Time filter

Source Type

Colditz R.R.,National Commission for the Knowledge and Use of Biodiversity CONABIO
MultiTemp 2013 - 7th International Workshop on the Analysis of Multi-Temporal Remote Sensing Images: "Our Dynamic Environment", Proceedings

To eliminate atmospheric disturbances and sensor malfunctions many remote sensing products are offered in form of composites that combine multiple observations over a defined period. However, selecting specific observations result in varying interval lengths which affect and limit time series analysis techniques. This study provides a conceptual view on the possibilities to use or define specific composite days and its impact on time series. Often the starting or final day of the compositing period is selected, alternatively the middle day may be chosen. These time series are compared to a series that considers the actual day of observation. An experimental exercise employs 16-day MODIS VI composites of 1000m spatial resolution for a 1100×500km region in central Mexico. Statistical measures such as temporal cross-correlation, harmonic analysis, and difference in start, center, and end of season were used to characterize the temporal shift in time series. A shift of approximately seven days with a high variability is introduced when using the starting day of the compositing period, which is mitigated with assuming the middle day. However, only time series that take into account the day of observation can be used for correct estimation of temporal characteristics. © 2013 IEEE. Source

Cord A.F.,Helmholtz Center for Environmental Research | Klein D.,German Aerospace Center | Mora F.,National Commission for the Knowledge and Use of Biodiversity CONABIO | Dech S.,German Aerospace Center | Dech S.,University of Wurzburg
Ecological Modelling

Given the rapid loss of biodiversity worldwide and the resulting impacts on ecosystem functions and services, we more than ever rely on current and spatially continuous assessments of species distributions for biodiversity conservation and sustainable land management. Over the last decade, the usefulness of categorical land cover data to account for the human-induced degradation, transformation and loss of natural habitat in species distribution models (SDMs) has been questioned and the number of studies directly analyzing remotely sensed variables has lately multiplied. While several assumptions support the advantages of remote sensing data, an empirical comparison is still lacking. The objective of this study was to bridge this gap and compare the suitability of an existing categorical land cover classification and of continuous remote sensing variables for modeling the distribution patterns of 30 Mexican tree species. We applied the Maximum Entropy algorithm to predict species distributions based on both data types independently, quantified model performance and analyzed species-land cover relationships in detail. As part of this comparison, we focused on two particular aspects, namely the effects of (1) thematic detail and (2) spatial resolution of the land cover data on model performance. Our analysis revealed that remote sensing data were significantly better model predictors and that the main obstacle of the land cover-based SDMs were their bolder predictions, together with their overall overestimation of suitability. Among the land cover-based models, we found that thematic detail was more important than spatial resolution for SDM performance. However, our results also suggest that the suitability of land cover data differs largely among species and is dependent on their habitat distinctiveness. Our findings have relevant implications for future species distribution modeling studies which aim at complementing their set of topo-climatic predictors by data on land surface characteristics. © 2013 Elsevier B.V. Source

Molgora J.M.E.,National Commission for the Knowledge and Use of Biodiversity CONABIO
ACM International Conference Proceeding Series

Since 2000 CONABIO has processed and published in near real time geospatial information of occurring wild fires in Mexico and Central America as part of the " Hotspot detection with remote sensing techniques" program. Currently the Fire Alert System receives via direct-broadcast raw data from MODIS (Aqua and Terra platforms) and VIIRS (NPP). This raw data is processed to L1A, L1B, MOD14, MOD13 and MOD11 products (at the moment we only process VIIRS to L1B). MOD14 gives us a Fire-Mask raster that has the coordinates of the thermal anomalies called hotspots (Giglio, 2003). Each hotspot represents a 1 Km squared polygon that, with nearly 0.9 probability, has a uniformly distributed fire. The system automatically publishes information of county level locality name, area slope, ecosystem type and closeness to the nearest "Natural Protected Area". The system also has a WMS service provided by GeoServer that stores thematic map layers of Fire Propagation Index, a NDVI time-series analysis model that measures vegetation anomalies, and Moisture Percentage Map, based on dead vegetation fuel and rainfall duration raster products. Every 24 hrs we process a mean of 6 satellite passes during day and night. Over time, the amount of data has surpass the original capacity of the system. Although the first generation software has been upgraded to accomplish the needs of users, specially fire fighters, researches and decision-making people, the urgent need of a platform capable of forecasting, performing complex analysis and integrating it with next generation data sources has turned the use and administration of the old system an awkward task. A next-generation system is being developed at the moment and the release date is expected by the end of this year. The system lies on a Postgres9.1-Postgis2.0 with raster support engine that decreases dramatically the processing time. The design of a software based on a spatially-enable database has opened a new world of opportunities. This system will be based purely on Open Source Software. Fire Alert System homepage: http://www.conabio.gob.mx/incendios/. © 2012 Author. Source

Land cover mapping for large regions often employs satellite images of medium to coarse spatial resolution, which complicates mapping of discrete classes. Class memberships, which estimate the proportion of each class for every pixel, have been suggested as an alternative. This paper compares different strategies of training data allocation for discrete and continuous land cover mapping using classification and regression tree algorithms. In addition to measures of discrete and continuous map accuracy the correct estimation of the area is another important criteria. A subset of the 30 m national land cover dataset of 2006 (NLCD2006) of the United States was used as reference set to classify NADIR BRDF-adjusted surface reflectance time series of MODIS at 900 m spatial resolution. Results show that sampling of heterogeneous pixels and sample allocation according to the expected area of each class is best for classification trees. Regression trees for continuous land cover mapping should be trained with random allocation, and predictions should be normalized with a linear scaling function to correctly estimate the total area. From the tested algorithms random forest classification yields lower errors than boosted trees of C5.0, and Cubist shows higher accuracies than random forest regression. © 2015 by the authors; licensee MDPI, Basel, Switzerland. Source

Colditz R.R.,National Commission for the Knowledge and Use of Biodiversity CONABIO
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

Many ecological and agricultural studies require accurate information about the phenological state of the vegetation such as the start, peak/plateau, and end of the growing season. This study investigates the impact of the day of observation in image composites on time-series generation and analysis. It employs daily Moderate Resolution Imaging Spectroradiometer (MODIS) surface reflectance and 16-day vegetation index (VI) composites and combines data streams from Terra and Aqua. Results of a time series that take into account the day of observation and time series that assume the starting, ending, or middle day of the compositing period were compared to a reference of daily observations. A temporal shift of approximately 7.5 days with a high variation in error is introduced if the start of the compositing period is assumed. The middle day mitigates the mean error close to zero but cannot fully compensate for temporal delays. Only a time series that takes into account the actual day of observation can be used for the correct estimation of temporal characteristics. In addition, the study addresses that the day of observation for time-series generation from image composites is imperative when combining data of different data streams with phased production cycles such as MODIS VI composites from Terra and Aqua. © 2014 IEEE. Source

Discover hidden collaborations