Automatic delineation algorithm for site-specific management zones based on satellite remote sensing data
Automatic delineation algorithm for site-specific management zones based on satellite remote sensing data
Claudia Georgi 0 1
Daniel Spengler 0 1
Sibylle Itzerott 0 1
Birgit Kleinschmit 0 1
0 Department of Landscape Architecture and Environmental Planning, Technische Universita ̈t Berlin , Berlin , Germany
1 GFZ German Research Centre for Geosciences Potsdam, Geodesy & Remote Sensing , Telegrafenberg, 14473 Potsdam , Germany
In light of the increasing demand for food production, climate change challenges for agriculture, and economic pressure, precision farming is an ever-growing market. The development and distribution of remote sensing applications is also growing. The availability of extensive spatial and temporal data-enhanced by satellite remote sensing and open-source policies-provides an attractive opportunity to collect, analyze and use agricultural data at the farm scale and beyond. The division of individual fields into zones of differing yield potential (management zones (MZ)) is the basis of most offline and mapoverlay precision farming applications. In the process of delineation, manual labor is often required for the acquisition of suitable images and additional information on crop type. The authors therefore developed an automatic segmentation algorithm using multi-spectral satellite data, which is able to map stable crop growing patterns, reflecting areas of relative yield expectations within a field. The algorithm, using RapidEye data, is a quick and probably low-cost opportunity to divide agricultural fields into MZ, especially when yield data is insufficient or non-existent. With the increasing availability of satellite images, this method can address numerous users in agriculture and lower the threshold of implementing precision farming practices by providing a preliminary spatial field assessment.
Remote sensing
NDVI
RapidEye
Segmentation
Crop patterns
Introduction
The major aim of precision agriculture is to optimize crop management by addressing
spatial variability, and thus optimize the use of farm inputs such as fertilizers and
herbicides
(Mulla 2013)
. In general, vast information is accumulated and used for the analysis of
field inventory, crop growth and yield patterns. With this information, customized inputs
can be applied to management zones (MZ), i.e. the units into which large farm fields are
divided
(Mulla 2013)
.
The delineation of MZ is the basis of most precision agriculture (PA) practices,
addressing the within-field variability of crop and crop yield. MZ are subdivisions of a
field, each characterized by relative homogeneity of crops and/or environmental
parameters
(Doerge 1999)
, which therefore differ in the need for specific input rates of treatment.
The more generic term ‘management unit’ was introduced by
Lark and Stafford (1997)
,
and numerous delineation methods have emerged since then. They are usually either based
on yield maps
(Pedroso et al. 2010; Lark 1998)
, soil and topographic properties
(MacMillan et al. 1999; van Alphen and Stoorvogel 1999)
, electrical conductivity data
(Kitchen et al. 2005; Cambouris et al. 2006)
, remote sensing and vegetation indices
(Ahn
et al. 1999; Song et al. 2009)
, or a combination of these methods
(Fridgen et al. 2003; De
Benedetto et al. 2013; Yao et al. 2014)
.
Every automatic method to determine MZ has several disadvantages in terms of
accuracy and applicability. Commonly, segmentation applications rely on yield maps,
which are acquired neither by every farmer nor for every field, even if the company is in
possession of the required technology. Yield data have significant error sources, such as via
sensor, georeferencing, operator or data processing errors
(Simbahan et al. 2004)
, and are
also complicated to prepare
(Blackmore and Marshall 1996)
. Moreover, the irregular
distribution of data points in regard to the spatial variation in yield can impede accurate
interpolation, which is a necessity for most spatial analyses.
The use of soil sampling data and soil maps for delineation of MZ is also a common
approach, especially if yield maps are not available. However, as for soil inventory maps, a
sufficient scale is needed for precision farming applications
(Franzen et al. 2002)
.
Additionally, the question of relevance arises, when for example the standard soil map—such as
the ‘‘Bodenscha¨tzung’’ in Germany—dates back to the 1930s. Intensive soil grid mapping
for status-quo maps is often not cost-effective and does not necessarily reflect the total
spatial variability of crop growth
(Hornung et al. 2006)
.
Electrical conductivity (EC) maps acquired with instruments like the standard ‘‘EM38’’
can also be used for successful MZ delineation
(Cambouris et al. 2006)
. They reflect soil
differences due to such factors as moisture content, salinity and texture. However, even if
these characteristics influence crop growth significantly, EC maps may not always give a
direct picture of in-s (...truncated)