Revista de Teledetección

Classification of land use and land cover through machine learning algorithms: a literature review

2023-04-28T17:41:32+02:00

Methodologies for land use and land cover (LULC) classification have demonstrated significant advances in recent years, such as the incorporation of machine learning (ML) classification techniques, which have gained popularity and acceptance of their capabilities. However, the lack of methodological consensus has led to a disorderly application of ML methods in the classification of LULC. Through the literature review, we identified some points in how the methods are being implemented as possible implications for the classification of LULC. For this review, only scientific articles published between 2000 and 2020 were analyzed that incorporated any of the following algorithms for LULC classification: K-nearest neighbor (KNN), random forest (RF), support vector machine (SVM), artificial neural network (ANN) and decision trees (DT). Using the results of the literature review, we were able to confirm the potential of the algorithms. We also identified areas for improvement in the application of machine learning to the classification of LULC. These areas include the integration of data sets, parameterization of algorithms, and evaluation of results. Consequently, we generated a selection of guidelines based on the recommendations of various authors that we consider will be useful for users interested in these methods.

Calibration of volumetric soil moisture using Landsat-8 and Sentinel-2 satellite imagery by Google Earth Engine

2023-05-26T19:11:13+02:00

Water scarcity for agriculture is increasingly evident due to climatic alterations and inadequate management of this resource. Therefore, developing digital models that help improve water resource management to provide solutions to agronomic problems in northern Mexico is necessary. In this context, the objective of the present research is to calibrate the Optical Trapezoid (OPTRAM) and Thermal-Optical Trapezoid (TOTRAM) models to estimate the volumetric soil moisture at different depths through vegetation indices derived from Landsat-8 and Sentinel-2 satellite images using Google Earth Engine (GEE). Agricultural areas under gravity irrigation and rainfed runoff in the Comarca Lagunera, the lower part of the Hydrological Region No. 36 of the Nazas and Aguanaval rivers were selected for in-situ measurements. The OPTRAM and TOTRAM normalized moisture content (W) estimates were compared with in-situ volumetric soil moisture (Ɵ) data. Results indicate that the predictions of OPTRAM errors using Sentinel-2 images showed RMSE between 0.033 to 0.043 cm³ cm^-3 and R² between 0.66 to 0.75, whereas Landsat-8 errors showed RSME from 0.036 to from 0.036 to 0.057 cm³ cm^-3and R² between 0.70 to 0.81. On the other hand, TOTRAM errors showed RMSE between 0.045 to 0.053 cm³ cm^-3 and R² between 0.62 to 0.85 through calibrations. This study made it possible to evaluate the most accurate combinations of the pixel distributions of each model and vegetation indices for the estimation of volumetric soil moisture within the different phenological stages of the crops.

Spatio-temporal analysis of harmful algal blooms in tropical crater-lake from MODIS data (2003-2020)

2023-06-26T12:12:00+02:00

The crater lake of Santa María del Oro in Nayarit, presents Algal Blooms (AB) in a cyclical annual manner, the blooming and subsequent decline of these populations creates color changes in the water, generally in the first half of the year. This work evaluated supervised classification algorithms that allow these changes to be identified using data from the MOD09GQ and MYD09GQ products of MODIS sensor in the period from January 2003 to December 2020. Based on a review of AB recorded in the literature and statistical analysis of dispersion graphs, a database of spectral information and lake color state labels were built to evaluate the different classification algorithms. The best classifier was Random Forest with an accuracy of 87.1%. The temporal analysis and spatial evaluation of the blooms incidence showed that may, april and march are the months with the greatest presence of color changes related to AB in the lake. The spatial analysis found that the highest incidence of blooms occurs in the southeast region of the lake and the largest amounts of events occurred in the years 2011, 2008 and 2012 respectively. The influence of the El Niño-Southern Oscillation (ENSO) phenomenon on the incidence of algal blooms in the crater lake is determined due to the temporal pattern between the anomalies in the AB and the Multivariate ENSO Index, where the greater number of AF events occurred in the cold phases of the ENSO.

