographic stratifica-
tion approach. Association of American Geographers.
Schaafsma, M., Morse–Jones, S., Posen, P., Swetnam, R. D.,
Balmford, A., Bateman, I. J., Burgess, N. D., Chamshama, S.
A. O., Fisher, B., Green, R. E., Hepelwa, A. S., Hernández–
Sirvent, A., Kajembe, G. C., Kulindwa, K., Lund, J. F.,
Mbwambo, L., Meilby, H., Ngaga, Y. M., Theilade, I., …
Turner, R. K. (2012). Towards transferable functions for
extraction of Non-timber forest Products: A case study on
charcoal production in Tanzania. Ecological Economics, 80,
48–62. https://doi.org/10.1016/j.ecolecon.2012.04.026
Serra, J. M., Cristobal, J., & Ninyerola, M. (2011). A classifi-
cation procedure for mapping topo-climatic conditions
for strategic vegetation planning. Environmental
Modeling & Assessment, 16(1), 77–89. https://doi.org/10.
1007/s10666-010-9232-4
Sleeter, B. M., Sohl, T. L., Loveland, T. R., Auch, R. F.,
Acevedo, W., Drummond, M. A., Sayler, K. L., &
Stehman, S. V. (2013). Land-cover change in the conter-
minous United States from 1973 to 2000. Global
Environmental Change, 23(4), 733–748. https://doi.org/
10.1016/j.gloenvcha.2013.03.006
Smiraglia, D., Capotorti, G., Guida, D., Mollo, B., Siervo, V.,
& Blasi, C. (2013). Land units map of Italy. Journal of
Maps, 9(2), 239–244. https://doi.org/10.1080/17445647.
2013.771290
Song, X.-P., Huang, C., & Townshend, J. R. (2017).
Improving global land cover characterization through
data fusion. Geo-spatial Information Science, 20(2), 141–
150. https://doi.org/10.1080/10095020.2017.1323522
Soto, S., & Pintó, J. (2010). Delineation of natural landscape
units for Puerto Rico. Applied Geography, 30(4), 720–730.
https://doi.org/10.1016/j.apgeog.2010.01.010
386 J. KOSKIKALA ET AL.
Trabucco, A., & Zomer, R. J. (2019). Global aridity index and
potential evapotranspiration (ET0) climate database v2.
UNDP. (2015). Tanzania Human Development Report 2014,
Economic Transformation for Human Development,
Economic and Social Research Foundation, United
Nations Development Programme, Government of the
United Republic of Tanzania, Ministry of Finance, Dar
Es Salaam.
Vrebos, D., Staes, J., Vandenbroucke, T., D׳Haeyer, T.,
Johnston, R., Muhumuza, M., Kasabeke, C., & Meire, P.
(2015). Mapping ecosystem service flows with land
cover scoring maps for data-scarce regions. Ecosystem
Services, 13, 28–40. https://doi.org/10.1016/j.ecoser.2014.
11.005
Ward, J. H. (1963). Hierarchical grouping to optimize an
objective function. Journal of the American Statistical
Association, 58(301), 236–244. https://doi.org/10.1080/
01621459.1963.10500845
White, M. A., Hoffman, F., Hargrove, W. W., & Nemani, R.
R. (2005). A global framework for monitoring phenologi-
cal responses to climate change. Geophysical Research
Letters, 32(4), L04705. https://doi.org/10.1029/2004GL0
21961
Willemen, L., Verburg, P. H., Hein, L., & van Mensvoort,
M. E. F. (2008). Spatial characterization of landscape
functions. Landscape and Urban Planning, 88(1), 34–43.
https://doi.org/10.1016/j.landurbplan.2008.08.004
Williams, C. L., Hargrove, W. W., Liebman, M., & James, D.
E. (2008). Agro-ecoregionalization of Iowa using multi-
variate geographical clustering. Agriculture Ecosystems &
Environment, 123(1–3), 161–174. https://doi.org/10.
1016/j.agee.2007.06.006
Wu, J., Jones, K. B., Li, H., & Loucks, O. L. (Eds.). (2006).
Scaling and uncertainty analysis in ecology: Methods and
applications, 3–15. Springer. Printed in the Netherlands.
Xiong, J., Thenkabail, P. S., Gumma, M. K., Teluguntla, P.,
Poehnelt, J., Congalton, R. G., Yadav, K., & Thau, D.
(2017). Automated cropland mapping of continental
Africa using Google Earth Engine cloud computing. Isprs
Journal of Photogrammetry and Remote Sensing, 126,
225–244. https://doi.org/10.1016/j.isprsjprs.2017.01.019
Zhang, Y., Hepner, G. F., &Dennison, P. E. (2012).Delineation
of phenoregions in geographically diverse regions using k-
means plus plus clustering: A case study in the Upper
Colorado River Basin. Giscience & Remote Sensing, 49(2),
163–181. https://doi.org/10.2747/1548-1603.49.2.163
Zomer, R. J., Trabucco, A., Metzger, M., & Oli, K. P. (2013).
Environmental stratification of Kailash Sacred Landscape
and projected climate change impacts on ecosystems and
productivity (ICIMODWorking Paper 2013/1). ICIMOD.
JOURNAL OF MAPS 387