Events Public engagement Jobs Give to Cambridge. Research at Cambridge. The word 'populism' has been announced as the Cambridge Dictionary Word of the Year. Read this next. Search news. Sign up to receive our newsletter. English language. Wendalyn Nichols. Cambridge University Press. Find out more. Study at Cambridge Undergraduate Postgraduate Continuing education Executive and professional education Courses in education.
The form was designed to ensure quality assurance with checks upon data entry, limiting the entry to a range of plausible values and formats. This included restricting entries to a specified range of integers for cassava severity scores 1—5 and whitefly count values 0— Upon syncing, the app data were immediately uploaded including per-plant and per-field level data for each surveyed field. These data comprise Dataset A which includes per-plant information collected from fields between and , excluding years no retrievable survey forms exist and no surveys were conducted.
Notably, one of the main differences between Dataset A and B is that they contain per-plant vs per-field data values respectively. Per-field summaries represent averages for any numeric variable recorded at plant level, including per-field disease incidence proportion of infected plants. Severity means are the mean of diseased plants only.
Thus, in infected fields, mean severity is the conditional average of severity scores for plants with a severity score above 1. If all 30 plants assessed in a field are free of CBSD symptoms, then the mean would be 1. Per-field summaries are a more limited representation of data. They lack representation of within-field variability of collected variables such as disease severity scores or adult whitefly numbers.
Lack of per-plant data makes it harder to identify potential outliers or typographical errors. A unified field level dataset, referred to as Dataset C, was derived by integrating the highly reliable digitised paper-based forms stored in Dataset A, with the field summaries from Dataset B Fig.
Notably, Dataset B contained much larger numbers of recorded data points compared with Dataset A in years — We first summarized data from each survey field-level record in Dataset A and supplemented this information with remaining non-overlapping records from Dataset B. Record matching between Datasets A and B was performed based on proximity of the spatial coordinates for each site and corresponding date of survey. Dataset C Fig.
Dataset C field data collection locations. CBSD foliar symptoms are classified as present red or absent blue. Dataset C is a union of the Dataset A cross that contains information at plant level in each field and is supplemented with additional information from the dataset B circle data.
Due to frequent transcription errors and multiple changes in administrative organization in Uganda over the survey period of 15 years, four columns with standardized administrative units were added to the Dataset A. Information about location within an administrative unit was derived using field geographic coordinates within a shapefile representing administrative units of Uganda, as of , on four levels of hierarchical division: region, sub-region, district and county.
Three levels of administrative division of Uganda in The datasets compiled in this study are recorded in English and within the geographic extent of Uganda. The csv files can be read with a wide variety of programs including Excel and R. The following validation process applies to surveys between and , for which data were recorded on paper forms and subsequently transcribed to a digital record.
Surveys in were recorded digitally, hence requiring minimal additional validation. Here, we provide details of the procedures involved during validation. A Python program was written with approximately 30 different functions designed to parse 42 unique latitude notation patterns and 36 longitude patterns in the raw coordinate notation in Dataset A to decimal degrees.
This program automated the conversion of coordinate records, of which were located within Uganda. The parsed coordinates outside Uganda were commonly caused by transcription errors.
For approximately remaining records, it was necessary to interpret the correct notation manually after visual inspection of the individual record. Visible outliers in Dataset B were manually rectified by identifying the same record in Dataset A. Subsequently, an R program was written to automatically plot the daily survey sites in both Dataset A and Dataset B to ensure that all records corresponded to a route that would be realistic in a single day.
As required, the original survey form and neighboring surveys were used to identify errors and infer corrections based on the single and multi-day spatial sequence of survey sites. All variables were subjected to screening to ensure their formats and ranges of values were plausible in each column. For example, severity scores identified as 0, were converted to 1, to follow the overarching survey protocol Notably, field sizes were historically recorded in various formats including hectares, square meters, square kilometers and acres.
Final field size outputs were converted to hectares. In some cases, the recorded units were difficult to infer and in those instances original values were retained, as the typical default value was hectares. In all columns with expected numeric outputs, in instances where additional text symbols appeared in the same entry, these letters and symbols were removed.
Records where the values were not plausible were recorded as NA Not Available. For key variables in Dataset A, a manual comparison with visual inspection was performed between the digitized record and the scanned survey form. Coordinates were subjected to additional manual and automated validation in conjunction with the cleaning process described above.
Disease severity scores were in some cases not recorded on the paper forms if all plants scored 1 i. We have identified those fields and recorded plant severity scores as 1 for all 30 plants. For survey fields in which values of 1 or higher were recorded for individual plants, in addition to NAs, we did not rectify missing values for plants in those fields.
