Prithvi-EO-2.0: A Versatile Multitemporal Foundation Model for Earth Observation Applications

Land cover type is a crucial parameter that is required for various land surface models that simulate water and carbon cycles, ecosystem dynamics, and climate change. Many land use/land cover maps used in recent years have been derived from field investigations and remote-sensing observations. However, no land cover map that is derived from a single source (such as satellite observation) properly meets the needs of land surface simulation in China. This article presents a decision-fuse method to produce a higher-accuracy land cover map by combining multi-source local data based on the Dempster–Shafer (D–S) evidence theory. A practical evidence generation scheme was used to integrate multi-source land cover classification information. The basic probability values of the input data were obtained from literature reviews and expert knowledge. A Multi-source Integrated Chinese Land Cover (MICLCover) map was generated by combining multi-source land cover/land use classification maps including a 1:1,000,000 vegetation map, a 1:100,000 land use map for the year 2000, a 1:1,000,000 swamp-wetland map, a glacier map, and a Moderate-Resolution Imaging Spectroradiometer land cover map for China in 2001 (MODIS2001). The merit of this new map is that it uses a common classification system (the International Geosphere-Biosphere Programme (IGBP) land cover classification system), and it has a unified 1 km resolution. The accuracy of the new map was validated by a hybrid procedure. The validation results show great improvement in accuracy for the MICLCover map. The local-scale visual comparison validations for three regions show that the MICLCover map provides more spatial details on land cover at the local scale compared with other popular land cover products. The improvement in accuracy is true for all classes but particularly for cropland, urban, glacier, wetland, and water body classes. Validation by comparison with the China Forestry Scientific Data Center (CFSDC)–Forest Inventory Data (FID) data shows that overall forest accuracies in five provinces increased to between 42.19% and 88.65% for our MICLCover map, while those of the MODIS2001 map increased between 27.77% and 77.89%. The validation all over China shows that the overall accuracy of the MICLCover map is 71%, which is higher than the accuracies of other land cover maps. This map therefore can be used as an important input for land surface models of China. It has the potential to improve the modeling accuracy of land surface processes as well as to support other aspects of scientific land surface investigations in China.

Limitations of mapping land surface properties and their conversion into climate model boundary conditions are major sources of uncertainty in climate simulations. In this paper, the range of the largest possible uncertainty in satellite-derived land cover (LC) map is estimated and its impact on climate simulations is quantified with the Earth System Model of the Max-Planck Institute for Meteorology utilizing prescribed sea surface temperature and sea ice. Two types of uncertainty in the LC map are addressed: (i) uncertainty due to classification algorithm of spectral reflectance into LC classes, and (ii) uncertainty due to conversion of LC classes into the climate model vegetation distribution. For forest cover, each of them is about the same order of magnitude as the uncertainty range in recent observations (∼± 700 Mha). Superposing two sources of uncertainty results in LC maps that feature the range of vegetation deviation that is about the same order of magnitude as the recent (since year 1700) forest loss due to agriculture (forest cover uncertainty range ∼± 1700 Mha). These uncertainties in vegetation distribution lead to noticeable variations in near-surface climate variables, local, regional, and global climate forcing. Temperature does not show significant uncertainty in global mean, but rather exhibits regional deviations with an opposite response to LC uncertainty that compensate each other in the global mean (e.g., albedo feedback controls temperature in boreal North America resulting in cooling (warming) with decrease (increase) of vegetation while evaporative cooling controls temperature in South America and sub-Saharan Africa resulting in cooling (warming) with increase (decrease) of vegetation). Large-scale circulation is also affected by the LC uncertainty, and consequently precipitation pattern as well. It is demonstrated that precipitation uncertainty in the monsoonal regions are about the same order of magnitude as in previous studies with idealized perturbations of vegetation. These findings indicate that the range of uncertainty in satellite-derived vegetation maps for climate models is about the same order of magnitude as the uncertainty in recent observations of forest cover or as the forest lost due to agriculture. Consequently, climate simulations have a similar range of uncertainty in variables representing near-surface climate as the observed climate change due to land use. Hence, more accurate methods are needed for mapping and converting LC properties into model vegetation in order to increase reliability of climate model simulations.

