Browsing by Subject "Landsat"

Sort by: Order: Results:

Now showing items 1-9 of 9
  • Tang, Zhipeng; Adhikari, Hari; Pellikka, Petri; Heiskanen, Janne (2021)
    Combined with increasing computing ability, the free and open access to Landsat archive has enabled the changes on the Earth’s surface to be monitored for almost 50 years. However, due to missing observations that result from clouds, cloud shadows, and scan line corrector failure, the Landsat data record is neither a continuous nor consistent time series. We present a new gap-filling method, Missing Observation Prediction based on Spectral-Temporal Metrics (MOPSTM), which uses spectral-temporal metrics computed from Landsat one-year time series and the k-Nearest Neighbor (k-NN) regression. Herein, we demonstrate the performance of MOPSTM by using five, nearly cloud-free, full scene Landsat images from Kenya, Finland, Germany, the USA, and China. Cloud masks from the images with extensive cloud cover were used to simulate large-area gaps, with the highest value we tested being 92% of missing data. The gap-filling accuracy was assessed quantitatively considering all five sites and different land use/land cover types, and the MOPSTM algorithm performed better than the spectral angle-mapper based spatiotemporal similarity (SAMSTS) gap-filling algorithm. The mean RMSE values of MOPSTM were 0.010, 0.012, 0.025, 0.012, and 0.018 for the five sites, while those of SAMSTS were 0.011, 0.017, 0.038, 0.014, and 0.023, respectively. Furthermore, MOPSTM had mean coefficient of determination (R2) values of 0.90, 0.86, 0.78, 0.92, and 0.89, which were higher than those for SAMSTS (0.84, 0.75, 0.55, 0.89, and 0.83). The performance of MOPSTM was not considerably affected by image gap sizes as images ranging from gap sizes of 51% of the image all the way to 92% of the image yielded similar gap-filling accuracy. Also, MOPSTM does not require local parametertuning except for the k values in the k-NN regression, and it can make a gap-free image from any acquisition date. MOPSTM provides a new spectral-temporal approach to generate the gap-free imagery for typical Landsat applications, such as land use, land cover, and forest monitoring.
  • Erlandsson, Rasmus; Bjerke, Jarle W.; Finne, Eirik A.; Myneni, Ranga B.; Piao, Shilong; Wang, Xuhui; Virtanen, Tarmo; Rasanen, Aleksi; Kumpula, Timo; Kolari, Tiina H. M.; Tahvanainen, Teemu; Tommervik, Hans (2022)
    Although generally given little attention in vegetation studies, ground-dwelling (terricolous) lichens are major contributors to overall carbon and nitrogen cycling, albedo, biodiversity and biomass in many high-latitude ecosystems. Changes in biomass of mat-forming pale lichens have the potential to affect vegetation, fauna, climate and human activities including reindeer husbandry. Lichens have a complex spectral signature and terricolous lichens have limited growth height, often growing in mixtures with taller vegetation. This has, so far, prevented the development of remote sensing techniques to accurately assess lichen biomass, which would be a powerful tool in ecosystem and ecological research and rangeland management. We present a Landsat based remote sensing model developed using deep neural networks, trained with 8914 field records of lichen volume collected for > 20 years. In contrast to earlier proposed machine learning and regression methods for lichens, our model exploited the ability of neural networks to handle mixed spatial resolution input. We trained candidate models using input of 1 x 1 (30 x 30 m) and 3 x 3 Landsat pixels based on 7 reflective bands and 3 indices, combined with a 10 m spatial resolution digital elevation model. We normalised elevation data locally for each plot to remove the region-specific variation, while maintaining informative local variation in topography. The final model predicted lichen volume in an evaluation set (n = 159) reaching an R2 of 0.57. NDVI and elevation were the most important predictors, followed by the green band. Even with moderate tree cover density, the model was efficient, offering a considerable improvement compared to earlier methods based on specific reflectance. The model was in principle trained on data from Scandinavia, but when applied to sites in North America and Russia, the predictions of the model corresponded well with our visual interpretations of lichen abundance. We also accurately quantified a recent historic (35 years) change in lichen abundance in northern Norway. This new method enables further spatial and temporal studies of variation and changes in lichen biomass related to multiple research questions as well as rangeland management and economic and cultural ecosystem services. Combined with information on changes in drivers such as climate, land use and management, and air pollution, our model can be used to provide accurate estimates of ecosystem changes and to improve vegetation-climate models by including pale lichens.
