Pantanal deforestation forecast

this research aims at predicting Pantanal’s deforestation until 2030 using modern Machine Learning techniques under different scenarios that vary in terms of agricultural production, the number of heads of cattle produced, and the area burned during the years. The model chosen for the study is Extreme Gradient Boosting (XGBoost), an algorithm that combines many simple models sequentially cleverly and yields state-of-the-art results. Our main objective is to simulate different scenarios of agricultural/cattle decisions and burned areas (realistic, optimistic, and pessimistic) that can aid in decision-making for strategic land use decisions in Brazil. As a secondary objective, we want to build a robust and scalable code base so others can contribute to the project, increasing its reach and impact. Finally, hopefully, this project helps ensure this critical biome’s integrity and guarantee future generations’ well-being. The model uses two data sources: The Brazilian Geography and Statistics Institute (IBGE) (https://www.ibge.gov.br/), from which data for land use (ha) by and quantity produced (ton) of temporary and permanent crops from 1985 to 2021 are extracted. From IBGE we also extract the headcount of caddle (in number of heads) in Brazil, from 1985 to 2021. Data for temporary crops can be found at https://sidra.ibge.gov.br/tabela/1612. For permanent crops, in https://sidra.ibge.gov.br/tabela/1613 and for caddle head count in https://sidra.ibge.gov.br/tabela/3939. In this repository, data for these variables are stored in five Excel files: Divided into "culturas_temporarias_1", "culturas_temporarias_2" and "culturas_temporarias_3" (temporary crops), "culturas_permanentes" (permanent crops) and "pecuaria" for headcount. Within crops files, the sheet "Área destinada à colheita (H..." contains the land use information per year in hectares, and the sheet "Quantidade produzida (Tonela..." contains the quantity produced information per year. For caddle heads, there is only one sheet that contains information in number of heads from 1985 to 2021. the second data information is Mpbiomas (https://brasil.mapbiomas.org/), from which we obtain information on land use transitions (in hectares) and forest fires (in hectares). Data can be downloaded from https://brasil.mapbiomas.org/wp-content/uploads/sites/4/2023/08/TABELA-GERAL-MAPBIOMAS-COL8.0-BIOMASxESTADOS-1.xlsx (land use transitions) and https://mapbiomas-br-site.s3.amazonaws.com/downloads/MB-Fogo-2-Biome-State.xlsx (fires). In this repository, data for land use transitions is called "mapbiomas-2021" and for fires is called "mapbiomas-queimadas-2021. The code is also included in this repository (pantanal_deforestation-main.zip) and can be found at https://github.com/iglesiascaio/pantanal_deforestation. To run the model, please read the "READ ME" file. Simply add all Excel files in a single folder and run the code from the same folder. More info on data can be found in the file "metadata".本研究旨在借助现代机器学习技术，针对不同情景下潘塔纳尔湿地（Pantanal）至2030年的森林砍伐情况开展预测，这些情景在农业产量、牛只存栏量以及年度过火面积维度存在差异。本研究选用的模型为极限梯度提升树（Extreme Gradient Boosting, XGBoost），该算法通过巧妙地将多个简单模型进行串行结合，可取得当前顶尖的建模效果。 本研究的核心目标是模拟农业/畜牧决策与过火面积的三类情景（现实、乐观与悲观），以期为巴西战略性土地利用决策提供辅助支撑。次要目标则是构建一套健壮且可扩展的代码库，吸引其他开发者为项目贡献力量，扩大项目的影响力与覆盖范围。最后，本项目期望能够助力保护这一关键生物群落的完整性，为后代的福祉提供保障。 本模型采用两类数据源：其一为巴西地理与统计研究所（Instituto Brasileiro de Geografia e Estatística, IBGE，https://www.ibge.gov.br/），从中提取1985年至2021年间的土地利用面积（公顷）、临时性与永久性农作物产量（吨）数据；同时从IBGE获取1985年至2021年巴西牛只存栏量数据。其中，临时性作物数据可通过https://sidra.ibge.gov.br/tabela/1612 获取，永久性作物数据可通过https://sidra.ibge.gov.br/tabela/1613 获取，牛只存栏量数据则可通过https://sidra.ibge.gov.br/tabela/3939 获取。 本仓库中，上述变量的数据存储于5个Excel文件中：分为"culturas_temporarias_1""culturas_temporarias_2""culturas_temporarias_3"（临时性作物）、"culturas_permanentes"（永久性作物）以及用于存储牛只存栏量的"pecuaria"。在作物类文件中，工作表"Área destinada à colheita (H..."包含各年度以公顷为单位的土地利用信息，工作表"Quantidade produzida (Tonela..."包含各年度的产量信息；牛只存栏量文件仅包含1个工作表，存储了1985年至2021年的牛只存栏数量数据。 第二类数据源为MapBiomas生物群系项目（Mpbiomas，https://brasil.mapbiomas.org/），从中获取土地利用转型（以公顷为单位）与森林火灾（以公顷为单位）相关数据。相关数据可分别通过https://brasil.mapbiomas.org/wp-content/uploads/sites/4/2023/08/TABELA-GERAL-MAPBIOMAS-COL8.0-BIOMASxESTADOS-1.xlsx（土地利用转型数据）与https://mapbiomas-br-site.s3.amazonaws.com/downloads/MB-Fogo-2-Biome-State.xlsx（火灾数据）下载。本仓库中，土地利用转型数据命名为"mapbiomas-2021"，火灾数据命名为"mapbiomas-queimadas-2021"。 本仓库同时包含代码文件（pantanal_deforestation-main.zip），代码托管于https://github.com/iglesiascaio/pantanal_deforestation。如需运行模型，请阅读"READ ME"文件：仅需将所有Excel文件置于同一文件夹内，并在该文件夹中运行代码即可。 更多数据相关信息可查阅"metadata"文件。