# 经济代写|发展经济学代写Development Economics代考|ECON3110

## 经济代写|发展经济学代写Development Economics代考|Deep Learning to Extract Covariates from Satellite Images

For the purpose of this case, VAM focused on only one data source used to extract features from satellite images. These images are comprised of the red, green and blue (RGB) bands from Sentinel 2 and Google Maps for the area of interest (AOI). The processing is inspired by the transfer learning approach developed by Neal et al. CNNs are trained to predict nightlight intensities (as a proxy of poverty) from the aforementioned images. “Convolutional” means the neural networks are invariant to rotation and translation of images; hence, $\mathrm{CNNs}$ are commonly used for image recognition. In the process of learning how to predict nightlight intensity, the CNN creates features from the satellite imagery. These features are extracted (hence, the term “transfer learning”) to be used as covariates by the prediction model of step (5) in the pipeline mentioned above. Step (3) of the pipeline trains two models: one for Google images and the other for Sentinel images.
The nightlight data is classified into three categories according to their luminosity: low, medium and high. These values become the labels for training our models. Nightlight data from four countries has been used so far to train the model: Malawi, Nigeria, Senegal and Uganda. To reduce class imbalance, the nightlights are masked with Furopean Space Agency’s land use product to take luminosity only from populated areas. Google and Sentinel images are then downloaded (Figure 2.8).

The images and the nightlight classes are then fed to $\mathrm{CNNs}$. Compared with the transfer learning approaches mentioned above, VAM’s method uses a much smaller model architecture for two reasons: first, it relies upon fewer images for training, and second, VAM desired a light model that would be able to score images fast. The final accuracy on validation sets ranges from $60 \%$ to $70 \%$ (Figure $2.9$ ).

## 经济代写|发展经济学代写Development Economics代考|Using and Sharing Real-Time Data During

There is general recognition in the evaluation sector about the need to effectively communicate evaluation findings to key stakeholders (Bamberger, Rugh and Mabry, 2012). Gertler et al. (2011) argue that programme participants should also be included in dissemination efforts. Heinemann, Van Hemelrijck and Guijt (2017) and Van Hemelrijck (2017) provide examples of impact evaluations in which feedback and dialogue with key constituents and beneficiaries is part of the evaluation design. USAID (2018) suggests that helping stakeholders understand their data can lead to higher-quality data, more robust analysis and improved adaptation. McGee et al. (2018) recognize that the right to information is critical for accountability and that data needs to be made available in accessible, usable and actionable ways for the user. Therefore, it is widely seen as valuable in theory, and yet it is not common practice in the evaluation sector to share evaluation findings with survey respondents and targeted audiences.

Bamberger, Raftree and Olazahal (2016) make the case that ICTs are being rapidly introduced into development evaluations, bringing new opportunities for participation, as well as ethical, political and methodological challenges that evaluation needs to address. Holland (2013) argues that ICT has blown wide open the opportunities for participatory statistics.

This short case study presents Oxfam GB’s experience in using ICTs to share real-time data during fieldwork, including practical learning and considerations of when and how this can be carried out as part of a survey data collection process.

Since 2015, Oxfam GB’s team of impact evaluation advisers has been using digital devices in more than 18 countries to conduct individual and household surveys for Oxfam’s effectiveness reviews (expost, QIEs) (Oxfam, 2016; Tomkys and Lombardini, 2015). In many of these data collection processes, Oxfam took advantage of features provided by digital devices to increase data quality, knowledge dissemination, community participation and engagement. One of these features is the ability to process data in real time, while data collection is still under way.

## 经济代写|发展经济学代写Development Economics代考|Using and Sharing Real-Time Data During

Bamberger、Raftree 和 Olazahal（2016 年）证明 ICT 正在迅速被引入发展评估，带来了新的参与机会，以及评估需要解决的伦理、政治和方法挑战。Holland（2013 年）认为，ICT 为参与式统计打开了广阔的大门。

myassignments-help数学代考价格说明

1、客户需提供物理代考的网址，相关账户，以及课程名称，Textbook等相关资料~客服会根据作业数量和持续时间给您定价~使收费透明，让您清楚的知道您的钱花在什么地方。

2、数学代写一般每篇报价约为600—1000rmb，费用根据持续时间、周作业量、成绩要求有所浮动(持续时间越长约便宜、周作业量越多约贵、成绩要求越高越贵)，报价后价格觉得合适，可以先付一周的款，我们帮你试做，满意后再继续，遇到Fail全额退款。

3、myassignments-help公司所有MATH作业代写服务支持付半款，全款，周付款，周付款一方面方便大家查阅自己的分数，一方面也方便大家资金周转，注意:每周固定周一时先预付下周的定金，不付定金不予继续做。物理代写一次性付清打9.5折。

Math作业代写、数学代写常见问题

myassignments-help擅长领域包含但不是全部: