Reliable Crop Identification with Satellite Imagery in the Context of Common Agriculture Policy Subsidy Control

Jonas Schmedtmann,Manuel Campagnolo

doi:10.3390/rs70709325

Abstract

Agricultural subsidies in the context of the Common Agricultural Policy (CAP) represent over 40% of the EU’s yearly budget. To ensure that funds are properly spent, farmers are controlled by National Control and Paying Agencies (NCPA) using tools, such as computer-assisted photo interpretation (CAPI), which aims at identifying crops via remotely-sensed imagery. CAPI is time consuming and requires a large team of skilled photo interpreters. The objective of this study was to develop a reliable control system to partially replace CAPI for crop identification, with the overreaching goal of reducing control costs and completion time. Validated control data provided by the Portuguese Control and Paying Agency and an atmospherically-corrected Landsat ETM+ time series were used to perform parcel-based crop classification, leading to an accuracy of only 68% due to high similarity between crops’ spectral signatures. To address this problem, we propose an automatic control system (ACS) that couples crop classification to a reliability requirement. This allows the decision-maker to set a reliability level, which restricts automatic crop identification to parcels that are classified with high certainty. While higher reliability levels reduce the risk of misclassifications, lower levels increase the proportion of automatic control decisions (ACP). With a reliability level of 80%, more than half of the parcels in our study area are automatically identified with an overall accuracy of 84%. In particular, this allows automatically controlling over 85% of all parcels classified as maize, rice, wheat or vineyard.

Highlights

The problem we address is described, and the main motivations and goals of this research are discussed
The automatic control system (ACS) itself is divided into three steps: classification, which is done by an SVM classifier using 10-fold cross-validation (Section 3.2); calibration, which is at the core of our approach and is described in Section 3.3; and application in an operational context (Section 3.4)
RMSE values show that the overall standard deviation of pixel reflectance values with respect to the parcels’ average range between 0.011 and 0.037 units of reflectance, i.e., just 1.1% to 3.7%, for the combination of bands and dates we used for classification

Summary

Introduction

The problem we address is described, and the main motivations and goals of this research are discussed. The reader can find a list of the most important acronyms at the end of the paper. Use of Remote Sensing for CAP Subsidy Control. The Common Agricultural Policy (CAP) is a system of European Union (EU) agricultural subsidies and programs that is very significant in financial terms, representing over 40% of the EU’s budget, equivalent to e58 billion in 2011 [1]. To ensure that CAP funds are spent appropriately, Member. State Authorities have to comply with legal management and control mechanisms [2]. European Commission (EC)’s Joint Research Centre (JRC) provides technical support to Member. Control with Remote Sensing (CwRS), Digital Land Parcel Identification System (LPIS)

Objectives

Methods

Results

Conclusion