Biblio

Performance of neural network for estimating rainfall over {Mato Grosso State, Brazil}

TitlePerformance of neural network for estimating rainfall over {Mato Grosso State, Brazil}
Publication TypeConference Paper
Year of Publication2014
AuthorsMachado NGomes, Ventura TMeirelles, Danelichen VHugo de Mo, Biudes MSacardi
Conference NameDailyMeteo.org/2014
Date Published06/2014
PublisherFaculty of Civil Engineering, University of Belgrade
Conference LocationBelgrade
Abstract

Rainfall is the key element in regional water balance, and has direct influence over economic activity. There is increasing interest of climatic and meteorological data in large spatial and temporal scale. Likewise, computational methods and techniques for climatic and meteorological estimates in large areas with small dataset are growing. Thus, we evaluated neural network performance for rainfall estimates over Mato Grosso State located in the Brazilian Midwest region. The technique of neural network allows the algorithm identifies patterns in the data series, allowing estimates to make new datasets. In this preliminary study, a dataset obtained from 12 meteorological stations was used to train the neural network and then it was run to perform estimates, which allowed comparing with TRMM satellite estimates. We chose TRMM satellite estimates because it has estimated appropriately the annual accumulated rainfall in the Brazilian Midwest region. In general, there was an overestimation of total rainfall estimates by neural network of 21.9% in January and 26,219% in September for the year 2010. The higher overestimated rainfall values in January occurred in Pantanal (39.5%) and Amazon forest (25.4%) than in Cerrado (14%); while the higher overestimated values in September occurred in Cerrado (31,225%) and Amazon forest (25,645%) than in Pantanal (1,424%). The rainfall estimates by neural network had better performance in January (wet season) than in September (dry season) which means that neural network was weak to predict lack of rainfall probably due to use just latitude and longitude as auxiliary variables. The better performance of rainfall estimates by neural network was in the Brazilian Savanna in January than in Amazon forest and Pantanal. Bad estimates of rainfall using neural network in Mato Grosso state were due to (i) a short temporal dataset, (ii) few stations with poor spatial variability, (iii) few auxiliary variables to build neural network. The next step will be to analyze the rainfall and other climatic estimates for the whole year for several years on the Midwest region of Brazil by neural network including other auxiliary variables besides latitude and longitude or by other computational frameworks developed by DailyMeteo group.

Event Calendar

S M T W T F S
1
 
2
 
3
 
4
 
5
 
6
 
7
 
8
 
9
 
10
 
11
 
12
 
13
 
14
 
15
 
16
 
17
 
18
 
19
 
20
 
21
 
22
 
23
 
24
 
25
 
26
 
27
 
28
 
29
 
30
 
31
 
 
 
 
 

Meteo package

Recent comments

No comments available.