Biomass and carbon estimation with remote sensing tools in tropical dry forests of Tolima, Colombia

2023-06-16T11:53:22+02:00

Forests store a large amount of carbon in biomass, which constitutes an option for climate change mitigation. This research focused on the estimation of aboveground biomass and carbon using remote sensing and mathematical modeling tools in dry forests of the Centro Universitario Regional del Norte (CURDN) of the University of Tolima: gallery and riparian forest (152.2 ha) and secondary or transitional vegetation (329.1 ha). Fifty-nine temporary sampling plots were established and the aboveground biomass and carbon were estimated by measuring trees and using allometric models and a carbon fraction of 0.47. Four vegetation indexes (NDVI, EVI, SAVI and OSAVI) were estimated from two Sentinel 2A satellite images from rainy and dry season. The NDVI from the rainy season showed the best R² (0.87), which allowed the development of a model for estimation of aboveground biomass. Biomass and carbon distribution mapping was generated in the study area, yielding an average value of 95.1 and 44.1 t/ha of aboveground biomass and carbon, respectively. These results made it possible to spatialize the biomass content and carbon sinks within the CURDN and serve as a first step to manage the territory and establish mechanisms for the preservation of the bs-T in the department of Tolima.

Characterization of terrestrial ecosystems state based on interannual variations of RUE (Rain Use Efficiency)

2023-07-12T09:32:18+02:00

Ecosystems degradation has increased in recent decades and climate change is expected to increase the risk of such processes in the coming years, especially in arid and semi-arid ecosystems. The purpose of this work is to characterize the state of the terrestrial ecosystems of the Spanish mainland and the Balearic Islands through the temporal analysis of the variable RUE (Rain Use Efficiency) during the period 2004-2018. Annual RUE images have been calculated as the quotient between annual gross primary production (GPP) and annual cumulative precipitation (PPT) in a 1 km spatial resolution, and the values have been later normalized. The annual GPP is derived from the daily GPP, obtained using an optimization of the Monteith model and the PPT from daily precipitation images, which are computed by applying a kriging to the data from AEMet network stations. Temporal analysis of the RUE has been made by calculating the slope from a Mann-Kendall test and Sen-Theil method. RUE has been analyzed at three levels of study: at regional level, by vegetation types and at pixel level. The results have shown a negative trend of the normalized RUE (between -0.05 and -0.25 year^-¹) for most of the area, for the 9 classes of vegetation (the forest classes being the ones that have presented the steepest slopes) and in 5 of the 8 ecosystems analyzed at pixel level. A decline in the RUE indicates some degree of degradation in vegetation cover. From the analysis of the results it has been extracted that the interannual variability of the RUE is largely mediated by precipitation, presenting a negative correlation. On the other hand, it has been observed that GPP has experienced a progressive increase in recent years known as greening process.

Semiautomatic detection of burnt areas in Chimborazo-Ecuador using dNBR mean composites with adjusted thresholds

2023-07-25T11:33:06+02:00

A semi-automatic methodology was implemented for the delimitation of burning areas in the province of Chimborazo in Ecuador, during the period 2018-2021, by using the database of forest fires provided by the “Amazonia sin fuego” program of the “Ministerio de Ambiente Agua y Trnacisión Ecológica” (MAATE). The collections of atmospherically corrected Landsat 7 and Landsat 8 images available on the Google Earth Engine (GEE) platform were used. To delimit the burning areas, mean composite images of normalized burning indices (NBR) were calculated in GEE and the most appropriate thresholds of the difference of normalized burning indices (dNBR) were evaluated above which the burning for paramo ecosystem is delimited. The results show: (a) the dNBR threshold value, based on Landsat 7 and Landsat 8 composite mean images, that best fits to identify burning areas in the study area is 0.15; (b) nine events with areas equal to or greater than 100 ha were found, but only seven could be located; (c) most of the burned areas recorded in the MAATE database were overestimated from 25.2% to 84.9% compared to the burn areas digitized on satellite images. These findings provide information that contributes to the strengthening of national fire statistics, useful for the construction of public policies for monitoring and managing forest fires in Ecuador.