Cloud Computing, Web-Based GIS, Terrain Analysis, Data Fusion, and Multivariate Statistics for Precision Conservation in the 21st Century Tom Mueller, Surendran Neelakantan, Eduardo Rienzi, and Blazan Mijatovic

vanillaoliveInternet και Εφαρμογές Web

3 Νοε 2013 (πριν από 3 χρόνια και 11 μήνες)

87 εμφανίσεις

Cloud Computing,
Web
-
Based
GIS, Terrain Analysis, Data Fusion, and
1

Multivariate
Statistics
for Precision Conservation in the 21st Century

2


3


4

Tom Mueller, Surendran Neelakantan, Eduardo Rienzi, and Blazan
5

Mijatovic


6


7

Plant and Soil Science Department

8

University of Kentucky

9

Lexington, Kentucky

10


11

Annamaria Castrignano

12


13

Agricultural Research Council
-

Research

14

Unit for Cropping Systems in Dry Environments


15

Bari, Italy

16


17

Adam Pike


18


19

LiDAR group

20

PhotoScience

21

Lexington, Kentucky

22


23

Cody Bumgardner

24


25

Enterprise Architecture

26

University of Kentucky

27

Lexington, Kentucky

28


29


30

INTRODUCTION

31


32

Precision conservation
involves utilizing a
set
of
tools
in agriculture
that can help
33

meet
the
great challenges
of
this century including
a
growing world population,
34

increasing
demand for food and biofuel
s
, global climate change, and the global
35

water crisis.

Precision c
onservation seeks to
efficiently
target conservation
36

practices
at
specific
locations
within landscapes where they can help ma
intain or
37

improve soil quality, increase crop productivity, and protect water and air
38

resources.

The objective of this paper is to discuss
how
cutting edge techniques
39

(e.g., cloud computing, geospatial analysis, environmental modeling, and
40

precision tech
nologies)
could be used together to
help
facilitate
more effective
41

precision conservation management.


42


43

TECHNOLOGY ENABLED PRECISION CONSERVATION

44


45

Wireless Technologies and Cloud Computing

46


47

Equipment manufacturers are designing tractors and combines with wireless
48

communication
, allowing transmission of yield monitor

and sensor data to “the
49

cloud” where farmers
,
consultants
, and researchers

can access information for
50

real
-
time decision making.

The concept of “the cloud” developed when
51

companies like Amazon had extra server
capacity
and they realized
they could
52

“time sharing”
their server
resources when they were not in use

(e.g., at night)
.

53

One advantageous characteristic of cloud applications is that they are allocated
54

hardware resources from
the cloud

as software requests increase. Resources are
55

released when they are no longer needed. This effectively maximizes performance
56

while minimizing

costs. Load balancers constantly distribute
and redistribute
57

Internet traffic across multiple servers thus optimizing performance. Cloud
58

computing techniques could substantially reduce the infrastructure costs required
59

to perform
management and
conservati
on planning in agriculture today

because
60

contracting national budgets for government conservation programs
will
require

61

more

efficient use of resources.

We have developed a
sample
cloud computing
62

application for soil survey data for an MLRA
-
121 (located i
n parts of Kentucky,
63

Indiana, and Ohio) that can be found at go.uky.edu/LandUse. The architecture for
64

this website is shown in
Figure

1.

65


66


67




Figure 1.

Cloud computing architecture for a Land Use and Management
website
.

Data Fusion and Multivariate
Statistical
Analyses

68


69

D
ata fusion techniques will be
helpful
for
fast
, efficient, accurate, and automatic
70

data analysis

for conservation planning

because data are collected at a wide
71

variety of spatial and temporal scales
.

For example,
LiDAR
generally does not
72

require repeated measurements and are usually provided at a resolution
of
less
73

than 1 m.

Yield is measured once a growing season and for a
12
-
row combine
74

trave
l
ling at 5 mph harvesting corn planted on 2.5
-
ft centers, yield data will have
75

a resolution of approximately 7 ft in the direction of travel and 30 ft between
76

passes.

Soil EC
a

is generally
only needs to be
measured once
and this occurs

in
77

the spring or fall
. T
he spatial resolution of the EC survey data depends on ground
78

speed and the distance between passes (generally 40 and 60 feet).

Soil samples
79

are often collected in the fall or spring, every 2 to 4 years with a grid sampling
80

(e.g.

330 foot spacing) or with a regular or irregular zone sampling approach (e.g.,
81

5 to 20 acre zones).

Landsat 7 imagery has a resolution of about 100
-
ft in the
82

visible and near infrared ranges and 200
-
ft in the thermal infrared (TIR) range.
83


Soil survey da
ta is available for many areas within the US at scales between
84

1:12,000 and 1:26,000, but the intensity of ground soil profile observations varied
85

substantially and depended on many factors.

