A Framework for Large-Area Mapping of Past and Present Cropping Activity Using Seasonal Landsat Images and Time Series Metrics

Schmidt, Michaël; Pringle, M. J.; Devadas, Rakhesh; Denham, Robert; Tindall, D.

doi:10.3390/rs8040312

Cited by 51 publications

(34 citation statements)

References 59 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Multi-temporal images can also offer a further area of research towards better discriminating and classifying temporally dynamic LULC classes, such as crops (Schmidt, Pringle, Devadas, Denham, & Tindall, 2016).…”

Section: Resultsmentioning

confidence: 99%

Fusion of Sentinel-1A and Sentinel-2A data for land cover mapping: a case study in the lower Magdalena region, Colombia

2017

View full text Add to dashboard Cite

Land cover-land use (LCLU) classification tasks can take advantage of the fusion of radar and optical remote sensing data, leading generally to increase mapping accuracy. Here we propose a methodological approach to fuse information from the new European Space Agency Sentinel-1 and Sentinel-2 imagery for accurate land cover mapping of a portion of the Lower Magdalena region, Colombia. Data pre-processing was carried out using the European Space Agency's Sentinel Application Platform and the SEN2COR toolboxes. LCLU classification was performed following an object-based and spectral classification approach, exploiting also vegetation indices. A comparison of classification performance using three commonly used classification algorithms was performed. The radar and visible-near infrared integrated dataset classified with a Support Vector Machine algorithm produce the most accurate LCLU map, showing an overall classification accuracy of 88.75%, and a Kappa coefficient of 0.86. The proposed mapping approach has the main advantages of combining the all-weather capability of the radar sensor, spectrally rich information in the visible-near infrared spectrum, with the short revisit period of both satellites. The mapping results represent an important step toward future tasks of aboveground biomass and carbon estimation in the region. ARTICLE HISTORY

show abstract

Section: Resultsmentioning

confidence: 99%

Fusion of Sentinel-1A and Sentinel-2A data for land cover mapping: a case study in the lower Magdalena region, Colombia

2017

View full text Add to dashboard Cite

show abstract

“…Arguably, given a suitable data structure, there is no reason why a gap pixel should not be filled with a value that is very far distant. Conceptually this is similar to supervised land cover classification where a sample of training data are used to classify very large area Landsat time series [32,87,88]. Using larger tiles (or groups of tiles) may help overcome the main limitation of the SAMSTS algorithm: if there are no existing alternative similar pixels, then a gap cannot be filled reliably.…”

Section: Discussionmentioning

confidence: 99%

“…The SAMSTS algorithm provides gap-filled data with five-band reflective-wavelength root-mean-square differences less the 0.02, which is comparable to the OLI reflectance calibration accuracy.Temporal interpolation (TI) gap-filling approaches have been developed that fit time series statistical models to predict reflectance or vegetation index values on a given day. Linear, logistic, or sum of sinusoidal models have been used [32][33][34][35][36][37][38][39]. The model fits are conducted on individual pixel time series.…”

mentioning

confidence: 99%

Large-Area Gap Filling of Landsat Reflectance Time Series by Spectral-Angle-Mapper Based Spatio-Temporal Similarity (SAMSTS)

Yan

Roy

2018

Remote Sensing

View full text Add to dashboard Cite

Landsat time series commonly contain missing observations, i.e., gaps, due to the orbit and sensing geometry, data acquisition strategy, and cloud contamination. A spectral-angle-mapper (SAM) based spatio-temporal similarity (SAMSTS) gap-filling algorithm is presented that is designed to fill small and large area gaps in Landsat data, using one year or less of data and without using other satellite data. Each gap pixel is filled by an alternative similar pixel that is located in a non-missing region of the image. The alternative similar pixel locations are identified by comparison of reflectance time series using a SAM metric revised to be adaptive to missing observations. A time series segmentation-and-clustering approach is used to increase the search efficiency. The SAMSTS algorithm is demonstrated using six months of Landsat 8 Operational Land Imager (OLI) reflectance time series over three 150 × 150 km (5000 × 5000 30 m pixels) areas in California, Minnesota and Kansas. The three areas contain different land cover types, especially crops that have different phenology and abrupt changes due to agricultural harvesting, which make gap filling challenging. Fillings on simulated gaps, which are equivalent to 36% of 5000 × 5000 images in each test area, are presented. The gap filling accuracy is assessed quantitatively, and the SAMSTS algorithm is shown to perform better than the simple closest temporal pixel substitution gap filling approach and the sinusoidal harmonic model-based gap filling approach. The SAMSTS algorithm provides gap-filled data with five-band reflective-wavelength root-mean-square differences less the 0.02, which is comparable to the OLI reflectance calibration accuracy.Temporal interpolation (TI) gap-filling approaches have been developed that fit time series statistical models to predict reflectance or vegetation index values on a given day. Linear, logistic, or sum of sinusoidal models have been used [32][33][34][35][36][37][38][39]. The model fits are conducted on individual pixel time series. Model fitting is sensitive to the quality, number of available observations, and the seasonality of missing observations. TI methods are typically less reliable for surfaces that have abrupt changes, for example, due to land cover change, flooding, fire, or for surfaces with complex phenology, such as double or triple agricultural cropping [35,40,41]. Time series change detection methods that account for both abrupt and gradual trends in coarse spatial resolution data have been developed [41], but are less well suited for Landsat application as they require a higher observation frequency than provided by Landsat.The ASP gap-filling approach fills a gap pixel with the values of one or more alternative pixels selected from non-missing pixels found usually in the same image. Alternative similar pixel locations are identified from a reference image that may be the same, a previous, or a subsequent image that has no gap at the location to be filled [42][43][44][45][46][47][48][49][50]. The ASP method was ...

