Improving Parcel-Level Mapping of Smallholder Crops from VHSR Imagery: An Ensemble Machine-Learning-Based Framework

Abstract

Explicit spatial information about crop types on smallholder farms is important for the development of local precision agriculture. However, due to highly fragmented and heterogeneous cropland landscapes, fine-scale mapping of smallholder crops, based on low- and medium-resolution satellite images and relying on a single machine learning (ML) classifier, generally fails to achieve satisfactory performance. This paper develops an ensemble ML-based framework to improve the accuracy of parcel-level smallholder crop mapping from very high spatial resolution (VHSR) images. A typical smallholder agricultural area in central China covered by WorldView-2 images is selected to demonstrate our approach. This approach involves the task of distinguishing eight crop-level agricultural land use types. To this end, six widely used individual ML classifiers are evaluated. We further improved their performance by independently implementing bagging and stacking ensemble learning (EL) techniques. The results show that the bagging models improved the performance of unstable classifiers, but these improvements are limited. In contrast, the stacking models perform better, and the Stacking #2 model (overall accuracy = 83.91%, kappa = 0.812), which integrates the three best-performing individual classifiers, performs the best of all of the built models and improves the classwise accuracy of almost all of the land use types. Since classification performance can be significantly improved without adding costly data collection, stacking-ensemble mapping approaches are valuable for the spatial management of complex agricultural areas. We also demonstrate that using geometric and textural features extracted from VHSR images can improve the accuracy of parcel-level smallholder crop mapping. The proposed framework shows the great potential of combining EL technology with VHSR imagery for accurate mapping of smallholder crops, which could facilitate the development of parcel-level crop identification systems in countries dominated by smallholder agriculture.

Multi-Temporal Passive and Active Remote Sensing for Agricultural Mapping and Acreage Estimation in Context of Small Farm Holds in Ethiopia

“…However, the free availability of 10 m Sentinel-2 data and an advanced processing platform allows for efficient processing of high spatial resolution data, making crop area maps feasible [7,181,182]. The high spatial resolution satellite allows for fewer mixed pixels in these smallholder agricultural landscapes, resulting in mosaics of fields that are often heterogeneously mixed at lower resolution satellite data [183][184][185]. In this study, the high-resolution freely available sentinel data were evaluated, and the performances of commonly used MLAs were tested concerning the small and fragmented farmlands.…”

Evaluation of Multiple Classifier Systems for Mapping Different Hierarchical Levels of Forest Ecosystems in the Mediterranean Region Using Sentinel-2, Sentinel-1, and ICESat-2 Data

“…In RS image classification, MCSs have been proposed as a means to improve the accuracy and reduce the variance in classification results [29]. However, studies have shown mixed results regarding the effectiveness of MCSs, with some reporting only slight improvements compared to the best individual classifiers [20,24], while others demonstrate significant increases of up to 6% in classification accuracy [74][75][76]. Based on the findings of this study, it appears that the use of MCSs only marginally improved the OA by 1.2% in the L1 case, while in L2 and L3, the best MCS achieved slightly lower accuracy compared to the most accurate base classifier (SVM).…”

“…The lack of classifier diversity could be a possible explanation for the marginal or lack of improvement in OA, as diversity is an important parameter for the improvement of classification accuracy via MCSs. Previous studies have demonstrated that the success of any MCS is related to the accuracy and diversity of the base classifiers included in the system [74]. Thus, an ensemble of classifiers could improve the accuracy of any of its individual members if they have a low error rate (are accurate) and their errors are not coincident (are diverse).…”

The Classification Method Study of Crops Remote Sensing with Deep Learning, Machine Learning, and Google Earth Engine

“…Agricultural areas in China are characterized by scattered land use types, broken agricultural landscape, and complex crops planting structures; this brings great challenges to remote sensing extraction and classification of crop types [1][2][3]. Currently, information on rice fields relies on field survey data that are incomplete, time-consuming, and lacking spatial details.…”