86


87

These scale mismatches can be addressed with
d
ata fusion techniques
such as
88

block interpolation

(e.g., kriging) and multi
-
collocated cokriging
.
Multi
-
colocated
89

factorial co
-
kriging is a data fusion technique that also allows efficient synthesis
90

of
many variables in ways

that
would allow the
creation
of
management zones
91

for soil and water conservation

(Castrignano et al., 2009)
.
This procedure was
92

conducted for a field in Central Kentucky
(Figure 2)
.

Input data included LiDAR
93

derived terrain attributes, soil EC data, and grid soil sample data.
The first factor
94

(F1) at short range (140 m) and the first factor (F1) at long range (500 m)
were
95

the only factors that were
statistically significant
.
The map of F1

(
140

m)

mostly
96

summarized
the distribution of EC and Mg and Ca, with the highest values on

the
97

lower parts of the

field, whereas the map of F1 (500
-
m
) is
characterized

by a wide
98

middle area
with low values
corresponding to
ridge tops with
higher contents of
99

phosphorus
.

Isofrequency
maps could be created for the short and long range
F1
100

maps

and
the delineation in homogeneous
zones could be used for
site
-
specific
101

management

and
could serve as the basis for sampling for
conservation
soil
102

fertility
management.

Potentially, multi
-
colocated factorial co
-
kriging
, using as
103

finer
-
scale auxiliary v
ariable a terrain attribute or remote/proximal sensing image,

104

could be conducted in the cloud with GSLib, the source code for which is freely
105

available.

106


107

Terrain Analysis
for Soil and Water Conservation

108


109

Terrain analysis can be used to identify environmentally sensitive areas within
110

landscapes.

Several studies have used terrain attributes to determine the locations
111

where concentrated flow erosion (e.g., ephemeral gully erosion) is likely to occur

112

(Thorne

et al., 1986; Moore et al., 1988; Srivastava and Moore, 1989; Berry et al.,
113

2005). These studies overlaid maps depicting the locations where eroded channels
114

were observed in the field on the top of terrain attribute maps. The authors of
115

these studies visu
ally determined terrain attribute thresholds above which
116

concentrated flow erosion was likely to occur. The practical problems that
117

conservation planners would have employing this type of approach over large
118

areas are that it 1) requires site
-
specific thre
sholds, 2) would be very time
119

consuming, 3) requires a high level of GIS knowledge, and 4) does not easily lend
120

itself to automation.

121


122

To overcome these problems, we developed a multivariate statistical approach to
123

determine where concentrated flow erosion was likely to occur in agricultural
124

fields, based on terrain attributes (Pike et al., 2009).

We c
ompared logistic
125

regression and artificial neural network analysis as prediction methods and found
126

that both procedures performed very well as determined with a leave
-
one
-
field
-
127

out validation procedure.

Next, we evaluated how these procedures perform
ed

128

wi
th 10
-
m US Geological Survey (USGS) digital elevation models (DEMs; freely
129

available on the Internet for most of the United States) with those made from
130

elevation measurements created using Real
-
Time Kinematic (RTK) survey
-
grade
131

Global Positioning System (
GPS) (Pike et al., 2010).

Although we found that
132

RTK predictions were superior to those made with the USGS data
.
USGS
133

predictions would be adequate for many agricultural applications.

For example,
134

m
ost of the waterways that are visually apparent in Figur
e
3
a
could be predicted
135

by the 10
-
m USGS map (Figure
3
b
), but were more clearly delineated by the RTK
136

data

(Figure 3c)
.

In another study, we found that the model developed for the
137

farm in Shelby County, Kentucky (Pike et al., 2010) also worked well 80
miles
138

away in Hopkins County, where soils differed substantially in properties
139

including texture (Luck et al., 2010). We have observed that the waterway
140

prediction models also work well in other Kentucky counties.

141



Fig. 2. Multivariate co
-
located factorial co
-
kriging analysis to summarize
nutrient, soil EC
a
, and LiDAR derived ter
rain attributes

maps created at
different scales.



142


143

These conservation analyses should
be
available to conservation planners on hand
144

held computing devices in the field. Cloud computing technologies could be used
145

to distribute data efficiently by adding
ArcGIS servers
t
o
cloud
systems similar to
146

the one depicted in Figure 1. Elevation data could be pulled directly from a
147

national USGS server or a local database.

148


149

Processed Based Environmental Models

150


151

Environmental models

(e.g., SWAT, WEPP,
RUSLE2, and
ANNAGNPS
)
could
152

provide valuable information for assessing conservation practices

in the field
.