Characterization of climate uncertainties due to different land cover maps in regional climate models is essential for adaptation strategies. The spatiotemporal heterogeneity in surface characteristics is considered to play a key role in terrestrial surface processes. Here, we quantified the sensitivity of model results to changes in land cover input data (GlobCover 2009, GLC 2000, CCI, and ECOCLIMAP) in the regional climate model (RCM) COSMO-CLM (v5.0_clm16). We investigated land cover changes due to the retrieval year, number, fraction and spatial distribution of land cover classes by performing convection-permitting simulations driven by ERA5 reanalysis data over Germany from 2002 to 2011. The role of the surface parameters on the surface turbulent fluxes and temperature is examined, which is related to the land cover classes. The bias of the annual temperature cycle of all the simulations compared with observations is larger than the differences between simulations. The latter is well within the uncertainty of the observations. The land cover class fractional differences are small among the land cover maps. However, some land cover types, such as croplands and urban areas, have greatly changed over the years. These distribution changes can be seen in the temperature differences. Simulations based on the CCI retrieved in 2000 and 2015 revealed no accreditable difference in the climate variables as the land cover changes that occurred between these years are marginal, and thus, the influence is small over Germany. Increasing the land cover types as in ECOCLIMAP leads to higher temperature variability. The largest differences among the simulations occur in maximum temperature and from spring to autumn, which is the main vegetation period. The temperature differences seen among the simulations relate to changes in the leaf area index, plant coverage, roughness length, latent and sensible heat fluxes due to differences in land cover types. The vegetation fraction was the main parameter affecting the seasonal evolution of the latent heat fluxes based on linear regression analysis, followed by roughness length and leaf area index. If the same natural vegetation (e.g. forest) or pasture grid cells changed into urban types in another land cover map, daily maximum temperatures increased accordingly. Similarly, differences in climate extreme indices are strongest for any land cover type change to urban areas. The uncertainties in regional temperature due to different land cover datasets were overall lower than the uncertainties associated with climate projections. Although the impact and their implications are different on different spatial and temporal scales as shown for urban area differences in the land cover maps. For future development, more attention should be given to land cover classification in complex areas, including more land cover types or single vegetation species and regional representative classification sample selection. Including more sophisticated urban and vegetation modules with synchronized input data in RCMs would improve the underestimation of the urban and vegetation effect on local climate.

Precise global/regional land cover mapping is of fundamental importance in studies of land surface processes and modelling. Quantitative assessments of the map quality and classification accuracy for existing land cover maps will help to improve accuracy in future land cover mapping. We compare and evaluate four land cover datasets over China. The datasets include the Version 2 global land cover dataset of IGBP, MODIS land cover map 2001, a global land cover map produced by the University of Maryland, and the land cover map produced by the global land cover for the year 2000 (GLC 2000) project coordinated by the Global Vegetation Monitoring Unit of the European Commission Joint Research Centre. The four maps used different classification systems, which made the comparison difficult. So we first aggregated these maps by reclassifying them using a unified legend system. A large-scale, i.e. 1:100 000 land cover map of China was used as the reference data to validate the four maps. The results show that the GLC2000 land cover map represents the highest accuracy. However, it has obvious local labelling errors and a zero labelling accuracy for the wetland type. The MODIS land cover map ranks second for type area consistency and third for sub-fraction overall accuracy compared with reference data, which may be affected by the local labelling error. The IGBP land cover map has good labelling accuracy, although it has a local labelling error and third consistency for type area. The labelling accuracy and type area consistency for the reference data of UMd land cover map is low. We conclude that the accuracies of all the datasets cannot meet the requirements of land surface modelling. For the reference data, i.e. the 1:100 000 land cover map, the classification system needs to be transferred to a well recognized one that has been used commonly in land surface modelling. In addition, we propose an information fusion strategy to produce a more accurate land cover map of China whose classification system should be compatible with the well-accepted classification system used in land surface modelling.

Two decades of land cover mapping in the Río de la Plata grassland region: The MapBiomas Pampa initiative

This paper proposed a new land cover mapping algorithm for remote sensing images under the presence of clouds. Here, we modelled the observed image as the weighted sum between the reflectances from the cloud-free image and clouds where weights depend on the degree of cloud contamination called the ‘cloud thickness’ map. Next, we represented the cloud thickness map using a level set function. Similarly, we also represented the land cover map as functions of different level set functions whose values indicate the class label. Next, the maximum a posteriori (MAP) criterion is employed where the most likely land cover and cloud thickness maps are chosen. Under the MAP criterion, the corresponding energy function whose values depend on the level set functions can be derived where the minimum energy point corresponds to the optimum land cover and cloud thickness maps. Since the level set functions are continuous, the optimum solution can be obtained by applying the calculus of variation where the gradient descent algorithm is employed. From the observation model, the cloud-free data can only be reconstructed only if thin to medium clouds are present since thick clouds can completely block the reflectance from the land cover materials. Thus, our algorithm is more suitable for thin to medium than thick cloud covers. Our synthesis and real cloud-contamination examples support our statement since our algorithm achieves significantly higher overall accuracies over other classification techniques, especially, in the thin to medium cloud contamination, and fails to recover the true land cover classes over the thick clouds.