  • White, Joanne C.; Saarinen, Ninni; Wulder, Michael A.; Kankare, Ville; Hermosilla, Txomin; Coops, Nicholas C.; Holopainen, Markus; Hyyppä, Juha; Vastaranta, Mikko (2019)
    Information regarding the nature and rate of forest recovery is required to inform forest management, monitoring, and reporting activities. Delayed establishment or return of forests has implications to harvest rotations and carbon uptake, among others, creating a need for spatially-explicit, large-area, characterizations of forest recovery. Landsat time series (LTS) has been demonstrated as a means to quantitatively relate forest recovery, noting that there are gaps in our understanding of the linkage between spectral measures of forest recovery and manifestations of forest structure and composition. Field plots provide a means to better understand the linkage between forest characteristics and spectral recovery indices. As such, from a large set of existing field plots, we considered the conditions present for the year in which the co-located pixel was considered spectrally recovered using the Years to Recovery (Y2R) metric. Y2R is a long-term metric of spectral recovery that indicates the number of years required for a pixel to return to 80% of its pre-disturbance Normalized Burn Ratio value. Absolute and relative metrics of recovery at 5 years post-disturbance were also considered. We used these three spectral recovery metrics to predict the stand development class assigned by the field crew for 284 seedling plots with an overall accuracy of 73.59%, with advanced seedling stands more accurately discriminated (omission error, OE = 15.74%) than young seedling stands (OE = 49.84%). We then used field-measured attributes (e.g. height, stem density, dominant species) from the seedling plots to classify the plots into three spectral recovery groups, which were defined using the Y2R metric: spectral recovery in (1) 1–5 years, (2) 6–10 years, or (3) 11–15 years. Overall accuracy for spectral recovery groups was 61.06%. Recovery groups 1 and 3 were discriminated with greater accuracy (producer’s and user’s accuracies > 66%) than recovery group 2 (<50%). The top field-measured predictors of spectral recovery were mean height, dominant species, and percentage of stems in the plot that were deciduous. Variability in stand establishment and condition make it challenging to accurately discriminate among recovery rates within 10 years post-harvest. Our results indicate that the long-term metric Y2R relates to forest structure and composition attributes measured in the field and that spectral development post-disturbance corresponds with expectations of structural development, particularly height, for different species, site types, and deciduous abundance. These results confirm the utility of spectral recovery measures derived from LTS data to augment landscape-level assessments of post-disturbance recovery.
  • Bhattacharjee, Joy; Marttila, Hannu; Launiainen, Samuli; Lepistö, Ahti; Kløve, Bjørn (Elsevier, 2021)
    Science of The Total Environment 779 (2021), 146419
    Maintaining and improving surface water quality requires knowledge of nutrient and sediment loads due to past and future land-use practices, but historical data on land cover and its changes are often lacking. In this study, we tested whether land-use-specific export coefficients can be used together with satellite images (Landsat) and/or regional land-use statistics to estimate riverine nutrient loads and concentrations of total nitrogen (TN), total phosphorus (TP), and suspended solids (SS). The study area, Simojoki (3160 km2) in northern Finland, has been intensively drained for peatland forestry since the 1960s. We used different approaches at multiple sub-catchment scales to simulate TN, TP, and SS export in the Simojoki catchment. The uncertainty in estimates based on specific export coefficients was quantified based on historical land-use changes (derived from Landsat data), and an uncertainty boundary was established for each land-use. The uncertainty boundary captured at least 60% of measured values of TN, TP, and SS loads or concentrations. However, the uncertainty in estimates compared with measured values ranged from 7% to 20% for TN, 0% to 18% for TP, and 13% to 43% for SS for different catchments. Some discrepancy between predicted and measured loads and concentrations was expected, as the method did not account for inter-annual variability in hydrological conditions or river processes. However, combining historical land-use change estimates with simple export coefficients can be a practical approach for evaluating the influence on water quality of historical land-use changes such as peatland drainage for forest establishment.
