Climate is the set of average atmospheric conditions that characterizes a region. It directly influences the majority of human activities, especially agriculture. Climate classification systems (CCSs) are important tools in the study of agriculture, enabling knowledge of the climatic characteristics of a region. Thus, we aimed to perform the climatic characterization of the State of Paraná using the methods proposed by Köppen and Geiger (1928), modified by Trewartha (1954) (KT), Thornthwaite (1948) (TH) and Camargo (1991) and modified by Maluf (2000) (CM), using data from the European Center for Medium-Range Weather Forecast (ECMWF) model. The results of spatial interpolation (virtual stations) were performed using the Kriging method in spherical shape with one neighbour and resolution of 0.25°. The CCSs displayed the ability to separate the warm and dry from cold and wet regions. The most predominant climates were Cfa (temperate humid with hot summers), C₁rA'a' (sub-humid with little water deficiency, megathermal) and ST-UMi (humid subtropical with dry winter), according to KT, TH and CM, respectively. CM is an intermediate CCS between KT and TH.

Accurate and timely regional estimates of agricultural production are key for decision makers. This study aims to understand how different machine learning techniques impact soybean yield estimation in extracting maximum information from remotely sensed MODIS enhanced vegetation index (EVI) that is constrained by phenology. Specifically, a methodology is developed for incorporating phenological information aligned with EVI acquisition for each pixel and selecting the most significant predictors out of 36 predictors using feature selection. These predictors were then used in four machine learning algorithms (MLA) to obtain soybean yield estimates for observed farms in the Paraná State, Brazil. The optimal MLA was then implemented for the whole state to obtain regional soybean yield. The gradient boosting model (GBM) with all 36 predictors performed well with a mean difference of 3.5 kg ha^−  1, an RMSD of 373 kg ha^−  1, and Willmott’s d of 0.85, however, the random forest (RF) algorithm using five optimal EVI predictors presented similar results, but with considerably less computational time. Both GBM and RF provided higher regional yields compared to the officially reported yields by 1775 × 10³ and 2059 × 10³ metric tons, respectively. The RF with five EVI predictors provided the best results for regional soybean estimations, considering the accuracy and computational performances.

Brazilian farming influences directly the worldwide economy. Thus, fast and reliable information on areas sown with the main crops is essential for planning logistics and public or private commodity market policies. Recent farming practices have embraced remote sensing to provide fast and reliable information on commodity dynamics. Medium-to-low resolution free orbital images, such as those from Landsat 8 and Sentinel 2, have been used for crop mapping; however, satellite image processing requires high computing power, especially when monitoring vast areas. Therefore, cloud data processing has been the only feasible option to deal with a large amount of orbital data and its processing and analysis. Thus, our goal was to develop a method to map the two main crops (soybeans and corn) in Paraná, one of the major Brazilian state producers. Landsat-8, Sentinel-2, SRTM+, and field data from 2016 to 2018 were used with the Simple Non-Iterative Clustering segmentation method and the Continuous Naive Bayes classifier, to identify cropped areas. A minimum global accuracy of 90% was found for both crops. Comparison with field data showed correlations of 0.96 and agreement coefficients no lower than 0.86. This ensures mapping quality when using Sentinel and/or Landsat imagery on the GEE platform.

In the last decades, several methodologies for estimating crop phenology based on remote sensing data have been developed and used to create different algorithms. Although many studies have been conducted to evaluate the different methodologies, a comprehensive understanding of the potential of the different current algorithms to detect changes in the growing season is still lacking, especially in large regions and with more than one crop per season. Therefore, this work aimed to evaluate different phenological metrics extraction methodologies. Using data from over 1500 fields distributed across Brazil’s central area, six algorithms, including CropPhenology, Digital Earth Australia tools package (DEA), greenbrown, phenex, phenofit, and TIMESAT, to extract soybean crop phenology were applied. To understand how robust the algorithms are to different input sources, the NDVI and EVI2 time series derived from MODIS products (MOD13Q1 and MOD09Q1) and from Sentinel-2 satellites were used to estimate the sowing date (SD) and harvest date (HD) in each field. The algorithms produced significantly different phenological date estimates, with Spearman’s R ranging between 0.26 and 0.82 when comparing sowing and harvesting dates. The best estimates were obtained using TIMESAT and phenex for SD and HD, respectively, with R greater than 0.7 and RMSE of 16–17 days. The DEA tools and greenbrown packages showed higher sensitivity when using different data sources. Double cropping is an added challenge, with no method adequately identifying it.