Land cover (LC) information plays an important role in different geoscience applications such as land resources and ecological environment monitoring. Enhancing the automation degree of LC classification and updating at a fine scale by remote sensing has become a key problem, as the capability of remote sensing data acquisition is constantly being improved in terms of spatial and temporal resolution. However, the present methods of generating LC information are relatively inefficient, in terms of manually selecting training samples among multitemporal observations, which is becoming the bottleneck of application-oriented LC mapping. Thus, the objectives of this study are to speed up the efficiency of LC information acquisition and update. This study proposes a rapid LC map updating approach at a geo-object scale for high-spatial-resolution (HSR) remote sensing. The challenge is to develop methodologies for quickly sampling. Hence, the core step of our proposed methodology is an automatic method of collecting samples from historical LC maps through combining change detection and label transfer. A data set with Chinese Gaofen-2 (GF-2) HSR satellite images is utilized to evaluate the effectiveness of our method for multitemporal updating of LC maps. Prior labels in a historical LC map are certified to be effective in a LC updating task, which contributes to improve the effectiveness of the LC map update by automatically generating a number of training samples for supervised classification. The experimental outcomes demonstrate that the proposed method enhances the automation degree of LC map updating and allows for geo-object-based up-to-date LC mapping with high accuracy. The results indicate that the proposed method boosts the ability of automatic update of LC map, and greatly reduces the complexity of visual sample acquisition. Furthermore, the accuracy of LC type and the fineness of polygon boundaries in the updated LC maps effectively reflect the characteristics of geo-object changes on the ground surface, which makes the proposed method suitable for many applications requiring refined LC maps.

Approaches to fractional land cover and continuous field mapping: A comparative assessment over the BOREAS study region

Achieving food security in particular in Africa continues to pose a major challenge to humankind. It is clear that the future agricultural potential of Africa plays a critical role in meeting this challenge. Although crop yield can be estimated with a degree of reliability using a limited sample of ground observations, the exact crop acreage and the spatial distribution are rarely available. Even though remote sensing offers the ability to produce a rapid and up-to-date land use and land cover database for agricultural monitoring, there are only a few countries in Africa where higher resolution satellite data such as Landsat have been used for land cover map production at the national level. However a number of global products have been produced which contain information on cropland extent. This paper will outline a comparison of four sources of land cover data to determine which product is the most suitable for agricultural monitoring and for the subsequent development of a crop mask. The land cover products used are the Global Land Cover Map (GLC-2000), the Moderate Resolution Imaging Spectroradiometer (MODIS) land cover product (MOD12V1), the SAGE (Center for Sustainability and the Global Environment) and the AFRICOVER dataset from the Food and Agriculture Organisation (FAO). Both the GLC-2000 and MODIS land cover products are at a resolution of 1 km2 while AFRICOVER, based on visual interpretation of 30-m resolution Landsat images, is available at a much finer resolution. The four land cover products are first aggregated to the same resolution so that they can be compared. The legend categories of the four land cover products in this study are reconciled using the method developed in Fritz and See [Fritz, S. and See, L., 2005, Comparison of land cover maps using fuzzy agreement. International Journal of Geographic Information Science, 19, pp. 787–807.] and See and Fritz [See, L.M. and Fritz, S., 2005, A user-defined fuzzy logic approach to comparing global land cover products. In 14th European Colloquium on Theoretical and Quantitative Geography, 9–12 September 2005, Lisbon, Portugal.] that allows overlap between legend definitions to be taken into account. Once the legend definitions between the different land cover products are reconciled, the maps are then compared with national and sub-national statistics. Analysis is undertaken at both continental and national scales as well as sub-national for Sudan and Eritrea. The study generally concludes that MODIS has the tendency to underestimate cropland cover when compared with FAO statistics or AFRICOVER data, whereas GLC-2000 tends to overestimate cropland cover in those countries that are located at the northern transition zone of subtropical shrubland and semi-desert areas. In this area MODIS and SAGE show a relatively similar cropland distribution. Even though the SAGE database has been calibrated with national statistics, it does not perform better than the other two datasets overall, and has highlighted the fact that the SAGE data show regional weaknesses and should be replaced in certain regions by more recent datasets such as GLC-2000 and MODIS, or ideally by a hybrid product that combines the best of the three products, depending upon the region and country.