153

Unfortunately, these models may be too complex and cumbersome to
be operated
154

in real time
in the cloud

in the field by conservation planners on mobile
155

computing devices. An

alternative approach would be to adapt environmental
156



Figure
3
. The outlines of the locati
ons of observed waterways overlaid on a) an
aerial photograph, b) USGS and c) Survey grade GPS derived maps
of the probability of the occurrence of eroded waterways (from Pike
et al., 2009, 2010). The probability analysis is from a leave
-
one
-
field out val
idation analysis. The logistic regression equations to
calculate the likelihood of erosion were
)
WET
84
.
0
LS
94
.
0
55
.
7
(
USGS
1
1

Erosion

of
y
Probabilit








e

and
)
WET
12
.
1
LS
25
.
2
53
.
12
(
RTK
1
1

Erosion

of
y
Probabilit








e

where LS and WET are both
terrain attributes and LS is the estimated length slope factor from the
USLE
equation and WET is the topographic wetness index.


models for particular watersheds or regions and then to develop simple
157

relationships to predict the environmental impacts of conservation practices

which

158

would be more suited to for rapid decision makin
g in the field via the internet.
159

For example Dosskey et al. (2011) developed a design aid for sizing filter strips
160

based on the
process
-
based Vegetative Filter Strip Model
.

161


162

CONCLUSIONS

163


164

There is a new wave of the

information

technology revolution
in
ag
riculture today
165

including
web
-
based mapping and GIS, cloud computing, and wireless
166

technologies. Conservation professionals
could utilize these technologies as
167

environmental and agricultural modeling tools to improve the efficacy and
168

economics of conserva
tion planning in a way that will help meet the challenges of
169

the 21
st

century.

170


171

ACKNOWLEDGEMENTS

172


173

The authors gratefully acknowledge support from the NRCS, Kentucky SB
-
271
174

Water Quality program, and the University of


Kentucky

Division of Information
175

Technology. We would personally like to thank Steve Workman,
Steve Crabtree,
176

Steve Blanford,
Paul Finnell, Jon Hempel, Doyle Friskney, Victoria Banks,
Nick
177

McClur
e
and Carey Johnson.

178


179

REFERENCES

180

Berry, J.D., F.J. Pierce, and R. Kh
osla. 2005. Applying spatial analysis for
181

precision conservation across the landscape. Journal of Soil and Water
182

Conservation 60(6):363
-
370.

183

Castrignano, A., Costantini, E., Barbetti, R., Sollitto, D. (2009). Accounting for
184

extensive topographic and pedo
logic secondary information to improve
185

soil mapping. CATENA, vol. 77; p. 28
-
38.

186

Dosskey, M.G., M.J. Helmers, and D.E. Eisenhauer. 2011. A design aid for sizing
187

filter strips using buffer area ratio. Journal of Soil and Water
188

Conservation. 66:1
-
29.

189

Luck, J
.D., T.G. Mueller, S.A. Shearer, and A.C. Pike. 2010. Grassed waterway
190

planning model evaluated for agricultural fields in the western coal field
191

physiographic region of Kentucky. Journal of Soil and Water
192

Conservation 65:280
-
288.

193

Moore, I.D., G.J. Burch,

and D.H. Mackenzie. 1988. Topographic effects on the
194

distribution of surface soil water and the location of ephemeral
195

gullies.Transactions of the American Society of Agricultural Engineers
196

31(4):1098
-
1119.

197

Neelakantan, S., T.G. Mueller, B. Lee, B. Lee, P.

Finnell, V. Bumgardner, and D.
198

Carey. 2011. Web 2.0 spatial data browser for visualizing land
-
use
199

assessment information from soil surveys. JSWCS. 66:37A
-
39A.

200

Srivastava, K.P., and I.D. Moore. 1989. Application of terrain analysis to land
201

resource investi
gations of small catchments in the Caribbean. In
202

Proceedings of the 20th International Conference of the Erosion Control
203

Association, Vancouver, BC, Canada: 229
-
249, International Erosion
204

Control Association, Steamboat Springs, CO.

205

Pike, A.C., T.G.Mueller
, A. Schörgendorfer, J.D. Luck, S.A. Shearer, and A.D.
206

Karathanasis. 2009. Erosion index derived from terrain attributes using
207

logistic regression and neural networks. Agronomy Journal 101:1068
-
208

1079.

209

Pike, A.C., T.G. Mueller, A.

Schörgendorfer, J.D. Luck.
S.A. Shearer, A.D.
210

Karathanasis. 2010.

Eroded Waterways with USGS Elevation Data.
211

Agronomy Journal. 102:1269
-
1273.

212

Thorne, C.R., L.W. Zezenbergen, E.H. Grissinger, and J.B. Murphey.

1986.
213

Ephemeral gullies as sources of sediment. In Proceedings of the 4th
214

Federal Interagency Sedimentation Conference, Las Vegas, Nevada,Vol.
215

1:301
-
309. Washington, DC, Government Printing Office.

216