All variables were subjected to screening to ensure their formats and ranges of values were plausible in each category. The cassava variety names were assessed based on visual judgement of trained field assistants or supplied by the farmer and were not genetically confirmed. The names of varieties vary locally and are not standardized in the form. In Dataset A those names might be additionally mistranscribed due to difficulty in reading local variety names.
Based on the level of verification possible, Table 2 highlights the verification level we have assigned to a given variable. Level 1 refers to a high degree of verification, involving both automated and manual values checks against paper forms. Level 2 refers to a medium level of verification, based largely on automated checks. Level 3 variables are unverified original transcriptions from paper forms. Level 4 variables were derived from external sources based on GPS surveyed field location coordinates.
The administrative unit names were derived from shapefiles provided by the Geo-Information Services Division, Uganda Bureau of Statistics. Dataset B incorrectly reported disease in 1. We also investigated the ratio of mean adult whitefly count records differing above an arbitrarily selected threshold of 0.
Nearly one third of the matched records The largest proportion, however The remaining records had deviations of 1 to 5 6. The distribution of deviations between matched records in Datasets A and B. A total of out of These datasets can be used to investigate the characteristics and changes in the spatio-temporal distribution of the CBSD pandemic in Uganda between and including, and not limited to, changes in disease prevalence, incidence, severity, density of the vector and varietal distribution.
It is a highly valuable dataset from the epidemiological perspective and can be used to parameterize and test epidemiological spread distribution models, which can be applied to understand the rate of disease spread and areas of the highest risk of invasion outside the current pandemic zones in Africa and globally wherever cassava is grown. Thus, the first CBSD occurrence points appear only in in the attached dataset.
We found that disease presence and absence data in Dataset B, after verification of the matched records with dataset A, have minor error rates that should be accounted for in the analysis. The false positive rate of CBSD presence is 0. Nearly a third Most of those discrepancies We are unable to assess any variation in recorded variables that could stem from differences in the interpretation of field symptoms by individual field surveyors. Epidemiological models of CBSD spread trained on these data to be reported elsewhere enable exploration of disease control and management within affected area, and best management practices outside the current pandemic zone in preparation for when the disease arrives.
The data cleaning, summarizing, merging and supplementing with additional columns were done in Python and R and the custom code used for this project can be provided upon request. Alicai, T. Brazil experienced its second-highest rate of tree cover loss in , after a prominent spike in The rise comes despite declining deforestation rates , and is mainly due to fires in the Amazon.
While the southern Amazon did face a drought in , almost all fires in the region were set by people to clear land for pasture or agriculture. Experts are also concerned that high levels of fires and forest degradation are becoming the new normal in the Amazon. Climate change combined with human-caused deforestation is increasing the prevalence of drought, making the landscape more vulnerable to fires. Unlike most tropical forests, Indonesia experienced a drop in tree cover loss in , including a 60 percent decline in primary forest loss.
While some provinces in Sumatra still saw increased primary forest loss—including 7, hectares 18, acres in the Kerinci Seblat National Park— provinces in Kalimantan and Papua experienced a reduction. The decrease is likely due in part to the national peat drainage moratorium , in effect since Primary forest loss in protected peat areas went down by 88 percent between and , reaching the lowest level ever recorded. Educational campaigns and increased enforcement of forest laws from local police have also helped prevent land-clearing by fire.
Tree cover loss in the Democratic Republic of Congo DRC reached a record high in , increasing 6 percent from Agriculture, artisanal logging and charcoal production drove the tree cover loss, with nearly 70 percent of it occurring in agricultural areas known as the rural complex.
While shifting cultivation does not necessarily indicate expansion into primary forest, growing populations can intensify agricultural practices, thus reducing fallow periods where trees regrow naturally. Our analysis also showed that in , 3 percent of overall tree cover loss occurred in protected areas and 10 percent within logging concessions. For the past 16 years, DRC has had a moratorium on new industrial logging concessions, but the government reinstated concessions to two companies in While the moratorium applied only to industrial logging, artisanal logging, often illegal, also soared.
Given the increasing trends observed in and , it is critical that DRC move ahead with improved land use planning and forest law enforcement while enforcing better management practices.
The island of Dominica lost 32 percent of its tree cover in , while Puerto Rico lost 10 percent, including significant losses in El Yunque National Forest. While tropical forests in cyclone zones are naturally resilient to storms , scientists worry about their ability to recover in the face of more powerful storms due to climate change.
0コメント