Using the three-layer variable infiltration capacity (VIC-3L) hydrological model and the successive interpolation approach (SIA) of climate factors, the authors studied the effect of different land cover types on the surface hydrological cycle. Daily climate data from 1992 to 2001 and remotely-sensed leaf area index (LAI) are used in the model. The model is applied to the Baohe River basin, a subbasin of the Yangtze River basin, China, with an area of 2500 km2. The vegetation cover types in the Baohe River basin consist mostly of the mixed forest type (∼85%). Comparison of the modeled results with the observed discharge data suggests that: (1) Daily discharges over the period of 1992–2001 simulated with inputs of remotely-sensed land cover data and LAI data can generally produce observed discharge variations, and the modeled annual total discharge agrees with observations with a mean difference of 1.4%. The use of remote sensing images also makes the modeled spatial distributions of evapotranspiration physically meaningful. (2) The relative computing error (RCE) of the annual average discharge is −24.8% when the homogeneous broadleaf deciduous forestry cover is assumed for the watershed. The error is 21.8% when a homogeneous cropland cover is assumed and −14.32% when an REDC (Resource and Environment Database of China) land cover map is used. The error is reduced to 1.4% when a remotely-sensed land cover at 1000-m resolution is used.

Very-high-resolution (VHR) optical imaging satellites can offer precise, accurate, and direct measurements of snow-covered areas (SCA) with sub-meter to meter-scale resolution in regions of complex land cover and terrain. We explore the potential of Maxar WorldView-2 and WorldView-3 in-track stereo images (WV) for land and snow cover mapping at two sites in the Western U.S. with different snow regimes, topographies, vegetation, and underlying geology. We trained random forest models using combinations of multispectral bands and normalized difference indices (i.e., NDVI) to produce land cover maps for priority feature classes (snow, shaded snow, vegetation, water, and exposed ground). We then created snow-covered area products from these maps and compared them with coarser resolution satellite fractional snow-covered area (fSCA) products from Landsat (~30 m) and MODIS (~500 m). Our models generated accurate classifications, even with limited combinations of available multispectral bands. Models trained on a single image demonstrated limited model transfer, with best results found for in-region transfers. Coarser-resolution Landsat and MODSCAG fSCA products identified many more pixels as completely snow-covered (100% fSCA) than WV fSCA. However, while MODSCAG fSCA products also identified many more completely snow-free pixels (0% fSCA) than WV fSCA, Landsat fSCA products only slightly underestimated the number of completely snow-free pixels. Overall, our results demonstrate that strategic image observations with VHR satellites such as WorldView-2 and WorldView-3 can complement the existing operational snow data products to map the evolution of seasonal snow cover.

National land cover map with 30m resolution, an important database for studying the interaction between human and the environment, is a tedious work. The rise of deep learning technique provides a new idea for the work. This paper reports a novel method based on deep convolutional neural networks for the national land cover mapping task. The proposed method has four major parts: classification system, data sources, training samples selection, and training & inferencing. The produced deep learning (DL)-based land cover map is compared with two highly accepted land cover maps (the reference land cover map and the GLC30). Overall accuracies of GLC30 and DL-land cover map are 76.45% and 82.59% when considering the reference land cover map as the ground truth. Overall accuracies of the reference land cover map and the DL-land cover map are 74.25% and 78.87% when the GLC30 is treated as the ground truth.

Enhancing MODIS land cover product with a spatial–temporal modeling algorithm

Abstract. Land Cover (LC) maps represent key resources to understand, model and address many global and local dynamics affecting our planet. They are usually derived from the classification of satellite imagery, after which a validation or intercomparison process is performed to assess their accuracy. This paper presents the project “Capacity Building for High-Resolution Land Cover Intercomparison and Validation”, an educational initiative funded by the International Society for Photogrammetry and Remote Sensing (ISPRS) and mainly targeting developing countries. First, with the help of two open surveys, an analysis of the state of the art was performed which assessed the overall good awareness on LC maps and the needs and requirements for validating and comparing them, as well as the rich availability of educational material on this topic. The second task, currently under finalization, is the development of new educational material, based on open source software and released under an open access license, consisting of: an introduction to the GlobeLand30 (GL30) LC map and its online platform; a desktop GIS procedure showing two use cases on GL30 validation; and an application to collect LC data on the field to be used for validation. Finally, this educational material will be tested in practice in three workshops during the second half of the project, two of which held in developing countries: Dar es Salaam, Tanzania and Nairobi, Kenya.

Remote sensing based land cover mapping at large scale is time consuming when using either supervised or unsupervised classification approaches. This article used a fast clustering method—Clustering by Eigen Space Transformation (CBEST) to produce a land cover map for China. Firstly, 508 Landsat TM scenes were collected and processed. Then, TM images were clustered by combining CBEST and K-means in each pre-defined ecological zone (50 in total for China). Finally, the obtained clusters were visually interpreted as land cover types to complete a land cover map. Accuracy evaluation using 2159 test samples indicates an overall accuracy of 71.7% and a Kappa coefficient of 0.64. Comparisons with two global land cover products (i.e., Finer Resolution Observation and Monitoring of Global Land Cover (FROM-GLC) and GlobCover 2009) also indicate that our land cover result using CBEST is superior in both land cover area estimation and visual effect for different land cover types.