show abstract

“…Resultados similares a esses foram encontrados por Santos et al (2014) Para melhorar a separação das classes pode-se aumentar o número de amostras de treinamento para pastagem, ou mesmo, agrupar as amostras em classes como agricultura e não-agricultura, por exemplo, afim de aumentar os vetores de separação e melhorar a acurácia da classificação e evitar a superestimativa para a classe de agricultura (Ouzemou et al, 2018). Segundo Schmidt et al (2016), a identificação de culturas agrícolas, por meio de aprendizado de máquina, tem maior precisão com uso de imagens no inverno, já que áreas de pastagem não irrigadas apresentam menor fitomassa e não seriam confundidas com culturas agrícolas. Porém, o noroeste de Minas Gerais destaca-se por apresentar grande produção de grãos, com plantios concentrados no verão devido a ocorrência de chuvas, sendo, dessa forma, necessário avaliar a melhor época de aquisição de imagem para identificação da agricultura.…”

Section: Mapeamento De áReas Agrícolas Com Máquina De Vetor De Suportunclassified

Mapping of Agricultural Areas with Support Vector Machine in the Northwest of Minas Gerais, Brazil

et al. 2020

View full text Add to dashboard Cite

ResumoA agricultura é um dos setores que mais se destaca na economia do Brasil, sendo necessário muitas vezes o emprego do sensoriamento remoto, para identificação da expansão das áreas agrícolas e estimativas da sua produção. Esse trabalho tem por objetivo mapear as áreas agrícolas do noroeste de Minas Gerais por meio de máquina de vetor de suporte e comparar os resultados obtidos com o censo estatístico do Instituto Brasileiro de Geografia e Estatística. Para identificação das áreas agrícolas foi utilizado o algoritmo Service Vector Machine e imagens dos satélites e sensores Landsat 8 / OLI e Terra / MODIS. As amostras de treinamento do algoritmo foram obtidas por meio de imagem de alta resolução espacial, disponível no programa Google Earth Pro, nas classes rios, floresta, agricultura, pastagem e silvicultura. O mapeamento utilizando imagem do sensor OLI apresentou melhor Acurácia Global (0,81) e Kappa (0,66). A classificação com imagem OLI e MODIS apresentaram maior precisão na classe agricultura quando comparada as demais classes, apresentando confusão com pastagem, em decorrência da alta fitomassa da pastagem na época de aquisição das imagens (verão). O cálculo das áreas agrícolas demonstra superestimativa do Service Vector Machine (SVM) na classificação das imagens OLI e MODIS, com forte relação dos dados MODIS com o censo do IBGE (R²=0,83). Apenas municípios com áreas agrícolas superiores a 50.000 ha apresentaram menor erro na estimativa das áreas agrícolas. A aplicação do algoritmo mostra-se potencial para mapeamento da agricultura por meio de imagens dos sensores MODIS e OLI, porém deve-se avaliar a época de aquisição das imagens orbitais e variações nos parâmetros do algoritmo para melhorar a acurácia da classificação. Palavras-chave: agricultura; classificação supervisionada; aprendizado de máquina AbstractAgriculture is one of the sectors that stands out most in the Brazilian economy, often requiring the use of remote sensing to identify the expansion of agricultural areas and estimates of their production. This work aims to map the agricultural areas of the northwest of Minas Gerais by means of a support vector machine and compare the results obtained with the statistical census of the Brazilian Institute of Geography and Statistics. For the identification of the agricultural areas, the Service Vector Machine algorithm and images of the Landsat 8 / OLI and Terra / MODIS satellites and sensors were used. The training samples of the algorithm were obtained by high resolution spatial image, available in Google Earth Pro software, in the categories rivers, forest, agriculture, pasture and forestry. The OLI image mapping showed better Global Accuracy (0.81) and Kappa (0.66). The classification with OLI and MODIS images showed greater precision in the agriculture class when compared to the other classes, presenting confusion with pasture, due to the high phytomass of the pasture at the time of acquisition of the images (summer). The calculation of the agricultural areas shows an overestimation of the Serv...

show abstract

A Framework for Large-Area Mapping of Past and Present Cropping Activity Using Seasonal Landsat Images and Time Series Metrics

Cited by 51 publications

References 59 publications

Fusion of Sentinel-1A and Sentinel-2A data for land cover mapping: a case study in the lower Magdalena region, Colombia

Fusion of Sentinel-1A and Sentinel-2A data for land cover mapping: a case study in the lower Magdalena region, Colombia

Large-Area Gap Filling of Landsat Reflectance Time Series by Spectral-Angle-Mapper Based Spatio-Temporal Similarity (SAMSTS)

Mapping of Agricultural Areas with Support Vector Machine in the Northwest of Minas Gerais, Brazil

Contact Info

Product

Resources

About