  • White, Joanne C.; Saarinen, Ninni; Kankare, Ville; Wulder, Michael A.; Hermosilla, Txomin; Coops, Nicholas C.; Pickell, Paul D.; Holopainen, Markus; Hyyppä, Juha; Vastaranta, Mikko (2018)
    Landsat time series (LTS) enable the characterization of forest recovery post-disturbance over large areas; however, there is a gap in our current knowledge concerning the linkage between spectral measures of recovery derived from LTS and actual manifestations of forest structure in regenerating stands. Airborne laser scanning (ALS) data provide useful measures of forest structure that can be used to corroborate spectral measures of forest recovery. The objective of this study was to evaluate the utility of a spectral index of recovery based on the Normalized Burn Ratio (NBR): the years to recovery, or Y2R metric, as an indicator of the return of forest vegetation following forest harvest (clearcutting). The Y2R metric has previously been defined as the number of years required for a pixel to return to 80% of its pre-disturbance NBR (NBRpre) value. In this study, the Composite2Change (C2C) algorithm was used to generate a time series of gap-free, cloud-free Landsat surface reflectance composites (1985–2012), associated change metrics, and a spatially-explicit dataset of detected changes for an actively managed forest area in southern Finland (5.3 Mha). The overall accuracy of change detection, determined using independent validation data, was 89%. Areas of forest harvesting in 1991 were then used to evaluate the Y2R metric. Four alternative recovery scenarios were evaluated, representing variations in the spectral threshold used to define Y2R: 60%, 80%, and 100% of NBRpre, and a critical value of z (i.e. the year in which the pixel's NBR value is no longer significantly different from NBRpre). The Y2R for each scenario were classified into five groups: recovery within 17 years, and not recovered. Measures of forest structure (canopy height and cover) were obtained from ALS data. Benchmarks for height (>5 m) and canopy cover (>10%) were applied to each recovery scenario, and the percentage of pixels that attained both of these benchmarks for each recovery group, was determined for each Y2R scenario. Our results indicated that the Y2R metric using the 80% threshold provided the most realistic assessment of forest recovery: all pixels considered in our analysis were spectrally recovered within the analysis period, with 88.88% of recovered pixels attaining the benchmarks for both cover and height. Moreover, false positives (pixels that had recovered spectrally, but not structurally) and false negatives (pixels that had recovered structurally, but not spectrally) were minimized with the 80% threshold. This research demonstrates the efficacy of LTS-derived assessments of recovery, which can be spatially exhaustive and retrospective, providing important baseline data for forest monitoring.
  • Hiremath, Santosh; Wittke, Samantha; Palosuo, Taru; Kaivosoja, Jere; Tao, Fulu; Proll, Maximilian; Puttonen, Eetu; Peltonen-Sainio, Pirjo; Marttinen, Pekka; Mamitsuka, Hiroshi (Public Library of Science, 2021)
    PLoS ONE
    Identifying crop loss at field parcel scale using satellite images is challenging: first, crop loss is caused by many factors during the growing season; second, reliable reference data about crop loss are lacking; third, there are many ways to define crop loss. This study investigates the feasibility of using satellite images to train machine learning (ML) models to classify agricultural field parcels into those with and without crop loss. The reference data for this study was provided by Finnish Food Authority (FFA) containing crop loss information of approximately 1.4 million field parcels in Finland covering about 3.5 million ha from 2000 to 2015. This reference data was combined with Normalised Difference Vegetation Index (NDVI) derived from Landsat 7 images, in which more than 80% of the possible data are missing. Despite the hard problem with extremely noisy data, among the four ML models we tested, random forest (with mean imputation and missing value indicators) achieved the average AUC (area under the ROC curve) of 0.688±0.059 over all 16 years with the range [0.602, 0.795] in identifying new crop-loss fields based on reference fields of the same year. To our knowledge, this is one of the first large scale benchmark study of using machine learning for crop loss classification at field parcel scale. The classification setting and trained models have numerous potential applications, for example, allowing government agencies or insurance companies to verify crop-loss claims by farmers and realise efficient agricultural monitoring.
  • doninck, Jasper Van; Jones, Mirkka M.; Zuquim, Gabriela; Ruokolainen, Kalle; Moulatlet, Gabriel M.; Sirén, Anders; Cárdenas, Glenda; Lehtonen, Samuli; Tuomisto, Hanna (2020)
    Species distribution models are required for the research and management of biodiversity in the hyperdiverse tropical forests, but reliable and ecologically relevant digital environmental data layers are not always available. We here assess the usefulness of multispectral canopy reflectance (Landsat) relative to climate data in modelling understory plant species distributions in tropical rainforests. We used a large dataset of quantitative fern and lycophyte species inventories across lowland Amazonia as the basis for species distribution modelling (SDM). As predictors, we used CHELSA climatic variables and canopy reflectance values from a recent basin-wide composite of Landsat TM/ETM+ images both separately and in combination. We also investigated how species accumulate over sites when environmental distances were expressed in terms of climatic or surface reflectance variables. When species accumulation curves were constructed such that differences in Landsat reflectance among the selected plots were maximised, species accumulated faster than when climatic differences were maximised or plots were selected in a random order. Sixty-nine species were sufficiently frequent for species distribution modelling. For most of them, adequate SDMs were obtained whether the models were based on CHELSA data only, Landsat data only or both combined. Model performance was not influenced by species’ prevalence or abundance. Adding Landsat-based environmental data layers overall improved the discriminatory capacity of SDMs compared to climate-only models, especially for soil specialist species. Our results show that canopy surface reflectance obtained by multispectral sensors can provide studies of tropical ecology, as exemplified by SDMs, much higher thematic (taxonomic) detail than is generally assumed. Furthermore, multispectral datasets complement the traditionally used climatic layers in analyses requiring information on environmental site conditions. We demonstrate the utility of freely available, global remote sensing data for biogeographical studies that can aid conservation planning and biodiversity management.
  • Ershov, Dmitry V.; Gavrilyuk, Egor A.; Koroleva, Natalia V.; Belova, Elena I.; Tikhonova, Elena V.; Shopina, Olga V.; Titovets, Anastasia V.; Tikhonov, Gleb N. (2022)
    Remote monitoring of natural afforestation processes on abandoned agricultural lands is crucial for assessments and predictions of forest cover dynamics, biodiversity, ecosystem functions and services. In this work, we built on the general approach of combining satellite and field data for forest mapping and developed a simple and robust method for afforestation dynamics assessment. This method is based on Landsat imagery and index-based thresholding and specifically targets suitability for limited field data. We demonstrated method's details and performance by conducting a case study for two bordering districts of Rudnya (Smolensk region, Russia) and Liozno (Vitebsk region, Belarus). This study area was selected because of the striking differences in the development of the agrarian sectors of these countries during the post-Soviet period (1991-present day). We used Landsat data to generate a consistent time series of five-year cloud-free multispectral composite images for the 1985-2020 period via the Google Earth Engine. Three spectral indices, each specifically designed for either forest, water or bare soil identification, were used for forest cover and arable land mapping. Threshold values for indices classification were both determined and verified based on field data and additional samples obtained by visual interpretation of very high-resolution satellite imagery. The developed approach was applied over the full Landsat time series to quantify 35-year afforestation dynamics over the study area. About 32% of initial arable lands and grasslands in the Russian district were afforested by the end of considered period, while the agricultural lands in Belarus' district decreased only by around 5%. Obtained results are in the good agreement with the previous studies dedicated to the agricultural lands abandonment in the Eastern Europe region. The proposed method could be further developed into a general universally applicable technique for forest cover mapping in different growing conditions at local and regional spatial levels.
  • Maeda, Eduardo Eiji; Lisboa, Filipe; Kaikkonen, Laura Maria; Koponen, Sampsa; Kallio, Kari; Brotas, Vanda; Kuikka, Olli Sakari (2019)
    Monitoring temporal changes in phytoplankton dynamics in high latitude lakes is particularly timely for understanding the impacts of warming on aquatic ecosystems. In this study, we analyzed 33-years of high resolution (30 m) Landsat (LT) data for reconstructing seasonal patterns of chlorophyll a (chl a) concentration in four lakes across Finland, between 60°N and 64°N. Chl a models based on LT spectral bands were calibrated using 17-years (2000–2016) of field measurements collected across the four lakes. These models were then applied for estimating chl a using the entire LT-5 and 7 archives. Approximately 630 images, from 1984 to 2017, were analyzed for each lake. The chl a seasonal patterns were characterized using phenology metrics, and the time-series of LT-based chl a estimates were used for identifying temporal shifts in the seasonal patterns of chl a concentration. Our results showed an increase in the length of phytoplankton growth season in three of the lakes. The highest increase was observed in Lake Köyliönjärvi, where the length of growth season has increased by 28 days from the baseline period of 1984–1994 to 2007–2017. The increase in the length of season was mainly attributed to an earlier start of phytoplankton blooms. We further analyzed surface temperature (Ts) and precipitation data to verify if climatic factors could explain the shifts in the seasonal patterns of chl a. We found no direct relationship between Ts and chl a seasonal patterns. Similarly, the phenological metrics of Ts, in particular length of season, did not show significant temporal trends. On the other hand, we identify potential links between changes in precipitation patterns and the increase in the phytoplankton season length. We verified a significant increase in the rainfall contribution to the total precipitation during the autumn and winter, accompanied by a decline in snowfall volumes. This could indicate an increasing runoff volume during the beginning of spring, contributing to an earlier onset of the phytoplankton blooms, although further assessments are needed to analyze historical streamflow values and nearby land cover data. Likewise, additional studies are needed to better understand why chl a patterns in some lakes seem to be more resilient than in others.